1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5 2.6.16.46-0.14 (SLES 10)
6 2.6.18.8 vanilla (kernel.org)
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.9-42.0.10.EL (RHEL 4)
10 2.6.16 - 2.6.21 vanilla (kernel.org)
11 * Recommended e2fsprogs version: 1.39.cfs8
12 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
16 Description: don't use GFP_* in generic Lustre code.
17 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
21 Description: obdclass is limited by single OBD_ALLOC(idarray)
22 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
26 Description: warnings with build patchless client with vanila 2.6.19 and up
27 Details : change old ctl_table style and replace ctl_table/ctl_table_header
28 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
30 --------------------------------------------------------------------------------
32 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
34 * Support for kernels:
37 2.6.16.46-0.14 (SLES 10)
38 2.6.18.8 vanilla (kernel.org)
39 * Client support for unpatched kernels:
40 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
41 2.6.16 - 2.6.22 vanilla (kernel.org)
42 * Due to recently discovered recovery problems, we do not recommend
43 using patchless RHEL 4 clients with this or any earlier release.
44 * Recommended e2fsprogs version: 1.39.cfs8
45 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
46 * Starting with this release, the ldiskfs backing filesystem required
47 by Lustre is now in its own package, lustre-ldiskfs. This package
48 should be installed. It is versioned separately from Lustre and
49 may be released separately in future.
51 Severity : enhancement
53 Description: add optional extra BUILD_VERSION info
54 Details : add a new environment variable (namely LUSTRE_VERS) which allows
55 to override the lustre version.
58 Frequency : 2.6.18 servers only
60 Description: ll_kern_mount() doesn't release the module reference
61 Details : The ldiskfs module reference count never drops down to 0
62 because ll_kern_mount() doesn't release the module reference.
67 Description: server LBUG when using old ost_num_threads parameter
68 Details : Accept the old ost_num_threads parameter but warn that it
69 is deprecated, and fix an off-by-one error that caused an LBUG.
74 Description: Transient SCSI error results in persistent IO issue
75 Details : iobuf->dr_error is not reinitialized to 0 between two
79 Frequency : sometimes when underlying device returns I/O errors
81 Description: OSTs not going read-only during write failures
82 Details : OSTs are not remounted read-only when the journal commit threads
83 get I/O errors because fsfilt_ext3 calls journal_start/stop()
84 instead of the ext3 wrappers.
88 Description: poor connect scaling with increasing client count
89 Details : Don't run filter_grant_sanity_check for more than 100 exports
90 to improve scaling for large numbers of clients.
93 Frequency : SLES10 only
95 Description: sanity-quota.sh quotacheck failed: rc = -22
96 Details : Quotas cannot be enabled on SLES10.
99 Frequency : liblustre clients only
101 Description: getdirentries does not give error when run on compute nodes
102 Details : getdirentries does not fail when the size specified as an argument
103 is too small to contain at least one entry
105 Severity : enhancement
107 Description: Add LNET router traceability for debug purposes
108 Details : If a checksum failure occurs with a router as part of the
109 IO path, the NID of the last router that forwarded the bulk data
110 is printed so it can be identified.
115 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
116 Details : Due to a race condition, liblustre clients were occasionally
119 Severity : enhancement
121 Description: lfs setstripe use optional parameters instead of postional
124 Severity : enhancement
126 Description: Nanosecond timestamp support for ldiskfs
127 Details : The on-disk ldiskfs filesystem has added support for nanosecond
128 resolution timestamps. There is not yet support for this at
129 the Lustre filesystem level.
132 Frequency : during server recovery
134 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
135 Details : request with rq_no_resend flag not awake l_wait_event if they get a
139 Frequency : nfs export on patchless client
141 Description: connectathon hang when test nfs export over patchless client
142 Details : Disconnected dentry cannot be found with lookup, so we do not need
143 to unhash it or make it invalid
146 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
147 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
148 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
153 Description: open req refcounting wrong on reconnect
154 Details : If reconnect happened between getting open reply from server and
155 call to mdc_set_replay_data in ll_file_open, we will schedule
156 replay for unreferenced request that we are about to free.
157 Subsequent close will crash in variety of ways.
158 Check that request is still eligible for replay in
159 mdc_set_replay_data().
164 Description: disable writes to filesystem when reading health_check file
165 Details : the default for reading the health_check proc file has changed
166 to NOT do a journal transaction and write to disk, because this
167 can cause reads of the /proc file to hang and block HA state
168 checking on a healthy but otherwise heavily loaded system. It
169 is possible to return to the previous behaviour during configure
170 with --enable-health-write.
172 Severity : enhancement
174 Description: 64-bit inode version
175 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
176 to the inode. This will be required for version-based recovery.
181 Description: MDS fails to start if a duplicate client export is detected
182 Details : in some rare cases it was possible for a client to connect to
183 an MDS multiple times. Upon recovery the MDS would detect this
184 and fail during startup. Handle this more gracefully.
186 Severity : enhancement
188 Description: Add -o localflock option to simulate old noflock
190 Details : This will achieve local-only flock/fcntl locks
196 Description: log_commit_thread vs filter_destroy race leads to crash
197 Details : Take import reference before releasing llog record semaphore
202 Description: Wrong request locking in request set processing
203 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
204 imp_delayed_list, in this place should be used imp_lock.
207 Frequency : when reconnection
209 Description: Grant Leak when osc reconnect to OST
210 Details : When osc reconnect ost, OST(filter) should check whether it
211 should grant more space to client by comparing fed_grant and
212 cl_avail_grant, and return the granted space to client instead
213 of "new granted" space, because client will call osc_init_grant
214 to update the client grant space info.
217 Frequency : when client reconnect to OST
219 Description: Grant Leak when osc do resend and replay bulk write
220 Details : When osc reconnect to OST, OST(filter)should clear grant info of
221 bulk write request, because the grant info while be sync between
222 OSC and OST when reconnect, and we should ignore the grant info
223 these of resend/replay write req.
228 Description: Grant space more than avaiable left space sometimes.
229 Details : When then OST is about to be full, if two bulk writing from
230 different clients came to OST. Accord the avaliable space of the
231 OST, the first req should be permitted, and the second one
232 should be denied by ENOSPC. But if the seconde arrived before
233 the first one is commited. The OST might wrongly permit second
234 writing, which will cause grant space > avaiable space.
237 Frequency : when client is evicted
239 Description: Grant might be wrongly erased when osc is evicted by OST
240 Details : when the import is evicted by server, it will fork another
241 thread ptlrpc_invalidate_import_thread to invalidate the
242 import, where the grant will be set to 0. While the original
243 thread will update the grant it got when connecting. So if
244 the former happened latter, the grant will be wrongly errased
245 because of this race.
250 Description: Checking Stale with correct fid
251 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
252 to check whether it is stale, because sometimes, we want the
253 enqueue happened anyway, and op_data.fid2 will not be initialized.
255 Severity : enhancement
257 Description: update patchless client
258 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
261 Frequency : only with 2.4 kernel
263 Description: random memory corruption
264 Details : size of struct ll_inode_info is to big for union inode.u and this
265 can be cause of random memory corruption.
270 Description: Memory leak in recovery
271 Details : Lov_mds_md was not free in an error handler in mds_create_object.
272 It should also check obd_fail before fsfilt_start, otherwise if
273 fsfilt_start return -EROFS,(failover mds during mds recovery).
274 then the req will return with repmsg->transno = 0 and rc = EROFS.
275 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
276 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
277 be freed no matter whether fsfilt_commit success or not.
280 Frequency : only with huge count clients
282 Description: Prevents from taking the superblock lock in llap_from_page for
284 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
285 for llap_from_page call in ll_removepage prevents from taking the
286 superblock lock for a soon died page.
291 Description: Not check open intent error before release open handle
292 Details : in some rare cases, the open intent error is not checked before
293 release open handle, which may cause
294 ASSERTION(open_req->rq_transno != 0), because it tries to release
295 the failed open handle.
300 Description: Set cat log bitmap only after create log success.
301 Details : in some rare cases, the cat log bitmap is set too early. and it
302 should be set only after create log success.
306 Description: Accessing a block bevice can re-enable I/O when Lustre is
307 tearing down a device.
308 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
313 Description: service threads may hog cpus when there are a lot of requests
315 Details : Insert cond_resched to give other threads a chance to use some of
321 Description: the cat log was not initialized in recovery
322 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
323 unlink log on mds will not be initialized until mds post
324 recovery. And also in mds post recovery, the unlink log will
325 initialization will be done asynchronausly, so there will be race
326 between add unlink log and unlink log initialization.
330 Description: brw_stats were being printed incorrectly
331 Details : brw_stats were being printed as log2 but all of them were not
332 recorded as log2. Also remove some code duplication arising from
333 filter_tally_{read,write}.
337 Frequency : rare, only in recovery.
338 Description: ASSERTION(req->rq_type != LI_POISON) failed
339 Details : imp_lock should be held while iterating over imp_sending_list for
340 prevent destroy request after get timeout in ptlrpc_queue_wait.
344 Description: replay-single.sh test 52 fails
345 Details : A lock's skiplist need to be cleanup when it being unlinked
346 from its resource list.
350 Description: Short directio read returns full requested size rather than
352 Details : Direct I/O operations should return actual amount of bytes
353 transferred rather than requested size.
355 Severity : enhancement
357 Description: metadata RPC reduction (e.g. for rm performance)
358 Details : decrease the amount of synchronous RPC between clients and servers
359 by canceling conflicing lock before the operation on the client side
360 and packing thier handles into the main operation RPC to server.
362 --------------------------------------------------------------------------------
364 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
369 Frequency : on some architectures
371 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
372 Details : Uninitialized flags sometimes cause configuration commands to
376 Frequency : patchless clients only
378 Description: missing __iget() symbol export
379 Details : The __iget() symbol export is missing. To avoid the need for
380 this on patchless clients the deathrow inode reaper is turned
381 off, and we depend on the VM to clean up old inodes. This
382 dependency was during via the fix for bug 12181.
384 --------------------------------------------------------------------------------
386 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
388 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
389 INTEROPERATE with older versions automatically. In many cases a
390 special upgrade step is needed. Please read the
391 user documentation before upgrading any part of a live system.
392 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
393 version will not interoperate with 1.6 betas before beta5 (1.5.95).
394 * WARNING: Lustre configuration and startup changes are required with
395 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
400 Severity : enhancement
402 Description: MountConf
403 Details : Lustre configuration is now managed via mkfs and mount
404 commands instead of lmc and lconf. New obd types (MGS, MGC)
405 are added for dynamic configuration management. See
406 https://mail.clusterfs.com/wikis/lustre/MountConf for
409 Severity : enhancement
411 Description: dynamic OST addition
412 Details : OSTs can now be added to a live filesystem
414 Severity : enhancement
416 Description: startup order invariance
417 Details : MDTs and OSTs can be started in any order. Clients only
418 require the MDT to complete startup.
420 Severity : enhancement
422 Description: parallel, asynchronous orphan cleanup
423 Details : orphan cleanup is now performed in separate threads for each
424 OST, allowing parallel non-blocking operation.
426 Severity : enhancement
428 Description: optimized stripe assignment
429 Details : stripe assignments are now made based on ost space available,
430 ost previous usage, and OSS previous usage, in order to try
431 to optimize storage space and networking resources.
433 Severity : enhancement
435 Description: Permanently set tunables
436 Details : All writable /proc/fs/lustre tunables can now be permanently
437 set on a per-server basis, at mkfs time or on a live
440 Severity : enhancement
442 Description: Lustre message v2
443 Details : Add lustre message format v2.
445 Severity : enhancement
447 Description: client OST exclusion list
448 Details : Clients can be started with a list of OSTs that should be
449 declared "inactive" for known non-responsive OSTs.
453 Description: SPEC SFS validation failure on NFS v2 over lustre.
454 Details : Changes the blocksize for regular files to be 2x RPC size,
455 and not depend on stripe size.
457 Severity : enhancement
459 Description: Multiple MD RPCs in flight.
460 Details : Further unserialise some read-only MDS RPCs - learn about intents.
461 To avoid overly-overloading MDS, introduce a limit on number of
462 MDS RPCs in flight for a single client and add /proc controls
463 to adjust this limit.
465 Severity : enhancement
467 Description: client read/write statistics
468 Details : Add client read/write call usage stats for performance
469 analysis of user processes.
470 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
471 file access. extents_stats shows chunk size distribution.
472 extents_stats_per_process show chunk size distribution per
475 Severity : enhancement
477 Description: mds statistics
478 Details : Add detailed mds operations statistics in
479 /proc/fs/lustre/mds/*/stats.
483 Description: Failure of copying files with lustre special EAs.
484 Details : Client side always return success for setxattr call for lustre
485 special xattr (currently only "trusted.lov").
490 Description: Refcount LNET uuids
491 Details : The global LNET uuid list grew linearly with every startup;
492 refcount repeated list entries instead of always adding to
495 Severity : enhancement
497 Description: Dynamic service threads
498 Details : Within a small range, start extra service threads
499 automatically when the request queue builds up.
502 Frequency : mixed-endian client/server environments
504 Description: mixed-endian crashes
505 Details : The new msg_v2 system had some failures in mixed-endian
508 Severity : enhancement
510 Description: Easy OST removal
511 Details : OSTs can be permanently deactivated with e.g. 'lctl
512 conf_param lustre-OST0001.osc.active=0'
514 Severity : enhancement
516 Description: MGS proc entries
517 Details : Added basic proc entries for the MGS showing what filesystems
520 Severity : enhancement
522 Description: provide MGS failover
523 Details : Added config lock reacquisition after MGS server failover.
525 Severity : enhancement
527 Description: add Linux 2.4 support
528 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
532 Description: a large application tries to do I/O to the same resource and dies
534 Details : Check the req->rq_arrival time after the call to
535 ost_brw_lock_get(), but before we do anything about
536 processing it & sending the BULK transfer request. This
537 should help move old stale pending locks off the queue as
538 quickly as obd_timeout.
541 Frequency : when an incorrect nid is specified during startup
543 Description: ptlrpc connect to non-existant node causes kernel crash
544 Details : LNET can't be re-entered from an event callback, which
545 happened when we expire a message after the export has been
546 cleaned up. Instead, hand the zombie cleanup off to another
549 Severity : enhancement
551 Description: plain/inodebits lock performance improvement
552 Details : Grouping plain/inodebits in granted list by their request modes
553 and bits policy, thus improving the performance of search through
557 Frequency : only if OST filesystem is corrupted
559 Description: client incorrectly hits assertion in ptlrpc_replay_req()
560 Details : for a short time RPCs with bulk IO are in the replay list,
561 but replay of bulk IOs is unimplemented. If the OST filesystem
562 is corrupted due to disk cache incoherency and then replay is
563 started it is possible to trip an assertion. Avoid putting
564 committed RPCs into the replay list at all to avoid this issue.
567 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
569 Description: System hang on startup
570 Details : This bug allowed the liblustre (e.g. catamount) client to
571 return to the app before handling all startup RPCs. This
572 could leave the node unresponsive to lustre network traffic
573 and manifested as a server ptllnd timeout.
575 Severity : enhancement
577 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
578 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
579 boolean to control whether to print peer debug info when a
580 client's RPC times out.
583 Frequency : only for kernels with patches from Lustre below 1.4.3
585 Description: Remove old rdonly API
586 Details : Remove old rdonly API which unused from at least lustre 1.4.3
589 Frequency : only for devices with external journals
591 Description: Set external device read-only also
592 Details : During a commanded failover stop, we set the disk device
593 read-only while the server shuts down. We now also set any
594 external journal device read-only at the same time.
597 Frequency : when upgrading from 1.4 while trying to change parameters
599 Description: The wrong (new) MDC name was used when setting parameters for
600 upgraded MDT's. Also allows changing of OSC (and MDC)
601 parameters if --writeconf is specified at tunefs upgrade time.
604 Frequency : when setting specific ost indicies
606 Description: QOS code breaks on skipped indicies
607 Details : Add checks for missing OST indicies in the QOS code, so OSTs
608 created with --index need not be sequential.
610 Severity : enhancement
612 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
613 Details : The uninit_groups feature works in conjunction with the kernel
614 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
615 up the pass1 processing of e2fsck. This is a read-only feature
616 in ldiskfs2 only, so older kernels and current ldiskfs cannot
617 mount filesystems that have had this feature enabled.
619 Severity : enhancement
621 Description: Improve multi-block allocation algorithm to avoid fragmentation
622 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
623 allocation locality and avoid filesystem fragmentation.
625 ------------------------------------------------------------------------------
627 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
629 * Support for kernels:
630 2.4.21-47.0.1.EL (RHEL 3)
632 2.6.9-42.0.10.EL (RHEL 4)
633 2.6.12.6 vanilla (kernel.org)
634 2.6.16.27-0.9 (SLES 10)
635 * Recommended e2fsprogs version: 1.39.cfs5
637 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
641 Frequency : occasional, depends on client load and configuration
642 Bugzilla : 12181, 12203
643 Description: data loss for recently-modified files
645 Details : In some cases it is possible that recently written or created
646 files may not be written to disk in a timely manner (this should
647 normally be within 30s unless client IO load is very high).
648 The problem appears as zero-length files or files that are a
649 multiple of 1MB in size after a client crash or client eviction
650 that are missing data at the end of the file.
652 This problem is more likely to be hit on clients where files are
653 repeatedly created and unlinked in the same directory, clients
654 have a large amount of RAM, have many CPUs, the filesystem has
655 many OSTs, the clients are rebooted frequently, and/or the files
656 are not accessed by other nodes after being written.
658 The presence of the problem can be detected by looking at
659 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
660 smaller than the second (nr_unused) then dirty files will not
661 be flushed automatically to disk. "sync; sleep 10" should be
662 run several times on the node before unmounting it to update
663 Lustre (this is also safe to run on nodes without this problem).
665 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
666 that can cause this same problem, so customers using that kernel
667 also need to update the kernel in addition to Lustre. In order
668 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
670 It is normal that files written just before a client crash (less
671 than 30s) may not yet have been flushed to disk, even for local
675 Frequency : frequent on thin XT3 nodes
677 Description: UUID collision on thin XT3 Linux nodes
678 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
679 randomly, since we relied on an insufficiently-seeded PRNG.
684 Description: OSS hangs after "All ost request buffers busy"
685 Details : A deadlock between quota and journal operations caused OSS
686 hangs after printing "All ost request buffers busy."
689 Frequency : always on liblustre builds
691 Description: Cleanup compiler warnings on liblustre
694 Frequency : always on liblustre builds on XT3
696 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
700 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
701 > 32000 subdirectories
702 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
703 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
704 subdirectory count crosses 32000. This will aid e2fsck to
705 correctly handle more than 32000 subdirectories.
708 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
710 Description: System hang on startup
711 Details : This bug allowed the liblustre (e.g. catamount) client to
712 return to the app before handling all startup RPCs. This
713 could leave the node unresponsive to lustre network traffic
714 and manifested as a server ptllnd timeout.
716 Severity : enhancement
718 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
719 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
720 boolean to control whether to print peer debug info when a
721 client's RPC times out.
726 Description: make O_SYNC working on 2.6 kernels
727 Details : 2.6 kernels use different method for mark pages for write,
728 so need add a code to lustre for O_SYNC work.
733 Description: Failure to close file and release space on NFS
734 Details : Put inode details into lock acquired in ll_intent_file_open.
735 Use mdc_intent_lock in ll_intent_open to properly
736 detect all kind of errors unhandled by mdc_enqueue.
741 Description: proc file read during shutdown sometimes raced obd removal,
743 Details : Add lock to prevent obd access after proc file removal.
746 Frequency : Only for files larger than 4GB on 32-bit clients.
748 Description: improperly doing page alignment of locks
749 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
750 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
755 Description: RPCs being resent when they shouldn't be
756 Details : Some RPCs that should not be resent are being resent. This
757 can cause inconsistencies in the RPC state machine. Do not
758 resend such requests.
761 Frequency : rare, only with NFS export
763 Description: Crash on NFS re-export node
764 Details : under very unusual load conditions an assertion is hit in
765 ll_intent_file_open()
768 Frequency : only if OST filesystem is corrupted
770 Description: client incorrectly hits assertion in ptlrpc_replay_req()
771 Details : for a short time RPCs with bulk IO are in the replay list,
772 but replay of bulk IOs is unimplemented. If the OST filesystem
773 is corrupted due to disk cache incoherency and then replay is
774 started it is possible to trip an assertion. Avoid putting
775 committed RPCs into the replay list at all to avoid this issue.
780 Description: large O_DIRECT requests fail under memory pressure/fragmentation
781 Details : Large single O_DIRECT read and write calls can fail to allocate
782 a sufficiently large buffer to process the request. In case of
783 allocation failure the allocation is retried with a smaller
784 buffer and broken into smaller requests.
786 Severity : enhancement
788 Description: Add -o localflock option to simulate old noflock behaviour.
789 Details : This will achieve local-only flock/fcntl locks coherentness.
794 Description: versioning check is incomplete
795 Details : Checking the version difference of client vs. server, report
796 error if the gap is too big.
801 Description: add support PG_writeback bit
802 Details : add support for PG_writeback bit for Lustre, for more carefull
803 work with page cache in 2.6 kernel. This also fix some deadlocks
804 and remove hack for work O_SYNC with 2.6 kernel.
806 Severity : enhancement
808 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
809 Details : The uninit_groups feature works in conjunction with the kernel
810 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
811 up the pass1 processing of e2fsck. This is a read-only feature
812 in ldiskfs2 only, so older kernels and current ldiskfs cannot
813 mount filesystems that have had this feature enabled.
815 Severity : enhancement
817 Description: Improve multi-block allocation algorithm to avoid fragmentation
818 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
819 allocation locality and avoid filesystem fragmentation.
821 ------------------------------------------------------------------------------
823 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
825 * Support for kernels:
826 2.6.9-42.0.3.EL (RHEL 4)
828 2.4.21-47.0.1.EL (RHEL 3)
829 2.6.12.6 vanilla (kernel.org)
830 2.6.16.21-0.8 (SLES10)
831 * Recommended e2fsprogs version: 1.39.cfs2-0
833 * The backwards-compatible /proc/sys/portals symlink has been removed
834 in this release. Before upgrading, please ensure that you change
835 any configuration scripts or /etc/sysctl.conf files that access
836 /proc/sys/portals/* or sysctl portals.* to use the corresponding
837 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
838 in advance of the upgrade on any system running Lustre 1.4.6 or
839 newer, since /proc/sys/lnet was added in that version.
840 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
844 Frequency : only when quota is used
846 Description: avoid scanning export list for quota master
847 Details : Change the algorithms to avoid scanning export list in order
848 to improve the efficiency.
851 Frequency : MDS failover only, very rarely
853 Description: "went back in time" messages on mds failover
854 Details : The greatest transno may be lost when the current operation
855 finishes with an error (transno==0) and the client's last_rcvd
856 record is over-written. Save the greatest transno in the
857 mds_last_transno for this case.
860 Frequency : always for specific kernels and striping counts
862 Description: client may get "Matching packet too big" without ACL support
863 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
864 when trying to access files in certain configurations. The
865 clients should in fact be denied when mounting because they do
869 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
871 Description: many clients connecting with IO in progress causes connect timeouts
872 Details : Avoid synchronous journal commits to avoid delays caused by many
873 clients connecting/disconnecting when bulk IO is in progress.
874 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
875 OST_IO_PORTAL to avoid delays behind potentially many pending
879 Frequency : occasionally with multiple writers to a single file
881 Description: shared writes to file may result in wrong size reported by stat()
882 Details : Allow growing of kms when extent lock is cancelled
885 Frequency : always with random mmap IO to multi-striped file
887 Description: mmap write might be lost if we are writing to a 'hole' in stripe
888 Details : Only if the hole is at the end of OST object so that kms is too
889 small. Fix is to increase kms accordingly in ll_nopage.
892 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
894 Description: writes to a missing object would leak memory on the OST
895 Details : If there is an inconsistency between the MDS and OST filesystems,
896 such that the MDS references an object that doesn't exist, writes
897 to that object will leak memory due to incorrect cleanup in the
898 error handling path, eventually running out of memory on the OST.
903 Description: Creating too long symlink causes lustre errors
904 Details : Check symlink and name lengths before sending requests to MDS.
907 Frequency : only if flock is enabled (not on by default)
909 Description: posix locks not released on fd closure on 2.6.9+
910 Details : We failed to add posix locks to list of inode locks on 2.6.9+
911 kernels, this caused such locks not to be released on fd close and
912 then assertions on fs unmount about still used locks.
915 Frequency : MDS failover only, very rarely
917 Description: clients may get ASSERTION(granted_lock != NULL)
918 Details : When request was taking a long time, and a client was resending
919 a getattr by name lock request. The were multiple lock requests
920 with the same client lock handle and
921 mds_getattr_name->fixup_handle_for_resent_request found one of the
922 lock handles but later failed with ASSERTION(granted_lock != NULL).
927 Description: handle->h_buffer_credits > 0, assertion failure
928 Details : h_buffer_credits is zero after truncate, causing assertion
929 failure. This patch extends the transaction or creates a new
933 Frequency : NFS re-export or patchless client
934 Bugzilla : 11179, 10796
935 Description: Crash on NFS re-export node (__d_move)
936 Details : We do not want to hash the dentry if we don't have a lock.
937 But if this dentry is later used in d_move, we'd hit uninitialised
938 list head d_hash, so we just do this to init d_hash field but
939 leave dentry unhashed.
942 Frequency : NFS re-export or patchless client
944 Description: NFS exports has problem with symbolic link
945 Details : lustre client didn't properly install dentry when re-exported
946 to NFS or running patchless client.
949 Frequency : NFS re-export or patchless client
951 Description: Various nfs/patchless fixes.
952 Details : fixes reuse disconected alias for lookup process - this fixes
953 warning "find_exported_dentry: npd != pd",
954 fix permission error with open files at nfs.
955 fix apply umask when do revalidate.
958 Frequency : occasional
960 Description: Crash on NFS re-export node
961 Details : calling clear_page() on the wrong pointer triggered oops in
962 generic_mapping_read().
965 Frequency : rarely, using O_DIRECT IO
967 Description: unaligned directio crashes client with LASSERT
968 Details : check for unaligned buffers before trying any requests.
971 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
973 Description: stale data returned from RAID cache
974 Details : If only a small amount of IO is done to the RAID device before
975 reading it again it is possible to get stale data from the RAID
976 cache instead of reading it from disk.
979 Frequency : always for sles10 kernel
981 Description: sles10 support
982 Details : ll_follow_link: compile fixes and using of nd_set_link
986 Frequency : depends on arch, kernel and compiler version, always on sles10
989 Description: recursive or deep enough symlinks cause stack overflow
990 Details : getting rid of large stack-allocated variable in
994 Frequency : depends on hardware
996 Description: lustre write performance loss in the SLES10 kernel
997 Details : the performance loss is caused by using of write barriers in the
998 ext3 code. The SLES10 kernel turns barrier support on by
999 default. The fix is to undo that change for ldiskfs.
1001 ------------------------------------------------------------------------------
1003 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1005 * Support for kernels:
1006 2.6.9-42.0.3EL (RHEL 4)
1007 2.6.5-7.276 (SLES 9)
1008 2.4.21-47.0.1.EL (RHEL 3)
1009 2.6.12.6 vanilla (kernel.org)
1013 Frequency : quota enabled and large files being deleted
1015 Description: releasing more than 4GB of quota at once hangs OST
1016 Details : If a user deletes more than 4GB of files on a single OST it
1017 will cause the OST to spin in an infinite loop. Release
1018 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1023 Description: statfs data retrieved from /proc may be stale or zero
1024 Details : When reading per-device statfs data from /proc, in the
1025 {kbytes,files}_{total,free,avail} files, it may appear
1026 as zero or be out of date.
1029 Frequency : systems with MD RAID1 external journal devices
1031 Description: lconf's call to blkid is confused by RAID1 journal devices
1032 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1033 of returning all block devices that match the journal UUID.
1036 Frequency : always, for aggregate stripe size over 4GB
1038 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1039 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1040 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1043 Frequency : always if "lfs find" used on a local file/directory
1045 Description: "lfs find" segfaults if used on a local file/directory
1046 Details : The case where a directory component was not specified wasn't
1047 handled correctly. Handle this properly.
1050 Frequency : always on ppc64
1052 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1053 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1054 perform endianness conversion.
1057 Frequency : rarely (truncate to non-zero file size after write under load)
1058 Bugzilla : 10730, 10687
1059 Description: Files padded with zeros to next 4K multiple
1060 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1061 it is possible that files that are truncated to a non-zero size
1062 immediately after being written are filled with zero bytes beyond
1063 the truncated size. No file data is lost.
1065 Severity : enhancement
1067 Description: Allow recovery/failover for liblustre clients.
1068 Details : liblustre clients were unaware of failover configurations until
1071 Severity : enhancement
1073 Description: user file locks should fail when not mounting with flock option
1074 Details : Set up an error-returning stub in ll_file_operations.lock field
1075 to prevent incorrect behaviour when client is mounted without
1076 flock option. Also, set up properly f_op->flock field for
1080 Frequency : always on ia64
1082 Description: "lfs df" loops on printing out MDS statfs information
1083 Details : The obd_ioctl_data was not initialized and in some systems
1084 this caused a failure during the ioctl that did not return
1085 an error. Initialize the struct and return an error on failure.
1088 Frequency : SLES 9 only
1090 Description: Error of copying files with lustre special EAs as root
1091 Details : Client side always return success for setxattr call for lustre
1092 special xattr (currently only "trusted.lov").
1095 Frequency : rarely on clusters with both ia64+i386 clients
1097 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1098 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1099 clients are doing concurrent writes to the same file it is
1100 possible that the ia64 clients may LASSERT because the OST
1101 extent locks are not PAGE_SIZE aligned. Ensure that grown
1102 locks are always aligned on the request boundary.
1105 Frequency : specific use, occasional
1107 Description: Overwriting in use executable truncates on-disk binary image
1108 Details : If one node attempts to overwrite an executable in use by
1109 another node, we now correctly return ETXTBSY instead of
1110 truncating the file.
1112 Severity : enhancement
1114 Description: Async OSC create to avoid the blocking unnecessarily.
1115 Details : If a OST has no remain object, system will block on the creating
1116 when need to create a new object on this OST. Now, ways use
1117 pre-created objects when available, instead of blocking on an
1118 empty osc while others are not empty. If we must block, we block
1119 for the shortest possible period of time.
1124 Description: chmod on Lustre root is propagated to other clients
1125 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1126 invalid by the follow_mount time.
1129 Frequency : liblustre clients only
1131 Description: Race in 'instant cancel' lock handling could lead to such locks
1132 never to be granted in case of SMP MDS
1133 Details : Do not destroy not yet granted but cbpending locks in
1137 Frequency : replay/resend of open
1139 Description: non null lock assetion failure in mds_intent_policy
1140 Details : Trying to replay/resend lockless open requests resulted in
1141 mds_open() returning 0 with no lock. Now it sets a flag if
1142 a lock is going to be returned.
1144 Severity : enhancement
1146 Description: Checksum enhancements
1147 Details : New checksum enhancements allow for resending RPCs that failed
1150 Severity : enhancement
1152 Description: Tunables on number of dirty pages in cacche
1153 Details : Allow to set limit on number of dirty pages cached.
1158 Description: client crash on unmount - lock still has references
1159 Details : In some error handling cases it was possible to leak a lock
1160 reference on a client while accessing a file. This was not
1161 harmful to the client during operation, but would cause the
1162 client to crash when the filesystem is unmounted.
1165 Frequency : specific case, rare
1167 Description: ETXTBSY on mds though file not in use by client
1168 Details : ETXTBSY is no longer incorrectly returned when attempting to
1169 chmod or chown a directory that the user previously tried to
1170 execute or a currently-executing binary.
1173 Frequency : extremely rare except on liblustre-based clients
1175 Description: Lustre space not freed when files are deleted
1176 Details : Clean up open-unlinked files after client eviction. Previously
1177 the unlink was skipped and the files remained as orphans.
1182 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1183 Details : In some cases it was possible to send a blocking callback to
1184 a client doing a glimpse, even though that client didn't get
1185 a lock granted. When the glimpse lock is cancelled on the OST
1186 the freed lock is left on the waiting list and corrupted the list.
1189 Frequency : all core dumps
1191 Description: Broke core dumps to lustre
1192 Details : Negative dentry may be unhashed if parent does not have UPDATE
1193 lock, but some callers, e.g. do_coredump, expect dentry to be
1194 hashed after successful create, hash it in ll_create_it.
1196 ------------------------------------------------------------------------------
1198 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1200 * Support for kernels:
1201 2.6.9-42.0.2.EL (RHEL 4)
1202 2.6.5-7.276 (SLES 9)
1203 2.4.21-40.EL (RHEL 3)
1204 2.6.12.6 vanilla (kernel.org)
1208 Frequency : always on RHEL 3
1210 Description: Number of open files grows over time
1211 Details : The number of open files grows over time, whether or not
1212 Lustre is started. This was due to a filp leak introduced
1213 by one of our kernel patches.
1215 ------------------------------------------------------------------------------
1217 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1219 * Support for kernels:
1220 2.6.9-42.EL (RHEL 4)
1221 2.6.5-7.276 (SLES 9)
1222 2.4.21-40.EL (RHEL 3)
1223 2.6.12.6 vanilla (kernel.org)
1228 Bugzilla : 5719, 9635, 9792, 9684
1229 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1230 Details : If a server is under heavy load and cannot reply to new
1231 connection requests before the client resends the (re)connect,
1232 the connection handling code can behave badly if two service
1233 threads are concurrently handing separate (re)connections from
1234 the same client. Add better locking to the connection handling
1235 code, and ensure that only a single connection will be processed
1236 for a given client UUID, even if the lock is dropped.
1238 Severity : enhancement
1240 Description: add TCP zero-copy support to kernel
1241 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1242 sends if the hardware supports scatter-gather and checksumming.
1243 This allows socklnd to do client-write and server-read more
1244 efficiently and reduce CPU utilization from skbuf copying.
1247 Frequency : only if NFS exporting from client
1249 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1250 Details : If setting attributes on a file created under NFS that had
1251 never been opened it would be possible to oops the client
1252 if the file had no objects.
1255 Frequency : always for liblustre
1257 Description: liblustre client does MDS+OSTs setattr RPC for each write
1258 Details : When doing a write from a liblustre client, the client
1259 incorrectly issued an RPC to the MDS and each OST the file was
1260 striped over in order to update the timestamps. When writing
1261 with small chunks and many clients this could overwhelm the MDS
1262 with RPCs. In all cases it would slow down the write because
1263 these RPCs are unnecessary.
1265 Severity : enhancement
1267 Description: allow number of MDS service threads to be changed at module load
1268 Details : It is now possible to change the number of MDS service threads
1269 running. Adding "options mds mds_num_threads={N}" to the MDS's
1270 /etc/modprobe.conf will set the number of threads for the next
1271 time Lustre is restarted (assuming the "mds" module is also
1272 reloaded at that time). The default number of threads will
1273 stay the same, 32 for most systems.
1278 Description: OST crash if filesystem is unformatted or corrupt
1279 Details : If an OST is started on a device that has never been formatted
1280 or if the filesystem is corrupt and cannot even mount then the
1281 error handling cleanup routines would dereference a NULL pointer.
1286 Description: NULL pointer deref in llap_from_page.
1287 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1288 incorrect) mapping if the page was truncated/reclaimed while it was
1289 searched for. Check for this condition and skip such pages when
1290 doing readahead. Introduce extra check to llap_from_page() to
1291 verify page->mapping->host is non-NULL (so page is not anonymous).
1294 Frequency : Sometimes when using sys_sendfile
1296 Description: "page not covered by a lock" warnings from ll_readpage
1297 Details : sendfile called ll_readpage without right page locks present.
1298 Now we introduced ll_file_sendfile that does necessary locking
1299 around call to generic_file_sendfile() much like we do in
1303 Frequency : with certain MDS communication failures at client mount time
1305 Description: NULL pointer deref after failed client mount
1306 Details : a client connection request may delayed by the network layer
1307 and not be sent until after the PTLRPC layer has timed out the
1308 request. If the client fails the mount immediately it will try
1309 to clean up before the network times out the request. Add a
1310 reference from the request import to the obd device and delay
1311 the cleanup until the network drops the request.
1314 Frequency : occasionally during client (re)connect
1316 Description: assertion failure during client (re)connect
1317 Details : processing a client connection request may be delayed by the
1318 client or server longer than the client connect timeout. This
1319 causes the client to resend the connection request. If the
1320 original connection request is replied in this interval, the
1321 client may trip an assertion failure in ptlrpc_connect_interpret()
1322 which thought it would be the only running connect process.
1325 Frequency : only with obd_echo servers and clients that are rebooted
1327 Description: kernel BUG accessing uninitialized data structure
1328 Details : When running an obd_echo server it did not start the ping_evictor
1329 thread, and when a client was evicted an uninitialized data
1330 structure was accessed. Start the ping_evictor in the RPC
1331 service startup instead of the OBD startup.
1333 Severity : enhancement
1334 Bugzilla : 10193 (patchless)
1335 Description: Remove dependency on various unexported kernel interfaces.
1336 Details : No longer need reparent_to_init, exit_mm, exit_files,
1337 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1340 Frequency : rare (only users of deprecated and unsupported LDAP config)
1342 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1343 Details : LDAP apparently contains 'lustreName' attributes instead of
1344 'name'. A simple remapping of the name is sufficient.
1347 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1349 Description: waiting_locks_callback trips kernel BUG if client is evicted
1350 Details : Running with the dump_on_timeout debug flag turned on makes
1351 it possible that the waiting_locks_callback() can try to dump
1352 the Lustre kernel debug logs from an interrupt handler. Defer
1353 this log dumping to the expired_lock_main() thread.
1355 Severity : enhancement
1357 Description: Support NFS exporting on 2.6 kernels.
1358 Details : Implement non-rawops metadata methods for NFS server to use without
1359 changing NFS server code.
1362 Frequency : very rare (synthetic metadata workload only)
1364 Description: two racing renames might cause an MDS thread to deadlock
1365 Details : Running the "racer" program may cause one MDS thread to rename
1366 a file from being the source of a rename to being the target of
1367 a rename at exactly the same time that another thread is doing
1368 so, and the second thread has already enqueued these locks after
1369 doing a lookup of the target and is trying to relock them in
1370 order. Ensure that we don't try to re-lock the same resource.
1373 Frequency : only very large systems with liblustre clients
1375 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1376 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1377 all OSTs in parallel. This allows the eviction of stale liblustre
1378 clients to proceed much faster than if they were done in series,
1379 and also offers similar improvements for other set_info RPCs.
1384 Description: excessive CPU usage during initial read phase on client
1385 Details : During the initial read phase on a client, it would agressively
1386 retry readahead on the file, consuming too much CPU and impacting
1387 performance (since 1.4.5.8). Improve the readahead algorithm
1388 to avoid this, and also improve some other common cases (read
1389 of small files in particular, where "small" is files smaller than
1390 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1395 Description: MDS crash when receiving packet with unknown intent.
1396 Details : Do not LBUG in unknown intent case, just return -EFAULT
1398 Severity : enhancement
1399 Bugzilla : 9293, 9385
1400 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1401 Details : Do not serialize getattr (non-intent version) and statfs.
1404 Frequency : occasional, when OST network is overloaded/intermittent
1406 Description: client evicted by OST after bulk IO timeout
1407 Details : If a client sends a bulk IO request (read or write) the OST
1408 may evict the client if it is unresposive to its data GET/PUT
1409 request. This is incorrect if the network is overloaded (takes
1410 too long to transfer the RPC data) or dropped the OST GET/PUT
1411 request. There is no need to evict the client at all, since
1412 the pinger and/or lock callbacks will handle this, and the
1413 client can restart the bulk request.
1416 Frequency : Always when mmapping file with no objects
1418 Description: client crashes when mmapping file with no objects
1419 Details : Check that we actually have objects in a file before doing any
1420 operations on objects in ll_vm_open, ll_vm_close and
1426 Description: Request leak when working with deleted CWD
1427 Details : Introduce advanced request refcount tracking for requests
1428 referenced from lustre intent.
1430 Severity : Enhancement
1432 Description: Cache open file handles on client.
1433 Details : MDS now will return special lock along with openhandle, if
1434 requested and client is allowed to hold openhandle, even if unused,
1435 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1436 closing files for every read/write openration.
1438 Severity : Enhancement
1440 Description: Cache open negative dentries on client when possible.
1441 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1442 negative dentries on lock revocation.
1447 Description: Remounting a client read-only wasn't possible with a zconf mount
1448 Details : It wasn't possible to remount a client read-only with llmount.
1450 Severity : enhancement
1451 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1452 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1453 MPICH is included to add Lustre-specific ADIO interfaces.
1454 This is based closely on the UFS ADIO layer and only differs
1455 in file creation, in order to allow the OST striping to be set.
1456 This is user-contributed code and not supported by CFS.
1461 Description: extended inode attributes (immutable, append-only) work improperly
1462 when 2.4 and 2.6 kernels are used on client/server or vice versa
1463 Details : Introduce kernel-independent values for these flags.
1465 Severity : enhancement
1468 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1469 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1470 granularity. Now, Fractional MB granularity is be supported,
1471 this is very useful for low memory system.
1473 Severity : enhancement
1475 Description: Getattr by fid
1476 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1477 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1478 have locks and avoids getattr rpc requests for it on every stat.
1481 Frequency : Always, for filesystems larger than 2TB
1483 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1484 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1485 filesystems larger than 2TB to be mounted with mballoc enabled.
1488 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1490 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1491 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1492 filesystems larger than 2TB is mounted then the kernel will
1493 silently corrupt the start of the filesystem. CONFIG_LBD is
1494 enabled for all CFS-supported kernels, but the possibility of
1495 this happening with a modified kernel config exists.
1497 Severity : enhancement
1499 Description: add client O_DIRECT support for 2.6 kernels
1500 Details : It is now possible to do O_DIRECT reads and writes to files
1501 in the Lustre client mountpoint on 2.6 kernel clients.
1503 Severity : enhancement
1505 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1506 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1507 parallel, not waiting for response from every OST before sending
1508 a rpc to the next OST.
1513 Description: setattr vs write race when updating file timestamps
1514 Details : Client processes that update a file timestamp into the past
1515 right after writing to the file (e.g. tar) it is possible that
1516 the updated file modification time can be reset to the current
1517 time due to a race between processing the setattr and write RPC.
1519 Severity : enhancement
1521 Description: Bring 'lfs find' closer in line with regular Linux find.
1522 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1523 -print0 options and obtains all the needed info through the lustre
1526 Severity : enhancement
1528 Description: support up to 1024 configured devices on one node
1529 Details : change obd_dev array from statically allocated to dynamically
1530 allocated structs as they are first used to reduce memory usage
1535 Description: Flush dirty partially truncated pages during truncate
1536 Details : Immediatelly flush partially truncated pages in filter_setattr,
1537 this way we completely avoid having any pages in page cache on OST
1538 and can retire ugly workarounds during writes to flush such pages.
1543 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1544 Details : protect inode from truncation within vfs_unlink() context
1545 just take a reference before calling vfs_unlink() and release it
1546 when parent's i_sem is free.
1551 Description: last_id value checked outside lock on OST caused LASSERT failure
1552 Details : If there were multiple MDS->OST object precreate requests in
1553 flight, it was possible that the OST's last object id was checked
1554 outside a lock and incorrectly tripped an assertion. Move checks
1555 inside locks, and discard old precreate requests.
1558 Frequency : always, if extents are used on OSTs
1560 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1561 Details : OSTs using the extents format would not zero the high 16 bits of
1562 the index physical block number. This is not a problem for any
1563 OST filesystems smaller than 16TB, and no kernels support ext3
1564 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1565 new/modified files) and can be fixed for existing filesystems
1566 with e2fsprogs-1.39-cfs1.
1571 Description: import connection selection may be incorrect if timer wraps
1572 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1573 import connections to be ignored if the 32-bit jiffies counter
1574 wraps. Use a 64-bit jiffies counter.
1577 Frequency : very large clusters immediately after boot
1579 Description: LNET request buffers exhausted under heavy short-term load
1580 Details : If a large number of client requests are generated on a service
1581 that has previously never seen so many requests it is possible
1582 that the request buffer growth cannot keep up with the spike in
1583 demand. Instead of dropping incoming requests, they are held in
1584 the LND until the RPC service can accept more requests.
1587 Frequency : Sometimes during replay
1589 Description: Assertion failure in ll_local_open after replay.
1590 Details : If replay happened on an open request reply before we were able
1591 to set replay handler, reply will become not swabbed tripping the
1592 assertion in ll_local_open. Now we set the handler right after
1593 recognising of open request
1596 Frequency : very rare
1598 Description: kernel reports "badness in vsnprintf"
1599 Details : Reading from the "recovery_status" /proc file in small chunks
1600 may cause a negative length in lprocfs_obd_rd_recovery_status()
1601 call to vsnprintf() (which is otherwise harmless). Exit early
1602 if there is no more space in the output buffer.
1604 Severity : enhancement
1606 Description: clear OBD RPC statistics by writing to them
1607 Details : It is now possible to clear the OBD RPC statistics by writing
1608 to the "stats" file.
1613 Description: Client mtime is not the same on different clients after utimes
1614 Details : In some cases, the client was using the utimes() syscall on
1615 a file cached on another node. The clients now validate the
1616 ctime from the MDS + OSTs to determine which one is right.
1621 Description: Inability to activate failout mode
1622 Details : lconf script incorrectly assumed that in pythong string's numeric
1623 value is used in comparisons.
1626 Frequency : always with multiple stripes per file
1628 Description: Inefficient object allocation for mutli-stripe files
1629 Details : When selecting which OSTs to stripe files over, for files with
1630 a stripe count that divides evenly into the number of OSTs,
1631 the MDS is always picking the same starting OST for each file.
1632 Return the OST selection heuristic to the original design.
1637 Description: mount failures may take full timeout to return an error
1638 Details : Under some heavy load conditions it is possible that a
1639 failed mount can wait for the full obd_timeout interval,
1640 possibly several minutes, before reporting an error.
1641 Instead return an error as soon as the status is known.
1643 Frequency : quota enabled and large files being deleted
1645 Description: releasing more than 4GB of quota at once hangs OST
1646 Details : If a user deletes more than 4GB of files on a single OST it
1647 will cause the OST to spin in an infinite loop. Release
1648 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1653 Description: statfs data retrieved from /proc may be stale or zero
1654 Details : When reading per-device statfs data from /proc, in the
1655 {kbytes,files}_{total,free,avail} files, it may appear
1656 as zero or be out of date.
1659 Frequency : systems with MD RAID1 external journal devices
1661 Description: lconf's call to blkid is confused by RAID1 journal devices
1662 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1663 of returning all block devices that match the journal UUID.
1666 Frequency : always, for aggregate stripe size over 4GB
1668 Description: assertion fails when trying to use 4GB stripe size
1669 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
1670 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
1673 Frequency : always on ppc64
1675 Description: the first write on an ext3 filesystem with mballoc got stuck
1676 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1677 perform endianness conversion.
1679 ------------------------------------------------------------------------------
1681 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1683 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1684 INTEROPERATE with older versions automatically. Please read the
1685 user documentation before upgrading any part of a live system.
1686 * WARNING: Lustre networking configuration changes are required with
1687 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1690 * Support for kernels:
1691 2.6.9-22.0.2.EL (RHEL 4)
1692 2.6.5-7.244 (SLES 9)
1693 2.6.12.6 vanilla (kernel.org)
1696 Severity : enhancement
1697 Bugzilla : 7981/8208
1698 Description: Introduced Lustre Networking (LNET)
1699 Details : LNET is new networking infrastructure for Lustre, it includes
1700 a reorganized network configuration mode (see the user
1701 documentation for full details) as well as support for routing
1702 between different network fabrics. Lustre Networking Devices
1703 (LNDs) for the supported network fabrics have also been
1704 created for this new infrastructure.
1706 Severity : enhancement
1707 Description: Introduced Access control lists
1708 Details : clients can set ACLs on files and directories in order to have
1709 more fine-grained permissions than the standard Unix UGO+RWX.
1710 The MDS must be started with the "-o acl" mount option.
1712 Severity : enhancement
1713 Description: Introduced filesystem quotas
1714 Details : Administrators may now establish per-user quotas on the
1717 Severity : enhancement
1719 Description: Configuration change for the XT3
1720 The PTLLND is now used to run Lustre over Portals on the XT3
1721 The configure option(s) --with-cray-portals are no longer used.
1722 Rather --with-portals=<path-to-portals-includes> is used to
1723 enable building on the XT3. In addition to enable XT3 specific
1724 features the option --enable-cray-xt3 must be used.
1729 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1730 Details : A race between allocating a new llcd and re-getting the llcd_lock
1731 allowed another thread to grab newly-allocated llcd.
1733 Severity : enhancement
1735 Description: 2.6 OST async journal commit and locking fix to improve performance
1736 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1737 2.6 kernels are now async as they already were in 2.4 kernels so
1738 that they can commit concurrently with the network bulk transfer.
1739 For block-allocated files the filter allocation semaphore is held
1740 to avoid filesystem fragmentation during allocation. BKL lock
1741 removed for 2.6 xattr operations where it is no longer needed.
1746 Description: lconf incorrectly determined whether two IP networks could talk
1747 Details : In some more complicated routing and multiple-network
1748 configurations, lconf will avoid trying to make a network
1749 connection to a disjoint part of the IP space. It was doing the
1750 math incorrectly for one set of cases.
1755 Description: Fix for potential infinite loop processing records in an llog.
1756 Details : If an llog record is corrupted/zeroed, it is possible to loop
1757 forever in llog_process(). Validate the llog record length
1758 and skip the remainder of the block on error.
1761 Frequency : occasional (liblustre only)
1763 Description: liblustre could not open files whose last component is a symlink
1764 Details : sysio_path_walk() would incorrectly pass the open intent to
1765 intermediate path components.
1768 Frequency : rare (liblustre only with non-standard tuning)
1769 Bugzilla : 7201 (7350)
1770 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1771 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1772 after it has been released, only do this for referenced locks
1774 Severity : enhancement
1776 Description: specifying an (invalid) directory default stripe_size of -1
1777 would reset the directory default striping
1778 Details : stripe_size -1 was used internally to signal directory stripe
1779 removal, now use "all default" to signal dir stripe removal
1780 as a directory striping of "all default" is not useful
1783 Frequency : common for large clusters running liblustre clients
1785 Description: doing an ls when liblustre clients are running is slow
1786 Details : sending a glimpse AST to a liblustre client waits for every AST
1787 to time out, as liblustre clients will not respond. Since they
1788 cannot cache data we refresh the OST lock LVB from disk instead.
1790 Severity : enhancement
1792 Description: doing an ls at the same time as file IO can be slow
1793 Details : enqueue and other "small" requests can be blocked behind many
1794 large IO requests. Create a new OST IO portal for non-IO
1795 requests so they can be processed faster.
1798 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1800 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1801 created file with mode 000
1802 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1803 allow an ftruncate() as a non-root user to a file with mode 000.
1804 Lustre now respects this flag to disable mode checks when
1805 truncating a file owned by the user
1808 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1811 Description: Revoking locks from clients that went dead or catatonic might take
1813 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1814 cancellation of such locks instant on servers without waiting for
1815 any reply from clients. Clients drops these locks when cancel
1816 notification from server is received without replying.
1819 Frequency : liblustre-only, when liblustre client dies or becomes busy
1821 Description: Doing ls on Linux clients can take a long time with active
1823 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1824 granting such locks to it in the first place if possible. Locks
1825 are taken by proxy on the OST during the read or write and
1826 dropped immediately afterward. Add connect flags handling, do
1827 not grant locks to liblustre clients for glimpse ASTs.
1829 Severity : enhancement
1831 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1832 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1833 sequential reads and behaves badly with applications doing random
1834 reads. Improve it by reading ahead at least read region, and
1835 avoiding excessive large RPC for small reads.
1837 Severity : enhancement
1839 Description: Creating more than 1000 files for a single job may cause a load
1840 imbalance on the OSTs if there are also a large number of OSTs.
1841 Details : qos_prep_create() uses an OST index reseed value that is an
1842 even multiple of the number of available OSTs so that if the
1843 reseed happens in the middle of the object allocation it will
1844 still utilize the OSTs as uniformly as possible.
1849 Description: OST or MDS may oops in ping_evictor_main()
1850 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1851 but doesn't restart at beginning of obd_exports_timed list
1854 Severity : enhancement
1856 Description: improve by-nid export eviction on the MDS and OST
1857 Details : allow multiple exports with the same NID to be evicted at one
1858 time without re-searching the exports list.
1861 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1863 Description: MDS may oops in groups_free()
1864 Details : in rare race conditions a newly allocated group_info struct is
1865 freed again, and this can be NULL. The 2.4 compatibility code
1866 for groups_free() checked for a NULL pointer, but 2.6 did not.
1869 Frequency : common for liblustre clients doing little filesystem IO
1870 Bugzilla : 9352, 7313
1871 Description: server may evict liblustre clients accessing contended locks
1872 Details : if a client is granted a lock or receives a completion AST
1873 with a blocking AST already set it would not reply to the AST
1874 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1877 Frequency : lfs setstripe, only systems with more than 160 OSTs
1879 Description: unable to set striping with a starting offset beyond OST 160
1880 Details : llapi_create_file() incorrectly limited the starting stripe
1881 index to the maximum single-file stripe count.
1884 Frequency : LDAP users only
1886 Description: lconf did not handle in-kernel recovery with LDAP properly
1887 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1889 Severity : enhancement
1891 Description: bind OST threads to NUMA nodes to improve performance
1892 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1893 node and do their allocations there to localize memory access
1895 Severity : enhancement
1897 Description: llmount can determine client NID directly from Myrinet (GM)
1898 Details : the client NID code from gmnalnid was moved directly into
1899 llmount, removing the need to use this or specifying the
1900 client NID explicitly when mounting GM clients with zeroconf
1903 Frequency : if client is started with down MDS
1905 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1906 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1907 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1908 ptlrpc_import_delayed_req() because we didn't check if the
1909 request was interrupted, and we also didn't break out of the
1910 event loop if there was no timeout
1915 Description: data loss during non-page-aligned writes to a single file from
1916 both multiple nodes and multiple threads on one node at same time
1917 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1918 inconsistency led to false short-reads, that were cached and later
1919 used by ->prepare_write() to fill in partially written page,
1920 leading to data loss.
1923 Frequency : always, if lconf --abort_recovery used
1925 Description: lconf --abort_recovery fails with 'Operation not supported'
1926 Details : lconf was attempting to abort recovery on the MDT device and not
1929 Severity : enhancement
1931 Description: remove cleanup logs
1932 Details : replace lconf-generated cleanup logs with lustre internal
1933 cleanup routines. Eliminates the need for client-cleanup and
1936 Severity : enhancement
1938 Description: add support for EAs (user and system) on lustre filesystems
1939 Details : it is now possible to store extended attributes in the Lustre
1940 client filesystem, and with the user_xattr mount option it
1941 is possible to allow users to store EAs on their files also
1943 Severity : enhancement
1945 Description: Add possibility (config option) to show minimal available OST free
1947 Details : When compiled with --enable-mindf configure option, statfs(2)
1948 (and so, df) will return least minimal free space available from
1949 all OSTs as amount of free space on FS, instead of summary of
1950 free spaces of all OSTs.
1952 Severity : enhancement
1954 Description: do not expand extent locks acquired on OST-side
1955 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1956 by server: they are not cached anyway.
1959 Frequency : when mmap is used/binaries executed from Lustre
1961 Description: Unmmap pages before throwing them away from read cache.
1962 Details : llap_shrink cache now attempts to unmap pages before discarding
1963 them (if unmapping failed - do not discard). SLES9 kernel has
1964 extra checks that trigger if this unmapping is not done first.
1969 Description: lconf didn't resolve symlinks before checking to see whether a
1970 given mountpoint was already in use
1973 Frequency : when migrating failover services
1974 Bugzilla : 6395, 9514
1975 Description: When migrating a subset of services from a node (e.g. failback
1976 from a failover service node) the remaining services would
1977 time out and evict clients.
1978 Details : lconf --force (implied by --failover) sets the global obd_timeout
1979 to 5 seconds in order to quickly disconnect, but this caused
1980 other RPCs to time out too quickly. Do not change the global
1981 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1983 Severity : enhancement
1984 Frequency : if MDS is started with down OST
1985 Bugzilla : 9439,5706
1986 Description: Allow startup/shutdown of an MDS without depending on the
1987 availability of the OSTs.
1988 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1989 Add appropriate locking and lov-osc refcounts for safe
1990 cleaning. Add osc abort_inflight calls in case the
1991 synchronize never started.
1994 Frequency : occasional (Cray XT3 only)
1996 Description: root not authorized to access files in CRAY_PORTALS environment
1997 Details : The client process capabilities were not honoured on the MDS in
1998 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1999 been accessed by an authorized user then root was able to access
2000 the file on the local client also. The root user capabilities
2001 are now allowed on the MDS, as this environment has secure UID.
2004 Frequency : occasional
2006 Description: ldiskfs "too long searching" message happens too often
2007 Details : A debugging message (otherwise harmless) prints too often on
2008 the OST console. This has been reduced to only happen when
2009 there are fragmentation problems on the filesystem.
2014 Description: Division by zero in statfs when all OSCs are inactive
2015 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2016 lov_statfs divides by value returned by lov_get_stripecnt().
2020 Bugzilla : 9489, 3273
2021 Description: First write from each client to each OST was only 4kB in size,
2022 to initialize client writeback cache, which caused sub-optimal
2023 RPCs and poor layout on disk for the first writen file.
2024 Details : Clients now request an initial cache grant at (re)connect time
2025 and so that they can start streaming writes to the cache right
2026 away and always do full-sized RPCs if there is enough data.
2027 If the OST is rebooted the client also re-establishes its grant
2028 so that client cached writes will be honoured under the grant.
2033 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2034 Details : Now I/O RPCs go to different portal number and (presumably) fast
2035 lock requests (and glimses) and other RPCs get their own service
2036 threads pool that should be able to service those RPCs
2039 Severity : enhancement
2041 Description: Ability to exchange lustre version between client and servers and
2042 issue warnings at client side if client is too old. Also for
2043 liblustre clients there is ability to refuse connection of too old
2045 Details : New 'version' field is added to connect data structure that is
2046 filled with version info. That info is later checked by server and
2050 Frequency : rare, liblustre only.
2051 Bugzilla : 9296, 9581
2052 Description: Two simultaneous writes from liblustre at offset within same page
2053 might proceed at the same time overwriting eachother with stale
2055 Details : I/O lock withing llu_file_prwv was released too early, before data
2056 actually was hitting the wire. Extended lock-holding time until
2057 server acknowledges receiving data.
2060 Frequency : extremely rare. Never observed in practice.
2062 Description: avoid generating lustre_handle cookie of 0.
2063 Details : class_handle_hash() generates handle cookies by incrementing
2064 global counter, and can hit 0 occasionaly (this is unlikely, but
2065 not impossible, because initial value of cookie counter is
2066 selected randonly). Value of 0 is used as a sentinel meaning
2067 "unassigned handle" --- avoid it. Also coalesce two critical
2068 sections in this function into one.
2070 Severity : enhancement
2072 Description: allow liblustre clients to delegate truncate locking to OST
2073 Details : To avoid overhead of locking, liblustre client instructs OST to
2074 take extent lock in ost_punch() on client's behalf. New connection
2075 flag is added to handle backward compatibility.
2077 Severity : enhancement
2078 Bugzilla : 4928, 7341, 9758
2079 Description: allow number of OST service threads to be specified
2080 Details : a module parameter allows the number of OST service threads
2081 to be specified via "options ost ost_num_threads={N}" in the
2082 OSS's /etc/modules.conf or /etc/modprobe.conf.
2086 Bugzilla : 6146, 9635, 9895
2087 Description: servers crash with bad pointer in target_handle_connect()
2088 Details : In rare cases when a client is reconnecting it was possible that
2089 the connection request was the last reference for that export.
2090 We would temporarily drop the export reference and get a new
2091 one, but this may have been the last reference and the export
2092 was just destroyed. Get new reference before dropping old one.
2094 Severity : enhancement
2095 Frequency : if client is started with failover MDS
2097 Description: Allow multiple MDS hostnames in the mount command
2098 Details : Try to read the configuration from all specified MDS
2099 hostnames during a client mount in case the "primary"
2102 Severity : enhancement
2104 Description: Stop sending data to evicted clients as soon as possible.
2105 Details : Check if the client we are about to send or are sending data to
2106 was evicted already. (Check is done every second of waiting,
2107 for which l_wait_event interface was extended to allow checking
2108 of exit condition at specified intervals).
2111 Frequency : rare, normally only when NFS exporting is done from client
2113 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2115 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2118 Frequency : always, if srand() is called before liblustre initialization
2120 Description: Liblustre uses system PRNG disturbing its usage by user application
2121 Details : Introduce internal to lustre fast and high-quality PRNG for
2122 lustre usage and make liblustre and some other places in generic
2123 lustre code to use it.
2125 Severity : enhancement
2126 Bugzilla : 9477, 9557, 9870
2127 Description: Verify that the MDS configuration logs are updated when xml is
2128 Details : Check if the .xml configuration logs are newer than the config
2129 logs stored on the MDS and report an error if this is the case.
2130 Request --write-conf, or allow starting with --old_conf.
2132 Severity : enhancement
2134 Description: Handle symlinks in the path when checking if Lustre is mounted.
2135 Details : Resolve intermediate symlinks when checking if a client has
2136 mounted a filesystem to avoid duplicate client mounts.
2141 Description: lconf can hit an error exception but still return success.
2142 Details : The lconf command catches the Command error exception at the top
2143 level script context and will exit with the associated exit
2144 status, but doesn't ensure that this exit status is non-zero.
2149 Description: failure of ptlrpc thread startup can cause oops
2150 Details : Starting a ptlrpc service thread can fail if there are a large
2151 number of threads or the server memory is very fragmented.
2152 Handle this without oopsing.
2155 Frequency : always, only if liblustre and non-default acceptor port was used
2157 Description: liblustre cannot connect to servers with non-default acceptor port
2158 Details : tcpnal_set_default_params() was not called and was therefore
2159 ignoring the environment varaible TCPNAL_PORT, as well as other
2160 TCPNAL_ environment variables
2165 Description: two objects could be created on the same OST for a single file
2166 Details : If an OST is down, in some cases it was possible to create two
2167 objects on a single OST for a single file. No problems other
2168 than potential performance impact and spurious error messages.
2172 Bugzilla : 5681, 9562
2173 Description: Client may oops in ll_unhash_aliases
2174 Details : Client dcache may become inconsistent in race condition.
2175 In some cases "getcwd" can fail if the current directory is
2181 Description: Inode refcounting problems in NFS export code
2182 Details : link_raw functions used to call d_instantiate without obtaining
2183 extra inode reference first.
2187 Bugzilla : 9942, 9903
2188 Description: Referencing freed requests leading to crash, memleaks with NFS.
2189 Details : We used to require that call to ll_revalidate_it was always
2190 followed by ll_lookup_it. Also with revalidate_special() it is
2191 possible to call ll_revalidate_it() twice for the same dentry
2192 even if first occurence returned success. This fix changes semantic
2193 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2194 reference on a request referred from the intent.
2195 ll_intent_release() then releases such a request.
2198 Frequency : rare, normally benchmark loads only
2200 Description: unlinked inodes were kept in memory on the client
2201 Details : If a client is repeatedly creating and unlinking files it
2202 can accumulate a lot of stale inodes in the inode slab cache.
2203 If there is no other client load running this can cause the
2204 client node to run out of memory. Instead flush old inodes
2205 from client cache that have the same inode number as a new inode.
2208 Frequency : SLES9 2.6.5 kernel and long filenames only
2209 Bugzilla : 9969, 10379
2210 Description: utime reports stale NFS file handle
2211 Details : SLES9 uses out-of-dentry names in some cases, which confused
2212 the lustre dentry revalidation. Change it to always use the
2216 Frequency : rare, unless heavy write-truncate concurrency is continuous
2217 Bugzilla : 4180, 6984, 7171, 9963, 9331
2218 Description: OST becomes very slow and/or deadlocked during object unlink
2219 Details : filter_destroy() was holding onto the parent directory lock
2220 while truncating+unlinking objects. For very large objects this
2221 may block other threads for a long time and slow overall OST
2222 responsiveness. It may also be possible to get a lock ordering
2223 deadlock in this case, or run out of journal credits because of
2224 the combined truncate+unlink. Solution is to do object truncate
2225 first in one transaction without parent lock, and then do the
2226 final unlink in a new transaction with the parent lock. This
2227 reduces the lock hold time dramatically.
2230 Frequency : rare, 2.4 kernels only
2232 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2233 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2234 before unmounting their filesystem in order to avoid deadlock.
2235 The kernel_locked() function in 2.4 kernels only checks whether
2236 the kernel lock is held, not whether it is this process that is
2237 holding it as 2.6 kernels do.
2242 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2243 Details : The client ptlrpc code may be trying to reconnect to a down
2244 server before a previous connection attempt has timed out.
2245 Increase the reconnect interval to be longer than the connection
2246 timeout interval to avoid sending duplicate connections to
2250 Frequency : echo_client brw_test command
2252 Description: fix echo_client to work with OST preallocated code
2253 Details : OST preallocation code (5137) didn't take echo_client IO path
2254 into account: echo_client calls filter methods outside of any
2255 OST thread and, hence, there is no per-thread preallocated
2256 pages and buffers to use. Solution: hijack pga pages for IO. As
2257 a byproduct, this avoids unnecessary data copying.
2261 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2262 Description: Client can oops in mdc_commit_close() after open replay
2263 Details : It was possible for the MDS to return an open request with no
2264 transaction number in mds_finish_transno() if the client was
2265 evicted, but without actually returning an error. Clients
2266 would later try to replay that open and may trip an assertion
2267 Simplify the client close codepath, and always return an error
2268 from the MDS in case the open is not successful.
2271 Frequency : rare, 2.6 OSTs only
2273 Description: OST may deadlock under high load on fragmented files
2274 Details : If there was a heavy load and highly-fragmented OST filesystems
2275 it was possible to have all the OST threads deadlock waiting on
2276 allocation of biovecs, because the biovecs were not released
2277 until the entire RPC IO was completed. Instead, release biovecs
2278 as soon as they are complete to ensure forward IO progress.
2280 Severity : enhancement
2282 Description: Support for specifying external journal device at mount
2283 Details : If an OST or MDS device is formatted with an external journal
2284 device, this device major/minor is stored in the ext3 superblock
2285 and may not be valid for failover. Allow detecting and
2286 specifying the external journal at mount time.
2291 Description: Mounting an MDS with pending unlinked files may cause oops
2292 Details : target_finish_recovery() calls mds_postrecov() which returned
2293 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2294 considers this an error and immediately begins cleaning up the
2295 lov, just after starting the mds_lov process
2297 Severity : enhancement
2299 Description: Implement 'lfs df' to report actual free space on per-OST basis
2300 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2301 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2302 usage of objects; '-h' to report in human readable format.
2304 ------------------------------------------------------------------------------
2306 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2313 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2314 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2315 at startup during orphan unlinking. Instead, skip buddy update
2316 before setup, as it will be regenerated after recovery is complete.
2321 Description: If an OST is inactive, its locks might reference stale inodes.
2322 Details : lov_change_cbdata() must iterate over all namespaces, even if
2323 they are inactive to clear inode references from the lock.
2325 Severity : enhancement
2326 Frequency : occasional, if non-standard max_dirty_mb used
2328 Description: Client will block write RPCs if not enough grant
2329 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2330 then the client will block writes while waiting for another RPC
2331 to complete instead of consuming its dirty limit. With change
2332 we get improved performance when max_dirty_mb is small.
2334 Severity : enhancement
2335 Bugzilla : 3389, 6253
2336 Description: Add support for supplementary groups on the MDS.
2337 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2338 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2339 lookups for user supplementary groups into a cache.
2343 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2344 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2345 for liblustre O_EXCL is enforced by MDS.
2349 Description: readdir never returns NULL in liblustre.
2350 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2351 dentry in struct dirent.
2355 Frequency : liblustre only, depends on application IO pattern
2356 Description: liblustre clients evicted if not contacting servers
2357 Details : Don't put liblustre clients into the ping_evictor list, so
2358 they will not be evicted by the pinger ever.
2360 Severity : enhancement
2362 Description: Add ability to evict clients by NID from MDS.
2363 Details : By echoing "nid:$NID" string into
2364 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2365 $NID would be instantly evicted from this MDS and from all active
2366 OSTs connected to it.
2370 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2371 Details : lookup_it_finish() used to query file size from OSTs that was not
2376 Description: service threads change working directory to that of init
2377 Details : Starting lustre service threads may pin the working directory
2378 of the parent thread, making that filesystem busy. Threads
2379 now change to the working directory of init to avoid this.
2383 Frequency : during shutdown only
2384 Description: shutdown with a failed MDS or OST can cause unmount to hang
2385 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2390 Frequency : 2.6 only
2391 Description: chmod/chown may include an extra supplementary group
2392 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2393 supplementary group and if none is specified this is used.
2396 Bugzilla : 5479 (6816)
2398 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2399 Details : If lookup is in progress on a file that is unlinked we might try
2400 to revalidate the inode and fail in revalidate after lookup is
2401 complete and ll_file_open() enqueues the open again but
2402 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2405 Frequency : always, if lconf --abort_recovery used
2407 Description: lconf --abort_recovery fails with 'Operation not supported'
2408 Details : lconf was attempting to abort recovery on the MDT device and not
2411 ------------------------------------------------------------------------------
2413 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2418 Frequency : rare (only unsupported configurations with a node running as an
2420 Bugzilla : 6514, 5137
2421 Description: Mounting a Lustre file system on a node running as an OST could
2423 Details : OSTs now preallocates memory needed to write out data at
2424 startup, instead of when needed, to avoid having to
2425 allocate memory in possibly low memory situations.
2426 Specifically, if the file system is mounted on on OST,
2427 memory pressure could force it to try to write out data,
2428 which it needed to allocate memory to do. Due to the low
2429 memory, it would be unable to do so and the node would
2430 become unresponsive.
2432 Severity : enhancement
2434 Description: Addition of lconf --service command line option
2435 Details : lconf now accepts a '--service <arg>' option, which is
2436 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2438 Severity : enhancement
2440 Description: Failover mode is now the default for OSTs.
2441 Details : By default, OSTs will now run in failover mode. To return to
2442 the old behaviour, add '--failout' to the lmc line for OSTs.
2444 Severity : enhancement
2446 Description: Health checks are now provided for MDS and OSTs
2447 Details : Additional detailed health check information on MSD and OSTs
2448 is now provided through the procfs health_check value.
2451 Frequency : occasional, depends on IO load
2453 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2454 after numerous create/delete cycles
2455 Details : The ext3 inode allocation policy would not allocate new inodes
2456 very well on the OSTs because there are no new directories
2457 being created. Instead we look for groups with free space if
2458 the parent directories are nearly full.
2463 Description: Network or server problems during mount may cause partially
2464 mounted clients instead of returning an error.
2465 Details : The config llog parsing code may overwrite the error return
2466 code during mount error handling, returning success instead
2472 Description: MDS can fail to allocate large reply buffers
2473 Details : After long uptimes the MDS can fail to allocate large reply
2474 buffers (e.g. zconf client mount config records) due to memory
2475 fragmentation or consumption by the buffer cache. Preallocate
2476 some large reply buffers so that these replies can be sent even
2477 under memory pressure.
2481 Frequency : rare (liblustre)
2482 Description: fsx running with liblustre complained that using truncate() to
2483 extend the file doesn't work. This patch corrects that issue.
2484 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2485 ATTR_SIZE and lsm use in llu_setattr_raw.
2489 Frequency : rare, only 2.6 kernels
2490 Description: Unusual file access patterns on the MDS may result in inode
2491 data being lost in very rare circumstances.
2492 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2493 code in the 2.6 kernel caused the inode and/or EA data not to
2494 be read from disk, causing single-file corruption.
2498 Frequency : rare, only 2.6 filesystems using extents
2499 Description: Heavy concurrent write and delete load may cause data corruption.
2500 Details : It was possible under high-load situations to have an extent
2501 metadata block in the block device cache from a just-unlinked
2502 file overwrite a newly-allocated data block. We now unmap any
2503 metadata buffers that alias just-allocated data blocks.
2507 Frequency : filesystems with default stripe_count larger than 77
2508 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2509 Details : lconf specifies an inode size of 4096 bytes when the default
2510 stripe_count is larger than 77. This conflicts with the default
2511 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2512 this case to avoid pinning too much memory for large EAs.
2514 ------------------------------------------------------------------------------
2516 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2521 Frequency : rare (extremely heavy IO load with hundreds of clients)
2523 Description: Client is evicted, gets IO error writing to file
2524 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2525 caused clients to be evicted for AST timeouts. The fixes in
2526 bug 5192 mean we no longer need to have such short AST timeouts
2527 so ldlm_timeout has been increased.
2530 Frequency : occasional during --force or --failover shutdown under load
2531 Bugzilla : 5949, 4834
2532 Description: Server oops/LBUG if stopped with --force or --failover under load
2533 Details : a collection of import/export refcount and cleanup ordering
2534 issues fixed for safer force cleanup
2537 Frequency : only filesystems larger than 120 OSTs
2538 Bugzilla : 5990, 6223
2539 Description: lfs getstripe would oops on a very large filesystem
2540 Details : lov_getconfig used kfree on vmalloc'd memory
2543 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2544 Bugzilla : 6242, 6243
2545 Description: reading from files that had been truncated to a non-zero size
2546 but never opened returned no data
2547 Details : ll_file_read() reads zeros from no-object files to EOF
2552 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2553 Details : There's little harm in aborting MDS/OSS recovery and letting it
2554 try again, so I removed the LASSERT and return an error instead.
2556 Severity : enhancement
2558 Description: New debugging infrastructure for tracking down data corruption
2559 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2560 (b) cover more of the client-side code path, and (c) try to narrow
2561 down where problems occurred
2565 Bugzilla : 3819, 4364, 4397, 6313
2566 Description: Racing close and eviction MDS could cause assertion in mds_close
2567 Details : It was possible to get multiple mfd references during close and
2568 client eviction, leading to one thread referencing a freed mfd.
2570 Severity: : enhancement
2571 Bugzilla : 3262, 6359
2572 Description: Attempts to reconnect to servers are now more aggressive.
2573 Details : This builds on the enhanced upcall-less recovery that was added
2574 in 1.4.2. When trying to reconnect to servers, clients will
2575 now try each server in the failover group every 10 seconds. By
2576 default, clients would previously try one server every 25 seconds.
2581 Description: After recovery, certain operations trigger a failed
2582 assertion on a client.
2583 Details : Failing over an mds, using lconf -d --failover, while a
2584 client was doing a readdir() call would cause the client to
2585 LBUG after recovery completed and the readdir() was resent.
2587 Severity : enhancement
2589 Description: Default groups are now added by lconf
2590 Details : You can now run lconf --group <servicename> without having to
2591 manually add groups with lmc.
2594 Frequency : occasional
2596 Description: Nodes with an elan id of 0 trigger a failed assertion
2599 Frequency : always when accessing e.g. tty/console device nodes
2601 Description: tty and some other devices nodes cannot be used on lustre
2602 Details : file's private_data field is used by device data and lustre
2603 values in there got lost. New field was added to struct file to
2604 store fs-specific private data.
2607 Frequency : when exporting Lustre via NFS
2609 Description: NFSD failed occasionally when looking up a path component
2610 Details : NFSD is looking up ".." which was broken in ext3 directories
2611 that had grown large enough to become hashed.
2614 Frequency : Clusters with multiple interfaces not on the same subnet
2616 Description: Nodes will repeatedly try to reconnect to an interface which it
2617 cannot reach and report an error to the log.
2618 Details : Extra peer list entries will be created by lconf with some peers
2619 unreachable. lconf now validates the peer before adding it.
2622 Frequency : Only if a default stripe is set on the filesystem root.
2624 Description: Setting a default stripe on the filesystem root prevented the
2625 filesystem from being remounted.
2626 Details : The client was sending extra request flags in the root getattr
2627 request and did not allocate a reply buffer for the dir EA.
2630 Frequency : occasional, higher if lots of files are accessed by one client
2631 Bugzilla : 6159, 6097
2632 Description: Client trips assertion regarding lsm mismatch/magic
2633 Details : While revalidating inodes the VFS looks up inodes with ifind()
2634 and in rare cases can find an inode that is being freed.
2635 The ll_test_inode() code will free the lsm during ifind()
2636 when it finds an existing inode and then the VFS later attaches
2637 this free lsm to a new inode.
2641 Bugzilla : 6422, 7030
2642 Description: MDS deadlock between mkdir and client eviction
2643 Details : Creating a new file via mkdir or mknod (starting a transaction
2644 and getting the ns lock) can deadlock with client eviction
2645 (gets ns lock and trying to finish a synchronous transaction).
2648 Frequency : occasional
2649 Description: While starting a server, the fsfilt_ext3 module could not be
2651 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2652 kernels. Previously, lconf would still use the ext3 name
2653 when trying to load modules. Now, it will correctly use
2654 ext3 on 2.4 and ldiskfs on 2.6.
2656 Severity : enhancement
2657 Description: The default stripe count has been changed to 1
2658 Details : The interpretation of the default stripe count (0, to lfs
2659 or lmc) has been changed to mean striping across a single
2660 OST, rather than all available. For general usage we have
2661 found a stripe count of 1 or 2 works best.
2663 Severity : enhancement
2664 Description: Add support for compiling against Cray portals.
2665 Details : Conditional compiling for some areas that are different
2669 Frequency : occasional
2670 Bugzilla : 6409, 6834
2671 Description: Creating files with an explicit stripe count may lead to
2672 a failed assertion on the MDS
2673 Details : If some OSTs are full or unavailable, creating files may
2674 trigger a failed assertion on the MDS. Now, Lustre will
2675 try to use other servers or return an error to the
2679 Frequency : occasional
2681 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2682 caused lock timeout during readahead (since 1.4.2).
2683 Details : Processes doing readahead might match a lock that hasn't been
2684 granted yet if there are overlapping and conflicting lock
2685 requests. The readahead process waits on ungranted lock
2686 (original lock is CBPENDING), while OST waits for that process
2687 to cancel CBPENDING read lock and eventually evicts client.
2689 Severity : enhancement
2691 Description: Initial enabling of flock support for clients
2692 Details : Implements fcntl advisory locking and file status functions.
2693 This feature is provided as an optional mount flag (default
2694 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2695 locking are implemented yet, and those that are are not guaranteed
2696 to be completely correct in production environments.
2697 mount -t lustre -o [flock|noflock] ...
2700 Frequency : occasional
2702 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2703 assertion in the ext3 JBD (journaling) layer.
2704 Details : The b_committed_data struct is protected by the big kernel lock
2705 in 2.4 kernels, serializing journal_commit_transaction() and
2706 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2707 access to this struct. In 2.6 kernels there is finer grained
2708 locking to improve SMP performance of the JBD layer.
2712 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2714 -----------------------------------------------------------------------------
2716 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2718 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2719 versions of Lustre. Please update all servers and clients to
2720 version 1.4.2 or later at the same time. You must also run
2721 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2722 to update the configuration logs.
2724 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2725 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2726 - swab llog records of type '0' so we get proper header size/idx (5861)
2727 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2728 - fix rename of one directory over another leaking an inode (5953)
2729 - avoid SetPageDirty on 2.6 (5981)
2730 - don't re-add just-being-destroyed locks to the waiting list (5653)
2731 - when creating new directories, inherit the parent's custom
2732 striping settings if present parent (3048)
2733 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2734 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2735 - don't hold client locks on temporary worklist from l_lru (5666)
2736 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2737 - automatically evict dead clients (5921)
2738 - Update file size properly in create+truncate+fstat case (6196)
2739 - Do not unhash mountpoint dentries, do not allow removal of
2741 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2742 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2743 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2744 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2745 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2746 - Correctly handle reads of files with no objects (6243)
2747 - lctl recover will also mark a device active if deactivate used (5933)
2749 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2750 - allow --write-conf on an MDS with different nettype than client (5619)
2751 - don't write config llogs to MDS for mounts not from that MDS (5617)
2752 - lconf should create multiple TCP connections from a client (5201)
2753 - init scripts are now turned off by default; run chkconfig --on
2754 lustre and chkconfig --on lustrefs to use them
2755 - upcalls are no longer needed for clients to recover to failover
2757 - add --abort-recovery option to lconf to abort recovery on device
2759 - add support for an arbitrary number of OSTs (3026)
2760 - Quota support protocol changes.
2761 - forward compatibility changes to wire structs (6007)
2762 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2763 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2764 - improved "lustre status" script
2765 - initialize blocksize for non-regular files (6062)
2766 - added --disable-server and --disable-client configure options (5782)
2767 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2768 - Vanilla 2.4.29 support
2769 - increase maximum number of obd devices to 520 (6242)
2770 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2771 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2772 SLES 9 2.6.5 kernels
2773 - limit stripes per file to 160 (the maximum EA size) (6093)
2775 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2778 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2779 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2780 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2781 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2782 - small ext3 extents cleanups and fixes (5733)
2783 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2784 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2785 - avoid needless client->OST connect, fix handle mismatch (5317)
2786 - fix DLM error path that led to out-of-sync client, long delays (5779)
2787 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2788 - fix several locking issues related to i_size (5492,5624,5654,5672)
2789 - don't move pending lock onto export if it is already evicted (5683)
2790 - fix kernel oops when creating .foo in unlinked directory (5548)
2791 - fix deadlock in obdfilter statistics vs. object create (5811)
2792 - use time_{before,after} to avoid timer jiffies wrap (5882)
2793 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2794 - Do not leak request if server was not able to process it (5154)
2795 - If mds_open unable to find parent dir, make that negative lookup(5154)
2796 - don't create new directories with extent-mapping (5909, 5936)
2798 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2799 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2800 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2801 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2802 - support for RHEL4 kernel on i686 (5773)
2803 - provide error messages when incompatible logs are encountered (5898)
2805 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2806 * version 1.4.0.10 (1.4.1 release candidate 1)
2808 - don't keep a lock reference when lock is not granted (4238)
2809 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2810 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2811 - drop import inflight refcount on signal_completed_replay error (5255)
2812 - unlock page after async write error during send (3677)
2813 - handle missing objects in filter_preprw_read properly (5265)
2814 - no transno return for symlink open, don't save no-trasno open (3440)
2815 - don't try to complete elan receive that already failed (4012)
2816 - free RPC server reply state on error (5406)
2817 - clean up thread from ptlrpc_start_thread() on error (5160)
2818 - readahead could read extra page into cache that wasn't ejected (5388)
2819 - prevent races in class_attach/setup/cleanup/detach (5260)
2820 - don't dereference de->d_inode after l_dput of de (5458)
2821 - use "int" for stripe value returned from lock_to_stripe (5544)
2822 - mballoc allocation and error-checking fixes in 2.6 (5504)
2823 - block device patches to fix I/O request sizes in 2.6 (5482)
2824 - look up hostnames for IB nals (5602)
2825 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2826 - don't start multiple acceptors for the same port (5277)
2827 - fix incorrect LASSERT in mds_getattr_name (5635)
2828 - export a proc file for general "ping" checking (5628)
2829 - fix "lfs check" to not block when the MDS is down (5628)
2831 - service request history (4965)
2832 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2833 - create an "evict_client" /proc entry on OSTs, like the MDS has
2834 - fix mount usage message, return errors per mount(8) (5168)
2835 - change grep [] to grep "[]" in tests so they work in more UMLs
2836 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2837 - remove ancient LOV_MAGIC_V0 EA support (5047)
2838 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2839 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2840 - allow readahead allocations to fail when low on memory (5383)
2841 - mmap locking landed again, after considerable improvement (2828)
2842 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2844 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2847 - send OST transaction number in read/write reply to free req (4966)
2848 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2849 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2850 - the watchdog thread now runs as interruptible (5246)
2851 - flock/lockf fixes (but it's still disabled, pending 5135)
2852 - don't use EXT3 constants in llite code (5094)
2853 - memory shortage at startup could cause assertion (5176)
2855 - reorganization of lov code
2856 - single portals codebase
2858 - add extents/mballoc support (5025)
2859 - direct I/O reads in the obdfilter (4048)
2860 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2862 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2865 - send OST transaction number in read/write reply to free req (4966)
2866 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2867 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2868 - the watchdog thread now runs as interruptible (5246)
2869 - handle missing objects in filter_preprw_read properly (5265)
2870 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2871 - drop import inflight refcount on signal_completed_replay error (5255)
2872 - unlock page after async write error during send (3677)
2873 - return original error code on reconstructed replies (3761)
2874 - no transno return for symlink open, don't save no-trasno open (3440)
2876 - add pid to ldlm debugging output (4922)
2877 - bump the watchdog timeouts -- we can't handle 30sec yet
2878 - extra debugging for orphan dentry/inode bug (5259)
2880 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2883 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2884 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2885 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2886 - fix NULL dereference in /proc/sys/portals/routes (4827)
2887 - allow failed mdc_close() operations to be interrupted (4561)
2888 - stop precreate on OST before MDS would time out on it (4778)
2889 - don't send partial-page writes before EOF from client (4410)
2890 - discard client grant for sub-page writes on large-page clients (4520)
2891 - don't free dentries not owned by NFS code, check generation (4806)
2892 - fix lsm leak if mds_create_objects() fails (4801)
2893 - limit debug_daemon file size, always print CERROR messages (4789)
2894 - use transno after validating reply (3892)
2895 - process timed out requests if import state changes (3754)
2896 - update mtime on OST during writes, return in glimpse (4829)
2897 - add mkfsoptions to LDAP (4679)
2898 - use ->max_readahead method instead of zapping global ra (5039)
2899 - don't interrupt __l_wait_event() during strace
2901 - add software watchdogs to catch hung threads quickly (4941)
2902 - make lustrefs init script start after nfs is mounted
2903 - fix CWARN/ERROR duplication (4930)
2904 - return async write errors to application if possible (2248)
2905 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2906 - print NAL number in %x format (4645)
2907 - update barely-supported suse-2.4.21-171 series (4842)
2908 - support for sles 9 %post scripts
2909 - support for building 2.6 kernel-source packages
2910 - support for sles km_* packages
2912 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2915 - ignore -ENOENT errors in osc_destroy (3639)
2916 - notify osc create thread that OSC is being cleaned up (4600)
2917 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2918 - reconstruct ost_handle() like mds_handle() (4657)
2919 - create a new thread to do import eviction to avoid deadlock (3969)
2920 - let lconf resolve symlinked-to devices (4629)
2921 - don't unlink "objects" from directory with default EA (4554)
2922 - hold socknal file ref over connect in case target is down (4394)
2923 - allow more than 32000 subdirectories in a single directory (3244)
2924 - fix blocks count for O_DIRECT writes (3751)
2925 - OST returns ENOSPC from object create when no space left (4539)
2926 - don't send truncate RPC if file size isn't changing (4410)
2927 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2928 - bind to privileged port in socknal and tcpnal (3689)
2930 - rate limit CERROR/CWARN console message to avoid overload (4519)
2931 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2932 - basic mmap support (3918)
2933 - kernel patch series update from b1_4 (4711)
2935 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2938 - avoid crash during MDS cleanup with OST shut down (2775)
2939 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2940 - don't use bad inodes on the MDS (3744)
2941 - dynamic object preallocation to improve recovery speed (4236)
2942 - don't hold spinlock over lock dumping or change debug flags (4401)
2943 - don't zero obd_dev when it is force cleaned (3651)
2944 - print grants to console if they go negative (4431)
2945 - "lctl deactivate" will stop automatic recovery attempts (3406)
2946 - look for existing locks in ldlm_handle_enqueue() (3764)
2947 - don't resolve lock handle twice in recovery avoiding race (4401)
2948 - revalidate should check working dir is a directory (4134)
2950 - don't always mark "slow" obdfilter messages as errors (4418)
2952 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2955 - don't close LustreDB during write_conf until it is done (3860)
2956 - fix typo in lconf for_each_profile (3821)
2957 - allow dumping logs from multiple threads at one time (3820)
2958 - don't allow multiple threads in OSC recovery (3812)
2959 - fix debug_size parameters (3864)
2960 - fix mds_postrecov to initialize import for llog ctxt (3121)
2961 - replace config semaphore with spinlock (3306)
2962 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2963 - don't allow enqueue to complete on a destroyed export (3822)
2964 - down write_lock before checking llog header bitmap (3825)
2965 - recover from lock replay timeout (3764)
2966 - up llog sem before sending rpc (3652)
2967 - reduce ns lock hold times when setting kms (3267)
2968 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2969 - fix NULL deref and obd_dev leak on setup error (3312)
2970 - replace some LBUG about llog ops with error handling (3841)
2971 - don't match INVALID dentries from d_lookup and spin (3784)
2972 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2973 - fix invalid assertion in ptlrpc_set_wait (3880)
2975 - add libwrap support for the TCP acceptor (3996)
2976 - add /proc/sys/portals/routes for non-root route listing (3994)
2977 - allow setting MDS UUID in .xml (2580)
2978 - print the stack of a process that LBUGs (4228)
2980 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2983 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2984 - make sure to unset replay flag from failed open requests (3440)
2985 - if default stripe count is 0, use OST count for inode size (3636)
2986 - update parent mtime/ctime on client for create/unlink (2611)
2987 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2988 - free recovery state on server during a forced cleanup (3571)
2989 - unregister_reply for resent reqs (3063)
2990 - loop back devices mounting and status check on 2.6 (3563)
2991 - fix resource-creation race that can provoke i_size == 0 (3513)
2992 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2993 - more debugging for page-accounting assertion (3746)
2994 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2995 - avoid deadlock after precreation failure (3758)
2996 - fix race and lock order deadlock in orphan handling (3450, 3750)
2997 - add validity checks when grabbing inodes from l_ast_data (3599)
2999 - add /proc/.../recovery_status to obdfilter (3428)
3000 - lightweight CDEBUG infrastructure, debug daemon (3668)
3001 - change default OSC RPC parameters to be better on small clusters
3002 - turn off OST read cache for files smaller than 32MB
3003 - install man pages and include them in rpms (3100)
3004 - add new init script for (un)mounting lustre filesystems (2593)
3005 - run chkconfig in %post for init scripts (3701)
3006 - drop scimac NAL (unmaintained)
3008 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3011 - clean kiobufs before and after use (3485)
3012 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3013 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3014 - change init script to fail more clearly if not run as root (1528)
3015 - allow clients to reconnect during replay (1742)
3016 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3017 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3018 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3019 - fix cond_resched (3554)
3020 - only evict unfinished clients after recovery (3515)
3021 - allow bulk resend, prevent data loss (3570)
3022 - dynamic ptlrpc request buffer allocation (2102)
3023 - don't allow unlinking open directory if it isn't empty (2904)
3024 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3025 - remove extraneous obd dereference causing LASSERT failure (3334)
3026 - don't use get_cycles() when creating temp. files on the mds (3156)
3027 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3028 - handle EEXIST for set-stripe, set proper directory name (3336)
3030 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3031 - fix ksocknal_fmb_callback() error messages (2918)
3033 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3036 - don't copy lvb into (possibly NULL) reply on error (2983)
3037 - don't deref dentry after dput, don't free lvb on error (2922)
3038 - use the kms to determine writeback rpc length (2947)
3039 - increment oti_logcookies when osc is inactive (2948)
3040 - update client's i_blocks count via lvb messages (2543)
3041 - handle intent open/close of special files properly (1557)
3042 - mount MDS with errors=remount-ro, like obdfilter (2009)
3043 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3044 - don't use cancelling-locks' kms values (2947)
3045 - use highest lock extent for kms, not last one (2925)
3046 - don't dereference ERR_PTR() dentry in error handling path (3107)
3047 - fix thread race in portals_debug_dumplog() (3122)
3048 - create lprocfs device entries at setup instead of at attach (1519)
3049 - common AST error handler, don't evict client on completion race (3145)
3050 - zero nameidata in detach_mnt in 2.6 (3118)
3051 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3052 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3053 - initialize RPC timeout timer earlier for 2.6 (3219)
3054 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3055 - print nal/nid for unknown nid (3258)
3056 - additional checks for oscc recovery before doing precreate (3284)
3057 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3058 - don't crash in mdc_close for bad permissions on open (3285)
3059 - zero i_rdev for non-device files (3147)
3060 - clear page->private before handing to FS, better assertion (3119)
3061 - tune the read pipeline (3236)
3062 - fix incorrect decref of invalidated dentry (2350)
3063 - provide read-ahead stats and refine rpc in flight stats (3328)
3064 - don't hold journal transaction open across create RPC (3313)
3065 - update atime on MDS at close time (3265)
3066 - close LDAP connection when recovering to avoid server load (3315)
3067 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3068 - don't leak open file on MDS after open resend (3325)
3069 - serialize filter_precreate and filter_destroy_precreated (3329)
3070 - loop device shouldn't call sync_dev() for nul device (3092)
3071 - clear page cache after eviction (2766)
3072 - resynchronize MDS->OST in background (2824)
3073 - refuse to mount the same filesystem twice on same mountpoint (3394)
3074 - allow llmount to create routes for mounting behind routers (3320)
3075 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3076 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3077 - fix rare problem with rename on htree directories (3417)
3079 - allow default OST striping configuration per directory (1414)
3080 - fix compilation for qswnal for 2.6 kernels (3125)
3081 - increase maximum number of MDS request buffers for large systems
3082 - change liblustreapi to be useful for external progs like lfsck (3098)
3083 - increase local configuration timeout for slow disks (3353)
3084 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3086 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3089 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3090 - don't overwrite extent policy data in reply if lock was blocked (2901)
3091 - drop filter export grants atomically with removal from device (2663)
3092 - del obd_self_export from work_list in class_disconnect_exports (2908)
3093 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3094 - swab reply message in mdc_close, other PPC fixes (2464)
3095 - fix destroying of named logs (2325)
3096 - overwrite old logs when running lconf --write_conf (2264)
3097 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3098 - fix race in target_handle_connect (2898)
3099 - mds_reint_create() should take same inode create lock (2926)
3100 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3101 - don't close files for self_export to avoid uninitialized obd (2936)
3102 - allow MDS with the same name as client node (2939)
3103 - hold dentry reference for closed log files for unlink (2325)
3104 - reserve space for all logs during transactions (2059)
3105 - don't evict page beyond end of stripe extent (2925)
3106 - don't oops on a deleted current working directory (2399)
3107 - handle hard links to targets without a parent properly (2517)
3108 - don't dereference NULL lock when racing during eviction (2867)
3109 - don't grow lock extents when lots of conflicting locks (2919)
3111 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3114 - account for cache space usage on clients to avoid data loss (974)
3115 - lfsck support in lustre kernel code (2349)
3116 - reduce journal credits needed for BRW writes (2370)
3117 - orphan handling to avoid losing space on client/server crashes
3118 - ptlrpcd can be blocked, stopping ALL progress (2477)
3119 - use lock value blocks to assist in proper KMS, faster stat (1021)
3120 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3121 - recovery for initial connections (2355)
3122 - fixes for mds_cleanup_orphans (1934)
3123 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3124 - block all file creations until orphan recovery completes (1901)
3125 - client remove rq_connection from request struct (2423)
3126 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3127 - recovery timer race (2670)
3128 - mdc_close recovey bug (2532)
3129 - ptlrpc cleanup bug (2710)
3130 - mds timeout on local locks (2588)
3131 - namespace lock held during RPCs (2431)
3132 - handle interrupted sync write properly (2503)
3133 - don't try to handle a message that hasn't been replied to (2699)
3134 - client assert failure during cleanup after abort recovery (2701)
3135 - leak mdc device after failed mount (2712)
3136 - ptlrpc_check_set allows timedout requests to complete (2714)
3137 - wait for inflight reqs when ptlrpcd finishes (2710)
3138 - make sure unregistered services are removed from the srv_list
3139 - reset bulk XID's when resending them (caught by 1138 test)
3140 - unregister_bulk after timeout
3141 - fix lconf error (2694)
3142 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3143 - readahead locks pages, leaves pending causing memory pressure (2673)
3144 - increase OST request buffers to 4096 on large machines (2729)
3145 - fix up permission of existing directories in simple_mkdir (2661)
3146 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3147 - don't assign transno to errored transactions (2742)
3148 - don't delete objects on OST if given a bogus objid from MDS (2751)
3149 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3150 - if rq_no_resend, then timeout request after recovery (2432)
3151 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3152 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3153 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3154 - MDS recovery shouldn't delete objects that it has given out (2730)
3155 - if enqueue arrives after completion, don't clobber LVB (2819)
3156 - don't unlock pages twice when trigger_group_io returns error (2814)
3157 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3158 - don't write pages to disk if there was an error (1450)
3159 - don't ping imports that have recovery disabled (2676)
3160 - take buffered bytes into account when balancing socknal conn (2817)
3161 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3162 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3163 - remove little-endian swabbing of llog records (1987)
3164 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3165 - retry reposting request buffers if they fail (1191)
3166 - grow extent at grant time to avoid granting a revoked lock (2809)
3167 - lock revoke doesn't evict page if covered by a second lock (2765)
3168 - disable VM readahead to avoid reading outside lock extents (2805)
3170 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3171 - updated kernel patches for hp-2.4.20 kernel (2681)
3173 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3176 - fix truncated write corruption (2366)
3177 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3179 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3180 - ensure len(uuid) < 37 in lmc (1171)
3181 - fix ia64 OOPS in llog_test (2255)
3182 - zero end of page at obdfilter for partial page writes (2648)
3183 - don't leave stale dentries around after renames (bug 2428)
3184 - fix timeouts when evicting a client with a single lock held (2642)
3185 - set deadline for the initial HELLO message to drain (2634)
3186 - print out dotted-quad IP addresses in the socknal (2302)
3188 - additional debugging for MDS client eviction problem (2443)
3189 - fix mkfsoptions support for osts (2603, 2604)
3191 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3194 - add series for the vanilla 2.6.0 kernel
3195 - add series for the vanilla 2.4.24 kernel
3196 - add series for a cray x86/64 UL kernel drop
3197 - fix xattr patches for the vanilla 2.4.19 series
3199 - generate true UUIDs in lmc (1171)
3200 - have portals stack dumping break in UML (2466)
3201 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3202 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3203 - wake callback waiting threads on client eviction (2460)
3204 - Add --ptldebug and --subsystem to lmc (1719)
3205 - update assertion to allow safe interrupt allocation
3206 - set rq_no_resend for cancel requests (2432)
3207 - recalculate ptlrpcd timeout after resend (2494)
3208 - call vfs_rmdir when removing pending directories (2368)
3209 - fix renaming a file to itself (2429)
3210 - lmc creates a default one-stripe lov (2454)
3211 - expand procfs space to handle large clusters (2326)
3212 - increase UML stack to avoid overflow
3213 - update lconf's list of debug and subsystem masks
3214 - fix lfs find --obd (2510)
3215 - /proc tunable for disabling filter read caching (2591)
3216 - stop rpm packages from altering slapd.conf (2301)
3217 - disable nagle in the socknal under 0conf (2578)
3218 - choose mds inode size based on stripe count (2572)
3219 - fix kernel-source rpm problems (2516)
3221 - add --disable-doc to avoid pdf generation (2421)
3222 - update documentation, tests, type-os, comments
3223 - avoid format warnings on ia64
3224 - remove the TOE NAL
3225 - tiny code cleanups by removing unused fields
3227 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3230 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3231 - remove the most blatant lies from BUILDING file (2371)
3232 - change default debug level to reasonable production setting
3233 - reduce client side cache size to reduce cache flush time
3234 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3235 - make TCP zerocopy and pinger support enabled by default (2476)
3236 - sync writes completed after process exits caused crashes (2319)
3237 - maintain correct mount count on the MDS (2356)
3238 - backout 1557, because 2316 wasn't really fixed
3239 - better file I/O statistics gathering in /proc
3240 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3241 - another kernel patch to fix zero-copy TCP function export
3242 - don't take duplicate lock when processing re-sent getattr (2420)
3243 - lctl uses obd_self_export instead of creating new conn (2353)
3244 - MDS/OST recovery case which requires object creation asserted (2425)
3245 - move lfs from /usr/sbin to /usr/bin in packages
3246 - fix race between mds_client_add and mds_client_free (2417)
3247 - use kmalloc instead of slabs in portals (2430)
3248 - don't create duplicate records when a failover MDS is present (2442)
3249 - remove unnecessary mount age check (2332)
3250 - don't remove directory inodes from locks prematurely (2451)
3251 - don't break if MDS service name is the same as hostname (2103)
3252 - fix races in client write RPC generation when cache full (2482)
3254 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3257 - remove now-unused request->rq_obd (278)
3258 - if an allocation fails, print out how much memory we've used (1933)
3259 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3260 - add the "configurable stack size" patch to most series files (1256)
3261 - ability to write large log records, for 100+ OST configs (2306)
3262 - fix NULL deref when filter_prep fails (2314)
3263 - fix operator precedence error in filter_sync
3264 - dynamic allocation of socknal TX descriptors (2315)
3265 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3266 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3267 - ensure that CWARN messages in Portals always get to the syslog
3268 - __init/__exit are not for prototype decls (ldlm_init/exit)
3269 - x86-64 compile warning fixes
3270 - fix gateway LMC keyword conflict (2318)
3271 - fix MDS lock inversions in getattr/reint paths (1844)
3272 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3273 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3274 - DLM race condition prevented some lock evictions (2328)
3275 - ENOMEM detection and retry on socknal sends (2230)
3276 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3277 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3278 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3279 - better error messages when a client is rejected during recovery (1505)
3280 - avoid cancelling locks which were never granted, after failure (2330)
3281 - fix i_sem/journal inversion in mds_client_add (2333)
3282 - fix truncate/getattr lock cycle deadlock (2334)
3283 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3284 - fix two rare exit paths which could leak an l_lock() ref (2321)
3285 - fencepost error in MDS/OST orphan recovery (2226)
3286 - make log record alignment 8 bytes (1988)
3287 - lstripe now fails when requested offset > ost_count (2237)
3288 - ensure that all kernel series have a complete list.h (1607)
3289 - fix crashes in special-file operations (2316)
3290 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3292 - allow configurable automake binary, for testing new versions
3293 - small update to the lfs documentation
3295 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3297 * fix negative export reference count in fsfilt_sync (2312)
3299 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3300 * release candidate 0.9.1
3302 - orphans are moved into the PENDING directory for possible recovery
3303 - replayed opens now open by fid for orphan/rename safety (1042)
3304 - last close of an orphan inode generates a transno (683)
3305 - chdir() and mount() now pin the directory entry (1020)
3306 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3307 - discard very old requests without processing them (1502)
3308 - remove some common, well-understood CERRORs (1505)
3309 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3310 - clear "grant" flags in OST replies until OST grant code lands (1644)
3311 - fix read performance by not clobbering i_blksize on client (1598)
3312 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3313 - make LRU size a /proc tunable, clears locks when reduced (707)
3314 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3315 - server locks take references on exports now (1558)
3316 - build fixes for 2.4.20-rh trees (1663)
3317 - return an error from lov_create if all OSCs are inactive (1751)
3318 - fix import levels when a reconnect happens without a timeout (1597)
3319 - exit early from mds_open if we get a lookup error (1749)
3320 - partial page read at EOF wouldn't wait for disk before sending (1642)
3321 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3322 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3323 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3324 - write/truncate lock inversion (1639)
3325 - Don't auto-load obdclass, portals modules during cleanup (1495)
3326 - fix timestamps from jumping to "now" (1763)
3327 - extra journal assertions (1648)
3328 - add an extra multiunlink test (1771)
3329 - fix read_record/write_record API (1776)
3330 - fix leak of offset_extent, possible incorrect i_size later (1772)
3331 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3332 - Debugging for the kqswnal_get_idle_tx problems (1820)
3333 - Allow recovery to be attempted multiple times (1536)
3334 - Write out MDS last_rcvd file after it is first created (1600)
3335 - Fix tx_descriptor leak in failed transmit situations (1827)
3336 - ext3 journaling fixes for assertion failure after IO error (1871)
3337 - class_export_put() on freed export after completion AST error (1896)
3338 - Fix revalidate looping in VFS (1322)
3339 - Don't access a freed export during MDS_REINT timeout (1521)
3340 - Add open-unlink recovery support on the MDS (1673,1764)
3341 - Return an error if no MDS data was read from last_rcvd (1946)
3342 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3343 - Don't setup a disk device that doesn't match exported UUID (317)
3344 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3345 - avoid committing NULL handle in force close
3346 - local.sh is now a one-stripe LOV configuration
3347 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3348 - fix ext3 htree duplicate directory entry corruption (1516)
3349 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3350 - update to new LOV EA format (2097)
3351 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3353 2003-06-15 Phil Schwan <phil@clusterfs.com>
3356 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3357 - per-import/export recovery handling (958, 931, 959)
3358 - multiple last-rcvd slots, for serving multiple FSes (949)
3359 - connections are again shared between multiple imp/exports (963, 964)
3360 - "umount -f" would hang if any requests needed to be sent (393, 978)
3361 - avoid pinning large req buffer by copying for queued messages (989)
3362 - add "uuid" to "lctl device" command to help upcalls (991)
3363 - "open" RPCs with transnos would confuse recovery counters (1037)
3364 - do proper endian conversion of all wire messages (288, 340, 891)
3365 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3366 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3367 - fix ptlrpc_connection leak in target_handle_connect (1174)
3368 - fix import refcounting bug in OST and MDS cleanup (1134)
3369 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3370 - fix very unlikely obd_types race condition (501)
3371 - remove osc_open hack for echo_client (1187)
3372 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3373 - a failure in read_inode2 leads to deadlock (1139)
3374 - cancel ack-locks as soon as transaction is committed (1072)
3375 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3376 - make sure to commitrw after any preprw to avoid deadlock (1162)
3377 - failing to execute a file in a lustre FS would lock inode (1203)
3378 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3379 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3380 - obd_uuid2tgt would walk off the end of the list (1255)
3381 - on IA64 the osc would give portals incorrect bulk size (1258)
3382 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3383 - fix lock inversion caused by new llite matching code (1282)
3384 - limit the number of dirty pages on a client to 10MB (1286)
3385 - timed out locks were not being corrected cancelled (1289)
3386 - fix O_DIRECT above 4GB on IA-32 (1292)
3387 * major user-visible changes
3388 - fail out/fail over policy now controlled by the upcall (993)
3390 - add OBD_PING to check server availability and failure (954)
3391 - lustre messages are now sent in sending host order (288, 340, 891)
3392 - add eadatalen to MDS getattr reply (340)
3393 - OST read replies may contain second buffer, with per-page status (593)
3395 2003-03-11 Phil Schwan <phil@clusterfs.com>
3398 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3399 - failed open()s could cause deadlock; fixed (867, 869)
3400 - stop cancelling OST locks when files are closed (481)
3401 - overlapping XID spaces caused network corruption (851, 853)
3402 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3403 - setattr_raw added, to do single-RPC, server-side setattrs
3404 - lmc/lconf syntax change for OST UUIDs
3405 - fix crashy race condition between ptlrpc_free_req and osc_close
3406 - don't use request in mdc_enqueue if we hit a timeout (889)
3407 - don't set the inode i_size for regular files from the MDS (896)
3408 - handle out of order completion AST (842)
3409 - don't LBUG if a lock request times out after receiving AST (913)
3410 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3411 - if a bad lock AST arrives, send an error instead of dropping entirely
3412 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3413 - fix leak in bulk IO when only partially completed (899, 900, 926)
3414 - fix O_DIRECT for ia64 (55)
3415 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3416 - C-z after timeout could hang a process forever; fixed (977)
3418 - client-side I/O cache (678, 924, 929, 941, 970)
3420 - READPAGE and SETATTRs which don't take server-side locks get
3423 2003-02-11 Phil Schwan <phil@clusterfs.com>
3426 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3427 - Fix fsfilt_extN_readpage() to read a full page of directory
3428 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3429 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3430 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3431 - Most intent operations were removed, in favour of a new RPC mode
3432 that does a single RPC to the server and bypasses most of the VFS
3433 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3434 - Aggressively cancel local locks on DLM servers
3435 - mds_reint_unlink sends EA to the client if it's the last nlink.
3436 client uses that EA to unlink OST objects.
3437 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3438 - recursive symlinks were fixed (439)
3439 - fixed NULL deref in DEBUG_REQ
3440 - filter_update_lastobjid no longer calls sync, which annoyed extN
3441 - fixed multi-client small-writes to a single file problem (445)
3442 - fixed mtime updates during file writes (607)
3443 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3444 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3445 - fixed Linux OST size reporting problem (444, 656)
3446 - OST now updates object mtime with writes or setattr (607, 619)
3447 - client verifies file size before zeroing page past EOF (445)
3448 - OST now writes last allocated objid to disk with allocation (108)
3449 - LOV on echo now works (409)
3451 - mds_reint_unlink sends a new buffer, with the EA included. this
3452 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3453 set if a regular file was being unlinked, and it was the last link
3454 - use PtlGet from the target for bulk writes (315)
3455 - OST now updates object mtime with writes or setattr (607, 619)
3456 - LDLM now has a grant-time callback to revalidate locked items, if
3458 - Many MDS operations were reorganized to combat race conditions
3460 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3461 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3463 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3466 - Fully reactivate OST imports after reconnection (512, others)
3467 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3468 - More graceful error handling for truncating on dead OST (515)
3469 - Don't error out unless we're actually accessing dead stripes (474)
3470 - Fix garbage sizes when stripes are missing (410)
3471 - LRU counters were broken, causing constant lock purge (433, 432)
3472 - garbage on read from stripes with failed OSTs (441)
3473 - mark OSCs as active before reconnecting during recovery (438)
3474 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3475 - lfind did not preserve OST order in output (443)
3476 - symlinks cause hung clients, incorrect data (439)
3477 - stop dereferencing request after dropping refcount (457)
3478 - don't LASSERT(spin_is_locked) on non-SMP (455)
3479 - fixes for many rename() bugs
3480 - fstat didn't correctly synchronize attributes (399)
3481 - server must handle lock cancellation during blocking AST prep (487)
3482 - bulk descriptors were free()d too soon (511)
3483 - fix paths in lconf, which would load incorrect modules (451, 507)
3484 - fix confusing lconf 'host not found' error message (386)
3485 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3486 - fix race condition in mdc_blocking_ast() for inode access (526)
3487 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3488 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3489 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3490 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3491 - fix lstripe in several ways (kernel side) (527)
3492 - fix request leak in ldlm_cli_enqueue (262)
3493 - incorrect OSC was marked inactive after OST failure
3494 - call mds_fs_cleanup before unmounting filesystem (524)
3495 - fix races between taking ns_lock and ldlm_lock_change_resource
3496 - fix races updating LOV export open file list
3497 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3498 - fix recovery NULL deref in ldlm_cli_cancel_unused
3499 - fix some DLM races by using new hash table for lock handles (419)
3500 - permit the client to specify desired inodes, at replay
3501 - duplicate requests when we queue them for replay reintegration
3502 - fix last_rcvd offset calculation
3503 - sync after each recovered transaction, so we always make progress
3504 - never, not always, ERESTART requests without transnos
3505 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3507 - skip replay if the MDS says that the client is already connected
3508 - don't check for a recovery-enabled export to match lctl's UUID
3509 - don't INC_USE_COUNT for phantom exports
3510 - don't crash when cleaning up phantom exports (567)
3511 - don't double-finish or set replay data for errored mdc_open requests
3512 - abort requests when they time out, so we don't get old replies
3513 - send/receive replies for AST messages again
3514 - if the client says that it doesn't have the lock, cancel it on the
3516 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3517 mark it as destroyed, it will all work out when decref is called
3518 - fix module use counts (22, 581)
3520 - ASTs now expect a reply (server cancels lock on error reply)
3522 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3525 - fix many simultaneous client startup (392)
3526 - fix dentry->d_it clobbering
3527 - credentials weren't being shipped for readdir/getattr operations
3528 - remove invalid assertions triggered during some concurrent MD
3530 - proper Lustre versions added (336, 389)
3531 - fix memory leak for create error case (398)
3532 - fix LOV locking bug that would get cli/srv out of sync
3533 - fix echo client over LOV (409)
3534 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3535 - fix double-O_EXCL intent crash (424)
3536 - avoid sending multiple lock CANCELs (352)
3538 - MDS can do multi-client recovery (modulo bugs in new code)
3540 - many updates, edits, cleanups
3542 2002-11-18 Phil Schwan <phil@clusterfs.com>
3545 - fix null d_it dereference (346)
3546 - fix full OST/dbench hang (333)
3547 - fix permission problem with file removal (286)
3548 - fix removal of OSCs from LOV when they fail
3549 - fix NULL deref during bulk timeout (214)
3550 - fix problems related to multiple filesystems on one MDS (241)
3551 - fixed serious subtle metadata locking bugs
3552 - free locks on clients when inodes are removed due to memory
3554 - fix inode pointer in lock data (285)
3555 - partial support for multiple MDS on a single host (241)
3556 - data locks weren't cancelled at clear_inode time (290, 311)
3557 - intent locks could lead to unbounded lock growth (205)
3558 - added a maximum lock count, an LRU list, and a flusher
3559 - fix multiple rename (365)
3560 - properly abstracted the echo client
3561 - OSC locked 1 byte too many; fixed
3562 - rewrote brw callback code:
3563 - fixed recovery bugs related to LOVs (306)
3564 - fixed too-many-pages-in-one-write crash (191)
3565 - fixed (again) crash in sync_io_timeout (214)
3566 - probably fixed callback-related race (385)
3568 - Add capability to MDS protocol
3569 - LDLM cancellations and callbacks on different portals
3571 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3574 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3575 - defer open object destruction to close time (601981, 138)
3576 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3577 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3578 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3579 - don't call obd_setattr in ll_file_release for destroyed objects
3580 * protocol change to lustre_msg: move |version| and add |flags|
3581 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3582 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3583 * added replay of create, unlink, link and rename operations during
3584 MDS failover; recovery should be much more robust now
3585 * remove failed OSCs from LOVs (only lov_create uses this so far)
3586 * the lustre-HOWTO was brought (more) up to date (582544)
3588 2002-10-23 Phil Schwan <phil@clusterfs.com>
3591 - in-use dentries weren't being reused properly (617851)
3592 - prevent multiple LDLM setup (599178)
3593 - fix LOV size calculations for truncate (617853)
3594 - fix client handling of MDS intent errors (POSIX)
3595 - fix permission bug in lovstripe.c test (624321)
3596 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3597 - truncate past end of file could corrupt data
3598 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3599 - a race in recovery could return ETIMEDOUT to apps (623947)
3600 - building outside the source directory was fixed
3601 * the lustre-HOWTO was brought (more) up to date (582544)
3602 * major progress was made on recovery functionality
3604 2002-10-10 Phil Schwan <phil@clusterfs.com>
3607 - recovery deadlock fix
3608 - rm -rf causes LBUG fix (617817)
3609 - file open by multiple tasks fix (618962)
3610 - directory permissions bugs (602707 and 620007)
3611 - journal_stop fixed with locking (611313)
3612 - O_APPEND failures resolved (618273, perhaps 614459)
3613 - lconf PATH fix (619770)
3614 - IA64 build fix (621450)
3615 - RPC buffer sizes scale with amount of memory
3617 2002-10-01 Phil Schwan <phil@clusterfs.com>
3620 - locks would be cancelled without throwing away data pages,
3621 resulting in inconsistent data (605627)
3622 - inode attributes were not always being refreshed (605627, 612449)
3623 - lconf now continues to cleanup after lctl reports an error
3624 - MDS now enforces user permissions (602707)
3625 - lprocfs cleanup fixed, but not yet enabled (614157)
3626 - fixed infinite server hang, should a client not respond to an AST
3627 - avoid going into recovery if user calls readlink() with a buffer
3628 that's too small (613941)
3629 - AST RPCs no longer require replies (614867) -- this may be changed
3630 - don't crash server if client sends an IOV that's too big (611336)
3631 - fixed lock conversion deadlock (611892)
3632 - fixed the following of symlinks (614622)
3633 * recovery: the server can remove locks from a client that dies, other
3634 clients can make progress
3635 * more extN patch fixes
3636 * compile-time configurable ptlrpc buffer allocations
3638 - collaborative read cache document
3639 - Lustre Lite Performance CDR document-in-progress
3641 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3644 - fix typo in patch-2.4.18
3646 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3649 - clear ptlrpc request each time in handle_incoming_request()
3650 - unlink of files now destroys the object on the OST
3652 2002-09-19 Peter Braam <braam@clusterfs.com>
3654 * add hard link support
3655 * change obdfile creation method
3656 * kernel patch changed
3658 2002-09-19 Peter Braam <braam@clusterfs.com>
3661 - stack overflow bug in extN fixed
3663 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3665 * documentation updates
3666 - add man pages for config tools
3667 - update tests/README to describe testing with new config tools
3668 - finish metadata API descriptions
3669 * bug fixes and cleanups
3670 - statfs workaround for 16TB limit
3671 - LOV stripe allocation improved, can stripe on subset of OSTs
3672 - LOV file size/IO offset was wrong for files > 4GB in size
3673 - object EA data was being dropped, caused files to be unreadable
3674 - memory overflow with non-LOV OST caused memory corruption
3675 - fixed regression tests to work with new config tools, obdfilter
3676 - fixed bug when directory size became larger than 1 block
3677 - fixed bug (for single client case) when PWD was deleted
3678 - invalidate local directory pages when doing intent-based ops
3679 - avoid LDLM oops when lock callback contained bad data
3681 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3683 * documentation updates
3684 * bug fixes and cleanups
3685 - configuration tools
3688 - 64-bit compile warnings
3689 - 64-bit internal statfs data
3691 * test_brw on persistent OST devices
3693 * lprocfs (disabled)
3695 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3697 * documentation updates
3698 * bug fixes and cleanups
3699 * configuration tools
3701 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3704 * many small fixes to 0_5_4
3705 * io/network handling
3706 * thinkos in MDS operations
3708 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3711 * crucial basic fixes to 0.5.3
3712 * IOR, Iozone work over Elan
3715 2002-08-07 Phil Schwan <phil@clusterfs.com>
3716 * version 0_5_3, our first alpha
3717 * we use the new Portals iovs
3718 * documentation updates
3719 * bug fixes and cleanups
3720 * small changes in the DLM wire protocol
3722 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3723 * version 0_5_1 with some initial stability,
3724 * locking on MD and file I/O.
3725 * documentation updates
3726 * several bug fixes since 0.5.0
3727 * small changes in wire protocol
3729 2002-07-18 Phil Schwan <phil@clusterfs.com>
3731 * delivered as Lustre Light Alpha
3732 * fixed a crash after handling invalid MDS requests
3733 * fixed directory pages for architectures with non-4k pages sizes
3735 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3736 * release version v0_4_4
3737 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3739 * Fixes a number of interruption problems with OST operations.
3740 * Update documentation for portals header changes
3741 * Move all wire protocol structs/defines to lustre_idl.h
3742 * Fixes symlink length bug.
3743 * Add tcpdump to repository.
3745 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3746 * release version v0_4_3
3747 * Fixes statfs for inodes on extN.
3748 * Fixes bug in runtests which would delete /etc/hosts.
3749 * Use 64-bit object IDs wherever possible (not into VFS though)
3750 Remove ost_get_info, which is unused by lustre, and out of date.
3752 2002-07-03 Peter Braam <braam@clusterfs.com>
3753 * release version v0_4_2 Fixes a lookup error (type not passed)
3754 * move forward to head of Portals
3755 * move forward to latest Lustre kernel
3757 2002-06-25 Peter Braam <braam@clusterfs.com>
3758 * release version v0_4_1. Hopefully stable on single node use.