4 2.6.16.60-0.42.8 (SLES 10),
5 2.6.27.39-0.3.1 (SLES11),
6 2.6.18-194.3.1.el5 (RHEL 5)
7 2.6.18-194.3.1.0.1.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: Update RHEL5.5 kernel to 2.6.18-194.3.1.el5 and OEL5.5 kernel to
18 2.6.18-194.3.1.0.1.el5.
20 Severity : enhancement
22 Description: using inkernel OFED stack for rhel5 & oel5.
24 Severity : enhancement
26 Description: Add "lfs_migrate" script from manual into lustre/scripts and RPMs
27 Details : lfs_migrate does a "poor man's" migration of files from their
28 current OST layout to a new OST layout as chosen by the MDS.
32 Description: mds_orphan_add_link()) error linking orphan to PENDING
33 Details : quota limits might disallow linking orphans to PENDING
34 when unlinking a file - temporary raise threads' privileges
35 when processing unlinks.
37 Severity : enhancement
39 Description: add conf-param -d option to remove permanent settings.
40 Details : Add the ability to remove permanent lctl conf_param settings.
41 (Previously conf_param settings could only be changed, not
42 removed.) This also provides a method to change failover
43 nid locations. Improve lctl man page.
45 Severity : enhancement
47 Description: add list_param to b1_8 and add "-R" option to list params
50 Severity : enhancement
52 Description: lfs quota output is not very convenient for awk/sed-parsing
53 Details : Some positions in lfs quota output table could be empty or
54 non-empty which made it hard to parse it with scripts, now a dash
55 is put instead of space where there is not supposed to be any data.
57 Severity : enhancement
59 Description: fix obdfilter-survey script to work properly with remote oss-s
61 Severity : enhancement
63 Description: add new OBDFILTER_SURVEY test suite
65 Severity : enhancement
67 Description: add new multiple mount protection (MMP) test suite
69 Severity : enhancement
71 Description: add support for async journal commit in echo client
73 Severity : enhancement
75 Description: allow userland programs to include <lustre/lustre_idl.h>
76 from stardard include directories
78 Severity : enhancement
80 Description: The prune-icache-use-trylock is no longer needed now that
81 the patch from bug 20008 is landed.
85 Description: The shrink grant feature is still active on the client although the
86 connect flag is not set.
90 Description: Don't leak grant space if the write failed with quota exceeded.
94 Description: Don't consume grant space twice on recoverable resent.
98 Description: a race condition could lead to SIGBUS being sent to an
99 application using mmap-ped files from Lustre
100 Details : truncate_complete_page implementation for the patchless
101 client could arbitrarily unset PG_Uptodate flag for a
102 page being kicked from the page cache, an uptodate check
103 right after a readpage call in filemap_fault could fail
104 because of that as though the page read had been unsuccessful.
108 Description: dlm lock slab shrinking is not efficient
109 Details : The dlm_locks slab can grow significantly and consumes a lot of
110 memory on the server. Set a hardlimit to grant_plan.
114 Description: Lustre does not do 1MB IOs to HW RAID
115 Details : Bump MAX_PHYS/HW_SEGMENTS and SG_ALL to 256 in the RHEL5 kernel.
116 This is what we do already for SLES kernels.
120 Description: bump maximum number of phys/hw segments in the SLES11 kernel
121 until s/g chaining works properly.
126 Description: LSI Fusion MPT driver hacks to improve performance
127 Details : Set CONFIG_FUSION_MAX_SGE to 256 for RHEL5
129 Severity : enhancement
131 Description: increase default md stripe_cache_size to 16k
134 Bugzilla : 15587/21439
135 Description: don't handle security.capability xattr
136 Details : CONFIG_SECURITY_FILE_CAPABILITIES is enabled by default on SLES11.
137 This results in additional getxattr calls, causing VBR test
138 failures as well as a preformance drop when writing.
142 Description: obdfilter-survey is no longer working
143 Details : revert patch from bug 20355 to resolve an issue with lctl --threads
144 not working correctly with $(PTHREAD_LIBS) being linked to lctl.
148 Description: ll_shrink_cache does not handle __GFP_FS properly
152 Description: lfs getstripe shows wrong info for directories
153 Details : Set correct LOVEA default values for filesystem-wide.
157 Description: FSX checksum false positves due to mmap IO
158 Details : Use OBD_FL_MMAP flag for IOs on a memory mapped file. Do not print
159 checksum errors, if the flag is set on a request.
163 Description: file operations after eviction have successful return values
164 Details : use vfs ->flush callback to return any pending async errors
169 Description: mdsrate fails to write after 1.3+M files opened
170 Details : decrease memory usage on clients by recycling dentries and
175 Description: obdfilter-survey gives unreasonably high numbers
176 Details : Wait for all threads to complete when running test_brw.
180 Description: do not set lustre read_only device when server umount and keep
181 client records for recoverable ones
185 Description: move sync_on_lock_cancel tunable to the obdfilter layer
186 Details : move the tunable to trigger a journal flush on lock cancel from
187 the ost layer to the obdfilter layer. This tunable is useful
188 when using the async journal commit feature.
192 Description: exp->exp_nid_stats == NULL in filter_tally()
193 Details : fix race with per-nid stats by delaying procfs cleanup until
198 Description: extent lock cancellation on client can keep the cpu busy for too
203 Description: Do not fail OST activation when a llog is not found, just
204 issue an error message.
208 Description: Don't enable extents by default for MDT.
212 Description: Protect bitfield access to ptlrpc_request's rq_flags, since
213 the AT code can access it concurrently while sending early
218 Description: Disable lockless truncate by default since it is sometimes flawed
219 and causes the write_disjoint test to fail.
223 Description: OSSs which don't have the patch from bug 20278 can trigger an
226 Severity : enhancement
228 Description: don't print message to the console when we have not managed to
233 Description: The MDS fails to synchronize OSTs which registered with the MGS
234 after the MDT. The problem is that OBD_NOTIFY_CREATE events are
235 raised too early and thus discarded by the MDT stack.
236 The fix consists of issuing OBD_NOTIFY_CREATE event in the lov
241 Description: Fix race when the ping evictor and a service thread execute
242 target_recovery_check_and_stop() concurrently.
246 Description: quota broadcast can trigger a LBUG on the MDT if there are
251 Description: Resetting the lov_objid values to last_id reported by the OST
252 during orphan recovery is incorrect and can cause the same
253 objects to be allocated twice.
255 Severity : enhancement
257 Description: "weak-modules" support
258 Details : Implement "weak-modules" support which enables kernel modules
259 to be used with any kernel that implements the same kABI. In
260 order to achieve this modules are now installed in
261 /lib/modules/$(uname -r)/updates/kernel on all distributions.
263 Severity : enhancement
265 Description: add writeconf as mount option
267 Severity : enhancement
269 Description: produce debuginfo packages for SLES.
271 Severity : enhancement
273 Description: add failover nidlist to the import proc file.
275 Severity : enhancement
277 Description: fix LUSTRE_SEQ_MAX_WIDTH for interoperability between 1.8
278 clients and 2.0 servers.
280 Severity : enhancement
282 Description: lfs find -s does not work correctly because of a bug in
287 Description: ll_read_ahead_page() must validate the dlm lock before using
292 Description: Prevent failover nids from registering with MGS first.
296 Description: fix lock inversion in ll_setattr_raw().
300 Description: object allocation is not balanced across OSTs.
301 Details : osc_precreate() should return 0, if there are enough objects left.
303 -------------------------------------------------------------------------------
305 2010-04-30 Oracle, Inc.
307 * Support for kernels:
308 2.6.16.60-0.42.8 (SLES 10),
309 2.6.27.39-0.3.1 (SLES11),
310 2.6.18-164.11.1.el5 (RHEL 5)
311 2.6.18-164.11.1.0.1.el5 (OEL 5)
312 * Client support for unpatched kernels:
313 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
314 2.6.16 - 2.6.30 vanilla (kernel.org)
315 * Recommended e2fsprogs version: 1.41.10-sun2
316 * The async journal commit feature (bug 19128) and the cancel
317 lock before replay feature (bug 16774) are disabled by default.
321 Description: open-unlinked directories trigger MDS LBUG
322 Details : Fix regression introduced by the patch from bug 19640.
323 ext3_inc_count() can reset nlink to 1 when the directory
324 is indexed and inode->i_nlink == 2. Work around the problem
325 by incrementing nlink by 2 instead of 1.
329 Description: Reconnects are not throttled
330 Details : Don't wake up pinger on reconnect failures and rely on regular
331 pings to trigger the next reconnection. Please note that the
332 pinger already uses a smaller interval if the import is
336 Frequency : only with NFS export
338 Description: Console flooded with error message from ll_inode_from_lock()
340 Details : in mds_open, initialize the child_res_id before enqueuing
341 the OPEN lock for the child inode, then to avoid senting
342 wrong ldlm_res_id to client.
346 Description: allow multiple instances of the same nid in NID hash
347 Details : Case of multiple separate clients from the same NID (as
348 with liblustre) is legitimate and so we should allow
349 multiple instances of the same NID in nid hash.
353 Description: the readahead code can sleep on a semaphore while holding a
355 Details : in ras_update, "lov_get_info" could be called during increasing
356 readahead windows, which tries to get the mutex lock "lov_lock"
357 while holding the spin_lock "ras_lock", then causes system
362 Description: ASSERTION(cli->cl_avail_grant >= 0) failed
363 Details : fix assertion failure in the grant code.
367 Description: Use CNETERR (which is rate limited) in specific places in
368 the portal's LNET driver to avoid flooding the console.
372 Description: include last created object in precreate slow case
376 Description: don't do rep-ack if not created anything
377 Details : mds_open currently always put a lock into a rep-ack regardless
378 if something was created or not. This is pointless and only
379 creates needless contention. In fact the entire idea was to do
380 this for real creates as a recovery protection.
384 Description: Spurious error messages from smp_processor_id() on preemptible
386 Details : Disable a preemption by grabbing the lock in fs_trace_get_tcd()
387 first. The function fs_trace_get_tcd() was moved up.
391 Description: interval_erase() fix
392 Details : interval_erase() calls update_maxhigh() properly when child
397 Description: "lfs df" does not print stats for all mountpoints
398 Details : Print all mounted lustre filesystems with "lfs df"
402 Description: lfs setstripe -p not longer work with a relative pathname
403 Details : Use realpath() to provide absolute pathname.
407 Description: fix for truncated reply buffer
408 Details : reply buffer could be referred by reply_in_callback after released
413 Description: lustre.lov error when backing up symlinks with extended attributes
414 Details : Improved logic in ll_listxattr()
418 Description: properly handle null value for setattr -n lustre.lov
419 Details : Running "setfattr -n trusted.lov ." causes a NULL dereference
420 in ll_setxattr() due to no checking if "value" is NULL.
421 This command now resets to the default striping when executed
426 Description: stack overflow on lock cancellation due to fsync call
427 Details : sync_on_lock_cancel is needed for recovery when async journal
428 is enabled, but we actually just need to make sure that
429 metadata blocks have hit the journal, so doing a fs sync
430 should be enough and should consume less stack (just create an
431 empty handle and commmit it).
435 Description: using current->journal_info to store per-thread data leads
436 to problem under memory pressure
437 Details : disable the per-thread data (current->journal_info) containing
438 the lock info during I/O to work around the issue for short tem
442 Description: control DCACHE_LUSTRE_INVALID flag with MDS_INODELOCK_LOOKUP lock
443 Details : DCACHE_LUSTRE_INVALID is controlled by MDS_INODELOCK_LOOKUP
444 lock which is corresponding to "IT_LOOKUP", do not skip invalidate
449 Description: Cannot send after transport shutdown
450 Details : Clear imp_vbr_failed flag upon eviction
454 Description: soft lock in request set code during recovery
455 Details : during recovery, uses req->rq_set itself to replay the request
456 instead of ptlrpcd_recovery_pc
460 Description: Use CFS_ALLOC_IO instead of _STD in llap_from_page_with_lockh
461 Details : During an ll_readahead under ll_readpage, we have seen the the
462 OBD_SLAB_ALLOC hang under ldlm_pools_shrink when trying to lock
463 a page that is already locked by the readahead code.
468 Description: stop waitting for next replay transno if shutdown
469 Details : if the system is shutting down, wake up service thread blocked
470 to wait for next replay transno during recovery, then all the
471 references held by queued requests can be dropped and device
476 Description: lov_merge_lvb()) ASSERTION(spin_is_locked(&lsm->lsm_lock)) failed
477 Details : Protect lli->lli_smd pointer updates with lli->lli_lock.
481 Description: per-nid stats should not access lustre-hash internal structures
486 Description: mount.lustre fails to pass some options to mount()
491 Description: ext4 extent allocation is slower than in ext3
492 Details : Increase the default value of MB_DEFAULT_ORDER2_REQS to 8,
493 enlarge ext4 preallocation table for 2048 4K blocks extents
498 Description: incorrect triggering of synchronous IO
499 Details : The OSC can mistakenly fall back to synchronous IO when the
500 max_dirty_mb limit is reached and no write requests have yet
501 been issued. This can occur when the dirty pages are spread
502 over many files all of which are below the optimal request size.
506 Description: Optimize quota_ctl operations by sending requests in parallel
507 Details : Send MDS->OST quota_ctl requests in parallel, do not resend.
508 Compiled from two attachments in the ticket.
512 Description: ordering issue between transaction start & i_mutex
513 Details : start the transaction earlier in llog_lvfs_destroy to get
514 transaction start and inode mutex lock nested properly.
518 Description: lru resize SLV can get stuck
519 Details : calculate SLV with a greater precision to not lose small
520 changes due to interger math truncation; round up SLV only
521 if the amount of granted locks less than the limit to not
522 get stuck with this SLV
526 Description: avoid divide-by-zero in lprocfs_rd_import()
530 Description: lfs quota failed when OSTs are down
531 Details : really return approximate block/inode usage when OSTs are down
535 Description: llobdstat fix and enhancement
536 Details : add a counter to set a limit to how many samples will be returned
537 fix a wildcard in the path to limit to obdfilter stats only
541 Description: recovery-small 51 hang - MDT can not stop, Mount still busy with
543 Details : abort lock enqueue processing sooner when we umount is in progress.
545 Severity : enhancement
547 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and OEL5.4 kernel to
548 2.6.18-164.11.1.0.1.el5.
550 Severity : enhancement
552 Description: error message improvements
553 Details : Use INFO/WARN instead of WARN/ERROR for the slow messages.
554 Simplify MDT/OST service start message.
555 Suppress "changing the import ..." warning.
558 Bugzilla : 21961/17914
559 Description: ignore trailing -mdc when determining index number
562 Frequency : only with SLES11
564 Description: fix for a race condition in linux quotas implementation
565 Details : dq_flags(struct dquot) access is not properly locked which could
566 lead to certain inconsistencies when accessing it using non-atomic
567 bit operations like __set_bit in do_set_dqblk.
571 Description: return any pending async errors in close(2) using flush callback
573 -------------------------------------------------------------------------------
575 2010-01-29 Sun Microsystems, Inc.
577 * Support for kernels:
578 2.6.16.60-0.42.8 (SLES 10),
579 2.6.27.39-0.3.1 (SLES11),
580 2.6.18-164.11.1.el5 (RHEL 5)
581 2.6.18-164.6.1.0.1.el5 (OEL 5)
582 * Client support for unpatched kernels:
583 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
584 2.6.16 - 2.6.30 vanilla (kernel.org)
585 * Recommended e2fsprogs version: 1.41.6.sun1
586 * The async journal commit feature (bug 19128) and the cancel
587 lock before replay feature (bug 16774) are disabled by default.
589 Severity : enhancement
591 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and
592 OEL5.4 kernel to 2.6.18-164.11.1.0.1.el5.
594 Severity : enhancement
595 Bugzilla : 21511/19848
596 Description: Update SLES11 kernel to 2.6.27.39-0.3.1.
598 Severity : enhancement
600 Description: Update supported SLES10 kernel to 2.6.16.60-0.42.8.
602 Severity : enhancement
604 Description: Update kernel to RHEL5.4 2.6.18-164.6.1.el5 and
605 OEL5 2.6.18-164.6.1.0.1.el5(Both in-kernel OFED enabled).
607 Severity : enhancement
609 Description: Build kernels (RHEL5, OEL5 and SLES10/11) using the vendor's own
612 Severity : enhancement
614 Description: Vanilla kernel 2.6.30 patchless client support.
619 Description: bad entry in directory xxx: inode out of bounds
620 Details : fix locking issue in the rename path which could race with any
621 other operations updating the same directory.
623 Severity : enhancement
625 Description: Make watchdog timer messages to be more clear and descriptive.
629 Description: cp -p command does not preserve the dates and timestamp
630 Details : mtime could be spoiled by a write callback
634 Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
638 Description: Allow non-root access for "lfs check".
639 Details : Added a check in obd_class_ioctl() for OBD_IOC_PING_TARGET.
641 Severity : enhancement
643 Description: quotacheck performance/scaling issues
644 Details : reduce quotacheck time on empty filesystem by skipping uninit
647 Severity : enhancement
649 Description: Enhancement for lfs(1) command to use numeric uid/gid.
651 Severity : enhancement
653 Description: Adjust locks' extents on their first enqueue, so that at the time
654 they get granted, there is no need for another pass through the
655 queues since they are already shaped into the proper forms.
659 Description: Fix mds_shrink_intent_reply()/mds_intent_policy() to pass correct
660 arguments and prevent LBUG() in lustre_shrink_reply_v2().
664 Description: Change tunefs.lustre and mkfs.lustre --mountfsoptions so that
665 exactly the specified mount options are used. Leaving off
666 any "mandatory" mount options is an error. Leaving off any
667 default mount options causes a warning, but is allowed.
668 Change errors=remount-ro from mandatory to default. Sanitize
669 the mount string before storing it. Update man pages accordingly.
673 Description: mds_getattr() should return 0, even if mds_fid2entry() fails with
674 -ENOENT. Also fix in ptlrpc_expire_one_request() to print signed
677 Severity : enhancement
679 Description: Remove set_info(KEY_UNLINKED) from MDS/OSC
681 Severity : enhancement
683 Description: Clients can replay thousands of unused locks during recovery
684 Details : Don't replay unused locks (only read locks for now) during
685 recovery. This feature is disabled by default and can be
686 enabled by running the following command on the clients:
687 lctl get_param ldlm.cancel_unused_locks_before_replay
691 Description: can't stat file in some situation.
692 Details : improve initialize osc date when target is added to mds and
693 ability to resend too big getattr request is client isn't have info
698 Description: Prevent inconsistences between linux and lustre mount structures.
699 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
700 Make the sleep interruptible.
702 Severity : enhancement
704 Description: Communicate OST degraded/readonly state via statfs to MDS
705 Details : Flags in the statfs returned from OSTs indicate whether the
706 OST is in a degraded RAID state, or if the filesystem has
707 turned read-only after a filesystem error is detected.
712 Description: don't panic if EPROTO was hit when reading symlink
713 Details : correctly handling request reference in error cases.
718 Description: open sometimes returns ENOENT instead of EACCES
719 Details : checking permission should be part of open part of mds_open, not
720 lookup part. so server should be set DISP_OPEN_OPEN disposition
721 before starting permission check.
722 Also not need revalidate dentry if client already have LOOKUP lock.
726 Description: enable client interface failover
727 Frequency : on servers with multiple network interfaces
728 Details : When a child reconnects from another NID, properly update export
729 nid hash position and ldlm reverse import.
731 Severity : enhancement
733 Description: implemented direct I/O with arbitrary (nonaligned) memory
734 addresses and file offsets.
736 Severity : enhancement
738 Description: added more recovery timeout options.
740 Severity : enhancement
742 Description: added llapi_file_open, llapi_file_create, llapi_file_get_stripe
747 Description: Avoid deadlock for local client writes
748 Frequency : only on systems with clients writing to an OST on the same node
749 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
750 memory freeing context. This allows OST threads to set the
751 PF_MEMALLOC flag on task structures in order to allocate memory
752 from reserved pools and complete IO.
753 Use GFP_HIGHUSER for OST allocations for non-local client writes,
754 so that the OST threads generate memory pressure and allow
755 inactive pages to be reclaimed.
760 Descriptoin: lock ordering violation between &cli->cl_sem and _lprocfs_lock
761 Details : move ldlm namespace creation in setup phase to avoid grab
762 _lprocfs_lock with cli_sem held.
766 Frequency : only during format of test systems
767 Description: Unable to run several mkfs.lustre on loop devices at the same time
768 Details : mkfs.lustre returns error 256 on the concurrent loop devices
769 formatting. The solution is to proper handle the error.
771 Severity : enhancement
773 Description: implement async create (obd_async_create) method for osc, to avoid
774 too long waiting new ost objects with holding ldlm lock.
778 Frequency : occasionally during network problems
779 Description: client not allowed to reconnect to OST because of active request
780 Details : abort bulk requests received by the OST once the client has timed
781 out since the client will resend the request anyway. The client
782 also now retries to reconnect to the same server if a connect
783 request failed with EBUSY or -EAGAIN.
786 Frequency : rare, if used wide striped file and one ost in down.
788 Descriptoin: don't return error if we created a subset of objects for file.
789 Details : lov_update_create_set() uses set->set_success as index for created
790 objects, so if some requests failed, they will have hole at end of
791 array and we can use qos_shrink_lsm for allocate correct lsm.
795 Description: Slow stale export processing during normal start up
796 Details : The global mgc lock prevents OST setup to be run in parallel.
797 Replace the global lock with a per-config_llog_data semaphore.
801 Description: Out or order replies might be lost on replay
802 Details : In ptlrpc_retain_replayable_request if we cannot find retained
803 request with tid smaller then one currently being added, add it
804 to the start, not end of the list.
808 Description: BUG: soft lockup - CPU#1 stuck for 10s! [ll_mdt_07:4523]
809 Details : add cond_resched() calls to avoid hogging the cpu for too long
810 in the hash code. Make also lustre_hash_for_each_empty() more
813 Severity : enhancement
815 Description: Performance improvements for debug messages with D_RPCTRACE,
816 D_LDLM, D_QUOTA options.
819 Frequency : only with NFS export
821 Description: (lov_merge.c:74:lov_merge_lvb())
822 ASSERTION(spin_is_locked(&lsm->lsm_lock)) failed (SR 71691004)
823 Details : Fix a race in the nfs export code by populating inode
824 info while the new inode is still locked
826 Severity : enhancement
828 Description: add a new file in procfs called force_lbug. Writting to this
829 file triggers a LBUG. Only for test purpose.
833 Description: OOM killer causes node hang
834 Details : really interrupt the sleep in osc_enter_cache on signals
838 Description: LustreError: 9153:0:(quota_context.c:622:dqacq_completion()) LBUG
839 Details : fix race during quota release on the slave.
841 Severity : enhancement
843 Description: smaller hash bucket sizes, cleanups
844 Details : increase hash table sizes and enabled rehashing for pools, quota,
845 uuid, nid & per-nid stats.
847 Severity : enhancement
849 Description: Add ldiskfs maxdirsize mount option
850 Details : add max_dir size mount option
854 Description: panic in ll_statahead_thread
855 Details : prevent parent thread to be killed before its child
858 Frequency : only with 16TB device
860 Description: unable to perform "mount -t lustre" of 16TB OST device
861 Details : Mounting 16TB LUNs failed due to three bugs in mkfs.lustre.
865 Description: ASSERTION(atomic_read(&imp->imp_inflight) == 0) failed
866 Details : unregistering should be zero if no RPC inflight.
870 Description: hyperion: Oops during metabench
871 Details : Correct the refcount of lov_request_set
873 Severity : enhancement
875 Description: Add mptlinux and nxge drivers to Lustre builds
877 Severity : enhancement
879 Description: Fix watchdog timer message to be more clear
880 Details : Make watchdog timer messages more clear and descriptive.
884 Description: LNET soft lockups in socknal_cd thread
885 Details : don't hog CPU for active-connecting if another connd is
886 accepting connecting-requst from the same peer
890 Description: recovery-small test_17 hang
891 Details : Land several AT improvements & fixes.
895 Description: MDS panic and hanging client processes
896 Details : Replace exp_ops_stats with exp_nid_stats->nid_stats
900 Description: OSS stuck in recovery.
901 Details : fix race during recovery. class_unlink_export,
902 class_set_export_delayed and target_queue_last_replay_reply
903 may race while increasing/decreasing obd_recoverable_clients
904 and obd_delayed_clients, causing recovery to wait forever.
906 Severity : enhancement
908 Description: add cascading_rw.c to lustre/tests
912 Description: filter_last_id() NULL deref
913 Details : lprocfs_filter_rd_last_id() should check for the fully
914 setup obd device, before proceeding further.
916 Severity : enhancement
918 Description: Loadgen improvements
919 Details : stacksize and locking fixes for loadgen
923 Description: Quiet CERROR("dirty %d > system dirty_max %d\n"
924 Details : The atomic_read() allowing the atomic_inc() are not covered
925 by a lock. Thus they may safely race and trip this CERROR()
926 unless we add in a small fudge factor (+1).
928 Severity : enhancement
930 Description: shrink_slab: nr=-9223362083340912175
931 Details : fix spurious message from shrink_slab reporing negative nr
935 Description: Quiet bogus previously committed transno error
936 Details : suppress the "server went back in time" error message which
937 is always printed even in the common case after a client eviction
939 Severity : enhancement
941 Description: Parallel statfs() calls result in client eviction
942 Details : cache statfs data for 1s.
946 Description: parallel-scale test_compilebench: @@@@@@ FAIL: compilebench
948 Details : fix serveral issues in pinger code causing clients not to ping
949 servers for too long, resulting in evictions.
953 Description: e2fsck should warn when MMP update interval is extended
954 Details : print mmp_check_interval and make it possible to abort mount
955 operation in case it takes too long.
959 Description: mdsrate-create-large.sh, BUG: soft lockup - CPU#0 stuck for 10s!
960 Details : fix bug in the RHEL5's jbd2 callback patch.
964 Description: drop number of active requests when queued for recovery
965 Details : Now that we take a reference on the original request instead of
966 making a copy of it for recovery. We need to drop the number of
967 active requests or the queued requests will prevent all request
968 processing when they exceed (srv->srv_threads_running - 1).
970 Severity : enhancement
972 Description: refuse to invalidate operational quota files when they are in use
973 Details : an attempt to invalidate operational quota files on the quota
974 master is not actually permitted by VFS (returning -EPERM), but we
975 should not depend on that and should return the error earlier.
979 Description: Applications stuck in jbd2_log_wait_commit during exit
980 Details : fix deadlock between kjournald2 trying to acquire the page lock
981 owned by an ost_io thread waiting for journal commit.
983 -------------------------------------------------------------------------------
985 2009-10-16 Sun Microsystems, Inc.
987 * Support for kernels:
988 2.6.16.60-0.42.4 (SLES 10),
989 2.6.27.29-0.1 (SLES11, i686 & x84_64 only),
990 2.6.18-128.7.1.el5 (RHEL 5),
991 2.6.18-128.7.1.el5 (OEL 5),
992 * Client support for unpatched kernels:
993 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
994 2.6.16 - 2.6.27 vanilla (kernel.org)
995 * Recommended e2fsprogs version: 1.41.6.sun1
996 * ext4 support for RHEL5 is experimental and thus should not be
999 Severity : enhancement
1001 Description: Add OEL5 support.
1003 Severity : enhancement
1005 Description: Update kernel to SLES11 2.6.27.29-0.1.
1009 Description: File checksum failures with OST read cache on
1010 Details : Disable page poisoning when the bulk transfer has to be aborted
1011 because the client got evicted.
1015 Description: Don't allow make backward step on assiging osc next id.
1016 Details : race between allocation next id and ll_sync thread can be cause
1017 of set wrong osc next id and can be kill valid ost objects.
1019 Severity : enhancement
1021 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
1023 Severity : enhancement
1025 Description: Update kernel to SLES10 SP2 2.6.16.60-0.42.4.
1029 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
1033 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
1037 Description: Error handling in osc_statfs_interpret() has been improved.
1038 Details : Check in osc_statfs_interpret() for EBADR.
1042 Description: Do not send statfs() requests to OSTs disabled by administrator.
1043 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
1047 Description: Do not update ctime for the deleted inode.
1048 Details : Check in mds_reint_unlink() before calling fsfilt_setattr().
1052 Description: Increase of the size of the LDLM resource hash.
1053 Details : Bump up RES_HASH_BITS=12.
1057 Description: correctly send lsm on open replay
1058 Details : MDS is trust to LSM size on replay open, but client can set wrong size
1063 Description: Deadlock between filter_destroy() and filter_commitrw_write().
1064 Details : filter_destroy() does not hold the DLM lock over the whole
1065 operation. If the DLM lock is dropped, filter_commitrw() can go
1066 through, causing the deadlock between page lock and i_mutex.
1067 The i_alloc_sem should also be hold in filter_destroy() while
1068 truncating the file.
1072 Description: truncate starts GFP_FS allocation under transaction causing deadlock
1073 Details : ldiskfs_truncate calls grab_cache_page which may start page
1074 allocation under an open transaction. This may lead to
1075 calling prune_icache with consequent lustre reentrance.
1078 Frequency : only when down/upgrading the MDS to 1.6/1.8 while 1.8 clients are
1079 still up and when the OST pool feature is used
1081 Description: interop testing got LBUG when run dd with OST pool
1082 :LustreError: 30032:0:(llite_lib.c:1913:ll_replace_lsm()) LBUG
1083 Details : down/upgrading the MDS to a version that doesn't/does support OST
1084 pool can cause clients to crash because the lsm has changed
1089 Description: missing tree_status on 1.8.1 RPM build
1090 Details : make rpms failed due because the tree_status file is missing.
1094 Description: continuing LustreError "mds adjust qunit failed!"
1095 Details : don't print message on the console when ->adjust_qunit fails.
1099 Description: don't increase ldlm timeout if previous client was evicted
1100 Details : if a client doesn't respond to a blocking callback within the
1101 adaptive ldlm enqueue timeout, don't adjust the adaptive estimate
1102 when the lock is next granted.
1106 Description: ost is being unmounted w/o all writes to last_rcvd landing on disk.
1107 affects recovery negatively.
1108 Details : make sure all exports have been properly destroyed by the zombie
1109 thread processed before stopping the target.
1113 Description: Performance degradation with O_DIRECT between 1.6 & 1.8.1 b190
1114 Details : disable write barrier for ext4/SLES11.
1118 Description: Kernel panic - not syncing: Out of memory and no killable
1119 processes... on OSS when iozone
1120 Details : fix memory leak in the journal checksum patch.
1124 Description: group quota "too many blocks" OSS crashes
1125 Details : we should keep the same uid/gid for lquota_chkquota() and
1126 lquota_pending_commit()
1130 Description: LustreError: 9153:0:(quota_context.c:622:dqacq_completion()) LBUG
1131 Details : don't LBUG on release quota error. Just a workaround until the
1132 problem is understood.
1134 ------------------------------------------------------------------------------
1135 2009-07-31 Sun Microsystems, Inc.
1137 * Support for kernels:
1138 2.6.16.60-0.39.3 (SLES 10),
1139 2.6.27.23-0.1 (SLES11, i686 & x84_64 only),
1140 2.6.18-128.1.14.el5 (RHEL 5),
1141 * Client support for unpatched kernels:
1142 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1143 2.6.16 - 2.6.27 vanilla (kernel.org)
1144 * Recommended e2fsprogs version: 1.41.6.sun1
1145 * File join has been disabled in this release, refer to Bugzilla 16929.
1146 * NFS export disabled when stack size < 8192. Since the NFSv4 export
1147 of Lustre filesystem with 4K stack may cause a stack overflow. For
1148 more information, please refer to bugzilla 17630.
1149 * ext4 support for RHEL5 is experimental and thus should not be
1152 Severity : enhancement
1154 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
1157 Frequency : with 1.8 server and 1.6 clients
1159 Description: correctly shrink reply for avoid send too big message to client.
1160 Details : 1.8 mds is allocate to big buffer to LOV EA data and this produce
1161 some problems with sending this reply to 1.6 client.
1165 Description: Repeated atomic allocation failures.
1166 Details : Use GFP_HIGHUSER | __GFP_NOMEMALLOC flags for memory allocations
1167 to generate memory pressure and allow reclaiming of inactive pages.
1168 At the same time, do not allow to exhaust emergency pools.
1169 For local clients the use of GFP_NOFS will be introduced in 1.8.2
1171 Severity : enhancement
1172 Bugzilla : 19846, 18289
1173 Description: Update kernel to RHEL5 2.6.18-128.1.14.el5.
1175 Severity : enhancement
1176 Bugzilla : 19625, 16893, 18668, 19848
1177 Description: Add support for SLES11 2.6.27.23-0.1.
1179 Severity : enhancement
1181 Description: Update client support to vanila kernels up to 2.6.27.
1183 Severity : enhancement
1185 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
1187 Severity : enhancement
1189 Description: Compile with -Werror by default for i686 and x86_64.
1193 Description: resolve race between obd_disconnect and class_disconnect_exports
1194 Details : if obd_disconnect will be called to already disconnected export he
1195 forget release one reference and osc module can't unloaded.
1197 Severity : enhancement
1199 Description: move AT tunable parameters for more consistent usage
1200 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
1204 Descriptoin: correctly skip time estimate if in recovery
1205 Details : rq_send_state insn't bitmask so using bitwise ops is forbid.
1209 Descriptoin: OSS DeadLock
1210 Details : Use trylock to prevent deadlock when shrink icache.
1212 Severity : enhancement
1214 Description: Allow tuning service thread via /proc
1215 Details : For each service a new
1216 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
1217 created that can be used to set min/max thread counts, and get the
1218 current number of running threads.
1220 Severity : enhancement
1222 Description: Add state history info file, enhance import info file
1223 Details : Track import connection state changes in a new osc/mdc proc file;
1224 add overview-type data to the osc/mdc import proc file.
1228 Description: Reduce small size read RPC
1229 Details : Set read-ahead limite for every file and only do read-ahead when
1230 available read-ahead pages are bigger than 1M to avoid small size
1235 Description: free_entry erroneously used groups_free instead of put_group_info
1237 Severity : enhancement
1239 Description: Make read-ahead stripe size aligned.
1241 Severity : enhancement
1243 Description: MDS create should not wait for statfs RPC while holding DLM lock.
1246 Frequency : rare, connect and disconnect target at same time
1248 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
1249 Details : don't call obd_disconnect under lov_lock. this long time
1250 operation and can block ptlrpcd which answer to connect request.
1253 Frequency : start MDS on uncleanly shutdowned MDS device
1255 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
1256 Details : stay in waiting mds<>ost recovery finished produce random bugs
1257 due race between two ll_sync thread for one lov target. send
1258 ACTIVATE event only if connect realy finished and import have
1262 Frequency : start MDS on uncleanly shutdowned MDS device
1264 Descriptoin: aborting recovery hang on MDS
1265 Details : don't throttle destroy RPCs for the MDT.
1269 Description: Slow reads beyond 8Tb offsets.
1270 Details : Page index integer overflow in ll_read_ahead_page
1274 Description: Soft lockup on OSS after MDS failover
1275 Details : MSG_CONNECT_INITIAL is not set on the initial MDS->OST connect.
1276 As a conseqence, the patch from bug 18224 is not operational
1277 and the MDS export cannot be reused on the OSTs until it gets
1281 Frequency : rare, only if using MMP with Linux RAID
1283 Description: MMP doesn't work with Linux RAID
1284 Details : While using HA for Lustre servers with Linux RAID, it is possible
1285 that MMP will not detect multiple mounts. To make this work we
1286 need to unplug the device queue in RAID when the MMP block is being
1287 written. Also while reading the MMP block, we should read it from
1288 disk and not the cached one.
1291 Frequency : rare, during recovery
1293 Description: Assertion failure in ldlm_lock_put
1294 Details : Do not put cancelled locks into replay list, hold references on
1295 locks in replay list
1299 Description: 1.6.5 mdsrate performance is slower than 1.4.11/12 (MDS is not cpu bound!)
1300 Details : create_count always drops to the min value (=32) because grow_count
1301 is being changed before the precreate RPC completes.
1304 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
1307 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
1308 Details : add proper locking for creation of jbd_4k slab cache
1312 Description: MMP check in ext3_remount() fails without displaying any error
1313 Details : When multiple mount protection fails during remount, proper error
1318 Description: Rare Client crash on resend if the file was deleted.
1319 Details : When file is opened, but open reply is lost and file is
1320 subsequently deleted before resend, resend processing logic
1321 breaks trying to open the file again, should not try to open.
1325 Description: add check for >8TB ldiskfs filesystems
1326 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
1327 Don't allow >8TB ldiskfs filesystems to be mounted without
1328 force_over_8tb mount option
1332 Description: Client locked up when running multiple instances of an app. on
1333 multiple mount points
1334 Details : ll_shrink_cache() can sleep while holding the ll_sb_lock.
1335 Convert ll_sb_lock to a read/write semaphore to fix the problem.
1339 Description: Cannot acces an NFS-mounted Lustre filesystem
1340 Details : An NFS client cannot access the Lustre filesystem NFS-mounted
1341 from a Lustre-client exporting the Lustre filesystem via NFS.
1345 Description: panic in ll_statahead_thread
1346 Details : grab dentry reference in parent process.
1349 -------------------------------------------------------------------------------
1351 tbd Sun Microsystems, Inc.
1353 * Support for kernels:
1354 2.6.16.60-0.31 (SLES 10),
1355 2.6.18-128.1.6.el5 (RHEL 5),
1356 2.6.22.14 vanilla (kernel.org)
1357 * Client support for unpatched kernels:
1358 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1359 2.6.16 - 2.6.22 vanilla (kernel.org)
1360 * Recommended e2fsprogs version: 1.40.11-sun1
1361 * File join has been disabled, refer to Bugzilla 16929.
1362 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
1363 * NFS export disabled when stack size < 8192. Since the NFSv4 export of
1364 Lustre filesystem with 4K stack may cause a stack overflow. For more
1365 information, please refer to bugzilla 17630.
1367 Severity : enhancement
1369 Description: Update to RHEL5.3 kernel-2.6.18-128.1.6.el5.
1371 Severity : enhancement
1373 Description: Update OFED release to 1.4.1 RC4
1375 Severity : major, only with big OST
1377 Description: Very poor metadata performance on Infiniband lustre configuration
1378 Details : OST object precreation becomes very slow on big OSTs. This is due
1379 to the ialloc patch spending too much time scanning groups.
1382 Frequency : during recovery
1384 Description: don't mix llog inodes with normal.
1385 Details : allocate inodes for log in last inode group
1390 Description: fix lqs' reference which won't be put in some situations
1391 Details : This patch fixes:
1392 1. In quota_check_common(), this function will check quota
1393 for user and group, but only send one return via "pending".
1394 In most cases, the pendings should be same. But that is not
1396 2. If quotaoff runs between lquota_chkquota() and
1397 lquota_pending_commit(), the same thing will happen too.
1398 That is why it comes:
1399 - if (!ll_sb_any_quota_active(qctxt->lqc_sb))
1402 -------------------------------------------------------------------------------
1404 2008-12-31 Sun Microsystems, Inc.
1406 * Support for kernels:
1407 2.6.5-7.314 (SLES 9),
1408 2.6.9-67.0.22.EL (RHEL 4),
1409 2.6.16.60-0.31 (SLES 10),
1410 2.6.18-92.1.17.el5 (RHEL 5),
1411 2.6.22.14 vanilla (kernel.org)
1412 * Client support for unpatched kernels:
1413 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1414 2.6.16 - 2.6.22 vanilla (kernel.org)
1415 * Client support for unpatched kernels:
1416 we do not recommend using patchless RHEL4 clients with kernels
1417 prior to 2.6.9-55EL (RHEL4U5).
1418 * Recommended e2fsprogs version: 1.40.11-sun1
1419 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1420 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
1421 removed cwd "./" (refer to Bugzilla 14399).
1422 * A new quota file format has been introduced in 1.6.5.
1423 The format conversion from prior releases is handled transparently,
1424 but releases older than 1.4.12/1.6.5 don't understand this new
1425 format. The automatic format conversion can be avoided by running
1426 the following command on the MDS:
1427 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
1428 For more information, please refer to bugzilla 13904.
1429 * A new quota file format was introduced in 1.6.6/1.8.0 (kernels 2.6.16+).
1430 The format conversion from prior releases is handled transparently,
1431 but releases older than 1.6.6/1.8.0 don't understand this new
1432 format. The automatic format conversion can be avoided by running
1433 the following commands on the MDS and OSS servers (for
1434 pre 1.4.12-1.6.5 quota files):
1435 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV',
1436 'tunefs.lustre --param="ost.quota_type=ug1" $MDTDEV'
1437 or (for 1.4.12/1.6.5 quota files)
1438 'tunefs.lustre --param="mdt.quota_type=ug2" $MDTDEV',
1439 'tunefs.lustre --param="ost.quota_type=ug2" $MDTDEV'
1440 For more information, please refer to bugzilla 13904.
1441 * Output of lfs quota has been made less detailed by default,
1442 old (verbose) output can be obtained by using -v option.
1443 * File join has been disabled in this release, refer to Bugzilla 16929.
1444 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
1445 * NFS export disabled when stack size < 8192. Since the NFSv4 export of
1446 Lustre filesystem with 4K stack may cause a stack overflow. For more
1447 information, please refer to bugzilla 17630.
1449 Severity : enhancement
1451 Description: Caching OSS
1452 Details : introduce data caching on the OSS. The OSS now relies on the linux
1453 kernel page cache to keep recently accessed data in memory.
1454 It is worth noting that all write requests are still flushed
1455 synchronously as in lustre 1.6.
1457 Severity : enhancement
1459 Description: Version based recovery
1460 Details : introduce finer grained recovery able to detect transaction
1461 dependencies and can deal with transaction gaps caused by clients
1462 failing at the same time as the server.
1464 Severity : enhancement
1466 Description: Enable adaptive timeouts by default
1467 Details : The Lustre timeout value in /proc/sys/lustre/timeout is now
1468 managed dynamically based on server load and should not need
1469 to be tuned manually based on cluster size. This allows Lustre
1470 to work under a wider variety of system sizes and loads, without
1471 unnecessarily causing lengthy recovery times.
1473 Severity : enhancement
1475 Description: Add OST Pools support
1476 Details : File striping can now be set to use an arbitrary pool of OSTs
1478 Severity : enhancement
1480 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
1481 Details : allow skip disconnected ost for send statfs request and hide error
1485 Frequency : rare, on llog test 6
1487 Descriptoin: don't allow connect to already connected import
1488 Details : allowing connect to already connected import is hide connecting problem.
1491 Frequency : rare, on failed llog setup
1493 Descriptoin: don't leak obd reference on failed llog setup
1494 Details : for failed llog setup - mgc forget call class_destroy_import for
1495 client import, move destroy import to more generic place.
1500 Descriptoin: allow kill process which wait statahead result
1501 Details : for some reasons 'ls' can stick in waiting result from statahead,
1502 in this case need way for kill this process.
1505 Frequency : rare, at shutdown
1507 Descriptoin: panic at umount
1508 Details : llap_shrinker can be raced with killing super block from list and
1509 this produce panic with access to already freeded pointer
1514 Descriptoin: don't lose wakeup for imp_recovery_waitq
1515 Details : recover_import_no_retry or invalidate_import and import_close can
1516 both sleep on imp_recovery_waitq, but we was send only one wakeup
1522 Descriptoin: panic in mds_open
1523 Details : don't confuse mds_finish_transno() with PTR_ERR(-ENOENT)
1528 Descriptoin: stuck in cache_remove_extent() or panic with accessing to already
1530 Details : release lock refernce only after add page to pages list.
1533 Frequency : always with long access acl
1535 Descriptoin: mds can't pack reply with long acl.
1536 Details : mds don't control size of acl but they limited by reint/getattr
1541 Frequency : on remount
1543 Description: external journal device not working after the remount
1544 Details : clear dev_rdonly flag for external journal devices in
1550 Description: shutdown vs evict race
1551 Details : client_disconnect_export vs connect request race.
1552 if client will evicted at this time - we start invalidate
1553 thread without referece to import and import can be freed
1559 Description: shrink LOV EAs before replying
1560 Details : correctly adjust LOV EA buffer for reply.
1565 Description: don't skip ost target if they assigned to file
1566 Details : Drop slow OSCs if we can, but not for requested start idx.
1567 This means "if OSC is slow and it is not the requested
1568 start OST, then it can be skipped, otherwise skip it only
1569 if it is inactive/recovering/out-of-space.
1571 Severity : enhancement
1573 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
1576 Frequency : rare, need acl's on inode.
1578 Description: client can't handle ost additional correctly
1579 Details : if ost was added after client connected to mds client can have
1580 hit lnet_try_match_md ... to big messages to wide striped files.
1581 in this case need teach client to handle config events about add
1582 lov target and update client max ea size at that event.
1584 Severity : enhancement
1586 Description: Update to sles9 kernel-2.6.5-7.314.
1588 Severity : enhancement
1590 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
1593 Frequency : Create a symlink file with a very long name
1595 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
1596 Details : If there is no extra space in the request for early cancels,
1597 ldlm_req_handles_avail() returns 0 instead of a negative value.
1602 Description: mds is deadlocked
1603 Details : in rare cases, inode in catalog can have i_no less than have parent
1604 i_no, this produce wrong order for locking during open, and
1605 parallel unlink can be lock open. this need teach mds_open to grab
1606 locks in resource id order, not at parent -> child order.
1608 Severity : enhancement
1610 Description: Add /proc entry for import status
1611 Details : The mdc, osc, and mgc import directories now have
1612 an import directory that contains useful import data for debugging
1613 connection problems.
1615 Severity : enhancement
1617 Description: Re-disable certain /proc logging
1618 Details : Enable and disable client's offset_stats, extents_stats and
1619 extents_stats_per_process stats logging on the fly.
1622 Frequency : Only on FC kernels 2.6.22+
1624 Description: oops in statahead
1625 Details : Do not drop reference count for the dentry from VFS when lookup,
1626 VFS will do that by itself.
1628 Severity : enhancement
1630 Description: Generic /proc file permissions
1631 Details : Set /Proc file permissions in a more generic way to enable non-
1632 root users operate on some /proc files.
1636 Description: Hitting mdc_commit_close() ASSERTION
1637 Details : Properly handle request reference release in
1638 ll_release_openhandle().
1642 Frequency : only patchless client
1643 Description: add workaround for race between add/remove dentry from hash
1645 Severity : enhancement
1647 Description: Allow OST glimpses to return PW locks
1651 Description: LBUG when llog conf file is full
1652 Details : When llog bitmap is full, ENOSPC should be returned for plain
1657 Description: Prevent import from entering FULL state when server in recovery
1661 Description: service mount cannot take device name with ":"
1662 Details : Only when device name contains ":/" will mount treat it as
1668 Description: replace ptlrpcd with the statahead thread to interpret the async
1669 statahead RPC callback
1673 Frequency : on recovery
1674 Description: I/O failures after umount during fail back
1675 Details : if client reconnected to restarted server we need join to recovery
1676 instead of find server handler is changed and process self eviction
1677 with cancel all locks.
1679 Severity : enhancement
1681 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
1685 Description: Kernel BUG tries to release flock
1686 Details : Lustre does not destroy flock lock before last reference goes
1687 away. So always drop flock locks when client is evicted and
1688 perform unlock regardless of successfulness of speaking to MDS.
1690 Severity : enhancement
1692 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
1694 Severity : enhancement
1696 Description: Upcall on Lustre log has been dumped
1697 Details : Allow for a user mode script to be called once a Lustre log has
1698 been dumped. It passes the filename of the dumped log to the
1699 script, the location of the script can be specified via
1700 /proc/sys/lnet/debug_log_upcall.
1705 Description: avoid messages about idr_remove called for id that is not allocated
1706 Details : Move assigment s_dev for clustered nfs to end of initialization,
1707 for avoid problem with error handling.
1712 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
1713 Details : When connection is reused this not moved from CONN_UNUSED_HASH into
1714 CONN_USED_HASH and this prodice warning when put connection again
1720 Description: avoid ASSERTION(client_stat->nid_exp_ref_count == 0) failed
1721 Details : release reference to stats when client disconnected, not
1722 when export destroyed for avoid races when client destroyed
1723 after main ost export.
1727 Description: more cleanup in mds_lov
1728 Details : add workaround for get valid ost count for avoid warnings about
1729 drop too big messages, not init llog cat under semphore which
1730 can be blocked on reconnect and break normal replay, fix access
1733 Severity : enhancement
1735 Description: Export bytes_read/bytes_write count on OSC/OST.
1739 Description: Early reply size mismatch, MGC loses connection
1740 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
1741 the connect flags are properly negotiated.
1745 Description: Properly propagate oinfo flags from lov to osc for statfs
1746 Details : restore missing copy oi_flags to lov requests.
1750 Description: exports in /proc are broken
1751 Details : recreate /proc entries for clients when they reconnect.
1753 Severity : enhancement
1755 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
1756 : l_getgroups(8), lst(8), routerstat(8)
1757 Details : included man pages for llobdstat(8), llstat(8),
1758 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
1760 Severity : enhancement
1762 Description: Implement lustre ll_show_options method.
1764 Severity : enhancement
1766 Description: Update to SLES9 kernel-2.6.5-7.312.
1768 Severity : enhancement
1770 Description: Update to RHEL4 kernel-2.6.9-67.0.22.EL.
1774 Description: exports in /proc are broken
1775 Details : recreate /proc entries for clients when they reconnect.
1779 Description: don't fail open with -ERANGE
1780 Details : if client connected until mds will be know about real ost count
1781 get LOV EA can be fail because mds not allocate enougth buffer
1786 Description: Resolve device initialization race
1787 Details : Prevent proc handler from accessing devices added to the
1788 obd_devs array but yet be intialized.
1790 Severity : enhancement
1792 Description: configure's --enable-quota should check the
1793 : kernel .config for CONFIG_QUOTA
1794 Details : configure is terminated if --enable-quota is passed but
1795 : no quota support is in kernel
1797 Severity : enhancement
1799 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
1801 Severity : enhancement
1803 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
1807 Frequency : rare, on PPC clients
1808 Description: don't swab ost objects in response about directory, because
1810 Details : bug similar bug 14856, but in different function.
1812 Severity : enhancement
1814 Description: lfs quota tool enhancement
1815 Details : added units specifiers support for setquota, default to
1816 current uid/gid for quota report, short quota stats by
1817 default, nonpositional parameters for setquota, added
1818 llapi_quotactl manual page.
1820 Severity : enhancement
1822 Description: *optional* service tags registration
1823 Details : if the "service tags" package is installed on a Lustre node
1824 When the filesystem is mounted, a local-node service tag will
1825 be created. See http://inventory.sun.com/ for more information
1826 about the Service Tags asset management system.
1830 Description: Client runs out of low memory
1831 Details : Consider only lowmem when counting initial number of llap pages
1834 Frequency : occasional
1836 Description: add refcount for osc callbacks, so avoid panic on shutdown
1838 Severity : enhancement
1840 Description: Update to RHEL4 kernel-2.6.9-67.0.20.
1843 Frequency : testing only
1845 Description: sanity test 65a fails if stripecount of -1 is set
1846 Details : handle -1 striping on filesystem in ll_dirstripe_verify
1849 Frequency : only in unusual configurations
1851 Description: Kernel panic with find ost index.
1852 Details : lov_obd have panic if some OST's have sparse indexes.
1854 Severity : enhancement
1856 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
1859 Frequency : rarely, if filesystem is mounted with -o flock
1861 Description: do not process already freed flock
1862 Details : flock can possibly be freed by another thread before it reaches
1863 to ldlm_flock_completion_ast.
1866 Frequency : rarely, if filesystem is mounted with -o flock
1868 Description: LBUG during stress test
1869 Details : Need properly lock accesses the flock deadlock detection list.
1872 Frequency : rarely, if binaries are being run from Lustre
1874 Description: oops in page fault handler
1875 Details : kernel page fault handler can return two special 'pages' in
1876 error case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
1879 Frequency : rarely, during shutdown
1881 Description: timeout with invalidate import.
1882 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
1883 should be handled by ptlrpcd. This produce long age waiting and
1884 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
1890 Description: ASSERTION(CheckWriteback(page,cmd)) failed
1891 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
1895 Frequency : only with broken builds/installations
1897 Description: no LBUG if lquota.ko and fsfilt_ldiskfs.ko are different versions
1898 Details : just return an error to a user, put a console error message
1900 Severity : enhancement
1902 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
1904 Severity : enhancement
1906 Description: Update to RHEL4 kernel-2.6.9-67.0.15.
1908 Severity : enhancement
1910 Description: enable MGS and MDT services start separately
1911 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
1912 a co-located MGS without starting the MGS, which is a complement
1913 to 'nosvc' mount option.
1916 Frequency : always, on big-endian systems
1918 Description: cleanup in ptlrpc code, related to PPC platform
1919 Details : store magic in native order avoid panic's in recovery on PPC
1920 node and forbid from this error in future. Also fix posibily
1921 of twice swab data. Fix get lov striping to userpace.
1924 Frequency : rarely, if replay get lost on server
1926 Description: server incorrectly drop resent replays lead to recovery failure.
1927 Details : do not drop replay according to msg flags, instead we check the
1928 per-export recovery request queue for duplication of transno.
1931 Frequency : after recovery
1933 Description: precreate to many object's after del orphan.
1934 Details : del orphan st in oscc last_id == next_id and this triger growing
1935 count of precreated objects. Set flag LOW to skip increase count
1936 of precreated objects.
1939 Frequency : rare, on clear nid stats
1941 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
1942 Details : when clean nid stats sometimes try destroy live entry,
1943 and this produce panic in free.
1946 Frequency : occasionally since 1.6.4
1948 Description: Stack overflow during MDS log replay
1949 Details : ease stack pressure by using a thread dealing llog_process.
1952 Frequency : very rare
1954 Description: MDT cannot be unmounted, reporting "Mount still busy"
1955 Details : Mountpoint references were being leaked during open reply
1956 reconstruction after an MDS restart. Drop mountpoint reference
1957 in reconstruct_open() and free dentry reference also.
1962 Description: wait until IO finished before start new when do lock cancel.
1963 Details : VM protocol want old IO finished before start new, in this case
1964 need wait until PG_writeback is cleared until check dirty flag
1965 and call writepages in lock cancel callback.
1970 Description: mds_mfd_close() ASSERTION(rc == 0)
1971 Details : In mds_mfd_close(), we need protect inode's writecount change
1972 within its orphan write semaphore to prevent possible races.
1975 Frequency : rare, on shutdown ost
1977 Description: don't hit live lock with umount ost.
1978 Details : shrink_dcache_parent can be in long loop with destroy dentries,
1979 use shrink_dcache_sb instead.
1982 Frequency : only when echo_client is used
1984 Description: don't panic with use echo_client
1985 Details : echo client pass NULL as client nid pointer and this produce NULL
1986 pointer dereference.
1989 Frequency : Always on 32-bit PowerPC systems
1991 Description: fix build on PPC32
1992 Details : compile code with -m64 flag produce wrong object file for PPC32.
1997 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1998 Details : In reconstruct_* functions, LASSERTs on both the data supplied
1999 by a client, and the data on disk are dangerous and incorrect.
2000 Change them with client eviction.
2002 Severity : enhancement
2004 Description: skiplist implementation simplification
2005 Details : skiplists are used to group compatible locks on granted list
2006 that was implemented as tracking first and last lock of each lock
2007 group the patch changes that to using doubly linked lists
2011 Description: delete compatibility for 32bit qdata
2012 Details : as planned, when lustre is beyond b1_8, lquota won't support 32bit
2013 qunit. That means servers of b1_4 and servers of b1_8 can't be
2014 used together if users want to use quota.
2017 Frequency : only with administrator action
2019 Description: mount failure if config log has invalid conf_param setting
2020 Details : If administrator specified an incorrect configuration parameter
2021 with "lctl conf_param" this would cause an error during future
2022 client mounts. Instead, ignore the bad configuration parameter.
2025 Frequency : blocks per group < blocksize*8 and uninit_groups is enabled
2027 Description: ldiskfs error: XXX blocks in bitmap, YYY in gd
2028 Details : If blocks per group is less than blocksize*8, set rest of the
2032 Frequency : Application do stride read on lustre
2034 Description: The read performance will drop a lot if the application does
2036 Details : Because the stride_start_offset are missing in stride read-ahead,
2037 it will cause clients read a lot of unused pages in read-ahead,
2038 then the read-performance drops.
2042 Description: more ldlm soft lockups
2043 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
2044 starve other threads from the resource lock for a long time in
2045 case of long waiting queue, so change the debug level from
2046 D_OTHER to the less frequently used D_INFO.
2048 Severity : enhancement
2050 Description: add -gid, -group, -uid, -user options to lfs find
2052 Severity : enhancement
2054 Description: ll_recover_lost_found_objs - recover objects in lost+found
2055 Details : OST corruption and subsequent e2fsck can leave objects in the
2056 lost+found directory. Using the "ll_recover_lost_found_objs"
2057 tool, these objects can be retrieved and data can be salvaged
2058 by using the object ID saved in the fid EA on each object.
2063 Description: this bug _only_ happens when inode quota limitation is very low
2064 (less than 12), so that inode quota unit is 1 at initialization.
2065 Details : if remaining quota equates 1, it is a sign to demonstate that quota
2066 is effective now. So least quota qunit should be 2.
2070 Description: Hung threads in invalidate_inode_pages2_range
2071 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
2072 one is completed. As a result, some RPCs are stuck in the queue
2077 Description: Procfs and llog threads access destoryed import sometimes.
2078 Details : Sync the import destoryed process with procfs and llog threads by
2079 the import refcount and semaphore.
2083 Description: mds fails to respond, threads stuck in ldlm_completion_ast
2084 Details : Sort source/child resource pair after updating child resource.
2089 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
2090 Details : If insertion of an extent fails, then discard the inode
2091 preallocation and free data blocks else it can lead to duplicate
2096 Description: don't always update ctime in ext3_xattr_set_handle()
2097 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle()
2098 In some cases the ctime should not be updated, for example for
2099 2.0->1.8 compatibility it is necessary to delete an xattr and it
2100 should not update the ctime.
2104 Description: add quota statistics
2105 Details : 1. sort out quota proc entries and proc code.
2106 2. add quota statistics
2111 Description: quotas are not honored with O_DIRECT
2112 Details : all writes with the flag O_DIRECT will use grants which leads to
2113 this problem. Now using OBD_BRW_SYNC to guard this.
2117 Bugzilla : 15713/16362
2118 Description: Assertion in iopen_connect_dentry in 1.6.3
2119 Details : looking up an inode via iopen with the wrong generation number can
2120 populate the dcache with a disconneced dentry while the inode
2121 number is in the process of being reallocated. This causes an
2122 assertion failure in iopen since the inode's dentry list contains
2123 both a connected and disconnected dentry.
2127 Description: assertion failure in ldlm_handle2lock()
2128 Details : fix a race between class_handle_unhash() and class_handle2object()
2129 introduced in lustre 1.6.5 by bug 13622.
2131 Severity : enhancement
2133 Description: superblock lock contention with many SMP cores on one client
2134 Details : several client filesystem locks were highly contended on SMP
2135 NUMA systems with 8 or more cores. Per-CPU datastructures
2136 and more efficient locking implemented to reduce contention.
2141 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
2142 Details : remove the limit of 256 scsi disks in the sd_iostat patch
2147 Description: oops in sd_iostats_seq_show()
2148 Details : unloading/reloading the scsi low level driver triggers a kernel
2149 bug when trying to access the sd iostat file.
2154 Description: Kernel panics during QLogic driver reload
2155 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
2156 patch, causing memory corruption.
2161 Description: journal_dev option does not work in b1_6
2162 Details : pass mount option during pre-mount.
2164 Severity : enhancement
2166 Description: Add a FIEMAP(FIle Extent MAP) ioctl for ldiskfs
2167 Details : FIEMAP ioctl will allow an application to efficiently fetch the
2168 extent information of a file. It can be used to map logical blocks
2169 in a file to physical blocks in the block device.
2172 Frequency : only with adaptive timeout enabled
2174 Description: DEBUG_REQ() bad paging request
2175 Details : ptlrpc_at_recv_early_reply() should not modify req->rq_repmsg
2176 because it can be accessed by reply_in_callback() without the
2180 Frequency : only on Cray X2
2182 Description: X2 build failures
2183 Details : fix build failures on Cray X2.
2187 Description: xid & resent requests
2188 Details : Initialize RPC XID from clock at startup (randomly if clock is
2193 Description: quota recovery deadlock during mds failover
2194 Details : This patch includes att18982, att18236, att18237 in bz14840.
2196 1. fix osts hang when mds does failover with quotaon
2197 2. prevent watchdog storm when osts threads wait for the
2202 Description: kernel panic on racer
2203 Details : Do not access dchild->d_inode when IS_ERR(dchild) is true.
2205 Severity : enhancement
2207 Description: Add lustre_start utility to start or stop multiple Lustre servers
2212 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
2213 Details : In case of memory pressure, list_del() can be called twice on
2214 req->rq_history_list, causing a kernel oops.
2218 Description: (ptllnd_peer.c:557:kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
2219 Details : fix stack overflow in the distributed lock manager by defering
2220 export eviction after a failed ast to the elt thread instead of
2221 handling it in the dlm interpret routine.
2223 Severity : enhancement
2225 Description: More exported tunables for mballoc
2226 Details : Add support for tunable preallocation window and new tunables for
2227 large/small requests
2231 Description: Detect corruption of block bitmap and checking for preallocations
2232 Details : Checks validity of on-disk block bitmap. Also it does better
2233 checking of number of applied preallocations. When corruption is
2234 found, it turns filesystem readonly to prevent further corruptions.
2238 Frequency : only for big-endian servers
2239 Description: Check for big-endian system while mounting fs with extents feature
2240 Details : Mounting a filesystem with extents feature will fail on big-endian
2241 systems since ext3-based ldiskfs is not supported on big-endian
2242 systems. Can be over-ridden with "bigendian_extents" mount option.
2246 Description: Excessive recovery window
2247 Details : With AT enabled, the recovery window can be excessively long (6000+
2248 seconds). To address this problem, we no longer use
2249 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2250 timeout no longer depends on the service time, it is set to
2251 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2252 time via pb_service_time.
2255 Descriptoin: Don't sync journal after every i/o
2256 Details : Implement write RPC replay to allow server replies for write RPCs
2257 before data is on disk. However, this feature is disabled by
2258 default since some issues leading to data corruptions have been
2259 found during recovery (e.g. bug 19128). This feature can be enabled
2260 by running the following command on the OSSs:
2261 lctl set_param obdfilter.*.sync_journal=0
2265 Description: Watchdog triggered on MDS failover
2266 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2267 the OSTs know that the MDS "UUID" can be reused for the same export
2268 from a different NID, so we do not need to wait for the export to
2273 Descriptoin: Lustre detected file system corruption with inode out of bounds
2274 Details : don't update i_size on MDS_CLOSE for directories. This causes
2275 directory corruptions on the MDT.
2277 -------------------------------------------------------------------------------
2279 2008-05-26 Sun Microsystems, Inc.
2281 * Support for kernels:
2282 2.6.5-7.311 (SLES 9),
2283 2.6.9-67.0.7.EL (RHEL 4),
2284 2.6.16.54-0.2.5 (SLES 10),
2285 2.6.18-53.1.14.el5 (RHEL 5),
2286 2.6.22.14 vanilla (kernel.org)
2287 * Client support for unpatched kernels:
2288 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2289 2.6.16 - 2.6.22 vanilla (kernel.org)
2290 * Due to problems with nested symlinks and FMODE_EXEC (bug 12652),
2291 we do not recommend using patchless RHEL4 clients with kernels
2292 prior to 2.6.9-55EL (RHEL4U5).
2293 * Recommended e2fsprogs version: 1.40.7-sun1
2294 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2295 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
2296 removed cwd "./" (refer to Bugzilla 14399).
2297 * A new quota file format has been introduced in 1.6.5.
2298 The format conversion from prior releases is handled transparently,
2299 but releases older than 1.4.12/1.6.5 will not understand this new
2300 format. The automatic format conversion can be avoided by running
2301 the following command on the MDS before upgrading:
2302 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
2303 For more information, please refer to bugzilla 13904.
2307 Description: quota performance fix
2308 Details : quota data is written in journalled mode instead of ordered to
2309 increase performance
2313 Description: lfs support for human-readable quota grace time strings
2314 Details : lfs setquota -t and lfs quota -t represent quota grace times
2315 in "XXwXXdXXhXXmXXs" format instead of large values in seconds
2318 Frequency : always with o2ib 1.3 and sles10
2320 Description: fix build with SLES10 and o2ib v3.
2321 Details : sles10 uses diffrent name for Module.symver file but configure
2322 assume this file has same name on RHEL/SLES/vanila kernels.
2325 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
2327 Description: MDS may lose file striping (and hence file data) in some cases
2328 Details : If there are additional extended attributes stored on the MDS,
2329 in particular ACLs, SELinux, or user attributes (if user_xattr
2330 is specified for the client mount options) then there is a risk
2331 of attribute loss. Additionally, the Lustre file striping
2332 needs to be larger than default (e.g. striped over all OSTs),
2333 and an additional attribute must be stored initially in the
2334 inode and then increase in size enough to be moved to the
2335 external attribute block (e.g. ACL growing in size) for file
2338 Severity : enhancement
2340 Description: add message levels for liblustreapi
2345 Description: MDT cannot be unmounted, reporting "Mount still busy"
2346 Details : Mountpoint references were being leaked during open reply
2347 reconstruction after an MDS restart. Drop mountpoint reference
2348 in reconstruct_open() and free dentry reference also.
2353 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
2354 Details : Move the 'good_osts' check before the 'total_bavail' check. This
2355 will result in an -EAGAIN and in the exit call path we call
2356 alloc_rr() which will with increasing aggressiveness attempt to
2357 aquire precreated objects on the minimum number of required OSCs.
2361 Description: Use old size assignment to avoid deadlock
2362 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
2363 the scheduling while holding a spinlock. We do not need locking
2364 for size in ll_update_inode() because size is only updated from
2365 the MDS for directories or files without objects, so there is no
2366 other place to do the update, and concurrent access to such inodes
2367 are protected by the inode lock.
2371 Description: Use __u64 instead of int for valid bits
2375 Description: resolve "_IOWR redefined" build error on SLES10
2379 Description: dump the memory debugging after all modules are unloaded to
2380 suppress false negative in conf_sanity test 39
2384 Description: the recovery timer never expires
2385 Details : for new client connect request, the recovery timer should not be
2386 reset, otherwise recovery timer will never expired, if the old
2387 client never come. Only old client connect and first connection
2388 req should trigger recovery timer reset.
2392 Description: the min numbers of lproc stats are wrong
2393 Details : adding a new constant LC_MIN_INIT and use it for initialization
2397 Frequency : always with interactive lfs
2399 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
2402 Frequency : with multiple concurrent readdir processes in same directory
2403 Bugzilla : 15406, 15169, 15175
2404 Description: misc fixes for directory readahead.
2405 Details : prevent previous statahead async RPC callback from processing the
2406 current "statahead_info", race condition between async RPC callback
2407 add dentry into dentry hash table and "ls" thread revalidate such
2408 dentry, statahead his/miss control for hidden items, and so on.
2410 Severity : enhancement
2412 Description: build kernel-ib packages for OFED 1.3 in our release cycle
2416 Description: incore types cleaning in quota code (with respect to 64-bit limits)
2417 Details : several u32 variables declarations are replaced with u64 declarations
2422 Description: fix SLES kernel versioning
2423 Details : the kernel version for our SLES 10 kernel did not include a "-"
2424 before the "smp" at the end. while this was not a problem in
2425 general, it did mean that software trying to use the kernel
2426 version to try to detect a vendor specific kernel would fail.
2427 this was most evident by the OFED build scripts.
2432 Description: Don't update lov_desc members until making sure they are valid
2433 Details : When updating lov_desc members via proc fs, need fix their
2434 validities before doing the real update.
2437 Frequency : very rare
2439 Description: don't put request into delay list while invalidate in flight.
2440 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
2441 import in flight. this produce timeout for invalidate and sometimes
2442 can cause stale data.
2444 Severity : enhancement
2446 Description: Update kernel to SLES9 2.6.5-7.311.
2448 Severity : enhancement
2450 Description: Update kernel to RHEL4 2.6.9-67.0.7.
2455 Frequency : on PPC only
2456 Description: not convert ost objects for directory because it's not exist.
2457 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
2459 Severity : enhancement
2461 Description: Fix warnings with compile liblustre at sles10/rhel5 which have
2462 __u64 as usingied long long type.
2465 Frequency : rare, on shutdown
2467 Description: race process ast vs remove callback
2468 Details : removing callback before disconnect import open race with
2469 processing callback.
2471 Severity : enhancement
2473 Description: Update kernel to SLES9 2.6.5-7.311.
2475 Severity : enhancement
2477 Description: Files open for execute are not marked busy on SLES10
2478 Details : Add FMODE_EXEC to SLES10 SP1 server kernel series.
2480 Severity : enhancement
2482 Description: Add server support for vanilla-2.6.22.14.
2485 Frequency : occasional
2487 Description: Avoid lov_create() getting stuck in obd_statfs_rqset()
2488 Details : If an OST is down the MDS will hang indefinitely in
2489 obd_statfs_rqset() waiting for the statfs data. While for
2490 MDS QOS usage of statfs, it should not stuck in waiting.
2492 Severity : enhancement
2494 Description: Disable adaptive timeouts by default
2497 Frequency : on network error
2499 Description: panic with double free request if network error
2500 Details : mdc_finish_enqueue is finish request if any network error occuring,
2501 but it's true only for synchronus enqueue, for async enqueue
2502 (via ptlrpcd) this incorrect and ptlrpcd want finish request
2506 Frequency : rare, on recovery
2508 Description: read procfs can produce deadlock in some situation
2509 Details : Holding lprocfs lock with send rpc can produce block for destroy
2510 obd objects and this also block reconnect with -EALREADY.
2511 This isn't fix all lprocfs bugs - but make it rare.
2513 Severity : enhancement
2515 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
2518 Frequency : frequent on X2 node
2520 Description: mdc_set_open_replay_data LBUG
2521 Details : Set replay data for requests that are eligible for replay.
2526 Description: lustre_mgs: operation 101 on unconnected MGS
2527 Details : When MGC is disconnected from MGS long enough, MGS will evict the
2528 MGC, and late on MGC cannot successfully connect to MGS and a lot
2529 of the error messages complaining that MGS is not connected.
2532 Frequency : rare, depends on device drivers and load
2534 Description: MDS or OSS nodes crash due to stack overflow
2535 Details : Code changes in 1.6.4 increased the stack usage of some functions.
2536 In some cases, in conjunction with device drivers that use a lot
2537 of stack, the MDS (or possibly OSS) service threads could overflow
2538 the stack. One change which was identified to consume additional
2539 stack has been reworked to avoid the extra stack usage.
2541 Severity : enhancement
2543 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
2545 Severity : enhancement
2547 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
2549 Severity : enhancement
2551 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
2553 Serverity : enhancement
2555 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
2558 Frequency : rare on shutdown OST
2560 Description: Don't allow skipping OSTs if index has been specified.
2561 Details : Don't allow skipping OSTs if index has been specified, make locking
2562 in internal create lots better.
2567 Description: ASSERTION(!PageDirty(page)) failed
2568 Details : Wrong check could lead to an assertion failure under specific
2574 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
2575 Details : access to bitfield in structure is always rounded to long
2576 and this produce problem with not atomic change any bit.
2581 Description: Lustre make rpms failed.
2582 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
2583 builds Lustre rpms from tarball.
2585 Severity : enhancement
2587 Description: Update to SLES9 SP4 kernel-2.6.5-7.308.
2590 Frequency : rare on shutdown OST
2592 Description: If llog cancel was not send before clean_exports phase, this can
2593 produce deadlock in llog code.
2594 Details : If llog thread has last reference to obd and call class_import_put
2595 this produce deadlock because llog_cleanup_commit_master wait when
2596 last llog_commit_thread exited, but this never success because was
2597 called from llog_commit_thread.
2600 Frequency : only if OST index is skipped
2602 Description: NULL lov_tgts causing MDS oops
2603 Details : more safe checks for NULL lov_tgts for avoid oops.
2605 Severity : enhancement
2607 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
2609 Severity : enhancement
2611 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
2616 Description: make mgs_setparam() handle fsname containing dash
2617 Details : fsname containing a dash does not work with lctl conf_param
2619 Severity : enhancement
2621 Description: Update to RHEL4 Update-6 kernel-2.6.9-67.EL.
2624 Frequency : rare, in recovery and (or) destroy lovobjid file.
2626 Description: rewrite lov objid code.
2627 Details : Cleanup for lov objid code, remove scability problems and wrong
2628 locking. Fix sending last_id into ost.
2630 Severity : enhancement
2632 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
2634 Severity : enhancement
2636 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
2637 Details : Use d_move_locked instead of __d_move.
2640 Frequency : rare, at shutdown
2642 Description: access already free / zero obd_namespace.
2643 Details : if client_disconnect_export was called without force flag set,
2644 and exist connect request in flight, this can produce access to
2645 NULL pointer (or already free pointer) when connect_interpret
2646 store ocd flags in obd_namespace.
2649 Frequency : only at startup
2651 Description: not alloc memory with spinlock held.
2652 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
2653 if any spinlock held.
2658 Description: lfs find does not continue on file error
2659 Details : Continue other files processing when a file/dir is absent.
2663 Description: Inconsistent usage of lustre_pack_reply()
2664 Details : Standardize the usage of lustre_pack_reply() such that it
2665 always generate a CERROR on failure.
2668 Frequency : very rare
2670 Description: Fix replay if there is an un-replied request and open
2671 Details : In some cases, older replay request will revert the
2672 mcd->mcd_last_xid on MDS which is used to record the client's
2673 latest sent request.
2675 Severity : enhancement
2677 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
2679 Severity : enhancement
2681 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
2683 Severity : enhancement
2685 Description: Update to SLES9 kernel-2.6.5-7.287.3.
2687 Severity : enhancement
2689 Description: Update to RHEL4 kernel-2.6.9-55.0.12.EL.
2691 Severity : enhancement
2693 Description: Build SLES10 patchless client fails
2694 Details : The configure was broken by run ./configure with
2695 --with-linux-obj=.... argument for patchless client. When the
2696 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
2697 can't search header adequately. Use absolute path such as
2698 -I($LINUX)/include instead.
2700 Severity : enhancement
2702 Description: Lustre Page Accounting
2703 Details : New macros for page alloc and free which enable accounting
2704 of page allocation of Lustre. Use percpu counters to store memory
2705 and page statistics.
2708 Frequency : only if debugging is disabled
2710 Description: LASSERT_{REQ,REP}SWAB macros are buggy
2711 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
2712 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
2713 these macros and replace them with their definitions instead.
2718 Description: interrupt oig_wait produce painc on resend.
2719 Details : brw_redo_request can be used for resend requests from ptlrpcd and
2720 private set, and this produce situation when rq_ptlrpcd_data not
2721 copyed to new allocated request and triggered LBUG on assert
2722 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
2723 ptlrpcd set if request is changed and can be safety changed to use
2726 Severity : enhancement
2728 Description: organize the server-side client stats on per-nid basis
2729 Details : Change the structure of stats under obdfilter and mds to
2737 The "uuid"s file would list the uuids of _active_ exports.
2738 And the clear entry is to clear all stats and stale nids.
2743 Description: Processes looping in ll_readdir() on Lustre clients finally causing
2744 a full node pseudo-hang
2745 Details : Concurrent access to the same directory from multiple clients with
2746 intensive file creation/removal can cause a client node to spin in
2747 ll_readdir(). i_version must be increased every time the lock is
2748 cancelled to ensure a revalidate is done.
2753 Description: touch file failed when fs is not full
2754 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
2755 otherwise we can get ENOSP while fs is not full.
2758 Frequency : only for Cray XT3
2759 Bugzilla : 12829/13455
2760 Description: Changing primary group doesn't change the group lustre assigns to
2762 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
2763 overridden with the primary group provided by the group upcall,
2764 whereas the supplied fsgid can be trusted if it is in the list of
2765 supplementary groups returned by the group upcall.
2767 Severity : enhancement
2769 Description: Root Squash Functionality
2770 Details : Implementation of NFS-like root squash capability. Specifically,
2771 don't allow someone with root access on a client node to be able
2772 to manipulate files owned by root on a server node.
2774 Severity : enhancement
2776 Description: Slow trucate/writes to huge files at high offsets.
2777 Details : Directly associate cached pages to lock that protect those pages,
2778 this allows us to quickly find what pages to write and remove
2779 once lock callback is received.
2784 Description: Too many locks accumulating on client during NFS usage
2785 Details : mds_open improperly used accmode to find out access mode to a
2786 file. Also mdc_intent_lock logic to find out if we already have
2787 lock similar to just received was flawed since introduction of
2788 skiplists - locks are now added to the front of the granted
2793 Description: Hit ASSERTION(obd->obd_stopping == 1) failed in some setup failed
2795 Details : In obd setup failure handler, obd_stopping will not necessarily to
2796 be 1, and obd_set_up should also be checked to make sure whether
2797 obd is completely setup.
2799 Severity : enhancement
2801 Description: Allow masking D_WARNING, D_ERROR messages from console
2802 Details : Console messages can now be disabled via lnet.printk.
2807 Description: User code with malformed file open parameter crashes client node
2808 Details : Before packing join_file req, all the related reference should be
2809 checked carefully in case some malformed flags cause fake join_file
2815 Description: shrink/enlarge qunit size when needed; fix the problem of coarse
2816 grain of quota doing harm to quota's accuracy
2817 Details : qunit size will be changed when quota limitation is too low/high;
2818 record the pending quota write in order to get more accureate
2819 quota; delete the patch for bug12588, which is unnecessary when
2820 this patch is landed. This bug also contains fixes for bug 14526,
2821 14299, 14601 and 13794.
2825 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
2826 Details : ldlm_completion_ast() assumes that a lock is granted when the req
2827 mode is equal to the granted mode. However, it should also check
2828 that LDLM_FL_CP_REQD is not set.
2832 Description: Heavy nfs access might result in deadlocks
2833 Details : After ELC code landed, it is now improper to enqueue any mds
2834 locks under och_sem, because enqueue might want to decide to
2835 cancel open locks for same inode we are holding och_sem for.
2839 Description: 35% write performance drop with ldiskfs2 when quotas are on
2840 Details : Enable ext3 journalled quota by default to improve performance
2841 when quotas are turned on.
2845 Description: Client eviction while running blogbench
2846 Details : A lot of unlink operations with concurrent I/O can lead to a
2847 deadlock causing evictions. To address the problem, the number of
2848 oustanding OST_DESTROY requests is now throttled to
2849 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
2850 callbacks are processed in priority.
2853 Frequency : RHEL4 only
2855 Description: mkfs is very slow on IA64/RHEL4
2856 Details : A performance regression has been discovered in the MPT Fusion
2857 driver between versions 3.02.73rh and 3.02.99.00rh. As a
2858 consequence, we have downgraded the MPT Fusion driver in the RHEL4
2859 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
2862 Frequency : PPC/PPC64 only
2864 Description: conflicts between asm-ppc64/types.h and lustre_types.h
2865 Details : fix duplicated definitions between asm-ppc64/types.h and
2866 lustre_types.h on PPC.
2869 Frequency : PPC/PPC64 only
2871 Description: asm-ppc/segment.h does not exist
2872 Details : fix compile issue on PPC.
2876 Description: data checksumming impacts single node performance
2877 Details : add support for several checksum algorithms. Currently, CRC32 and
2878 Adler-32 are supported. The checksum type can be changed on the fly
2879 through /proc/fs/lustre/osc/*/checksum_type.
2883 Description: use adler32 for page checksums
2884 Details : when available, use the Adler-32 algorithm instead of CRC32 for
2889 Description: better handle error messages in extents code
2891 Severity : enhancement
2893 Description: SNMP support enhancement
2894 Details : Adding total number of sampled request for an MDS node in snmp
2897 Severity : enhancement
2899 Description: Optimize ldlm waiting list processing for PR extent locks
2900 Details : When processing waiting list for read extent lock and meeting read
2901 lock that is same or wider to it that is not contended, skip
2902 processing rest of the list and immediatelly return current
2903 status of conflictness, since we are guaranteed there are no
2904 conflicting locks in the rest of the list.
2908 Description: Time out and refuse to reconnect
2909 Details : When the failover node is the primary node, it is possible
2910 to have two identical connections in imp_conn_list. We must
2911 compare not conn's pointers but NIDs, otherwise we
2912 can defeat connection throttling.
2916 Description: Client not clear own cache if answer to reconnect is lost.
2917 Details : Client gets evicted from server. Now client also thinks it is
2918 disconnected (or gets ENOTCONN on its operation) and decides to
2919 reconnect. Server receives reconnect message, but cannot find
2920 export. New export is created that is fully valid (new cookie!),
2921 but reply is lost and not reported to client. Client reconnects
2922 again and gets back a just-created connection, but it is not new
2923 so client thinks it was not evicted and does not do recovery.
2927 Description: Detect stride IO mode in read-ahead
2928 Details : When a client does stride read, read-ahead should detect that and
2929 read-ahead pages according to the detected stride pattern.
2933 Description: build for x2 fails
2934 Details : fix compile issue on Cray systems.
2936 Severity : enhancement
2938 Description: implement readv/writev APIs(aio_read/aio_writes in newer kernels)
2939 Details : This greatly improves speed of NFS writes on 2.6 kernels.
2942 Frequency : only on PPC/SLES10
2944 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
2945 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
2946 be defined. Add a hack in mkfs_lustre.c to work around this compile
2951 Description: LASSERT on MDS when client holding flock lock dies
2952 Details : ldlm pool logic depends on number of granted locks equal to
2953 number of released locks which is not true for flock locks, so
2954 just exclude such locks from consideration.
2958 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
2959 Details : Use fsfilt_sync() for both the whole filesystem sync and
2960 individual file sync to eliminate dangerous inode locking
2961 with I_LOCK that can lead to a deadlock.
2965 Description: Update an obsolete wirecheck.c generator
2966 Details : Update wirecheck.c/wirehdr.c and regenerate wiretest.c
2970 Description: Client can panic on open sometimes
2971 Details : It is possible that we try to free already freed request in
2972 ll_file_open in some error cases when we send request from
2977 Description: performance in 1.6.3
2978 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
2979 sure that 1MB requests are not fragmented by the block layer.
2983 Description: LDLM soft lockups - improvement
2984 Details : It is be possible to send the lock handle along with each read
2985 or write request because the client is already doing a lock match
2986 itself so there isn't any reason the OST should have to re-do that
2992 Description: lfs quota fails with deactivated OSTS
2993 Details : With this patch, three improvements are included:
2994 1. detete the softlimit in mds and osts when use "lfs quota".
2995 2. display the inaccurate data in the output of "lfs quota".
2996 3. try to get quota info when "lfs quota" is executed.
3001 Description: Extent locks not granted with no conflicts sometimes.
3002 Details : When race occurs in glimpse handler and nothing is returned,
3003 we do not reprocess the queue after lock cancel, and that leads
3004 to a stall until next activity on a resource
3007 Frequency : failover with quotaon
3009 Description: during mds failovers with quota on, OSTs got into deadlock state
3010 and causing dumpstack.
3011 Details : for every quota slave, at any time, there is only one quota req
3012 is sent to quota master for every uid/gid. Before that quota req
3013 returns, all the thread relative to the same uid/gid will wait.
3014 So if the quota req is lost because mds failovers or any other
3015 reasons, this bug will be hit. Now, dqacq_interpret() will handle
3016 quota reqs who time out.
3018 Severity : enhancement
3021 Description: when quota slave checks if quota is enough, there is an unnecessary
3023 Details : place this wait on necessary place instead of always waiting.
3025 --------------------------------------------------------------------------------
3027 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
3029 * Support for kernels:
3030 2.6.5-7.286 (SLES 9),
3031 2.6.9-55.0.9.EL (RHEL 4),
3032 2.6.16.53-0.8 (SLES 10),
3033 2.6.18-8.1.14.el5 (RHEL 5),
3034 2.6.18.8 vanilla (kernel.org)
3035 * Client support for unpatched kernels:
3036 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3037 2.6.16 - 2.6.22 vanilla (kernel.org)
3038 * Due to recently discovered recovery problems, we do not recommend
3039 using patchless RHEL 4 clients with this or any earlier release.
3040 * Recommended e2fsprogs version: 1.40.2-cfs1
3041 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3044 Frequency : occasional
3046 Description: MDS hang or stay in waiting lock
3047 Details : If client receive lock with CBPENDING flag ldlm need send lock
3048 cancel as separate rpc, to avoid situation when cancel request
3049 can't processed due all i/o threads stay in wait lock.
3052 Frequency : occasional
3054 Description: Do not fail import if osc_interpret_create gets -EAGAIN
3055 Details : If osc_interpret_create got -EAGAIN it immediately exits and
3056 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
3057 oscc_has_objects and see OSC has no objests and call
3058 oscc_internal_create to resend create request.
3060 Severity : enhancement
3062 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
3063 Details : Update which_patch & target file for SLES10 latest kernel.
3065 Severity : enhancement
3067 Description: add --type and --size parameters to lfs find
3068 Details : Enhance lfs find by adding filetype and filesize parameters. Also
3069 multiple OBDs can now be specified for the --obd option.
3071 Severity : enhancement
3073 Description: eliminate client locks in face of contention
3074 Details : file contention detection and lockless i/o implementation
3075 for contended files.
3077 Severity : enhancement
3079 Description: Remove client patches from SLES 10 kernel.
3080 Details : This causes SLES 10 clients to behave as patchless clients
3081 even on a Lustre-patched (server) kernel.
3083 Severity : enhancement
3085 Description: use i_size_read and i_size_write in 2.6 port
3086 Details : replace inode->i_size access with i_size_read/write()
3089 Frequency : when removing large files
3091 Description: scheduling issue during removal of large Lustre files
3092 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
3093 It causes scheduling issues when removing large files (17TB in the
3099 Description: 1.4.11 Can't handle directories with stripe set and extended ACLs
3100 Details : Impossible (EPROTO is returned) to access a directory that has a
3101 non-default striping and ACLs.
3104 Frequency : only on ppc
3106 Description: /proc/fs/lustre/devices broken on ppc
3107 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
3108 We should make sure the type of 'index' is loff_t and then cast
3109 explicitly as needed below. Do not assign an explicitly cast
3113 Frequency : only for rhel5
3115 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
3116 Details : Modify the target file & which_kernel.
3119 Frequency : if the uninit_groups feature is enabled on ldiskfs
3121 Description: e2fsck reports "invalid unused inodes count"
3122 Details : If a new ldiskfs filesystem is created with the "uninit_groups"
3123 feature and only a single inode is created in a group then the
3124 "bg_unused_inodes" count is incorrectly updated. Creating a
3125 second inode in that group would update it correctly.
3130 Description: buffer overruns could theoretically occur
3131 Details : llapi_semantic_traverse() modifies the "path" argument by
3132 appending values to the end of the origin string, and a buffer
3133 overrun may occur. Adding buffer overrun check in liblustreapi.
3135 Severity : enhancement
3137 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla.
3140 Frequency : only if filesystem is inconsistent
3142 Description: handle "serious error: objid * already exists" more gracefully
3143 Details : If LAST_ID value on disk is smaller than the objects existing in
3144 the O/0/d* directories, it indicates disk corruption and causes an
3145 LBUG(). If the object is 0-length, then we should use the existing
3146 object. This will help to avoid a full fsck in most cases.
3148 Severity : enhancement
3150 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
3151 Details : Modify vm-tunables-rhel4.patch.
3153 Severity : enhancement
3155 Description: Kernel config for 2.6.18-vanilla.
3156 Details : Modify targets/2.6-vanilla.target.in.
3157 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
3158 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
3159 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
3160 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
3163 Frequency : occasional
3165 Description: improve handling recoverable errors
3166 Details : If request processed with error which can be recoverable on server
3167 request should be resend, otherwise page released from cache and
3173 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
3174 Details : Modify target file & which_patch.
3175 A flaw was found in the IA32 system call emulation provided
3176 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
3177 value could be stored in the %RAX register, which could trigger an
3178 out-of-bounds system call table access. An untrusted local user
3179 could exploit this flaw to run code in the kernel
3180 (ie a root privilege escalation). (CVE-2007-4573).
3184 Description: change order of libsysio includes
3185 Details : '#include sysio.h' should always come before '#include xtio.h'
3187 Severity : enhancement
3189 Description: adapt the lustre_config script to support the upgrade case
3190 Details : Add "-u" option for lustre_config script to support upgrading 1.4
3191 server targets to 1.6 in parallel.
3196 Description: To avoid grant space > avaible space when the disk is almost
3197 full. Without this patch you might see the error "grant XXXX >
3198 available" or some LBUG about grant, when the disk is almost
3200 Details : In filter_check_grant, for non_grant cache write, we should
3201 check the left space by if (*left > ungranted + bytes), instead
3202 of (*left > ungranted), because only we are sure the left space
3203 is enough for another "bytes", then the ungrant space should be
3204 increase. In client, we should update cl_avail_grant only there is
3205 OBD_MD_FLGRANT in the reply.
3210 Description: Update RHEL 4 kernel to fix local root privilege escalation.
3211 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
3212 described in CVE-2007-4573. This problem could allow untrusted
3213 local users to gain root access.
3216 Frequency : when using O_DIRECT and quotas
3218 Description: Incorrect file ownership on O_DIRECT output files
3219 Details : block usage reported by 'lfs quota' does not take into account
3220 files that have been written with O_DIRECT.
3224 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
3225 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
3226 window to 0. If it does hit cache miss, and read-pattern is still
3227 stride-io mode, does not reset the stride window, but also does not
3228 increase the stride window length in this case.
3230 --------------------------------------------------------------------------------
3232 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
3234 * Support for kernels:
3235 2.6.5-7.286 (SLES 9),
3236 2.6.9-55.0.2.EL (RHEL 4),
3237 2.6.16.46-0.14 (SLES 10),
3238 2.6.18-8.1.8.el5 (RHEL 5),
3239 2.6.18.8 vanilla (kernel.org)
3240 * Client support for unpatched kernels:
3241 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3242 2.6.16 - 2.6.21 vanilla (kernel.org)
3243 * Due to recently discovered recovery problems, we do not recommend
3244 using patchless RHEL 4 clients with this or any earlier release.
3245 * Recommended e2fsprogs version: 1.40.2-cfs1
3246 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3250 Description: Fix errors in lfs documentation
3251 Details : Fixes man pages
3253 Severity : enhancement
3255 Description: Adaptive timeouts
3256 Details : RPC timeouts adapt to changing server load and network
3257 conditions to reduce resend attempts and improve recovery time.
3259 Severity : enhancement
3261 Description: llapi_file_create() does not allow some changes
3262 Details : add llapi_file_open() that allows specifying the file creation
3263 mode and open flags, and also returns an open file handle.
3265 Severity : enhancement
3267 Description: df doesn't work properly if diskfs blocksize != 4K
3268 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
3271 Frequency : always on directories with default striping set
3273 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
3274 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
3276 Severity : enhancement
3278 Description: merge and cleanup kernel patches.
3279 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
3281 Severity : enhancement
3283 Description: RedHat Update kernel for RHEL5
3284 Details : Kernel config file for RHEL5.
3286 Severity : enhancement
3288 Description: OSS needs mutliple precreate threads
3289 Details : Add ability to start more than one create thread per OSS.
3292 Frequency : only with quota on the root user
3294 Description: mds_obd_create error creating tmp object
3295 Details : When the user sets quota on root, llog will be affected and can't
3296 create files and write files.
3299 Frequency : Always on ia64 patchless client, and possibly others.
3301 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
3302 Details : This allows the patchless client to be loaded on architectures
3303 without this export.
3308 Description: disorder of journal start and llog_add cause deadlock.
3309 Details : in llog_origin_connect, journal start should happen before
3310 llog_add keep the same order as other functions to avoid
3313 Severity : enhancement
3315 Description: RedHat Update kernel for RHEL5
3316 Details : Modify the kernel config file more closer RHEL5.
3318 Severity : enhancement
3320 Description: Build failure against Centos5 (RHEL5)
3321 Details : Define PAGE_SIZE when it isn't present.
3324 Frequency : occasionally when using NFS
3326 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
3327 Details : with NFS, the anon dentry's parent was set to itself in
3328 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
3329 corresponding dentry other than use rec->ur_name.
3331 Severity : enhancement
3333 Description: enable data checksumming by default
3334 Details : enable checksum by default, allow --disable-checksum
3335 configure option and "-o nochecksum" mount option. Checksums
3336 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
3339 Frequency : Occasionally with failover
3341 Description: Client eviction due to failover config
3342 Details : after a connection loss, the lustre client should attempt to
3343 reconnect to the last active server first before trying the
3344 other potential connections.
3346 Severity : enhancement
3348 Description: client-side metadata stat-ahead during readdir(directory readahead)
3349 Details : perform client-side metadata stat-ahead when the client detects
3350 readdir and sequential stat of dir entries therein
3355 Description: when mds and osts use different quota unit(32bit and 64bit),
3356 quota will be released repeatly.
3357 Details : void sending multiple quota reqs to mds, which will keep the
3358 status between the reqs.
3361 Frequency : only with liblustre clients on XT3
3363 Description: evictions taking too long
3364 Details : allow llrd to evict clients directly on OSTs
3368 Description: osts not allocated evenly to files
3369 Details : change the condition to increase offset_idx
3371 Severity : enhancement
3373 Description: self-adjustable client's lru lists
3374 Details : use adaptive algorithm for managing client cached locks lru
3375 lists according to current server load, other client's work
3376 pattern, memory activities, etc. Both, server and client
3377 side namespaces provide number of proc tunables for controlling
3382 Description: rewrite ext2-derived code in llite/dir.c and obdclass/uuid.c
3383 Details : rewrite inherited code (uuid parsing code from ext2 utils and
3384 readdir code from ext3) from scratch preserving functionality.
3388 Description: Only those disconnect error should be returned by rq_status.
3389 Details : In open/enqueue processs, Some errors, which will cause client
3390 disconnected, should be returned by rq_status, while other
3391 errors should still be returned by intent, then mdc or llite will
3394 Severity : enhancement
3396 Description: Tune the kernel for good SCSI performance.
3397 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
3398 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
3402 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
3403 Bugzilla : 13547, 13627
3404 Description: Data corruption for OSTs that are formatted larger than 2TB
3406 Details : When generating the bio request for lustre file writes the
3407 sector number would overflow a temporary variable before being
3408 used for the IO. The data reads correctly from Lustre (which
3409 will overflow in a similar manner) but other file data or
3410 filesystem metadata may be corrupted in some cases.
3414 Description: "lfs find -obd UUID" prints directories
3415 Details : "lfs find -obd UUID" will return all directory names instead
3416 of just file names. It is incorrect because the directories
3417 do not reside on the OSTs.
3420 Bugzilla : 13596, 13614
3421 Description: MDS hang after unclean shutdown of lots of clients
3422 Details : Never resend AST requests.
3426 Frequency : Always, for kernels after 2.6.16
3427 Description: Fix warning idr_remove called for id=.. which is not allocated.
3428 Details : Last kernels save old s_dev before kill super and not allow
3429 to restore from callback - restore it before call kill_anon_super.
3433 Description: Fix error on 'ls .' at the top of the Lustre mount.
3434 Details : Don't revalidate dentry if it is a root dentry.
3436 --------------------------------------------------------------------------------
3438 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
3440 * Support for kernels:
3441 2.6.5-7.286 (SLES 9),
3442 2.6.9-55.0.2.EL (RHEL 4),
3443 2.6.16.46-0.14 (SLES 10),
3444 2.6.18-8.1.8.el5 (RHEL 5),
3445 2.6.18.8 vanilla (kernel.org)
3446 * Client support for unpatched kernels:
3447 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3448 2.6.16 - 2.6.22 vanilla (kernel.org)
3449 * Due to recently discovered recovery problems, we do not recommend
3450 using patchless RHEL 4 clients with this or any earlier release.
3451 * Recommended e2fsprogs version: 1.39.cfs8
3452 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3457 Description: block reactivating mgc import until all deactivates complete
3458 Details : Fix race when failing back MDT/MGS to itself (testing)
3460 Severity : enhancement
3462 Description: lfs setstripe enhancement
3463 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
3466 Frequency : mds/oss recovery
3468 Description: llog ctxt is refrenced after it has been freed.
3469 Details : llog ctxt refcount was added to avoide the race between ctxt free
3470 and llog recovery process. Each llog user must hold ctxt refcount
3471 before it access the llog. And the llog ctxt can only be freed
3472 when its refcount is zero.
3474 Severity : enhancement
3476 Description: randomly memory allocation failure util
3477 Details : Make lustre randomly failed allocating memory for testing purpose.
3480 Frequency : only for SLES10
3482 Description: Update kernel patch for SLES10 SP1
3483 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
3485 Severity : enhancement
3487 Description: omit set fsid for export NFS
3488 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
3489 as FS_REQUIRES_DEV for avoid problems with generate fsid.
3492 Frequency : after network failures
3494 Description: Add sync option to mount_lustre.c
3495 Details : Client loses data written to lustre after a network interruption.
3497 Severity : enhancement
3499 Description: Error message improvement.
3500 Details : Merging of two LCONSOLE_ERROR_MSG into one.
3502 Severity : enhancement
3504 Description: don't use GFP_* in generic Lustre code.
3505 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
3507 Severity : enhancement
3509 Description: obdclass is limited by single OBD_ALLOC(idarray)
3510 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
3512 Severity : enhancement
3514 Description: updated patchess for new RHEL4 kernel
3515 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
3517 Severity : enhancement
3519 Description: warnings with build patchless client with vanila 2.6.19 and up
3520 Details : change old ctl_table style and replace ctl_table/ctl_table_header
3521 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
3523 Severity : enhancement
3525 Description: O_DIRECT bypasses client statistics.
3526 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
3527 read_ahead_stats were updated. Copied stats section from
3528 osc_send_oap_rpc() into async_internal().
3531 Frequency : only for Cray XT3
3533 Description: peer credits not enough on many OST per OSS systems.
3534 Details : Use new lnet way to add credits as we need those for pings and ASTs
3537 Frequency : only with liblustre
3539 Description: Liblustre is not releasing flock locks on file close.
3540 Details : Release flock locks on file close.
3543 Frequency : only for RHEL4
3545 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
3546 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
3547 Update target file and kernel config.
3551 Description: lustre support for RHEL5
3552 Details : Add support for RHEL5.
3557 Description: ASSERTION(export != NULL) failed in target_handle_connect
3558 Details : Assetion hit is result of rare race between disconnect and connet
3559 to same nid. target_handle_connect found old connect cockie and
3560 tried to reconnect, but can't find export for this cockie.
3565 Description: umount blocks forever on error
3566 Details : In result of wrong using obd_no_recov and obd_force flags client
3567 can hand if cancel or some other requests is lost.
3570 Frequency : Only for SLES
3572 Description: sanity_quota fail test_1
3573 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
3574 file, which makes TSTID[2] inunique.
3576 Severity : enhancement
3578 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
3579 Details : Update target/ChangeLog/which_patch .
3581 Severity : enhancement
3583 Description: jbd statistics
3584 Details : Port older jbd statistics patch for sles10
3589 Description: lvbo_init failed for resource with missing objects.
3590 Details : Fix returning error if we do stat for file with missing/corrupted
3591 objects and i_size set to all sum of size of all avaible objects.
3592 if we to truncate/write to missing object - it is recreated.
3597 Description: Oops in read and write path when failing to allocate lock.
3598 Details : Check if lock allocation failed and return error back.
3601 Frequency : When flocks are used.
3603 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
3604 Details : Flock locks might destroy just granted lock if it could be merged
3605 with another existing flock, this is done in completion handler,
3606 so teach ldlm_cli_enquque_fini that this is a valid case for
3612 Description: reply_lock_interpret crash due to race with it and lock cancel.
3613 Details : Do not replay locks that are being cancelled. Do not reference
3614 locks by their address during replay, just by their handle.
3617 Frequency : only with deactivated OSTs
3619 Description: lstripe command fails for valid OST index
3620 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
3621 instead of lov->desc.ld_active_tgt_count.
3623 Severity : enhancement
3625 Description: Build failure against Centos5 (RHEL5)
3626 Details : Use getpagesize() instead of PAGE_SIZE.
3628 --------------------------------------------------------------------------------
3630 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
3632 * Support for kernels:
3633 2.6.5-7.283 (SLES 9),
3634 2.6.9-55.EL (RHEL 4),
3635 2.6.16.46-0.14 (SLES 10),
3636 2.6.18.8 vanilla (kernel.org)
3637 * Client support for unpatched kernels:
3638 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3639 2.6.16 - 2.6.22 vanilla (kernel.org)
3640 * Due to recently discovered recovery problems, we do not recommend
3641 using patchless RHEL 4 clients with this or any earlier release.
3642 * Recommended e2fsprogs version: 1.39.cfs8
3643 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3644 * Starting with this release, the ldiskfs backing filesystem required
3645 by Lustre is now in its own package, lustre-ldiskfs. This package
3646 should be installed. It is versioned separately from Lustre and
3647 may be released separately in future.
3652 Description: server LBUG when shutting down
3653 Details : Block umount forever until the mount refcount is zero rather
3654 than giving up after an arbitrary timeout.
3656 Severity : enhancement
3658 Description: add optional extra BUILD_VERSION info
3659 Details : add a new environment variable (namely LUSTRE_VERS) which allows
3660 to override the lustre version.
3663 Frequency : 2.6.18 servers only
3665 Description: ll_kern_mount() doesn't release the module reference
3666 Details : The ldiskfs module reference count never drops down to 0
3667 because ll_kern_mount() doesn't release the module reference.
3672 Description: server LBUG when using old ost_num_threads parameter
3673 Details : Accept the old ost_num_threads parameter but warn that it
3674 is deprecated, and fix an off-by-one error that caused an LBUG.
3679 Description: Transient SCSI error results in persistent IO issue
3680 Details : iobuf->dr_error is not reinitialized to 0 between two
3684 Frequency : sometimes when underlying device returns I/O errors
3686 Description: OSTs not going read-only during write failures
3687 Details : OSTs are not remounted read-only when the journal commit threads
3688 get I/O errors because fsfilt_ext3 calls journal_start/stop()
3689 instead of the ext3 wrappers.
3692 Frequency : only with 10000 clients or more
3694 Description: poor connect scaling with increasing client count
3695 Details : Don't run filter_grant_sanity_check for more than 100 exports
3696 to improve scaling for large numbers of clients.
3699 Frequency : SLES10 only
3701 Description: sanity-quota.sh quotacheck failed: rc = -22
3702 Details : Quotas cannot be enabled on SLES10.
3705 Frequency : liblustre clients only
3707 Description: getdirentries does not give error when run on compute nodes
3708 Details : getdirentries does not fail when the size specified as an argument
3709 is too small to contain at least one entry
3711 Severity : enhancement
3713 Description: Add LNET router traceability for debug purposes
3714 Details : If a checksum failure occurs with a router as part of the
3715 IO path, the NID of the last router that forwarded the bulk data
3716 is printed so it can be identified.
3721 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
3722 Details : Due to a race condition, liblustre clients were occasionally
3723 evicted incorrectly.
3725 Severity : enhancement
3727 Description: lfs setstripe use optional parameters instead of postional
3730 Severity : enhancement
3732 Description: Nanosecond timestamp support for ldiskfs
3733 Details : The on-disk ldiskfs filesystem has added support for nanosecond
3734 resolution timestamps. There is not yet support for this at
3735 the Lustre filesystem level.
3738 Frequency : during server recovery
3740 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
3741 Details : request with rq_no_resend flag not awake l_wait_event if they get a
3745 Frequency : nfs export on patchless client
3747 Description: connectathon hang when test nfs export over patchless client
3748 Details : Disconnected dentry cannot be found with lookup, so we do not need
3749 to unhash it or make it invalid
3752 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
3753 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
3754 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
3759 Description: open req refcounting wrong on reconnect
3760 Details : If reconnect happened between getting open reply from server and
3761 call to mdc_set_replay_data in ll_file_open, we will schedule
3762 replay for unreferenced request that we are about to free.
3763 Subsequent close will crash in variety of ways.
3764 Check that request is still eligible for replay in
3765 mdc_set_replay_data().
3770 Description: disable writes to filesystem when reading health_check file
3771 Details : the default for reading the health_check proc file has changed
3772 to NOT do a journal transaction and write to disk, because this
3773 can cause reads of the /proc file to hang and block HA state
3774 checking on a healthy but otherwise heavily loaded system. It
3775 is possible to return to the previous behaviour during configure
3776 with --enable-health-write.
3778 Severity : enhancement
3780 Description: 64-bit inode version
3781 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
3782 to the inode. This will be required for version-based recovery.
3787 Description: MDS fails to start if a duplicate client export is detected
3788 Details : in some rare cases it was possible for a client to connect to
3789 an MDS multiple times. Upon recovery the MDS would detect this
3790 and fail during startup. Handle this more gracefully.
3792 Severity : enhancement
3794 Description: Add -o localflock option to simulate old noflock behaviour.
3795 Details : This will achieve local-only flock/fcntl locks coherentness.
3800 Description: log_commit_thread vs filter_destroy race leads to crash
3801 Details : Take import reference before releasing llog record semaphore
3806 Description: Wrong request locking in request set processing
3807 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
3808 imp_delayed_list, in this place should be used imp_lock.
3811 Frequency : when reconnecting
3813 Description: Grant leak when OSC reconnect to OST
3814 Details : When osc reconnect ost, OST(filter) should check whether it
3815 should grant more space to client by comparing fed_grant and
3816 cl_avail_grant, and return the granted space to client instead
3817 of "new granted" space, because client will call osc_init_grant
3818 to update the client grant space info.
3821 Frequency : when client reconnects to OST
3823 Description: Grant leak when OSC does a resend and replays bulk write
3824 Details : When osc reconnect to OST, OST(filter) should clear grant info of
3825 bulk write request, because the grant info while be sync between
3826 OSC and OST when reconnect, and we should ignore the grant info
3827 these of resend/replay write req.
3832 Description: Grant space more than avaiable space sometimes.
3833 Details : When then OST is about to be full, if two bulk writing from
3834 different clients came to OST. Accord the avaliable space of the
3835 OST, the first req should be permitted, and the second one
3836 should be denied by ENOSPC. But if the seconde arrived before
3837 the first one is commited. The OST might wrongly permit second
3838 writing, which will cause grant space > avaiable space.
3841 Frequency : when client is evicted
3843 Description: Grant might be wrongly erased when osc is evicted by OST
3844 Details : when the import is evicted by server, it will fork another
3845 thread ptlrpc_invalidate_import_thread to invalidate the
3846 import, where the grant will be set to 0. While the original
3847 thread will update the grant it got when connecting. So if
3848 the former happened latter, the grant will be wrongly errased
3849 because of this race.
3854 Description: Checking Stale with correct fid
3855 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
3856 to check whether it is stale, because sometimes, we want the
3857 enqueue happened anyway, and op_data.fid2 will not be initialized.
3859 Severity : enhancement
3861 Description: update patchless client
3862 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
3865 Frequency : only with 2.4 kernel
3867 Description: random memory corruption
3868 Details : size of struct ll_inode_info is to big for union inode.u and this
3869 can be cause of random memory corruption.
3874 Description: Memory leak in recovery
3875 Details : Lov_mds_md was not free in an error handler in mds_create_object.
3876 It should also check obd_fail before fsfilt_start, otherwise if
3877 fsfilt_start return -EROFS,(failover mds during mds recovery).
3878 then the req will return with repmsg->transno = 0 and rc = EROFS.
3879 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
3880 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
3881 be freed no matter whether fsfilt_commit success or not.
3884 Frequency : only with large numbers of cores on a single node
3886 Description: Prevents from taking the superblock lock in llap_from_page for
3888 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
3889 for llap_from_page call in ll_removepage() prevents from taking
3890 the superblock lock for a soon died page.
3895 Description: Not check open intent error before release open handle
3896 Details : in some rare cases, the open intent error is not checked before
3897 release open handle, which may cause
3898 ASSERTION(open_req->rq_transno != 0), because it tries to release
3899 the failed open handle.
3904 Description: Set cat log bitmap only after create log success.
3905 Details : in some rare cases, the cat log bitmap is set too early. and it
3906 should be set only after create log success.
3910 Description: Accessing a block bevice can re-enable I/O when Lustre is
3911 tearing down a device.
3912 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
3917 Description: service threads may hog cpus when there are a lot of requests
3918 Details : Insert cond_resched to give other threads a chance to use some CPU
3923 Description: the cat log was not initialized in recovery
3924 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
3925 unlink log on mds will not be initialized until mds post
3926 recovery. And also in mds post recovery, the unlink log will
3927 initialization will be done asynchronausly, so there will be race
3928 between add unlink log and unlink log initialization.
3932 Description: brw_stats were being printed incorrectly
3933 Details : brw_stats were being printed as log2 but all of them were not
3934 recorded as log2. Also remove some code duplication arising from
3935 filter_tally_{read,write}.
3939 Frequency : rare, only in recovery.
3940 Description: ASSERTION(req->rq_type != LI_POISON) failed
3941 Details : imp_lock should be held while iterating over imp_sending_list for
3942 prevent destroy request after get timeout in ptlrpc_queue_wait.
3946 Description: replay-single.sh test 52 fails
3947 Details : A lock's skiplist need to be cleanup when it being unlinked
3948 from its resource list.
3953 Description: Short directio read returns full requested size rather than
3955 Details : Direct I/O operations should return actual amount of bytes
3956 transferred rather than requested size.
3958 Severity : enhancement
3960 Description: metadata RPC reduction (e.g. for rm performance)
3961 Details : decrease the amount of synchronous RPC between clients and servers
3962 by canceling conflicing lock before the operation on the client
3963 and packing thier handles into the main operation RPC to server.
3965 Severity : enhancement
3967 Description: add #ifdef HAVE_KERNEL_CONFIG_H
3968 Details : kernels from 2.6.19 not need include linux/config.h, but add
3969 include linux/autoconf.h in commpiler command line.
3971 Severity : enhancement
3973 Description: patchless client support for 2.6.22 kernel
3974 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
3975 removed. In this case we need drop using os depended interface to
3976 kmem_cache and use cfs_mem_cache API.
3981 Description: fix mal-formatted messages
3982 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
3986 Frequency : always in liblustre
3987 Description: wrong IS_ERR implementation in liblustre.h
3988 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
3993 Description: Correct condition for output debug message.
3994 Details : inode i_nlink equal zero is not enough for output message about
3995 disk corruption, i_ctime and i_mode should be also checked.
3999 Frequency : always in patchless client
4000 Description: add configure check for truncate_complete_page
4001 Details : improve checks for exported symbols. This allow run check without
4002 sources, but with Module.symvers shipped with kernel distribution.
4003 add check for truncate_complete_page used by patchless client.
4007 Description: sanity.sh test_77h fails with "test_77h file compare failed"
4008 Details : test_77h uses a file which was messed by other test case.
4012 Description: Not Check whether lov_tgts is NULL in some lov functions
4013 Details : Checking whether lov_tgts is NULL in some functions.
4017 Description: replace obdo_alloc() with OBDO_ALLOC macro
4018 Details : nothing special is done in obdo_alloc() function, and for
4019 debugging purpose, it needs to be replaced with macros.
4023 Description: bad return value and errno from fcntl call
4024 Details : In liblustre API, errno should be a negative value if error
4029 Description: ptlrpc_check_set() LBUG
4030 Details : In case of positive reply from server and failed client bulk
4031 callback after bulk transfer shouldn't LBUG, but process this
4032 request as erroneous.
4034 Severity : enhancement
4036 Description: VFS operations stats tool.
4037 Details : Tool which collects stats by tracking value written in pid,
4038 ppid, gid and uses llstat to generate output to plot graph using
4040 Updated lustre/utils/Makefile.am
4041 Added lustre/utils/ltrack_stats.c
4043 Severity : enhancement
4045 Description: 2.6.18 server support (lustre 1.6.1)
4046 Details : Support for 2.6.18 kernels on the server side.
4051 Description: ASSERTION(imp->imp_conn_current) failed
4052 Details : an assertion failure is hit if a client node boots and attempts to
4053 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
4056 Frequency : only for i686
4058 Description: 1.4.11 RC1 build fails for RHEL 4, i686
4059 Details : Fixed config variable for build.
4064 Description: Updated patchess for new RHEL4 kernel
4065 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
4066 Updated patch jbd-stats-2.6.9.patch
4067 Updated patch qsnet-rhel4-2.6.patch
4068 Updated patch quota-deadlock-on-pagelock-core.patch
4069 Updated patch vfs_intent-2.6-rhel4.patch
4070 Updated patch vfs_races-2.6-rhel4.patch
4071 Updated series file 2.6-rhel4-titech.series
4072 Updated series file 2.6-rhel4.series
4073 Updated kernel config files
4078 Description: lquota slave complains LBUG when reconnecting with mds
4080 Details : quota slave depends qctxt->lqc_import to send its quota request.
4081 This pointer will be invalid if mds did failover or broke its
4082 connect to osts, which leads to LBUG.
4085 Frequency : when qunit size is too small(less than 20M)
4087 Description: write is stopped by improper -EDQUOT
4088 Details : If the master is busy and qunit size is small enough(let's say 1M),
4089 the slave can not get quota from master on time, which will lead
4090 slave to trigger a -EQUOTA to client.
4095 Description: Deadlock during metadata tests
4096 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
4097 called with the per-dentry lock held.
4100 Frequency : SLES9 only
4102 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
4103 Details : set CONFIG_NR_CPUS to 128 instead of 8.
4105 Severity : enhancement
4107 Description: remove fs_prep_san_write operation and related patches
4108 Details : remove the ext3-san-jdike patches which are no longer useful.
4113 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
4114 Details : a disk corruption can cause the mballoc code to assert on a
4115 double free or other extent corruptions. Handle these with
4116 ext3_error() instead of assertions.
4119 Frequency : only with mballoc3 code and deep extent trees
4121 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
4122 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
4123 can cause the filesystem to be remounted read-only.
4128 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
4129 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
4130 few external users of this ioctl number is lfsck's call to
4131 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
4132 Add the old ioctl number to the handler so both old and new
4137 Description: parallel lock callbacks
4138 Details : Instead of sending blocking and completion callbacks as separated
4139 requests, adding them to a set and sending in parallel.
4143 Description: Disable most debugging by default
4144 Details : To improve performance, disable most logging (for debug purposes)
4145 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
4146 default, and HA includes fewer messages.
4150 Frequency : only run on patchless client.
4151 Description: use do_facet on sanity.sh for test handling recoverables errors
4152 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
4156 Description: hash tables for lists of nids, connections and uuids
4157 Details : Hash tables noticeably help when a lot of clients connect to a
4158 server, to faster identify duplicate connections or reconnects,
4159 also to faster find export to evict in manual eviction case.
4163 Description: Sometimes, when the server evict a client, and the client will
4164 not be evicted as soon as possible.
4165 Details : In enqueue req, the error was returned by intent, instead of
4166 rq_status which make ptlrpc layer not detect this error, and
4167 does not evict the client. So enqueue error should be returned
4171 Frequency : only at startup
4173 Description: Delay client connections to MDT until fist MDT->OST connect
4174 Details : If a client tried to create a new file before the MDT had
4175 connected to any OSTs, the create would return EIO. Now
4176 the client will simply block until the MDT connects to the
4177 first OST and the create can succeed.
4180 Frequency : only for SLES9
4182 Description: Routinely utilize latest Quadrics drivers in CFS releases
4183 Details : Update patch qsnet-suse-2.6.patch.
4186 Frequency : only for sles10
4188 Description: Update patches for SLES 10 SP1 kernel.
4189 Details : Update the patch vfs_intent-2.6-sles10.patch.
4194 Description: Routinely utilize latest Quadrics drivers in CFS releases
4195 Details : Update patch qsnet-rhel4-2.6.patch.
4198 Frequency : at statup only
4200 Description: mds_lov_synchronize race leads to various problems
4201 Details : simultaneous MDT->OST connections at startup can cause the
4202 sync to abort, leaving the OSC in a bad state.
4207 Description: Using wrong pointer in osc_brw_prep_request
4208 Details : Access to array[-1] can produce panic if kernel compiled with
4209 CONFIG_PAGE_ALLOC enabled
4211 Severity : enhancement
4213 Description: Async OSC create to avoid the blocking unnecessarily.
4214 Details : If a OST has no remain object, system will block on the creating
4215 when need to create a new object on this OST. Now, ways use
4216 pre-created objects when available, instead of blocking on an
4217 empty osc while others are not empty. If we must block, we block
4218 for the shortest possible period of time.
4222 Frequency : only in recovery
4223 Description: Mark OST as early accessible if his start SYNC.
4224 Details : osc_precreate return flag early accessible if oscc marked as
4225 OSCC_FLAG_SYNC_IN_PROGRESS.
4230 Description: Sometimes precreate code can triger create object on wrong ost
4231 Details : Wrong protected or not not restored variables aftre precreate loop
4232 can produce creation object on wrong ost.
4235 Frequency : oss recovery
4237 Description: llog_commit_thread cleanup should sync with llog_commit_thread
4239 Details : llog_commit_thread_count should be synced between llog_commit
4240 start and cleanup, so new llog_commit thread should not be started
4241 when llog_commit threads being stopped to avoid accessing some
4244 Severity : enhancement
4246 Description: Add printing inode info into message about error in writepage.
4248 --------------------------------------------------------------------------------
4250 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
4255 Frequency : on some architectures
4257 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
4258 Details : Uninitialized flags sometimes cause configuration commands to
4262 Frequency : patchless clients only
4264 Description: missing __iget() symbol export
4265 Details : The __iget() symbol export is missing. To avoid the need for
4266 this on patchless clients the deathrow inode reaper is turned
4267 off, and we depend on the VM to clean up old inodes. This
4268 dependency was during via the fix for bug 12181.
4273 Description: sanity.sh fail: test_52b
4274 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
4275 incorrect inode's flags to client.
4277 --------------------------------------------------------------------------------
4279 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
4281 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
4282 INTEROPERATE with older versions automatically. In many cases a
4283 special upgrade step is needed. Please read the
4284 user documentation before upgrading any part of a 1.4.x system.
4285 * WARNING: Lustre configuration and startup changes are required with
4286 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
4288 * Support for kernels:
4289 2.4.21-47.0.1.EL (RHEL 3),
4290 2.6.5-7.283 (SLES 9),
4291 2.6.9-42.0.10.EL (RHEL 4),
4292 2.6.12.6 vanilla (kernel.org),
4293 2.6.16.27-0.9 (SLES10)
4294 * Client support for unpatched kernels:
4295 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
4296 2.6.16 - 2.6.19 vanilla (kernel.org),
4297 2.6.9-42.0.8EL (RHEL 4)
4298 * Recommended e2fsprogs version: 1.39.cfs6
4299 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
4302 Severity : enhancement
4304 Description: Async OSC create to avoid the blocking unnecessarily.
4305 Details : If a OST has no remain object, system will block on the creating
4306 when need to create a new object on this OST. Now, ways use
4307 pre-created objects when available, instead of blocking on an
4308 empty osc while others are not empty. If we must block, we block
4309 for the shortest possible period of time.
4311 Severity : enhancement
4313 Description: MountConf
4314 Details : Lustre configuration is now managed via mkfs and mount
4315 commands instead of lmc and lconf. New obd types (MGS, MGC)
4316 are added for dynamic configuration management. See
4317 https://mail.clusterfs.com/wikis/lustre/MountConf for
4320 Severity : enhancement
4322 Description: dynamic OST addition
4323 Details : OSTs can now be added to a live filesystem
4325 Severity : enhancement
4327 Description: startup order invariance
4328 Details : MDTs and OSTs can be started in any order. Clients only
4329 require the MDT to complete startup.
4331 Severity : enhancement
4333 Description: parallel, asynchronous orphan cleanup
4334 Details : orphan cleanup is now performed in separate threads for each
4335 OST, allowing parallel non-blocking operation.
4337 Severity : enhancement
4339 Description: optimized stripe assignment
4340 Details : stripe assignments are now made based on ost space available,
4341 ost previous usage, and OSS previous usage, in order to try
4342 to optimize storage space and networking resources.
4344 Severity : enhancement
4346 Description: Permanently set tunables
4347 Details : All writable /proc/fs/lustre tunables can now be permanently
4348 set on a per-server basis, at mkfs time or on a live
4351 Severity : enhancement
4353 Description: Lustre message v2
4354 Details : Add lustre message format v2.
4356 Severity : enhancement
4358 Description: client OST exclusion list
4359 Details : Clients can be started with a list of OSTs that should be
4360 declared "inactive" for known non-responsive OSTs.
4364 Description: ENOENT returned for valid filehandle during dbench.
4365 Details : Check if a directory has children when invalidating dentries
4366 associated with an inode during lock cancellation. This fixes
4367 an incorrect ENOENT sometimes seen for valid filehandles during
4368 testing with dbench.
4371 Frequency : SFS test only (otherwise harmless)
4373 Description: SPEC SFS validation failure on NFS v2 over lustre.
4374 Details : Changes the blocksize for regular files to be 2x RPC size,
4375 and not depend on stripe size.
4377 Severity : enhancement
4379 Description: fine-grained SMP locking inside DLM
4380 Details : Improve DLM performance on SMP systems by removing the single
4381 per-namespace lock and replace it with per-resource locks.
4383 Severity : enhancement
4385 Description: don't hold multiple extent locks at one time
4386 Details : To avoid client eviction during large writes, locks are not
4387 held on multiple stripes at one time or for very large writes.
4388 Otherwise, clients can block waiting for a lock on a failed OST
4389 while holding locks on other OSTs and be evicted.
4391 Severity : enhancement
4393 Description: Multiple MD RPCs in flight.
4394 Details : Further unserialise some read-only MDT RPCs - learn about intents.
4395 To avoid overly-overloading MDT, introduce a limit on number of
4396 MDT RPCs in flight for a single client and add /proc controls
4397 to adjust this limit.
4399 Severity : enhancement
4401 Description: client read/write statistics
4402 Details : Add client read/write call usage stats for performance
4403 analysis of user processes.
4404 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
4405 file access. extents_stats shows chunk size distribution.
4406 extents_stats_per_process show chunk size distribution per
4409 Severity : enhancement
4411 Description: per-client statistics on server
4412 Details : Add ldlm and operations statistics for each client in
4413 /proc/fs/lustre/mds|obdfilter/*/exports/
4415 Severity : enhancement
4417 Description: improved MDT statistics
4418 Details : Add detailed MDT operations statistics in
4419 /proc/fs/lustre/mds/*/stats
4421 Severity : enhancement
4423 Description: VFS operations stats
4424 Details : Add client VFS call stats, trackable by pid, ppid, or gid
4425 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
4430 Description: Fix client-side osc byte counters
4431 Details : The osc read/write byte counters in
4432 /proc/fs/lustre/osc/*/stats are now working
4435 Frequency : always as root on SLES
4437 Description: Failure of copying files with lustre special EAs.
4438 Details : Client side always return success for setxattr call for lustre
4439 special xattr (currently only "trusted.lov").
4444 Description: Refcount LNET uuids
4445 Details : The global LNET uuid list grew linearly with every startup;
4446 refcount repeated list entries instead of always adding to
4449 Severity : enhancement
4451 Description: Dynamic service threads
4452 Details : Within a small range, start extra service threads
4453 automatically when the request queue builds up.
4456 Frequency : mixed-endian client/server environments
4458 Description: mixed-endian crashes
4459 Details : The new msg_v2 system had some failures in mixed-endian
4462 Severity : enhancement
4464 Description: Easy OST removal
4465 Details : OSTs can be permanently deactivated with e.g. 'lctl
4466 conf_param lustre-OST0001.osc.active=0'
4468 Severity : enhancement
4470 Description: MGS proc entries
4471 Details : Added basic proc entries for the MGS showing what filesystems
4474 Severity : enhancement
4476 Description: provide MGS failover
4477 Details : Added config lock reacquisition after MGS server failover.
4479 Severity : enhancement
4481 Description: add Linux 2.4 support
4482 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
4486 Description: a large application tries to do I/O to the same resource and dies
4487 in the middle of it.
4488 Details : Check the req->rq_arrival time after the call to
4489 ost_brw_lock_get(), but before we do anything about
4490 processing it & sending the BULK transfer request. This
4491 should help move old stale pending locks off the queue as
4492 quickly as obd_timeout.
4495 Frequency : when an incorrect nid is specified during startup
4497 Description: ptlrpc connect to non-existant node causes kernel crash
4498 Details : LNET can't be re-entered from an event callback, which
4499 happened when we expire a message after the export has been
4500 cleaned up. Instead, hand the zombie cleanup off to another
4503 Severity : enhancement
4505 Description: plain/inodebits lock performance improvement
4506 Details : Grouping plain/inodebits in granted list by their request modes
4507 and bits policy, thus improving the performance of search through
4511 Frequency : only if OST filesystem is corrupted
4513 Description: client incorrectly hits assertion in ptlrpc_replay_req()
4514 Details : for a short time RPCs with bulk IO are in the replay list,
4515 but replay of bulk IOs is unimplemented. If the OST filesystem
4516 is corrupted due to disk cache incoherency and then replay is
4517 started it is possible to trip an assertion. Avoid putting
4518 committed RPCs into the replay list at all to avoid this issue.
4521 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
4523 Description: System hang on startup
4524 Details : This bug allowed the liblustre (e.g. catamount) client to
4525 return to the app before handling all startup RPCs. This
4526 could leave the node unresponsive to lustre network traffic
4527 and manifested as a server ptllnd timeout.
4529 Severity : enhancement
4531 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
4532 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
4533 boolean to control whether to print peer debug info when a
4534 client's RPC times out.
4537 Frequency : only for kernels with patches from Lustre below 1.4.3
4539 Description: Remove old rdonly API
4540 Details : Remove old rdonly API which unused from at least lustre 1.4.3
4543 Frequency : only for devices with external journals
4545 Description: Set external device read-only also
4546 Details : During a commanded failover stop, we set the disk device
4547 read-only while the server shuts down. We now also set any
4548 external journal device read-only at the same time.
4551 Frequency : when upgrading from 1.4 while trying to change parameters
4553 Description: The wrong (new) MDC name was used when setting parameters for
4554 upgraded MDT's. Also allows changing of OSC (and MDC)
4555 parameters if --writeconf is specified at tunefs upgrade time.
4558 Frequency : when setting specific ost indicies
4560 Description: QOS code breaks on skipped indicies
4561 Details : Add checks for missing OST indicies in the QOS code, so OSTs
4562 created with --index need not be sequential.
4564 Severity : enhancement
4566 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
4567 Details : The uninit_groups feature works in conjunction with the kernel
4568 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
4569 up the pass1 processing of e2fsck. This is a read-only feature
4570 in ldiskfs2 only, so older kernels and current ldiskfs cannot
4571 mount filesystems that have had this feature enabled.
4573 Severity : enhancement
4575 Description: Improve multi-block allocation algorithm to avoid fragmentation
4576 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
4577 allocation locality and avoid filesystem fragmentation.
4579 ------------------------------------------------------------------------------
4581 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
4583 * Support for kernels:
4584 2.4.21-47.0.1.EL (RHEL 3)
4585 2.6.5-7.283 (SLES 9)
4586 2.6.9-42.0.10.EL (RHEL 4)
4587 2.6.12.6 vanilla (kernel.org)
4588 2.6.16.27-0.9 (SLES 10)
4589 * Recommended e2fsprogs version: 1.39.cfs5
4591 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
4595 Frequency : occasional, depends on client load and configuration
4596 Bugzilla : 12181, 12203
4597 Description: data loss for recently-modified files
4599 Details : In some cases it is possible that recently written or created
4600 files may not be written to disk in a timely manner (this should
4601 normally be within 30s unless client IO load is very high).
4602 The problem appears as zero-length files or files that are a
4603 multiple of 1MB in size after a client crash or client eviction
4604 that are missing data at the end of the file.
4606 This problem is more likely to be hit on clients where files are
4607 repeatedly created and unlinked in the same directory, clients
4608 have a large amount of RAM, have many CPUs, the filesystem has
4609 many OSTs, the clients are rebooted frequently, and/or the files
4610 are not accessed by other nodes after being written.
4612 The presence of the problem can be detected by looking at
4613 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
4614 smaller than the second (nr_unused) then dirty files will not
4615 be flushed automatically to disk. "sync; sleep 10" should be
4616 run several times on the node before unmounting it to update
4617 Lustre (this is also safe to run on nodes without this problem).
4619 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
4620 that can cause this same problem, so customers using that kernel
4621 also need to update the kernel in addition to Lustre. In order
4622 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
4624 It is normal that files written just before a client crash (less
4625 than 30s) may not yet have been flushed to disk, even for local
4629 Frequency : frequent on thin XT3 nodes
4631 Description: UUID collision on thin XT3 Linux nodes
4632 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
4633 randomly, since we relied on an insufficiently-seeded PRNG.
4638 Description: OSS hangs after "All ost request buffers busy"
4639 Details : A deadlock between quota and journal operations caused OSS
4640 hangs after printing "All ost request buffers busy."
4643 Frequency : always on liblustre builds
4645 Description: Cleanup compiler warnings on liblustre
4648 Frequency : always on liblustre builds on XT3
4650 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
4654 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
4655 > 32000 subdirectories
4656 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
4657 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
4658 subdirectory count crosses 32000. This will aid e2fsck to
4659 correctly handle more than 32000 subdirectories.
4662 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
4664 Description: System hang on startup
4665 Details : This bug allowed the liblustre (e.g. catamount) client to
4666 return to the app before handling all startup RPCs. This
4667 could leave the node unresponsive to lustre network traffic
4668 and manifested as a server ptllnd timeout.
4670 Severity : enhancement
4672 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
4673 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
4674 boolean to control whether to print peer debug info when a
4675 client's RPC times out.
4680 Description: make O_SYNC working on 2.6 kernels
4681 Details : 2.6 kernels use different method for mark pages for write,
4682 so need add a code to lustre for O_SYNC work.
4687 Description: Failure to close file and release space on NFS
4688 Details : Put inode details into lock acquired in ll_intent_file_open.
4689 Use mdc_intent_lock in ll_intent_open to properly
4690 detect all kind of errors unhandled by mdc_enqueue.
4695 Description: proc file read during shutdown sometimes raced obd removal,
4697 Details : Add lock to prevent obd access after proc file removal.
4700 Frequency : Only for files larger than 4GB on 32-bit clients.
4702 Description: improperly doing page alignment of locks
4703 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
4704 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
4709 Description: RPCs being resent when they shouldn't be
4710 Details : Some RPCs that should not be resent are being resent. This
4711 can cause inconsistencies in the RPC state machine. Do not
4712 resend such requests.
4715 Frequency : rare, only with NFS export
4717 Description: Crash on NFS re-export node
4718 Details : under very unusual load conditions an assertion is hit in
4719 ll_intent_file_open()
4722 Frequency : only if OST filesystem is corrupted
4724 Description: client incorrectly hits assertion in ptlrpc_replay_req()
4725 Details : for a short time RPCs with bulk IO are in the replay list,
4726 but replay of bulk IOs is unimplemented. If the OST filesystem
4727 is corrupted due to disk cache incoherency and then replay is
4728 started it is possible to trip an assertion. Avoid putting
4729 committed RPCs into the replay list at all to avoid this issue.
4734 Description: large O_DIRECT requests fail under memory pressure/fragmentation
4735 Details : Large single O_DIRECT read and write calls can fail to allocate
4736 a sufficiently large buffer to process the request. In case of
4737 allocation failure the allocation is retried with a smaller
4738 buffer and broken into smaller requests.
4740 Severity : enhancement
4742 Description: Add -o localflock option to simulate old noflock behaviour.
4743 Details : This will achieve local-only flock/fcntl locks coherentness.
4748 Description: versioning check is incomplete
4749 Details : Checking the version difference of client vs. server, report
4750 error if the gap is too big.
4755 Description: add support PG_writeback bit
4756 Details : add support for PG_writeback bit for Lustre, for more carefull
4757 work with page cache in 2.6 kernel. This also fix some deadlocks
4758 and remove hack for work O_SYNC with 2.6 kernel.
4760 Severity : enhancement
4762 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
4763 Details : The uninit_groups feature works in conjunction with the kernel
4764 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
4765 up the pass1 processing of e2fsck. This is a read-only feature
4766 in ldiskfs2 only, so older kernels and current ldiskfs cannot
4767 mount filesystems that have had this feature enabled.
4769 Severity : enhancement
4771 Description: Improve multi-block allocation algorithm to avoid fragmentation
4772 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
4773 allocation locality and avoid filesystem fragmentation.
4775 ------------------------------------------------------------------------------
4777 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
4779 * Support for kernels:
4780 2.6.9-42.0.3.EL (RHEL 4)
4781 2.6.5-7.276 (SLES 9)
4782 2.4.21-47.0.1.EL (RHEL 3)
4783 2.6.12.6 vanilla (kernel.org)
4784 2.6.16.21-0.8 (SLES10)
4785 * Recommended e2fsprogs version: 1.39.cfs2-0
4787 * The backwards-compatible /proc/sys/portals symlink has been removed
4788 in this release. Before upgrading, please ensure that you change
4789 any configuration scripts or /etc/sysctl.conf files that access
4790 /proc/sys/portals/* or sysctl portals.* to use the corresponding
4791 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
4792 in advance of the upgrade on any system running Lustre 1.4.6 or
4793 newer, since /proc/sys/lnet was added in that version.
4794 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
4798 Frequency : only when quota is used
4800 Description: avoid scanning export list for quota master
4801 Details : Change the algorithms to avoid scanning export list in order
4802 to improve the efficiency.
4805 Frequency : MDS failover only, very rarely
4807 Description: "went back in time" messages on mds failover
4808 Details : The greatest transno may be lost when the current operation
4809 finishes with an error (transno==0) and the client's last_rcvd
4810 record is over-written. Save the greatest transno in the
4811 mds_last_transno for this case.
4814 Frequency : always for specific kernels and striping counts
4816 Description: client may get "Matching packet too big" without ACL support
4817 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
4818 when trying to access files in certain configurations. The
4819 clients should in fact be denied when mounting because they do
4820 not understand ACLs.
4823 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
4825 Description: many clients connecting with IO in progress causes connect timeouts
4826 Details : Avoid synchronous journal commits to avoid delays caused by many
4827 clients connecting/disconnecting when bulk IO is in progress.
4828 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
4829 OST_IO_PORTAL to avoid delays behind potentially many pending
4833 Frequency : occasionally with multiple writers to a single file
4835 Description: shared writes to file may result in wrong size reported by stat()
4836 Details : Allow growing of kms when extent lock is cancelled
4839 Frequency : always with random mmap IO to multi-striped file
4841 Description: mmap write might be lost if we are writing to a 'hole' in stripe
4842 Details : Only if the hole is at the end of OST object so that kms is too
4843 small. Fix is to increase kms accordingly in ll_nopage.
4846 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
4848 Description: writes to a missing object would leak memory on the OST
4849 Details : If there is an inconsistency between the MDS and OST filesystems,
4850 such that the MDS references an object that doesn't exist, writes
4851 to that object will leak memory due to incorrect cleanup in the
4852 error handling path, eventually running out of memory on the OST.
4857 Description: Creating too long symlink causes lustre errors
4858 Details : Check symlink and name lengths before sending requests to MDS.
4861 Frequency : only if flock is enabled (not on by default)
4863 Description: posix locks not released on fd closure on 2.6.9+
4864 Details : We failed to add posix locks to list of inode locks on 2.6.9+
4865 kernels, this caused such locks not to be released on fd close and
4866 then assertions on fs unmount about still used locks.
4869 Frequency : MDS failover only, very rarely
4871 Description: clients may get ASSERTION(granted_lock != NULL)
4872 Details : When request was taking a long time, and a client was resending
4873 a getattr by name lock request. The were multiple lock requests
4874 with the same client lock handle and
4875 mds_getattr_name->fixup_handle_for_resent_request found one of the
4876 lock handles but later failed with ASSERTION(granted_lock != NULL).
4881 Description: handle->h_buffer_credits > 0, assertion failure
4882 Details : h_buffer_credits is zero after truncate, causing assertion
4883 failure. This patch extends the transaction or creates a new
4887 Frequency : NFS re-export or patchless client
4888 Bugzilla : 11179, 10796
4889 Description: Crash on NFS re-export node (__d_move)
4890 Details : We do not want to hash the dentry if we don't have a lock.
4891 But if this dentry is later used in d_move, we'd hit uninitialised
4892 list head d_hash, so we just do this to init d_hash field but
4893 leave dentry unhashed.
4896 Frequency : NFS re-export or patchless client
4898 Description: NFS exports has problem with symbolic link
4899 Details : lustre client didn't properly install dentry when re-exported
4900 to NFS or running patchless client.
4903 Frequency : NFS re-export or patchless client
4905 Description: Various nfs/patchless fixes.
4906 Details : fixes reuse disconected alias for lookup process - this fixes
4907 warning "find_exported_dentry: npd != pd",
4908 fix permission error with open files at nfs.
4909 fix apply umask when do revalidate.
4912 Frequency : occasional
4914 Description: Crash on NFS re-export node
4915 Details : calling clear_page() on the wrong pointer triggered oops in
4916 generic_mapping_read().
4919 Frequency : rarely, using O_DIRECT IO
4921 Description: unaligned directio crashes client with LASSERT
4922 Details : check for unaligned buffers before trying any requests.
4925 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
4927 Description: stale data returned from RAID cache
4928 Details : If only a small amount of IO is done to the RAID device before
4929 reading it again it is possible to get stale data from the RAID
4930 cache instead of reading it from disk.
4933 Frequency : always for sles10 kernel
4935 Description: sles10 support
4936 Details : ll_follow_link: compile fixes and using of nd_set_link
4937 under newer kernels.
4940 Frequency : depends on arch, kernel and compiler version, always on sles10
4943 Description: recursive or deep enough symlinks cause stack overflow
4944 Details : getting rid of large stack-allocated variable in
4948 Frequency : depends on hardware
4950 Description: lustre write performance loss in the SLES10 kernel
4951 Details : the performance loss is caused by using of write barriers in the
4952 ext3 code. The SLES10 kernel turns barrier support on by
4953 default. The fix is to undo that change for ldiskfs.
4955 ------------------------------------------------------------------------------
4957 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
4959 * Support for kernels:
4960 2.6.9-42.0.3EL (RHEL 4)
4961 2.6.5-7.276 (SLES 9)
4962 2.4.21-47.0.1.EL (RHEL 3)
4963 2.6.12.6 vanilla (kernel.org)
4967 Frequency : quota enabled and large files being deleted
4969 Description: releasing more than 4GB of quota at once hangs OST
4970 Details : If a user deletes more than 4GB of files on a single OST it
4971 will cause the OST to spin in an infinite loop. Release
4972 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
4977 Description: statfs data retrieved from /proc may be stale or zero
4978 Details : When reading per-device statfs data from /proc, in the
4979 {kbytes,files}_{total,free,avail} files, it may appear
4980 as zero or be out of date.
4983 Frequency : systems with MD RAID1 external journal devices
4985 Description: lconf's call to blkid is confused by RAID1 journal devices
4986 Details : Use the "blkid -l" flag to locate the MD RAID device instead
4987 of returning all block devices that match the journal UUID.
4990 Frequency : always, for aggregate stripe size over 4GB
4992 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
4993 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
4994 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
4997 Frequency : always if "lfs find" used on a local file/directory
4999 Description: "lfs find" segfaults if used on a local file/directory
5000 Details : The case where a directory component was not specified wasn't
5001 handled correctly. Handle this properly.
5004 Frequency : always on ppc64
5006 Description: the write to an ext3 filesystem mounted with mballoc got stuck
5007 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
5008 perform endianness conversion.
5011 Frequency : rarely (truncate to non-zero file size after write under load)
5012 Bugzilla : 10730, 10687
5013 Description: Files padded with zeros to next 4K multiple
5014 Details : With filesystems mounted using the "extents" option (2.6 kernels)
5015 it is possible that files that are truncated to a non-zero size
5016 immediately after being written are filled with zero bytes beyond
5017 the truncated size. No file data is lost.
5019 Severity : enhancement
5021 Description: Allow recovery/failover for liblustre clients.
5022 Details : liblustre clients were unaware of failover configurations until
5025 Severity : enhancement
5027 Description: user file locks should fail when not mounting with flock option
5028 Details : Set up an error-returning stub in ll_file_operations.lock field
5029 to prevent incorrect behaviour when client is mounted without
5030 flock option. Also, set up properly f_op->flock field for
5034 Frequency : always on ia64
5036 Description: "lfs df" loops on printing out MDS statfs information
5037 Details : The obd_ioctl_data was not initialized and in some systems
5038 this caused a failure during the ioctl that did not return
5039 an error. Initialize the struct and return an error on failure.
5042 Frequency : SLES 9 only
5044 Description: Error of copying files with lustre special EAs as root
5045 Details : Client side always return success for setxattr call for lustre
5046 special xattr (currently only "trusted.lov").
5049 Frequency : rarely on clusters with both ia64+i386 clients
5051 Description: ia64+i686 clients doing shared IO on the same file may LBUG
5052 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
5053 clients are doing concurrent writes to the same file it is
5054 possible that the ia64 clients may LASSERT because the OST
5055 extent locks are not PAGE_SIZE aligned. Ensure that grown
5056 locks are always aligned on the request boundary.
5059 Frequency : specific use, occasional
5061 Description: Overwriting in use executable truncates on-disk binary image
5062 Details : If one node attempts to overwrite an executable in use by
5063 another node, we now correctly return ETXTBSY instead of
5064 truncating the file.
5066 Severity : enhancement
5068 Description: Async OSC create to avoid the blocking unnecessarily.
5069 Details : If a OST has no remain object, system will block on the creating
5070 when need to create a new object on this OST. Now, ways use
5071 pre-created objects when available, instead of blocking on an
5072 empty osc while others are not empty. If we must block, we block
5073 for the shortest possible period of time.
5078 Description: chmod on Lustre root is propagated to other clients
5079 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
5080 invalid by the follow_mount time.
5083 Frequency : liblustre clients only
5085 Description: Race in 'instant cancel' lock handling could lead to such locks
5086 never to be granted in case of SMP MDS
5087 Details : Do not destroy not yet granted but cbpending locks in
5091 Frequency : replay/resend of open
5093 Description: non null lock assetion failure in mds_intent_policy
5094 Details : Trying to replay/resend lockless open requests resulted in
5095 mds_open() returning 0 with no lock. Now it sets a flag if
5096 a lock is going to be returned.
5098 Severity : enhancement
5100 Description: Checksum enhancements
5101 Details : New checksum enhancements allow for resending RPCs that failed
5104 Severity : enhancement
5106 Description: Tunables on number of dirty pages in cacche
5107 Details : Allow to set limit on number of dirty pages cached.
5112 Description: client crash on unmount - lock still has references
5113 Details : In some error handling cases it was possible to leak a lock
5114 reference on a client while accessing a file. This was not
5115 harmful to the client during operation, but would cause the
5116 client to crash when the filesystem is unmounted.
5119 Frequency : specific case, rare
5121 Description: ETXTBSY on mds though file not in use by client
5122 Details : ETXTBSY is no longer incorrectly returned when attempting to
5123 chmod or chown a directory that the user previously tried to
5124 execute or a currently-executing binary.
5127 Frequency : extremely rare except on liblustre-based clients
5129 Description: Lustre space not freed when files are deleted
5130 Details : Clean up open-unlinked files after client eviction. Previously
5131 the unlink was skipped and the files remained as orphans.
5136 Description: OST failure "would be an LBUG" in waiting_locks_callback()
5137 Details : In some cases it was possible to send a blocking callback to
5138 a client doing a glimpse, even though that client didn't get
5139 a lock granted. When the glimpse lock is cancelled on the OST
5140 the freed lock is left on the waiting list and corrupted the list.
5143 Frequency : all core dumps
5145 Description: Broke core dumps to lustre
5146 Details : Negative dentry may be unhashed if parent does not have UPDATE
5147 lock, but some callers, e.g. do_coredump, expect dentry to be
5148 hashed after successful create, hash it in ll_create_it.
5150 ------------------------------------------------------------------------------
5152 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
5154 * Support for kernels:
5155 2.6.9-42.0.2.EL (RHEL 4)
5156 2.6.5-7.276 (SLES 9)
5157 2.4.21-40.EL (RHEL 3)
5158 2.6.12.6 vanilla (kernel.org)
5162 Frequency : always on RHEL 3
5164 Description: Number of open files grows over time
5165 Details : The number of open files grows over time, whether or not
5166 Lustre is started. This was due to a filp leak introduced
5167 by one of our kernel patches.
5169 ------------------------------------------------------------------------------
5171 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
5173 * Support for kernels:
5174 2.6.9-42.EL (RHEL 4)
5175 2.6.5-7.267 (SLES 9)
5176 2.4.21-40.EL (RHEL 3)
5177 2.6.12.6 vanilla (kernel.org)
5182 Bugzilla : 5719, 9635, 9792, 9684
5183 Description: OST (or MDS) trips assertions in (re)connection under heavy load
5184 Details : If a server is under heavy load and cannot reply to new
5185 connection requests before the client resends the (re)connect,
5186 the connection handling code can behave badly if two service
5187 threads are concurrently handing separate (re)connections from
5188 the same client. Add better locking to the connection handling
5189 code, and ensure that only a single connection will be processed
5190 for a given client UUID, even if the lock is dropped.
5192 Severity : enhancement
5194 Description: add TCP zero-copy support to kernel
5195 Details : Add support to the kernel TCP stack to allow zero-copy bulk
5196 sends if the hardware supports scatter-gather and checksumming.
5197 This allows socklnd to do client-write and server-read more
5198 efficiently and reduce CPU utilization from skbuf copying.
5201 Frequency : only if NFS exporting from client
5203 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
5204 Details : If setting attributes on a file created under NFS that had
5205 never been opened it would be possible to oops the client
5206 if the file had no objects.
5210 Bugzilla : 9326, 10402, 10897
5211 Description: client crash in ptlrpcd_wake() thread when sending async RPC
5212 Details : It is possible that ptlrpcd_wake() dereferences a freed async
5213 RPC. In rare cases the ptlrpcd thread alread processed the RPC
5214 before ptlrpcd_wake() was called and the request was freed.
5217 Frequency : always for liblustre
5219 Description: liblustre client does MDS+OSTs setattr RPC for each write
5220 Details : When doing a write from a liblustre client, the client
5221 incorrectly issued an RPC to the MDS and each OST the file was
5222 striped over in order to update the timestamps. When writing
5223 with small chunks and many clients this could overwhelm the MDS
5224 with RPCs. In all cases it would slow down the write because
5225 these RPCs are unnecessary.
5227 Severity : enhancement
5229 Description: allow number of MDS service threads to be changed at module load
5230 Details : It is now possible to change the number of MDS service threads
5231 running. Adding "options mds mds_num_threads={N}" to the MDS's
5232 /etc/modprobe.conf will set the number of threads for the next
5233 time Lustre is restarted (assuming the "mds" module is also
5234 reloaded at that time). The default number of threads will
5235 stay the same, 32 for most systems.
5240 Description: OST crash if filesystem is unformatted or corrupt
5241 Details : If an OST is started on a device that has never been formatted
5242 or if the filesystem is corrupt and cannot even mount then the
5243 error handling cleanup routines would dereference a NULL pointer.
5248 Description: NULL pointer deref in llap_from_page.
5249 Details : get_cache_page_nowait can return a page with NULL (or otherwise
5250 incorrect) mapping if the page was truncated/reclaimed while it was
5251 searched for. Check for this condition and skip such pages when
5252 doing readahead. Introduce extra check to llap_from_page() to
5253 verify page->mapping->host is non-NULL (so page is not anonymous).
5256 Frequency : Sometimes when using sys_sendfile
5258 Description: "page not covered by a lock" warnings from ll_readpage
5259 Details : sendfile called ll_readpage without right page locks present.
5260 Now we introduced ll_file_sendfile that does necessary locking
5261 around call to generic_file_sendfile() much like we do in
5265 Frequency : with certain MDS communication failures at client mount time
5267 Description: NULL pointer deref after failed client mount
5268 Details : a client connection request may delayed by the network layer
5269 and not be sent until after the PTLRPC layer has timed out the
5270 request. If the client fails the mount immediately it will try
5271 to clean up before the network times out the request. Add a
5272 reference from the request import to the obd device and delay
5273 the cleanup until the network drops the request.
5276 Frequency : occasionally during client (re)connect
5278 Description: assertion failure during client (re)connect
5279 Details : processing a client connection request may be delayed by the
5280 client or server longer than the client connect timeout. This
5281 causes the client to resend the connection request. If the
5282 original connection request is replied in this interval, the
5283 client may trip an assertion failure in ptlrpc_connect_interpret()
5284 which thought it would be the only running connect process.
5287 Frequency : only with obd_echo servers and clients that are rebooted
5289 Description: kernel BUG accessing uninitialized data structure
5290 Details : When running an obd_echo server it did not start the ping_evictor
5291 thread, and when a client was evicted an uninitialized data
5292 structure was accessed. Start the ping_evictor in the RPC
5293 service startup instead of the OBD startup.
5295 Severity : enhancement
5296 Bugzilla : 10193 (patchless)
5297 Description: Remove dependency on various unexported kernel interfaces.
5298 Details : No longer need reparent_to_init, exit_mm, exit_files,
5299 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
5302 Frequency : rare (only users of deprecated and unsupported LDAP config)
5304 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
5305 Details : LDAP apparently contains 'lustreName' attributes instead of
5306 'name'. A simple remapping of the name is sufficient.
5309 Frequency : rare (only with non-default dump_on_timeout debug enabled)
5311 Description: waiting_locks_callback trips kernel BUG if client is evicted
5312 Details : Running with the dump_on_timeout debug flag turned on makes
5313 it possible that the waiting_locks_callback() can try to dump
5314 the Lustre kernel debug logs from an interrupt handler. Defer
5315 this log dumping to the expired_lock_main() thread.
5317 Severity : enhancement
5319 Description: Support NFS exporting on 2.6 kernels.
5320 Details : Implement non-rawops metadata methods for NFS server to use without
5321 changing NFS server code.
5324 Frequency : very rare (synthetic metadata workload only)
5326 Description: two racing renames might cause an MDS thread to deadlock
5327 Details : Running the "racer" program may cause one MDS thread to rename
5328 a file from being the source of a rename to being the target of
5329 a rename at exactly the same time that another thread is doing
5330 so, and the second thread has already enqueued these locks after
5331 doing a lookup of the target and is trying to relock them in
5332 order. Ensure that we don't try to re-lock the same resource.
5335 Frequency : only very large systems with liblustre clients
5337 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
5338 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
5339 all OSTs in parallel. This allows the eviction of stale liblustre
5340 clients to proceed much faster than if they were done in series,
5341 and also offers similar improvements for other set_info RPCs.
5346 Description: excessive CPU usage during initial read phase on client
5347 Details : During the initial read phase on a client, it would agressively
5348 retry readahead on the file, consuming too much CPU and impacting
5349 performance (since 1.4.5.8). Improve the readahead algorithm
5350 to avoid this, and also improve some other common cases (read
5351 of small files in particular, where "small" is files smaller than
5352 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
5357 Description: MDS crash when receiving packet with unknown intent.
5358 Details : Do not LBUG in unknown intent case, just return -EFAULT
5360 Severity : enhancement
5361 Bugzilla : 9293, 9385
5362 Description: MDS RPCs are serialised on client. This is unnecessary for some.
5363 Details : Do not serialize getattr (non-intent version) and statfs.
5366 Frequency : occasional, when OST network is overloaded/intermittent
5368 Description: client evicted by OST after bulk IO timeout
5369 Details : If a client sends a bulk IO request (read or write) the OST
5370 may evict the client if it is unresposive to its data GET/PUT
5371 request. This is incorrect if the network is overloaded (takes
5372 too long to transfer the RPC data) or dropped the OST GET/PUT
5373 request. There is no need to evict the client at all, since
5374 the pinger and/or lock callbacks will handle this, and the
5375 client can restart the bulk request.
5378 Frequency : Always when mmapping file with no objects
5380 Description: client crashes when mmapping file with no objects
5381 Details : Check that we actually have objects in a file before doing any
5382 operations on objects in ll_vm_open, ll_vm_close and
5388 Description: Request leak when working with deleted CWD
5389 Details : Introduce advanced request refcount tracking for requests
5390 referenced from lustre intent.
5392 Severity : Enhancement
5394 Description: Cache open file handles on client.
5395 Details : MDS now will return special lock along with openhandle, if
5396 requested and client is allowed to hold openhandle, even if unused,
5397 until such a lock is revoked. Helps NFS a lot, since NFS is opening
5398 closing files for every read/write openration.
5400 Severity : Enhancement
5402 Description: Cache open negative dentries on client when possible.
5403 Details : Guard negative dentries with UPDATE lock on parent dir, drop
5404 negative dentries on lock revocation.
5409 Description: Remounting a client read-only wasn't possible with a zconf mount
5410 Details : It wasn't possible to remount a client read-only with llmount.
5412 Severity : enhancement
5413 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
5414 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
5415 MPICH is included to add Lustre-specific ADIO interfaces.
5416 This is based closely on the UFS ADIO layer and only differs
5417 in file creation, in order to allow the OST striping to be set.
5418 This is user-contributed code and not supported by CFS.
5423 Description: extended inode attributes (immutable, append-only) work improperly
5424 when 2.4 and 2.6 kernels are used on client/server or vice versa
5425 Details : Introduce kernel-independent values for these flags.
5427 Severity : enhancement
5430 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
5431 Details : Many of the /proc/ tunables can only be tuned at a megabyte
5432 granularity. Now, Fractional MB granularity is be supported,
5433 this is very useful for low memory system.
5435 Severity : enhancement
5437 Description: Getattr by fid
5438 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
5439 locks, avoids extra getattr rpc requests to MDS, allows '/' to
5440 have locks and avoids getattr rpc requests for it on every stat.
5443 Frequency : Always, for filesystems larger than 2TB
5445 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
5446 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
5447 filesystems larger than 2TB to be mounted with mballoc enabled.
5450 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
5452 Description: filesystem corruption for non-standard kernels and very large OSTs
5453 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
5454 filesystems larger than 2TB is mounted then the kernel will
5455 silently corrupt the start of the filesystem. CONFIG_LBD is
5456 enabled for all CFS-supported kernels, but the possibility of
5457 this happening with a modified kernel config exists.
5459 Severity : enhancement
5461 Description: add client O_DIRECT support for 2.6 kernels
5462 Details : It is now possible to do O_DIRECT reads and writes to files
5463 in the Lustre client mountpoint on 2.6 kernel clients.
5465 Severity : enhancement
5467 Description: parallel glimpse, setattr, statfs, punch, destroy requests
5468 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
5469 parallel, not waiting for response from every OST before sending
5470 a rpc to the next OST.
5475 Description: setattr vs write race when updating file timestamps
5476 Details : Client processes that update a file timestamp into the past
5477 right after writing to the file (e.g. tar) it is possible that
5478 the updated file modification time can be reset to the current
5479 time due to a race between processing the setattr and write RPC.
5481 Severity : enhancement
5483 Description: Bring 'lfs find' closer in line with regular Linux find.
5484 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
5485 -print0 options and obtains all the needed info through the lustre
5488 Severity : enhancement
5490 Description: support up to 1024 configured devices on one node
5491 Details : change obd_dev array from statically allocated to dynamically
5492 allocated structs as they are first used to reduce memory usage
5497 Description: Flush dirty partially truncated pages during truncate
5498 Details : Immediatelly flush partially truncated pages in filter_setattr,
5499 this way we completely avoid having any pages in page cache on OST
5500 and can retire ugly workarounds during writes to flush such pages.
5505 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
5506 Details : protect inode from truncation within vfs_unlink() context
5507 just take a reference before calling vfs_unlink() and release it
5508 when parent's i_sem is free.
5511 Frequency : always, if extents are used on OSTs
5513 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
5514 Details : OSTs using the extents format would not zero the high 16 bits of
5515 the index physical block number. This is not a problem for any
5516 OST filesystems smaller than 16TB, and no kernels support ext3
5517 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
5518 new/modified files) and can be fixed for existing filesystems
5519 with e2fsprogs-1.39-cfs1.
5524 Description: import connection selection may be incorrect if timer wraps
5525 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
5526 import connections to be ignored if the 32-bit jiffies counter
5527 wraps. Use a 64-bit jiffies counter.
5530 Frequency : during server recovery
5532 Description: crash after server is denying duplicate export
5533 Details : If clients are resending connect requests to the server, the
5534 server refuses to allow a client to connect multiple times.
5535 Fixed a bug in the handling of this case.
5538 Frequency : very large clusters immediately after boot
5540 Description: LNET request buffers exhausted under heavy short-term load
5541 Details : If a large number of client requests are generated on a service
5542 that has previously never seen so many requests it is possible
5543 that the request buffer growth cannot keep up with the spike in
5544 demand. Instead of dropping incoming requests, they are held in
5545 the LND until the RPC service can accept more requests.
5548 Frequency : Sometimes during replay
5550 Description: Assertion failure in ll_local_open after replay.
5551 Details : If replay happened on an open request reply before we were able
5552 to set replay handler, reply will become not swabbed tripping the
5553 assertion in ll_local_open. Now we set the handler right after
5554 recognising of open request
5557 Frequency : very rare
5559 Description: kernel reports "badness in vsnprintf"
5560 Details : Reading from the "recovery_status" /proc file in small chunks
5561 may cause a negative length in lprocfs_obd_rd_recovery_status()
5562 call to vsnprintf() (which is otherwise harmless). Exit early
5563 if there is no more space in the output buffer.
5565 Severity : enhancement
5567 Description: clear OBD RPC statistics by writing to them
5568 Details : It is now possible to clear the OBD RPC statistics by writing
5569 to the "stats" file.
5574 Description: Client mtime is not the same on different clients after utimes
5575 Details : In some cases, the client was using the utimes() syscall on
5576 a file cached on another node. The clients now validate the
5577 ctime from the MDS + OSTs to determine which one is right.
5582 Description: Inability to activate failout mode
5583 Details : lconf script incorrectly assumed that in python string's numeric
5584 value is used in comparisons.
5587 Frequency : always with multiple stripes per file
5589 Description: Inefficient object allocation for mutli-stripe files
5590 Details : When selecting which OSTs to stripe files over, for files with
5591 a stripe count that divides evenly into the number of OSTs,
5592 the MDS is always picking the same starting OST for each file.
5593 Return the OST selection heuristic to the original design.
5598 Description: mount failures may take full timeout to return an error
5599 Details : Under some heavy load conditions it is possible that a
5600 failed mount can wait for the full obd_timeout interval,
5601 possibly several minutes, before reporting an error.
5602 Instead return an error as soon as the status is known.
5604 ------------------------------------------------------------------------------
5606 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
5608 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
5609 INTEROPERATE with older versions automatically. Please read the
5610 user documentation before upgrading any part of a live system.
5611 * WARNING: Lustre networking configuration changes are required with
5612 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
5615 * Support for kernels:
5616 2.6.9-22.0.2.EL (RHEL 4)
5617 2.6.5-7.244 (SLES 9)
5618 2.6.12.6 vanilla (kernel.org)
5621 Severity : enhancement
5622 Bugzilla : 7981/8208
5623 Description: Introduced Lustre Networking (LNET)
5624 Details : LNET is new networking infrastructure for Lustre, it includes
5625 a reorganized network configuration mode (see the user
5626 documentation for full details) as well as support for routing
5627 between different network fabrics. Lustre Networking Devices
5628 (LNDs) for the supported network fabrics have also been
5629 created for this new infrastructure.
5631 Severity : enhancement
5632 Description: Introduced Access control lists
5633 Details : clients can set ACLs on files and directories in order to have
5634 more fine-grained permissions than the standard Unix UGO+RWX.
5635 The MDS must be started with the "-o acl" mount option.
5637 Severity : enhancement
5638 Description: Introduced filesystem quotas
5639 Details : Administrators may now establish per-user quotas on the
5642 Severity : enhancement
5644 Description: Configuration change for the XT3
5645 The PTLLND is now used to run Lustre over Portals on the XT3
5646 The configure option(s) --with-cray-portals are no longer used.
5647 Rather --with-portals=<path-to-portals-includes> is used to
5648 enable building on the XT3. In addition to enable XT3 specific
5649 features the option --enable-cray-xt3 must be used.
5654 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
5655 Details : A race between allocating a new llcd and re-getting the llcd_lock
5656 allowed another thread to grab newly-allocated llcd.
5658 Severity : enhancement
5660 Description: 2.6 OST async journal commit and locking fix to improve performance
5661 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
5662 2.6 kernels are now async as they already were in 2.4 kernels so
5663 that they can commit concurrently with the network bulk transfer.
5664 For block-allocated files the filter allocation semaphore is held
5665 to avoid filesystem fragmentation during allocation. BKL lock
5666 removed for 2.6 xattr operations where it is no longer needed.
5671 Description: lconf incorrectly determined whether two IP networks could talk
5672 Details : In some more complicated routing and multiple-network
5673 configurations, lconf will avoid trying to make a network
5674 connection to a disjoint part of the IP space. It was doing the
5675 math incorrectly for one set of cases.
5680 Description: Fix for potential infinite loop processing records in an llog.
5681 Details : If an llog record is corrupted/zeroed, it is possible to loop
5682 forever in llog_process(). Validate the llog record length
5683 and skip the remainder of the block on error.
5686 Frequency : occasional (liblustre only)
5688 Description: liblustre could not open files whose last component is a symlink
5689 Details : sysio_path_walk() would incorrectly pass the open intent to
5690 intermediate path components.
5693 Frequency : rare (liblustre only with non-standard tuning)
5694 Bugzilla : 7201 (7350)
5695 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
5696 Details : llu_lookup_finish_locks() tries to set lock data on a lock
5697 after it has been released, only do this for referenced locks
5699 Severity : enhancement
5701 Description: specifying an (invalid) directory default stripe_size of -1
5702 would reset the directory default striping
5703 Details : stripe_size -1 was used internally to signal directory stripe
5704 removal, now use "all default" to signal dir stripe removal
5705 as a directory striping of "all default" is not useful
5708 Frequency : common for large clusters running liblustre clients
5710 Description: doing an ls when liblustre clients are running is slow
5711 Details : sending a glimpse AST to a liblustre client waits for every AST
5712 to time out, as liblustre clients will not respond. Since they
5713 cannot cache data we refresh the OST lock LVB from disk instead.
5715 Severity : enhancement
5717 Description: doing an ls at the same time as file IO can be slow
5718 Details : enqueue and other "small" requests can be blocked behind many
5719 large IO requests. Create a new OST IO portal for non-IO
5720 requests so they can be processed faster.
5723 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
5725 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
5726 created file with mode 000
5727 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
5728 allow an ftruncate() as a non-root user to a file with mode 000.
5729 Lustre now respects this flag to disable mode checks when
5730 truncating a file owned by the user
5733 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
5736 Description: Revoking locks from clients that went dead or catatonic might take
5738 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
5739 cancellation of such locks instant on servers without waiting for
5740 any reply from clients. Clients drops these locks when cancel
5741 notification from server is received without replying.
5744 Frequency : liblustre-only, when liblustre client dies or becomes busy
5746 Description: Doing ls on Linux clients can take a long time with active
5748 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
5749 granting such locks to it in the first place if possible. Locks
5750 are taken by proxy on the OST during the read or write and
5751 dropped immediately afterward. Add connect flags handling, do
5752 not grant locks to liblustre clients for glimpse ASTs.
5754 Severity : enhancement
5756 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
5757 Details : Existing read-ahead algorithm is tuned for the case of streamlined
5758 sequential reads and behaves badly with applications doing random
5759 reads. Improve it by reading ahead at least read region, and
5760 avoiding excessive large RPC for small reads.
5762 Severity : enhancement
5764 Description: Creating more than 1000 files for a single job may cause a load
5765 imbalance on the OSTs if there are also a large number of OSTs.
5766 Details : qos_prep_create() uses an OST index reseed value that is an
5767 even multiple of the number of available OSTs so that if the
5768 reseed happens in the middle of the object allocation it will
5769 still utilize the OSTs as uniformly as possible.
5774 Description: OST or MDS may oops in ping_evictor_main()
5775 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
5776 but doesn't restart at beginning of obd_exports_timed list
5779 Severity : enhancement
5781 Description: improve by-nid export eviction on the MDS and OST
5782 Details : allow multiple exports with the same NID to be evicted at one
5783 time without re-searching the exports list.
5786 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
5788 Description: MDS may oops in groups_free()
5789 Details : in rare race conditions a newly allocated group_info struct is
5790 freed again, and this can be NULL. The 2.4 compatibility code
5791 for groups_free() checked for a NULL pointer, but 2.6 did not.
5794 Frequency : common for liblustre clients doing little filesystem IO
5795 Bugzilla : 9352, 7313
5796 Description: server may evict liblustre clients accessing contended locks
5797 Details : if a client is granted a lock or receives a completion AST
5798 with a blocking AST already set it would not reply to the AST
5799 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
5802 Frequency : lfs setstripe, only systems with more than 160 OSTs
5804 Description: unable to set striping with a starting offset beyond OST 160
5805 Details : llapi_create_file() incorrectly limited the starting stripe
5806 index to the maximum single-file stripe count.
5809 Frequency : LDAP users only
5811 Description: lconf did not handle in-kernel recovery with LDAP properly
5812 Details : lconf/LustreDB get_refs() is searching the wrong namespace
5814 Severity : enhancement
5816 Description: bind OST threads to NUMA nodes to improve performance
5817 Details : all OST threads are uniformly bound to CPUs on a single NUMA
5818 node and do their allocations there to localize memory access
5820 Severity : enhancement
5822 Description: llmount can determine client NID directly from Myrinet (GM)
5823 Details : the client NID code from gmnalnid was moved directly into
5824 llmount, removing the need to use this or specifying the
5825 client NID explicitly when mounting GM clients with zeroconf
5828 Frequency : if client is started with down MDS
5830 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
5831 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
5832 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
5833 ptlrpc_import_delayed_req() because we didn't check if the
5834 request was interrupted, and we also didn't break out of the
5835 event loop if there was no timeout
5840 Description: data loss during non-page-aligned writes to a single file from
5841 both multiple nodes and multiple threads on one node at same time
5842 Details : updates to KMS and lsm weren't protected by common lock. Resulting
5843 inconsistency led to false short-reads, that were cached and later
5844 used by ->prepare_write() to fill in partially written page,
5845 leading to data loss.
5848 Frequency : always, if lconf --abort_recovery used
5850 Description: lconf --abort_recovery fails with 'Operation not supported'
5851 Details : lconf was attempting to abort recovery on the MDT device and not
5854 Severity : enhancement
5856 Description: remove cleanup logs
5857 Details : replace lconf-generated cleanup logs with lustre internal
5858 cleanup routines. Eliminates the need for client-cleanup and
5861 Severity : enhancement
5863 Description: add support for EAs (user and system) on lustre filesystems
5864 Details : it is now possible to store extended attributes in the Lustre
5865 client filesystem, and with the user_xattr mount option it
5866 is possible to allow users to store EAs on their files also
5868 Severity : enhancement
5870 Description: Add possibility (config option) to show minimal available OST free
5872 Details : When compiled with --enable-mindf configure option, statfs(2)
5873 (and so, df) will return least minimal free space available from
5874 all OSTs as amount of free space on FS, instead of summary of
5875 free spaces of all OSTs.
5877 Severity : enhancement
5879 Description: do not expand extent locks acquired on OST-side
5880 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
5881 by server: they are not cached anyway.
5884 Frequency : when mmap is used/binaries executed from Lustre
5886 Description: Unmmap pages before throwing them away from read cache.
5887 Details : llap_shrink cache now attempts to unmap pages before discarding
5888 them (if unmapping failed - do not discard). SLES9 kernel has
5889 extra checks that trigger if this unmapping is not done first.
5894 Description: lconf didn't resolve symlinks before checking to see whether a
5895 given mountpoint was already in use
5898 Frequency : when migrating failover services
5899 Bugzilla : 6395, 9514
5900 Description: When migrating a subset of services from a node (e.g. failback
5901 from a failover service node) the remaining services would
5902 time out and evict clients.
5903 Details : lconf --force (implied by --failover) sets the global obd_timeout
5904 to 5 seconds in order to quickly disconnect, but this caused
5905 other RPCs to time out too quickly. Do not change the global
5906 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
5908 Severity : enhancement
5909 Frequency : if MDS is started with down OST
5910 Bugzilla : 9439,5706
5911 Description: Allow startup/shutdown of an MDS without depending on the
5912 availability of the OSTs.
5913 Details : Asynchronously call mds_lov_synchronize during MDS startup.
5914 Add appropriate locking and lov-osc refcounts for safe
5915 cleaning. Add osc abort_inflight calls in case the
5916 synchronize never started.
5919 Frequency : occasional (Cray XT3 only)
5921 Description: root not authorized to access files in CRAY_PORTALS environment
5922 Details : The client process capabilities were not honoured on the MDS in
5923 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
5924 been accessed by an authorized user then root was able to access
5925 the file on the local client also. The root user capabilities
5926 are now allowed on the MDS, as this environment has secure UID.
5929 Frequency : occasional
5931 Description: ldiskfs "too long searching" message happens too often
5932 Details : A debugging message (otherwise harmless) prints too often on
5933 the OST console. This has been reduced to only happen when
5934 there are fragmentation problems on the filesystem.
5939 Description: Division by zero in statfs when all OSCs are inactive
5940 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
5941 lov_statfs divides by value returned by lov_get_stripecnt().
5945 Bugzilla : 9489, 3273
5946 Description: First write from each client to each OST was only 4kB in size,
5947 to initialize client writeback cache, which caused sub-optimal
5948 RPCs and poor layout on disk for the first writen file.
5949 Details : Clients now request an initial cache grant at (re)connect time
5950 and so that they can start streaming writes to the cache right
5951 away and always do full-sized RPCs if there is enough data.
5952 If the OST is rebooted the client also re-establishes its grant
5953 so that client cached writes will be honoured under the grant.
5958 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
5959 Details : Now I/O RPCs go to different portal number and (presumably) fast
5960 lock requests (and glimses) and other RPCs get their own service
5961 threads pool that should be able to service those RPCs
5964 Severity : enhancement
5966 Description: Ability to exchange lustre version between client and servers and
5967 issue warnings at client side if client is too old. Also for
5968 liblustre clients there is ability to refuse connection of too old
5970 Details : New 'version' field is added to connect data structure that is
5971 filled with version info. That info is later checked by server and
5975 Frequency : rare, liblustre only.
5976 Bugzilla : 9296, 9581
5977 Description: Two simultaneous writes from liblustre at offset within same page
5978 might proceed at the same time overwriting eachother with stale
5980 Details : I/O lock withing llu_file_prwv was released too early, before data
5981 actually was hitting the wire. Extended lock-holding time until
5982 server acknowledges receiving data.
5985 Frequency : extremely rare. Never observed in practice.
5987 Description: avoid generating lustre_handle cookie of 0.
5988 Details : class_handle_hash() generates handle cookies by incrementing
5989 global counter, and can hit 0 occasionaly (this is unlikely, but
5990 not impossible, because initial value of cookie counter is
5991 selected randonly). Value of 0 is used as a sentinel meaning
5992 "unassigned handle" --- avoid it. Also coalesce two critical
5993 sections in this function into one.
5995 Severity : enhancement
5997 Description: allow liblustre clients to delegate truncate locking to OST
5998 Details : To avoid overhead of locking, liblustre client instructs OST to
5999 take extent lock in ost_punch() on client's behalf. New connection
6000 flag is added to handle backward compatibility.
6002 Severity : enhancement
6003 Bugzilla : 4928, 7341, 9758
6004 Description: allow number of OST service threads to be specified
6005 Details : a module parameter allows the number of OST service threads
6006 to be specified via "options ost ost_num_threads={N}" in the
6007 OSS's /etc/modules.conf or /etc/modprobe.conf.
6011 Bugzilla : 6146, 9635, 9895
6012 Description: servers crash with bad pointer in target_handle_connect()
6013 Details : In rare cases when a client is reconnecting it was possible that
6014 the connection request was the last reference for that export.
6015 We would temporarily drop the export reference and get a new
6016 one, but this may have been the last reference and the export
6017 was just destroyed. Get new reference before dropping old one.
6019 Severity : enhancement
6020 Frequency : if client is started with failover MDS
6022 Description: Allow multiple MDS hostnames in the mount command
6023 Details : Try to read the configuration from all specified MDS
6024 hostnames during a client mount in case the "primary"
6027 Severity : enhancement
6029 Description: Stop sending data to evicted clients as soon as possible.
6030 Details : Check if the client we are about to send or are sending data to
6031 was evicted already. (Check is done every second of waiting,
6032 for which l_wait_event interface was extended to allow checking
6033 of exit condition at specified intervals).
6036 Frequency : rare, normally only when NFS exporting is done from client
6038 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
6040 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
6043 Frequency : always, if srand() is called before liblustre initialization
6045 Description: Liblustre uses system PRNG disturbing its usage by user application
6046 Details : Introduce internal to lustre fast and high-quality PRNG for
6047 lustre usage and make liblustre and some other places in generic
6048 lustre code to use it.
6050 Severity : enhancement
6051 Bugzilla : 9477, 9557, 9870
6052 Description: Verify that the MDS configuration logs are updated when xml is
6053 Details : Check if the .xml configuration logs are newer than the config
6054 logs stored on the MDS and report an error if this is the case.
6055 Request --write-conf, or allow starting with --old_conf.
6057 Severity : enhancement
6059 Description: Handle symlinks in the path when checking if Lustre is mounted.
6060 Details : Resolve intermediate symlinks when checking if a client has
6061 mounted a filesystem to avoid duplicate client mounts.
6066 Description: lconf can hit an error exception but still return success.
6067 Details : The lconf command catches the Command error exception at the top
6068 level script context and will exit with the associated exit
6069 status, but doesn't ensure that this exit status is non-zero.
6074 Description: failure of ptlrpc thread startup can cause oops
6075 Details : Starting a ptlrpc service thread can fail if there are a large
6076 number of threads or the server memory is very fragmented.
6077 Handle this without oopsing.
6080 Frequency : always, only if liblustre and non-default acceptor port was used
6082 Description: liblustre cannot connect to servers with non-default acceptor port
6083 Details : tcpnal_set_default_params() was not called and was therefore
6084 ignoring the environment varaible TCPNAL_PORT, as well as other
6085 TCPNAL_ environment variables
6090 Description: two objects could be created on the same OST for a single file
6091 Details : If an OST is down, in some cases it was possible to create two
6092 objects on a single OST for a single file. No problems other
6093 than potential performance impact and spurious error messages.
6097 Bugzilla : 5681, 9562
6098 Description: Client may oops in ll_unhash_aliases
6099 Details : Client dcache may become inconsistent in race condition.
6100 In some cases "getcwd" can fail if the current directory is
6106 Description: Inode refcounting problems in NFS export code
6107 Details : link_raw functions used to call d_instantiate without obtaining
6108 extra inode reference first.
6112 Bugzilla : 9942, 9903
6113 Description: Referencing freed requests leading to crash, memleaks with NFS.
6114 Details : We used to require that call to ll_revalidate_it was always
6115 followed by ll_lookup_it. Also with revalidate_special() it is
6116 possible to call ll_revalidate_it() twice for the same dentry
6117 even if first occurence returned success. This fix changes semantic
6118 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
6119 reference on a request referred from the intent.
6120 ll_intent_release() then releases such a request.
6123 Frequency : rare, normally benchmark loads only
6125 Description: unlinked inodes were kept in memory on the client
6126 Details : If a client is repeatedly creating and unlinking files it
6127 can accumulate a lot of stale inodes in the inode slab cache.
6128 If there is no other client load running this can cause the
6129 client node to run out of memory. Instead flush old inodes
6130 from client cache that have the same inode number as a new inode.
6133 Frequency : SLES9 2.6.5 kernel and long filenames only
6134 Bugzilla : 9969, 10379
6135 Description: utime reports stale NFS file handle
6136 Details : SLES9 uses out-of-dentry names in some cases, which confused
6137 the lustre dentry revalidation. Change it to always use the
6141 Frequency : rare, unless heavy write-truncate concurrency is continuous
6142 Bugzilla : 4180, 6984, 7171, 9963, 9331
6143 Description: OST becomes very slow and/or deadlocked during object unlink
6144 Details : filter_destroy() was holding onto the parent directory lock
6145 while truncating+unlinking objects. For very large objects this
6146 may block other threads for a long time and slow overall OST
6147 responsiveness. It may also be possible to get a lock ordering
6148 deadlock in this case, or run out of journal credits because of
6149 the combined truncate+unlink. Solution is to do object truncate
6150 first in one transaction without parent lock, and then do the
6151 final unlink in a new transaction with the parent lock. This
6152 reduces the lock hold time dramatically.
6155 Frequency : rare, 2.4 kernels only
6157 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
6158 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
6159 before unmounting their filesystem in order to avoid deadlock.
6160 The kernel_locked() function in 2.4 kernels only checks whether
6161 the kernel lock is held, not whether it is this process that is
6162 holding it as 2.6 kernels do.
6167 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
6168 Details : The client ptlrpc code may be trying to reconnect to a down
6169 server before a previous connection attempt has timed out.
6170 Increase the reconnect interval to be longer than the connection
6171 timeout interval to avoid sending duplicate connections to
6175 Frequency : echo_client brw_test command
6177 Description: fix echo_client to work with OST preallocated code
6178 Details : OST preallocation code (5137) didn't take echo_client IO path
6179 into account: echo_client calls filter methods outside of any
6180 OST thread and, hence, there is no per-thread preallocated
6181 pages and buffers to use. Solution: hijack pga pages for IO. As
6182 a byproduct, this avoids unnecessary data copying.
6186 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
6187 Description: Client can oops in mdc_commit_close() after open replay
6188 Details : It was possible for the MDS to return an open request with no
6189 transaction number in mds_finish_transno() if the client was
6190 evicted, but without actually returning an error. Clients
6191 would later try to replay that open and may trip an assertion
6192 Simplify the client close codepath, and always return an error
6193 from the MDS in case the open is not successful.
6196 Frequency : rare, 2.6 OSTs only
6198 Description: OST may deadlock under high load on fragmented files
6199 Details : If there was a heavy load and highly-fragmented OST filesystems
6200 it was possible to have all the OST threads deadlock waiting on
6201 allocation of biovecs, because the biovecs were not released
6202 until the entire RPC IO was completed. Instead, release biovecs
6203 as soon as they are complete to ensure forward IO progress.
6205 Severity : enhancement
6207 Description: Support for specifying external journal device at mount
6208 Details : If an OST or MDS device is formatted with an external journal
6209 device, this device major/minor is stored in the ext3 superblock
6210 and may not be valid for failover. Allow detecting and
6211 specifying the external journal at mount time.
6216 Description: Mounting an MDS with pending unlinked files may cause oops
6217 Details : target_finish_recovery() calls mds_postrecov() which returned
6218 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
6219 considers this an error and immediately begins cleaning up the
6220 lov, just after starting the mds_lov process
6222 Severity : enhancement
6224 Description: Implement 'lfs df' to report actual free space on per-OST basis
6225 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
6226 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
6227 usage of objects; '-h' to report in human readable format.
6229 ------------------------------------------------------------------------------
6231 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
6238 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
6239 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
6240 at startup during orphan unlinking. Instead, skip buddy update
6241 before setup, as it will be regenerated after recovery is complete.
6246 Description: If an OST is inactive, its locks might reference stale inodes.
6247 Details : lov_change_cbdata() must iterate over all namespaces, even if
6248 they are inactive to clear inode references from the lock.
6250 Severity : enhancement
6251 Frequency : occasional, if non-standard max_dirty_mb used
6253 Description: Client will block write RPCs if not enough grant
6254 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
6255 then the client will block writes while waiting for another RPC
6256 to complete instead of consuming its dirty limit. With change
6257 we get improved performance when max_dirty_mb is small.
6259 Severity : enhancement
6260 Bugzilla : 3389, 6253
6261 Description: Add support for supplementary groups on the MDS.
6262 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
6263 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
6264 lookups for user supplementary groups into a cache.
6268 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
6269 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
6270 for liblustre O_EXCL is enforced by MDS.
6274 Description: readdir never returns NULL in liblustre.
6275 Details : Corrected llu_iop_getdirentries logic, to return offset of next
6276 dentry in struct dirent.
6280 Frequency : liblustre only, depends on application IO pattern
6281 Description: liblustre clients evicted if not contacting servers
6282 Details : Don't put liblustre clients into the ping_evictor list, so
6283 they will not be evicted by the pinger ever.
6285 Severity : enhancement
6287 Description: Add ability to evict clients by NID from MDS.
6288 Details : By echoing "nid:$NID" string into
6289 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
6290 $NID would be instantly evicted from this MDS and from all active
6291 OSTs connected to it.
6295 Description: Do not query file size twice, somewhat slowing stat(2) calls.
6296 Details : lookup_it_finish() used to query file size from OSTs that was not
6301 Description: service threads change working directory to that of init
6302 Details : Starting lustre service threads may pin the working directory
6303 of the parent thread, making that filesystem busy. Threads
6304 now change to the working directory of init to avoid this.
6308 Frequency : during shutdown only
6309 Description: shutdown with a failed MDS or OST can cause unmount to hang
6310 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
6315 Frequency : 2.6 only
6316 Description: chmod/chown may include an extra supplementary group
6317 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
6318 supplementary group and if none is specified this is used.
6321 Bugzilla : 5479 (6816)
6323 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
6324 Details : If lookup is in progress on a file that is unlinked we might try
6325 to revalidate the inode and fail in revalidate after lookup is
6326 complete and ll_file_open() enqueues the open again but
6327 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
6330 Frequency : always, if lconf --abort_recovery used
6332 Description: lconf --abort_recovery fails with 'Operation not supported'
6333 Details : lconf was attempting to abort recovery on the MDT device and not
6336 ------------------------------------------------------------------------------
6338 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
6343 Frequency : rare (only unsupported configurations with a node running as an
6345 Bugzilla : 6514, 5137
6346 Description: Mounting a Lustre file system on a node running as an OST could
6348 Details : OSTs now preallocates memory needed to write out data at
6349 startup, instead of when needed, to avoid having to
6350 allocate memory in possibly low memory situations.
6351 Specifically, if the file system is mounted on on OST,
6352 memory pressure could force it to try to write out data,
6353 which it needed to allocate memory to do. Due to the low
6354 memory, it would be unable to do so and the node would
6355 become unresponsive.
6357 Severity : enhancement
6359 Description: Addition of lconf --service command line option
6360 Details : lconf now accepts a '--service <arg>' option, which is
6361 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
6363 Severity : enhancement
6365 Description: Failover mode is now the default for OSTs.
6366 Details : By default, OSTs will now run in failover mode. To return to
6367 the old behaviour, add '--failout' to the lmc line for OSTs.
6369 Severity : enhancement
6371 Description: Health checks are now provided for MDS and OSTs
6372 Details : Additional detailed health check information on MSD and OSTs
6373 is now provided through the procfs health_check value.
6376 Frequency : occasional, depends on IO load
6378 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
6379 after numerous create/delete cycles
6380 Details : The ext3 inode allocation policy would not allocate new inodes
6381 very well on the OSTs because there are no new directories
6382 being created. Instead we look for groups with free space if
6383 the parent directories are nearly full.
6388 Description: Network or server problems during mount may cause partially
6389 mounted clients instead of returning an error.
6390 Details : The config llog parsing code may overwrite the error return
6391 code during mount error handling, returning success instead
6397 Description: MDS can fail to allocate large reply buffers
6398 Details : After long uptimes the MDS can fail to allocate large reply
6399 buffers (e.g. zconf client mount config records) due to memory
6400 fragmentation or consumption by the buffer cache. Preallocate
6401 some large reply buffers so that these replies can be sent even
6402 under memory pressure.
6406 Frequency : rare (liblustre)
6407 Description: fsx running with liblustre complained that using truncate() to
6408 extend the file doesn't work. This patch corrects that issue.
6409 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
6410 ATTR_SIZE and lsm use in llu_setattr_raw.
6414 Frequency : rare, only 2.6 kernels
6415 Description: Unusual file access patterns on the MDS may result in inode
6416 data being lost in very rare circumstances.
6417 Details : Bad interaction between the ea-in-inode patch and the "no-read"
6418 code in the 2.6 kernel caused the inode and/or EA data not to
6419 be read from disk, causing single-file corruption.
6423 Frequency : rare, only 2.6 filesystems using extents
6424 Description: Heavy concurrent write and delete load may cause data corruption.
6425 Details : It was possible under high-load situations to have an extent
6426 metadata block in the block device cache from a just-unlinked
6427 file overwrite a newly-allocated data block. We now unmap any
6428 metadata buffers that alias just-allocated data blocks.
6432 Frequency : filesystems with default stripe_count larger than 77
6433 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
6434 Details : lconf specifies an inode size of 4096 bytes when the default
6435 stripe_count is larger than 77. This conflicts with the default
6436 inode density of 1 per 4096 bytes. Allocate smaller inodes in
6437 this case to avoid pinning too much memory for large EAs.
6439 ------------------------------------------------------------------------------
6441 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
6446 Frequency : rare (extremely heavy IO load with hundreds of clients)
6448 Description: Client is evicted, gets IO error writing to file
6449 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
6450 caused clients to be evicted for AST timeouts. The fixes in
6451 bug 5192 mean we no longer need to have such short AST timeouts
6452 so ldlm_timeout has been increased.
6455 Frequency : occasional during --force or --failover shutdown under load
6456 Bugzilla : 5949, 4834
6457 Description: Server oops/LBUG if stopped with --force or --failover under load
6458 Details : a collection of import/export refcount and cleanup ordering
6459 issues fixed for safer force cleanup
6462 Frequency : only filesystems larger than 120 OSTs
6463 Bugzilla : 5990, 6223
6464 Description: lfs getstripe would oops on a very large filesystem
6465 Details : lov_getconfig used kfree on vmalloc'd memory
6468 Frequency : only filesystems exporting via NFS to Solaris 10 clients
6469 Bugzilla : 6242, 6243
6470 Description: reading from files that had been truncated to a non-zero size
6471 but never opened returned no data
6472 Details : ll_file_read() reads zeros from no-object files to EOF
6477 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
6478 Details : There's little harm in aborting MDS/OSS recovery and letting it
6479 try again, so I removed the LASSERT and return an error instead.
6481 Severity : enhancement
6483 Description: New debugging infrastructure for tracking down data corruption
6484 Details : The I/O checksum code was replaced to: (a) control it at runtime,
6485 (b) cover more of the client-side code path, and (c) try to narrow
6486 down where problems occurred
6490 Bugzilla : 3819, 4364, 4397, 6313
6491 Description: Racing close and eviction MDS could cause assertion in mds_close
6492 Details : It was possible to get multiple mfd references during close and
6493 client eviction, leading to one thread referencing a freed mfd.
6495 Severity: : enhancement
6496 Bugzilla : 3262, 6359
6497 Description: Attempts to reconnect to servers are now more aggressive.
6498 Details : This builds on the enhanced upcall-less recovery that was added
6499 in 1.4.2. When trying to reconnect to servers, clients will
6500 now try each server in the failover group every 10 seconds. By
6501 default, clients would previously try one server every 25 seconds.
6506 Description: After recovery, certain operations trigger a failed
6507 assertion on a client.
6508 Details : Failing over an mds, using lconf -d --failover, while a
6509 client was doing a readdir() call would cause the client to
6510 LBUG after recovery completed and the readdir() was resent.
6512 Severity : enhancement
6514 Description: Default groups are now added by lconf
6515 Details : You can now run lconf --group <servicename> without having to
6516 manually add groups with lmc.
6519 Frequency : occasional
6521 Description: Nodes with an elan id of 0 trigger a failed assertion
6524 Frequency : always when accessing e.g. tty/console device nodes
6526 Description: tty and some other devices nodes cannot be used on lustre
6527 Details : file's private_data field is used by device data and lustre
6528 values in there got lost. New field was added to struct file to
6529 store fs-specific private data.
6532 Frequency : when exporting Lustre via NFS
6534 Description: NFSD failed occasionally when looking up a path component
6535 Details : NFSD is looking up ".." which was broken in ext3 directories
6536 that had grown large enough to become hashed.
6539 Frequency : Clusters with multiple interfaces not on the same subnet
6541 Description: Nodes will repeatedly try to reconnect to an interface which it
6542 cannot reach and report an error to the log.
6543 Details : Extra peer list entries will be created by lconf with some peers
6544 unreachable. lconf now validates the peer before adding it.
6547 Frequency : Only if a default stripe is set on the filesystem root.
6549 Description: Setting a default stripe on the filesystem root prevented the
6550 filesystem from being remounted.
6551 Details : The client was sending extra request flags in the root getattr
6552 request and did not allocate a reply buffer for the dir EA.
6555 Frequency : occasional, higher if lots of files are accessed by one client
6556 Bugzilla : 6159, 6097
6557 Description: Client trips assertion regarding lsm mismatch/magic
6558 Details : While revalidating inodes the VFS looks up inodes with ifind()
6559 and in rare cases can find an inode that is being freed.
6560 The ll_test_inode() code will free the lsm during ifind()
6561 when it finds an existing inode and then the VFS later attaches
6562 this free lsm to a new inode.
6566 Bugzilla : 6422, 7030
6567 Description: MDS deadlock between mkdir and client eviction
6568 Details : Creating a new file via mkdir or mknod (starting a transaction
6569 and getting the ns lock) can deadlock with client eviction
6570 (gets ns lock and trying to finish a synchronous transaction).
6573 Frequency : occasional
6574 Description: While starting a server, the fsfilt_ext3 module could not be
6576 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
6577 kernels. Previously, lconf would still use the ext3 name
6578 when trying to load modules. Now, it will correctly use
6579 ext3 on 2.4 and ldiskfs on 2.6.
6581 Severity : enhancement
6582 Description: The default stripe count has been changed to 1
6583 Details : The interpretation of the default stripe count (0, to lfs
6584 or lmc) has been changed to mean striping across a single
6585 OST, rather than all available. For general usage we have
6586 found a stripe count of 1 or 2 works best.
6588 Severity : enhancement
6589 Description: Add support for compiling against Cray portals.
6590 Details : Conditional compiling for some areas that are different
6594 Frequency : occasional
6595 Bugzilla : 6409, 6834
6596 Description: Creating files with an explicit stripe count may lead to
6597 a failed assertion on the MDS
6598 Details : If some OSTs are full or unavailable, creating files may
6599 trigger a failed assertion on the MDS. Now, Lustre will
6600 try to use other servers or return an error to the
6604 Frequency : occasional
6606 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
6607 caused lock timeout during readahead (since 1.4.2).
6608 Details : Processes doing readahead might match a lock that hasn't been
6609 granted yet if there are overlapping and conflicting lock
6610 requests. The readahead process waits on ungranted lock
6611 (original lock is CBPENDING), while OST waits for that process
6612 to cancel CBPENDING read lock and eventually evicts client.
6614 Severity : enhancement
6616 Description: Initial enabling of flock support for clients
6617 Details : Implements fcntl advisory locking and file status functions.
6618 This feature is provided as an optional mount flag (default
6619 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
6620 locking are implemented yet, and those that are are not guaranteed
6621 to be completely correct in production environments.
6622 mount -t lustre -o [flock|noflock] ...
6625 Frequency : occasional
6627 Description: OSTs running 2.4 kernels but with extents enabled might trip an
6628 assertion in the ext3 JBD (journaling) layer.
6629 Details : The b_committed_data struct is protected by the big kernel lock
6630 in 2.4 kernels, serializing journal_commit_transaction() and
6631 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
6632 access to this struct. In 2.6 kernels there is finer grained
6633 locking to improve SMP performance of the JBD layer.
6637 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
6639 -----------------------------------------------------------------------------
6641 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
6643 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
6644 versions of Lustre. Please update all servers and clients to
6645 version 1.4.2 or later at the same time. You must also run
6646 "lconf --write-conf {config}.xml" on the MDS while it is stopped
6647 to update the configuration logs.
6649 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
6650 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
6651 - swab llog records of type '0' so we get proper header size/idx (5861)
6652 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
6653 - fix rename of one directory over another leaking an inode (5953)
6654 - avoid SetPageDirty on 2.6 (5981)
6655 - don't re-add just-being-destroyed locks to the waiting list (5653)
6656 - when creating new directories, inherit the parent's custom
6657 striping settings if present parent (3048)
6658 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
6659 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
6660 - don't hold client locks on temporary worklist from l_lru (5666)
6661 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
6662 - automatically evict dead clients (5921)
6663 - Update file size properly in create+truncate+fstat case (6196)
6664 - Do not unhash mountpoint dentries, do not allow removal of
6666 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
6667 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
6668 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
6669 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
6670 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
6671 - Correctly handle reads of files with no objects (6243)
6672 - lctl recover will also mark a device active if deactivate used (5933)
6674 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
6675 - allow --write-conf on an MDS with different nettype than client (5619)
6676 - don't write config llogs to MDS for mounts not from that MDS (5617)
6677 - lconf should create multiple TCP connections from a client (5201)
6678 - init scripts are now turned off by default; run chkconfig --on
6679 lustre and chkconfig --on lustrefs to use them
6680 - upcalls are no longer needed for clients to recover to failover
6682 - add --abort-recovery option to lconf to abort recovery on device
6684 - add support for an arbitrary number of OSTs (3026)
6685 - Quota support protocol changes.
6686 - forward compatibility changes to wire structs (6007)
6687 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
6688 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
6689 - improved "lustre status" script
6690 - initialize blocksize for non-regular files (6062)
6691 - added --disable-server and --disable-client configure options (5782)
6692 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
6693 - Vanilla 2.4.29 support
6694 - increase maximum number of obd devices to 520 (6242)
6695 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
6696 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
6697 SLES 9 2.6.5 kernels
6698 - limit stripes per file to 160 (the maximum EA size) (6093)
6700 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
6703 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
6704 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
6705 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
6706 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
6707 - small ext3 extents cleanups and fixes (5733)
6708 - improved mballoc code, several small races and bugs fixed (5733, 5638)
6709 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
6710 - avoid needless client->OST connect, fix handle mismatch (5317)
6711 - fix DLM error path that led to out-of-sync client, long delays (5779)
6712 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
6713 - fix several locking issues related to i_size (5492,5624,5654,5672)
6714 - don't move pending lock onto export if it is already evicted (5683)
6715 - fix kernel oops when creating .foo in unlinked directory (5548)
6716 - fix deadlock in obdfilter statistics vs. object create (5811)
6717 - use time_{before,after} to avoid timer jiffies wrap (5882)
6718 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
6719 - Do not leak request if server was not able to process it (5154)
6720 - If mds_open unable to find parent dir, make that negative lookup(5154)
6721 - don't create new directories with extent-mapping (5909, 5936)
6723 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
6724 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
6725 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
6726 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
6727 - support for RHEL4 kernel on i686 (5773)
6728 - provide error messages when incompatible logs are encountered (5898)
6730 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
6731 * version 1.4.0.10 (1.4.1 release candidate 1)
6733 - don't keep a lock reference when lock is not granted (4238)
6734 - unsafe list practices (rarely) led to infinite eviction loop (4908)
6735 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
6736 - drop import inflight refcount on signal_completed_replay error (5255)
6737 - unlock page after async write error during send (3677)
6738 - handle missing objects in filter_preprw_read properly (5265)
6739 - no transno return for symlink open, don't save no-trasno open (3440)
6740 - don't try to complete elan receive that already failed (4012)
6741 - free RPC server reply state on error (5406)
6742 - clean up thread from ptlrpc_start_thread() on error (5160)
6743 - readahead could read extra page into cache that wasn't ejected (5388)
6744 - prevent races in class_attach/setup/cleanup/detach (5260)
6745 - don't dereference de->d_inode after l_dput of de (5458)
6746 - use "int" for stripe value returned from lock_to_stripe (5544)
6747 - mballoc allocation and error-checking fixes in 2.6 (5504)
6748 - block device patches to fix I/O request sizes in 2.6 (5482)
6749 - look up hostnames for IB nals (5602)
6750 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
6751 - don't start multiple acceptors for the same port (5277)
6752 - fix incorrect LASSERT in mds_getattr_name (5635)
6753 - export a proc file for general "ping" checking (5628)
6754 - fix "lfs check" to not block when the MDS is down (5628)
6756 - service request history (4965)
6757 - put {ll,lov,osc}_async_page structs in a single slab (4699)
6758 - create an "evict_client" /proc entry on OSTs, like the MDS has
6759 - fix mount usage message, return errors per mount(8) (5168)
6760 - change grep [] to grep "[]" in tests so they work in more UMLs
6761 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
6762 - remove ancient LOV_MAGIC_V0 EA support (5047)
6763 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
6764 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
6765 - allow readahead allocations to fail when low on memory (5383)
6766 - mmap locking landed again, after considerable improvement (2828)
6767 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
6769 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
6772 - send OST transaction number in read/write reply to free req (4966)
6773 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
6774 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
6775 - the watchdog thread now runs as interruptible (5246)
6776 - flock/lockf fixes (but it's still disabled, pending 5135)
6777 - don't use EXT3 constants in llite code (5094)
6778 - memory shortage at startup could cause assertion (5176)
6780 - reorganization of lov code
6781 - single portals codebase
6783 - add extents/mballoc support (5025)
6784 - direct I/O reads in the obdfilter (4048)
6785 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
6787 tbd Cluster File Systems, Inc. <info@clusterfs.com>
6790 - send OST transaction number in read/write reply to free req (4966)
6791 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
6792 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
6793 - the watchdog thread now runs as interruptible (5246)
6794 - handle missing objects in filter_preprw_read properly (5265)
6795 - unsafe list practices (rarely) led to infinite eviction loop (4908)
6796 - drop import inflight refcount on signal_completed_replay error (5255)
6797 - unlock page after async write error during send (3677)
6798 - return original error code on reconstructed replies (3761)
6799 - no transno return for symlink open, don't save no-trasno open (3440)
6801 - add pid to ldlm debugging output (4922)
6802 - bump the watchdog timeouts -- we can't handle 30sec yet
6803 - extra debugging for orphan dentry/inode bug (5259)
6805 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
6808 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
6809 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
6810 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
6811 - fix NULL dereference in /proc/sys/portals/routes (4827)
6812 - allow failed mdc_close() operations to be interrupted (4561)
6813 - stop precreate on OST before MDS would time out on it (4778)
6814 - don't send partial-page writes before EOF from client (4410)
6815 - discard client grant for sub-page writes on large-page clients (4520)
6816 - don't free dentries not owned by NFS code, check generation (4806)
6817 - fix lsm leak if mds_create_objects() fails (4801)
6818 - limit debug_daemon file size, always print CERROR messages (4789)
6819 - use transno after validating reply (3892)
6820 - process timed out requests if import state changes (3754)
6821 - update mtime on OST during writes, return in glimpse (4829)
6822 - add mkfsoptions to LDAP (4679)
6823 - use ->max_readahead method instead of zapping global ra (5039)
6824 - don't interrupt __l_wait_event() during strace
6826 - add software watchdogs to catch hung threads quickly (4941)
6827 - make lustrefs init script start after nfs is mounted
6828 - fix CWARN/ERROR duplication (4930)
6829 - return async write errors to application if possible (2248)
6830 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
6831 - print NAL number in %x format (4645)
6832 - update barely-supported suse-2.4.21-171 series (4842)
6833 - support for sles 9 %post scripts
6834 - support for building 2.6 kernel-source packages
6835 - support for sles km_* packages
6837 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
6840 - ignore -ENOENT errors in osc_destroy (3639)
6841 - notify osc create thread that OSC is being cleaned up (4600)
6842 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
6843 - reconstruct ost_handle() like mds_handle() (4657)
6844 - create a new thread to do import eviction to avoid deadlock (3969)
6845 - let lconf resolve symlinked-to devices (4629)
6846 - don't unlink "objects" from directory with default EA (4554)
6847 - hold socknal file ref over connect in case target is down (4394)
6848 - allow more than 32000 subdirectories in a single directory (3244)
6849 - fix blocks count for O_DIRECT writes (3751)
6850 - OST returns ENOSPC from object create when no space left (4539)
6851 - don't send truncate RPC if file size isn't changing (4410)
6852 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
6853 - bind to privileged port in socknal and tcpnal (3689)
6855 - rate limit CERROR/CWARN console message to avoid overload (4519)
6856 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
6857 - basic mmap support (3918)
6858 - kernel patch series update from b1_4 (4711)
6860 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
6863 - avoid crash during MDS cleanup with OST shut down (2775)
6864 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
6865 - don't use bad inodes on the MDS (3744)
6866 - dynamic object preallocation to improve recovery speed (4236)
6867 - don't hold spinlock over lock dumping or change debug flags (4401)
6868 - don't zero obd_dev when it is force cleaned (3651)
6869 - print grants to console if they go negative (4431)
6870 - "lctl deactivate" will stop automatic recovery attempts (3406)
6871 - look for existing locks in ldlm_handle_enqueue() (3764)
6872 - don't resolve lock handle twice in recovery avoiding race (4401)
6873 - revalidate should check working dir is a directory (4134)
6875 - don't always mark "slow" obdfilter messages as errors (4418)
6877 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
6880 - don't close LustreDB during write_conf until it is done (3860)
6881 - fix typo in lconf for_each_profile (3821)
6882 - allow dumping logs from multiple threads at one time (3820)
6883 - don't allow multiple threads in OSC recovery (3812)
6884 - fix debug_size parameters (3864)
6885 - fix mds_postrecov to initialize import for llog ctxt (3121)
6886 - replace config semaphore with spinlock (3306)
6887 - be sure to send a reply for a CANCEL rpc with bad export (3863)
6888 - don't allow enqueue to complete on a destroyed export (3822)
6889 - down write_lock before checking llog header bitmap (3825)
6890 - recover from lock replay timeout (3764)
6891 - up llog sem before sending rpc (3652)
6892 - reduce ns lock hold times when setting kms (3267)
6893 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
6894 - fix NULL deref and obd_dev leak on setup error (3312)
6895 - replace some LBUG about llog ops with error handling (3841)
6896 - don't match INVALID dentries from d_lookup and spin (3784)
6897 - hold dcache_lock while marking dentries INVALID and hashing (4255)
6898 - fix invalid assertion in ptlrpc_set_wait (3880)
6900 - add libwrap support for the TCP acceptor (3996)
6901 - add /proc/sys/portals/routes for non-root route listing (3994)
6902 - allow setting MDS UUID in .xml (2580)
6903 - print the stack of a process that LBUGs (4228)
6905 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
6908 - don't cleanup request in ll_file_open() on failed MDS open (3430)
6909 - make sure to unset replay flag from failed open requests (3440)
6910 - if default stripe count is 0, use OST count for inode size (3636)
6911 - update parent mtime/ctime on client for create/unlink (2611)
6912 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
6913 - free recovery state on server during a forced cleanup (3571)
6914 - unregister_reply for resent reqs (3063)
6915 - loop back devices mounting and status check on 2.6 (3563)
6916 - fix resource-creation race that can provoke i_size == 0 (3513)
6917 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
6918 - more debugging for page-accounting assertion (3746)
6919 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
6920 - avoid deadlock after precreation failure (3758)
6921 - fix race and lock order deadlock in orphan handling (3450, 3750)
6922 - add validity checks when grabbing inodes from l_ast_data (3599)
6924 - add /proc/.../recovery_status to obdfilter (3428)
6925 - lightweight CDEBUG infrastructure, debug daemon (3668)
6926 - change default OSC RPC parameters to be better on small clusters
6927 - turn off OST read cache for files smaller than 32MB
6928 - install man pages and include them in rpms (3100)
6929 - add new init script for (un)mounting lustre filesystems (2593)
6930 - run chkconfig in %post for init scripts (3701)
6931 - drop scimac NAL (unmaintained)
6933 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
6936 - clean kiobufs before and after use (3485)
6937 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
6938 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
6939 - change init script to fail more clearly if not run as root (1528)
6940 - allow clients to reconnect during replay (1742)
6941 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
6942 - don't do DNS lookups on NIDs too small for IP addresses (3442)
6943 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
6944 - fix cond_resched (3554)
6945 - only evict unfinished clients after recovery (3515)
6946 - allow bulk resend, prevent data loss (3570)
6947 - dynamic ptlrpc request buffer allocation (2102)
6948 - don't allow unlinking open directory if it isn't empty (2904)
6949 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
6950 - remove extraneous obd dereference causing LASSERT failure (3334)
6951 - don't use get_cycles() when creating temp. files on the mds (3156)
6952 - hold i_sem when setting i_size in ll_extent_lock() (3564)
6953 - handle EEXIST for set-stripe, set proper directory name (3336)
6955 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
6956 - fix ksocknal_fmb_callback() error messages (2918)
6958 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
6961 - don't copy lvb into (possibly NULL) reply on error (2983)
6962 - don't deref dentry after dput, don't free lvb on error (2922)
6963 - use the kms to determine writeback rpc length (2947)
6964 - increment oti_logcookies when osc is inactive (2948)
6965 - update client's i_blocks count via lvb messages (2543)
6966 - handle intent open/close of special files properly (1557)
6967 - mount MDS with errors=remount-ro, like obdfilter (2009)
6968 - initialize lock handle to avoid ASSERT on error cleanup (3057)
6969 - don't use cancelling-locks' kms values (2947)
6970 - use highest lock extent for kms, not last one (2925)
6971 - don't dereference ERR_PTR() dentry in error handling path (3107)
6972 - fix thread race in portals_debug_dumplog() (3122)
6973 - create lprocfs device entries at setup instead of at attach (1519)
6974 - common AST error handler, don't evict client on completion race (3145)
6975 - zero nameidata in detach_mnt in 2.6 (3118)
6976 - verify d_inode after revalidate_special is valid in 2.6 (3116)
6977 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
6978 - initialize RPC timeout timer earlier for 2.6 (3219)
6979 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
6980 - print nal/nid for unknown nid (3258)
6981 - additional checks for oscc recovery before doing precreate (3284)
6982 - fix ll_extent_lock() error return code for 64-bit systems (3043)
6983 - don't crash in mdc_close for bad permissions on open (3285)
6984 - zero i_rdev for non-device files (3147)
6985 - clear page->private before handing to FS, better assertion (3119)
6986 - tune the read pipeline (3236)
6987 - fix incorrect decref of invalidated dentry (2350)
6988 - provide read-ahead stats and refine rpc in flight stats (3328)
6989 - don't hold journal transaction open across create RPC (3313)
6990 - update atime on MDS at close time (3265)
6991 - close LDAP connection when recovering to avoid server load (3315)
6992 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
6993 - don't leak open file on MDS after open resend (3325)
6994 - serialize filter_precreate and filter_destroy_precreated (3329)
6995 - loop device shouldn't call sync_dev() for nul device (3092)
6996 - clear page cache after eviction (2766)
6997 - resynchronize MDS->OST in background (2824)
6998 - refuse to mount the same filesystem twice on same mountpoint (3394)
6999 - allow llmount to create routes for mounting behind routers (3320)
7000 - push lock cancellation to blocking thread for glimpse ASTs (3409)
7001 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
7002 - fix rare problem with rename on htree directories (3417)
7004 - allow default OST striping configuration per directory (1414)
7005 - fix compilation for qswnal for 2.6 kernels (3125)
7006 - increase maximum number of MDS request buffers for large systems
7007 - change liblustreapi to be useful for external progs like lfsck (3098)
7008 - increase local configuration timeout for slow disks (3353)
7009 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
7011 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
7014 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
7015 - don't overwrite extent policy data in reply if lock was blocked (2901)
7016 - drop filter export grants atomically with removal from device (2663)
7017 - del obd_self_export from work_list in class_disconnect_exports (2908)
7018 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
7019 - swab reply message in mdc_close, other PPC fixes (2464)
7020 - fix destroying of named logs (2325)
7021 - overwrite old logs when running lconf --write_conf (2264)
7022 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
7023 - fix race in target_handle_connect (2898)
7024 - mds_reint_create() should take same inode create lock (2926)
7025 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
7026 - don't close files for self_export to avoid uninitialized obd (2936)
7027 - allow MDS with the same name as client node (2939)
7028 - hold dentry reference for closed log files for unlink (2325)
7029 - reserve space for all logs during transactions (2059)
7030 - don't evict page beyond end of stripe extent (2925)
7031 - don't oops on a deleted current working directory (2399)
7032 - handle hard links to targets without a parent properly (2517)
7033 - don't dereference NULL lock when racing during eviction (2867)
7034 - don't grow lock extents when lots of conflicting locks (2919)
7036 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
7039 - account for cache space usage on clients to avoid data loss (974)
7040 - lfsck support in lustre kernel code (2349)
7041 - reduce journal credits needed for BRW writes (2370)
7042 - orphan handling to avoid losing space on client/server crashes
7043 - ptlrpcd can be blocked, stopping ALL progress (2477)
7044 - use lock value blocks to assist in proper KMS, faster stat (1021)
7045 - takes i_sem instead of DLM locks internally on obdfilter (2720)
7046 - recovery for initial connections (2355)
7047 - fixes for mds_cleanup_orphans (1934)
7048 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
7049 - block all file creations until orphan recovery completes (1901)
7050 - client remove rq_connection from request struct (2423)
7051 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
7052 - recovery timer race (2670)
7053 - mdc_close recovey bug (2532)
7054 - ptlrpc cleanup bug (2710)
7055 - mds timeout on local locks (2588)
7056 - namespace lock held during RPCs (2431)
7057 - handle interrupted sync write properly (2503)
7058 - don't try to handle a message that hasn't been replied to (2699)
7059 - client assert failure during cleanup after abort recovery (2701)
7060 - leak mdc device after failed mount (2712)
7061 - ptlrpc_check_set allows timedout requests to complete (2714)
7062 - wait for inflight reqs when ptlrpcd finishes (2710)
7063 - make sure unregistered services are removed from the srv_list
7064 - reset bulk XID's when resending them (caught by 1138 test)
7065 - unregister_bulk after timeout
7066 - fix lconf error (2694)
7067 - handle write after unfinished setstripe, stripe-only getstripe (2388)
7068 - readahead locks pages, leaves pending causing memory pressure (2673)
7069 - increase OST request buffers to 4096 on large machines (2729)
7070 - fix up permission of existing directories in simple_mkdir (2661)
7071 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
7072 - don't assign transno to errored transactions (2742)
7073 - don't delete objects on OST if given a bogus objid from MDS (2751)
7074 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
7075 - if rq_no_resend, then timeout request after recovery (2432)
7076 - fix MDS llog_logid record size, 64-bit array alignment (2733)
7077 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
7078 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
7079 - MDS recovery shouldn't delete objects that it has given out (2730)
7080 - if enqueue arrives after completion, don't clobber LVB (2819)
7081 - don't unlock pages twice when trigger_group_io returns error (2814)
7082 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
7083 - don't write pages to disk if there was an error (1450)
7084 - don't ping imports that have recovery disabled (2676)
7085 - take buffered bytes into account when balancing socknal conn (2817)
7086 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
7087 - reconnect unlink llog connection after MDS reconnects to OST (2816)
7088 - remove little-endian swabbing of llog records (1987)
7089 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
7090 - retry reposting request buffers if they fail (1191)
7091 - grow extent at grant time to avoid granting a revoked lock (2809)
7092 - lock revoke doesn't evict page if covered by a second lock (2765)
7093 - disable VM readahead to avoid reading outside lock extents (2805)
7095 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
7096 - updated kernel patches for hp-2.4.20 kernel (2681)
7098 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
7101 - fix truncated write corruption (2366)
7102 - fix for failed assertion in iopen_connect_dentry (1792,2517)
7104 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
7105 - ensure len(uuid) < 37 in lmc (1171)
7106 - fix ia64 OOPS in llog_test (2255)
7107 - zero end of page at obdfilter for partial page writes (2648)
7108 - don't leave stale dentries around after renames (bug 2428)
7109 - fix timeouts when evicting a client with a single lock held (2642)
7110 - set deadline for the initial HELLO message to drain (2634)
7111 - print out dotted-quad IP addresses in the socknal (2302)
7113 - additional debugging for MDS client eviction problem (2443)
7114 - fix mkfsoptions support for osts (2603, 2604)
7116 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
7119 - add series for the vanilla 2.6.0 kernel
7120 - add series for the vanilla 2.4.24 kernel
7121 - add series for a cray x86/64 UL kernel drop
7122 - fix xattr patches for the vanilla 2.4.19 series
7124 - generate true UUIDs in lmc (1171)
7125 - have portals stack dumping break in UML (2466)
7126 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
7127 - allocate with _NOFS in ldlm to avoid deadlock (1933)
7128 - wake callback waiting threads on client eviction (2460)
7129 - Add --ptldebug and --subsystem to lmc (1719)
7130 - update assertion to allow safe interrupt allocation
7131 - set rq_no_resend for cancel requests (2432)
7132 - recalculate ptlrpcd timeout after resend (2494)
7133 - call vfs_rmdir when removing pending directories (2368)
7134 - fix renaming a file to itself (2429)
7135 - lmc creates a default one-stripe lov (2454)
7136 - expand procfs space to handle large clusters (2326)
7137 - increase UML stack to avoid overflow
7138 - update lconf's list of debug and subsystem masks
7139 - fix lfs find --obd (2510)
7140 - /proc tunable for disabling filter read caching (2591)
7141 - stop rpm packages from altering slapd.conf (2301)
7142 - disable nagle in the socknal under 0conf (2578)
7143 - choose mds inode size based on stripe count (2572)
7144 - fix kernel-source rpm problems (2516)
7146 - add --disable-doc to avoid pdf generation (2421)
7147 - update documentation, tests, type-os, comments
7148 - avoid format warnings on ia64
7149 - remove the TOE NAL
7150 - tiny code cleanups by removing unused fields
7152 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
7155 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
7156 - remove the most blatant lies from BUILDING file (2371)
7157 - change default debug level to reasonable production setting
7158 - reduce client side cache size to reduce cache flush time
7159 - reduce max RPCs in flight to avoid unnecessary file fragmentation
7160 - make TCP zerocopy and pinger support enabled by default (2476)
7161 - sync writes completed after process exits caused crashes (2319)
7162 - maintain correct mount count on the MDS (2356)
7163 - backout 1557, because 2316 wasn't really fixed
7164 - better file I/O statistics gathering in /proc
7165 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
7166 - another kernel patch to fix zero-copy TCP function export
7167 - don't take duplicate lock when processing re-sent getattr (2420)
7168 - lctl uses obd_self_export instead of creating new conn (2353)
7169 - MDS/OST recovery case which requires object creation asserted (2425)
7170 - move lfs from /usr/sbin to /usr/bin in packages
7171 - fix race between mds_client_add and mds_client_free (2417)
7172 - use kmalloc instead of slabs in portals (2430)
7173 - don't create duplicate records when a failover MDS is present (2442)
7174 - remove unnecessary mount age check (2332)
7175 - don't remove directory inodes from locks prematurely (2451)
7176 - don't break if MDS service name is the same as hostname (2103)
7177 - fix races in client write RPC generation when cache full (2482)
7179 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
7182 - remove now-unused request->rq_obd (278)
7183 - if an allocation fails, print out how much memory we've used (1933)
7184 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
7185 - add the "configurable stack size" patch to most series files (1256)
7186 - ability to write large log records, for 100+ OST configs (2306)
7187 - fix NULL deref when filter_prep fails (2314)
7188 - fix operator precedence error in filter_sync
7189 - dynamic allocation of socknal TX descriptors (2315)
7190 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
7191 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
7192 - ensure that CWARN messages in Portals always get to the syslog
7193 - __init/__exit are not for prototype decls (ldlm_init/exit)
7194 - x86-64 compile warning fixes
7195 - fix gateway LMC keyword conflict (2318)
7196 - fix MDS lock inversions in getattr/reint paths (1844)
7197 - fix a rare lock re-ordering bug, which caused deadlock (2322)
7198 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
7199 - DLM race condition prevented some lock evictions (2328)
7200 - ENOMEM detection and retry on socknal sends (2230)
7201 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
7202 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
7203 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
7204 - better error messages when a client is rejected during recovery (1505)
7205 - avoid cancelling locks which were never granted, after failure (2330)
7206 - fix i_sem/journal inversion in mds_client_add (2333)
7207 - fix truncate/getattr lock cycle deadlock (2334)
7208 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
7209 - fix two rare exit paths which could leak an l_lock() ref (2321)
7210 - fencepost error in MDS/OST orphan recovery (2226)
7211 - make log record alignment 8 bytes (1988)
7212 - lstripe now fails when requested offset > ost_count (2237)
7213 - ensure that all kernel series have a complete list.h (1607)
7214 - fix crashes in special-file operations (2316)
7215 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
7217 - allow configurable automake binary, for testing new versions
7218 - small update to the lfs documentation
7220 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
7222 * fix negative export reference count in fsfilt_sync (2312)
7224 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
7225 * release candidate 0.9.1
7227 - orphans are moved into the PENDING directory for possible recovery
7228 - replayed opens now open by fid for orphan/rename safety (1042)
7229 - last close of an orphan inode generates a transno (683)
7230 - chdir() and mount() now pin the directory entry (1020)
7231 - avoid CERROR in normal ll_setattr_raw() error case (1500)
7232 - discard very old requests without processing them (1502)
7233 - remove some common, well-understood CERRORs (1505)
7234 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
7235 - clear "grant" flags in OST replies until OST grant code lands (1644)
7236 - fix read performance by not clobbering i_blksize on client (1598)
7237 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
7238 - make LRU size a /proc tunable, clears locks when reduced (707)
7239 - fix some lprocfs rot that prevented ptlbd from loading (1732)
7240 - server locks take references on exports now (1558)
7241 - build fixes for 2.4.20-rh trees (1663)
7242 - return an error from lov_create if all OSCs are inactive (1751)
7243 - fix import levels when a reconnect happens without a timeout (1597)
7244 - exit early from mds_open if we get a lookup error (1749)
7245 - partial page read at EOF wouldn't wait for disk before sending (1642)
7246 - avoid NULL deref in obdfilter when reading page past EOF (1592)
7247 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
7248 - fix LBUG in ll_it_open_error with rc = -2 (1861)
7249 - write/truncate lock inversion (1639)
7250 - Don't auto-load obdclass, portals modules during cleanup (1495)
7251 - fix timestamps from jumping to "now" (1763)
7252 - extra journal assertions (1648)
7253 - add an extra multiunlink test (1771)
7254 - fix read_record/write_record API (1776)
7255 - fix leak of offset_extent, possible incorrect i_size later (1772)
7256 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
7257 - Debugging for the kqswnal_get_idle_tx problems (1820)
7258 - Allow recovery to be attempted multiple times (1536)
7259 - Write out MDS last_rcvd file after it is first created (1600)
7260 - Fix tx_descriptor leak in failed transmit situations (1827)
7261 - ext3 journaling fixes for assertion failure after IO error (1871)
7262 - class_export_put() on freed export after completion AST error (1896)
7263 - Fix revalidate looping in VFS (1322)
7264 - Don't access a freed export during MDS_REINT timeout (1521)
7265 - Add open-unlink recovery support on the MDS (1673,1764)
7266 - Return an error if no MDS data was read from last_rcvd (1946)
7267 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
7268 - Don't setup a disk device that doesn't match exported UUID (317)
7269 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
7270 - avoid committing NULL handle in force close
7271 - local.sh is now a one-stripe LOV configuration
7272 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
7273 - fix ext3 htree duplicate directory entry corruption (1516)
7274 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
7275 - update to new LOV EA format (2097)
7276 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
7278 2003-06-15 Phil Schwan <phil@clusterfs.com>
7281 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
7282 - per-import/export recovery handling (958, 931, 959)
7283 - multiple last-rcvd slots, for serving multiple FSes (949)
7284 - connections are again shared between multiple imp/exports (963, 964)
7285 - "umount -f" would hang if any requests needed to be sent (393, 978)
7286 - avoid pinning large req buffer by copying for queued messages (989)
7287 - add "uuid" to "lctl device" command to help upcalls (991)
7288 - "open" RPCs with transnos would confuse recovery counters (1037)
7289 - do proper endian conversion of all wire messages (288, 340, 891)
7290 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
7291 - call waiting locks callback from LDLM recovery thread (1127, 1151)
7292 - fix ptlrpc_connection leak in target_handle_connect (1174)
7293 - fix import refcounting bug in OST and MDS cleanup (1134)
7294 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
7295 - fix very unlikely obd_types race condition (501)
7296 - remove osc_open hack for echo_client (1187)
7297 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
7298 - a failure in read_inode2 leads to deadlock (1139)
7299 - cancel ack-locks as soon as transaction is committed (1072)
7300 - fix major leaks and crashes in the bulk I/O path (937, 1057)
7301 - make sure to commitrw after any preprw to avoid deadlock (1162)
7302 - failing to execute a file in a lustre FS would lock inode (1203)
7303 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
7304 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
7305 - obd_uuid2tgt would walk off the end of the list (1255)
7306 - on IA64 the osc would give portals incorrect bulk size (1258)
7307 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
7308 - fix lock inversion caused by new llite matching code (1282)
7309 - limit the number of dirty pages on a client to 10MB (1286)
7310 - timed out locks were not being corrected cancelled (1289)
7311 - fix O_DIRECT above 4GB on IA-32 (1292)
7312 * major user-visible changes
7313 - fail out/fail over policy now controlled by the upcall (993)
7315 - add OBD_PING to check server availability and failure (954)
7316 - lustre messages are now sent in sending host order (288, 340, 891)
7317 - add eadatalen to MDS getattr reply (340)
7318 - OST read replies may contain second buffer, with per-page status (593)
7320 2003-03-11 Phil Schwan <phil@clusterfs.com>
7323 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
7324 - failed open()s could cause deadlock; fixed (867, 869)
7325 - stop cancelling OST locks when files are closed (481)
7326 - overlapping XID spaces caused network corruption (851, 853)
7327 - fix unsafe fsfilt counter arithmetic; change to atomic_t
7328 - setattr_raw added, to do single-RPC, server-side setattrs
7329 - lmc/lconf syntax change for OST UUIDs
7330 - fix crashy race condition between ptlrpc_free_req and osc_close
7331 - don't use request in mdc_enqueue if we hit a timeout (889)
7332 - don't set the inode i_size for regular files from the MDS (896)
7333 - handle out of order completion AST (842)
7334 - don't LBUG if a lock request times out after receiving AST (913)
7335 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
7336 - if a bad lock AST arrives, send an error instead of dropping entirely
7337 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
7338 - fix leak in bulk IO when only partially completed (899, 900, 926)
7339 - fix O_DIRECT for ia64 (55)
7340 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
7341 - C-z after timeout could hang a process forever; fixed (977)
7343 - client-side I/O cache (678, 924, 929, 941, 970)
7345 - READPAGE and SETATTRs which don't take server-side locks get
7348 2003-02-11 Phil Schwan <phil@clusterfs.com>
7351 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
7352 - Fix fsfilt_extN_readpage() to read a full page of directory
7353 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
7354 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
7355 - Fix imbalanced LOV object allocation and out-of-bound access (469)
7356 - Most intent operations were removed, in favour of a new RPC mode
7357 that does a single RPC to the server and bypasses most of the VFS
7358 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
7359 - Aggressively cancel local locks on DLM servers
7360 - mds_reint_unlink sends EA to the client if it's the last nlink.
7361 client uses that EA to unlink OST objects.
7362 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
7363 - recursive symlinks were fixed (439)
7364 - fixed NULL deref in DEBUG_REQ
7365 - filter_update_lastobjid no longer calls sync, which annoyed extN
7366 - fixed multi-client small-writes to a single file problem (445)
7367 - fixed mtime updates during file writes (607)
7368 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
7369 - fixed bug in obd_brw_read/write() (under guise of testing 367)
7370 - fixed Linux OST size reporting problem (444, 656)
7371 - OST now updates object mtime with writes or setattr (607, 619)
7372 - client verifies file size before zeroing page past EOF (445)
7373 - OST now writes last allocated objid to disk with allocation (108)
7374 - LOV on echo now works (409)
7376 - mds_reint_unlink sends a new buffer, with the EA included. this
7377 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
7378 set if a regular file was being unlinked, and it was the last link
7379 - use PtlGet from the target for bulk writes (315)
7380 - OST now updates object mtime with writes or setattr (607, 619)
7381 - LDLM now has a grant-time callback to revalidate locked items, if
7383 - Many MDS operations were reorganized to combat race conditions
7385 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
7386 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
7388 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
7391 - Fully reactivate OST imports after reconnection (512, others)
7392 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
7393 - More graceful error handling for truncating on dead OST (515)
7394 - Don't error out unless we're actually accessing dead stripes (474)
7395 - Fix garbage sizes when stripes are missing (410)
7396 - LRU counters were broken, causing constant lock purge (433, 432)
7397 - garbage on read from stripes with failed OSTs (441)
7398 - mark OSCs as active before reconnecting during recovery (438)
7399 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
7400 - lfind did not preserve OST order in output (443)
7401 - symlinks cause hung clients, incorrect data (439)
7402 - stop dereferencing request after dropping refcount (457)
7403 - don't LASSERT(spin_is_locked) on non-SMP (455)
7404 - fixes for many rename() bugs
7405 - fstat didn't correctly synchronize attributes (399)
7406 - server must handle lock cancellation during blocking AST prep (487)
7407 - bulk descriptors were free()d too soon (511)
7408 - fix paths in lconf, which would load incorrect modules (451, 507)
7409 - fix confusing lconf 'host not found' error message (386)
7410 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
7411 - fix race condition in mdc_blocking_ast() for inode access (526)
7412 - fix lov_unpackmd() unpacking wrong number of stripes (537)
7413 - fix lov_set_osc_active() marking wrong OSC inactive (440)
7414 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
7415 - fix multiple writes of stripe MD to MDS (358, maybe 519)
7416 - fix lstripe in several ways (kernel side) (527)
7417 - fix request leak in ldlm_cli_enqueue (262)
7418 - incorrect OSC was marked inactive after OST failure
7419 - call mds_fs_cleanup before unmounting filesystem (524)
7420 - fix races between taking ns_lock and ldlm_lock_change_resource
7421 - fix races updating LOV export open file list
7422 - fix lov_enqueue error path, avoid decref-ing bad lock handle
7423 - fix recovery NULL deref in ldlm_cli_cancel_unused
7424 - fix some DLM races by using new hash table for lock handles (419)
7425 - permit the client to specify desired inodes, at replay
7426 - duplicate requests when we queue them for replay reintegration
7427 - fix last_rcvd offset calculation
7428 - sync after each recovered transaction, so we always make progress
7429 - never, not always, ERESTART requests without transnos
7430 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
7432 - skip replay if the MDS says that the client is already connected
7433 - don't check for a recovery-enabled export to match lctl's UUID
7434 - don't INC_USE_COUNT for phantom exports
7435 - don't crash when cleaning up phantom exports (567)
7436 - don't double-finish or set replay data for errored mdc_open requests
7437 - abort requests when they time out, so we don't get old replies
7438 - send/receive replies for AST messages again
7439 - if the client says that it doesn't have the lock, cancel it on the
7441 - if we timeout during I/O, don't try to cancel an in-use lock; instead
7442 mark it as destroyed, it will all work out when decref is called
7443 - fix module use counts (22, 581)
7445 - ASTs now expect a reply (server cancels lock on error reply)
7447 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
7450 - fix many simultaneous client startup (392)
7451 - fix dentry->d_it clobbering
7452 - credentials weren't being shipped for readdir/getattr operations
7453 - remove invalid assertions triggered during some concurrent MD
7455 - proper Lustre versions added (336, 389)
7456 - fix memory leak for create error case (398)
7457 - fix LOV locking bug that would get cli/srv out of sync
7458 - fix echo client over LOV (409)
7459 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
7460 - fix double-O_EXCL intent crash (424)
7461 - avoid sending multiple lock CANCELs (352)
7463 - MDS can do multi-client recovery (modulo bugs in new code)
7465 - many updates, edits, cleanups
7467 2002-11-18 Phil Schwan <phil@clusterfs.com>
7470 - fix null d_it dereference (346)
7471 - fix full OST/dbench hang (333)
7472 - fix permission problem with file removal (286)
7473 - fix removal of OSCs from LOV when they fail
7474 - fix NULL deref during bulk timeout (214)
7475 - fix problems related to multiple filesystems on one MDS (241)
7476 - fixed serious subtle metadata locking bugs
7477 - free locks on clients when inodes are removed due to memory
7479 - fix inode pointer in lock data (285)
7480 - partial support for multiple MDS on a single host (241)
7481 - data locks weren't cancelled at clear_inode time (290, 311)
7482 - intent locks could lead to unbounded lock growth (205)
7483 - added a maximum lock count, an LRU list, and a flusher
7484 - fix multiple rename (365)
7485 - properly abstracted the echo client
7486 - OSC locked 1 byte too many; fixed
7487 - rewrote brw callback code:
7488 - fixed recovery bugs related to LOVs (306)
7489 - fixed too-many-pages-in-one-write crash (191)
7490 - fixed (again) crash in sync_io_timeout (214)
7491 - probably fixed callback-related race (385)
7493 - Add capability to MDS protocol
7494 - LDLM cancellations and callbacks on different portals
7496 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
7499 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
7500 - defer open object destruction to close time (601981, 138)
7501 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
7502 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
7503 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
7504 - don't call obd_setattr in ll_file_release for destroyed objects
7505 * protocol change to lustre_msg: move |version| and add |flags|
7506 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
7507 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
7508 * added replay of create, unlink, link and rename operations during
7509 MDS failover; recovery should be much more robust now
7510 * remove failed OSCs from LOVs (only lov_create uses this so far)
7511 * the lustre-HOWTO was brought (more) up to date (582544)
7513 2002-10-23 Phil Schwan <phil@clusterfs.com>
7516 - in-use dentries weren't being reused properly (617851)
7517 - prevent multiple LDLM setup (599178)
7518 - fix LOV size calculations for truncate (617853)
7519 - fix client handling of MDS intent errors (POSIX)
7520 - fix permission bug in lovstripe.c test (624321)
7521 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
7522 - truncate past end of file could corrupt data
7523 - proper cleanup after timeouts, crashes, etc (592524, 550815)
7524 - a race in recovery could return ETIMEDOUT to apps (623947)
7525 - building outside the source directory was fixed
7526 * the lustre-HOWTO was brought (more) up to date (582544)
7527 * major progress was made on recovery functionality
7529 2002-10-10 Phil Schwan <phil@clusterfs.com>
7532 - recovery deadlock fix
7533 - rm -rf causes LBUG fix (617817)
7534 - file open by multiple tasks fix (618962)
7535 - directory permissions bugs (602707 and 620007)
7536 - journal_stop fixed with locking (611313)
7537 - O_APPEND failures resolved (618273, perhaps 614459)
7538 - lconf PATH fix (619770)
7539 - IA64 build fix (621450)
7540 - RPC buffer sizes scale with amount of memory
7542 2002-10-01 Phil Schwan <phil@clusterfs.com>
7545 - locks would be cancelled without throwing away data pages,
7546 resulting in inconsistent data (605627)
7547 - inode attributes were not always being refreshed (605627, 612449)
7548 - lconf now continues to cleanup after lctl reports an error
7549 - MDS now enforces user permissions (602707)
7550 - lprocfs cleanup fixed, but not yet enabled (614157)
7551 - fixed infinite server hang, should a client not respond to an AST
7552 - avoid going into recovery if user calls readlink() with a buffer
7553 that's too small (613941)
7554 - AST RPCs no longer require replies (614867) -- this may be changed
7555 - don't crash server if client sends an IOV that's too big (611336)
7556 - fixed lock conversion deadlock (611892)
7557 - fixed the following of symlinks (614622)
7558 * recovery: the server can remove locks from a client that dies, other
7559 clients can make progress
7560 * more extN patch fixes
7561 * compile-time configurable ptlrpc buffer allocations
7563 - collaborative read cache document
7564 - Lustre Lite Performance CDR document-in-progress
7566 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
7569 - fix typo in patch-2.4.18
7571 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
7574 - clear ptlrpc request each time in handle_incoming_request()
7575 - unlink of files now destroys the object on the OST
7577 2002-09-19 Peter Braam <braam@clusterfs.com>
7579 * add hard link support
7580 * change obdfile creation method
7581 * kernel patch changed
7583 2002-09-19 Peter Braam <braam@clusterfs.com>
7586 - stack overflow bug in extN fixed
7588 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
7590 * documentation updates
7591 - add man pages for config tools
7592 - update tests/README to describe testing with new config tools
7593 - finish metadata API descriptions
7594 * bug fixes and cleanups
7595 - statfs workaround for 16TB limit
7596 - LOV stripe allocation improved, can stripe on subset of OSTs
7597 - LOV file size/IO offset was wrong for files > 4GB in size
7598 - object EA data was being dropped, caused files to be unreadable
7599 - memory overflow with non-LOV OST caused memory corruption
7600 - fixed regression tests to work with new config tools, obdfilter
7601 - fixed bug when directory size became larger than 1 block
7602 - fixed bug (for single client case) when PWD was deleted
7603 - invalidate local directory pages when doing intent-based ops
7604 - avoid LDLM oops when lock callback contained bad data
7606 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
7608 * documentation updates
7609 * bug fixes and cleanups
7610 - configuration tools
7613 - 64-bit compile warnings
7614 - 64-bit internal statfs data
7616 * test_brw on persistent OST devices
7618 * lprocfs (disabled)
7620 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
7622 * documentation updates
7623 * bug fixes and cleanups
7624 * configuration tools
7626 2002-08-30 Peter J. Braam <braam@clusterfs.com>
7629 * many small fixes to 0_5_4
7630 * io/network handling
7631 * thinkos in MDS operations
7633 2002-08-24 Peter J. Braam <braam@clusterfs.com>
7636 * crucial basic fixes to 0.5.3
7637 * IOR, Iozone work over Elan
7640 2002-08-07 Phil Schwan <phil@clusterfs.com>
7641 * version 0_5_3, our first alpha
7642 * we use the new Portals iovs
7643 * documentation updates
7644 * bug fixes and cleanups
7645 * small changes in the DLM wire protocol
7647 2002-07-25 Peter J. Braam <braam@clusterfs.com>
7648 * version 0_5_1 with some initial stability,
7649 * locking on MD and file I/O.
7650 * documentation updates
7651 * several bug fixes since 0.5.0
7652 * small changes in wire protocol
7654 2002-07-18 Phil Schwan <phil@clusterfs.com>
7656 * delivered as Lustre Light Alpha
7657 * fixed a crash after handling invalid MDS requests
7658 * fixed directory pages for architectures with non-4k pages sizes
7660 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
7661 * release version v0_4_4
7662 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
7664 * Fixes a number of interruption problems with OST operations.
7665 * Update documentation for portals header changes
7666 * Move all wire protocol structs/defines to lustre_idl.h
7667 * Fixes symlink length bug.
7668 * Add tcpdump to repository.
7670 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
7671 * release version v0_4_3
7672 * Fixes statfs for inodes on extN.
7673 * Fixes bug in runtests which would delete /etc/hosts.
7674 * Use 64-bit object IDs wherever possible (not into VFS though)
7675 Remove ost_get_info, which is unused by lustre, and out of date.
7677 2002-07-03 Peter Braam <braam@clusterfs.com>
7678 * release version v0_4_2 Fixes a lookup error (type not passed)
7679 * move forward to head of Portals
7680 * move forward to latest Lustre kernel
7682 2002-06-25 Peter Braam <braam@clusterfs.com>
7683 * release version v0_4_1. Hopefully stable on single node use.