1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
6 2.6.16.46-0.14 (SLES 10)
7 2.6.18.8 vanilla (kernel.org)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.9-42.0.10.EL (RHEL 4)
11 2.6.16 - 2.6.21 vanilla (kernel.org)
12 * Recommended e2fsprogs version: 1.39.cfs8
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
17 Description: don't use GFP_* in generic Lustre code.
18 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
22 Description: obdclass is limited by single OBD_ALLOC(idarray)
23 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
27 Description: warnings with build patchless client with vanila 2.6.19 and up
28 Details : change old ctl_table style and replace ctl_table/ctl_table_header
29 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
33 Frequency : Only for SLES
34 Description: sanity_quota fail test_1
35 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
36 file, which makes TSTID[2] inunique.
40 Frequency : Only for SLES9
41 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
42 Details : Update target/ChangeLog/which_patch .
46 Frequency : Only for test_10 in sanity-quota.sh
47 Description: a bug in quota test code
48 Details : checking a test flag in a wrong place
52 Frequency : Only for test_13 in sanity-quota.sh
53 Description: a bug in quota test script
54 Details : shouldn't check overfull allocated quota
56 --------------------------------------------------------------------------------
58 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
60 * Support for kernels:
63 2.6.16.46-0.14 (SLES 10)
64 2.6.18.8 vanilla (kernel.org)
65 * Client support for unpatched kernels:
66 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
67 2.6.16 - 2.6.22 vanilla (kernel.org)
68 * Due to recently discovered recovery problems, we do not recommend
69 using patchless RHEL 4 clients with this or any earlier release.
70 * Recommended e2fsprogs version: 1.39.cfs8
71 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
72 * Starting with this release, the ldiskfs backing filesystem required
73 by Lustre is now in its own package, lustre-ldiskfs. This package
74 should be installed. It is versioned separately from Lustre and
75 may be released separately in future.
77 Severity : enhancement
79 Description: add optional extra BUILD_VERSION info
80 Details : add a new environment variable (namely LUSTRE_VERS) which allows
81 to override the lustre version.
84 Frequency : 2.6.18 servers only
86 Description: ll_kern_mount() doesn't release the module reference
87 Details : The ldiskfs module reference count never drops down to 0
88 because ll_kern_mount() doesn't release the module reference.
93 Description: server LBUG when using old ost_num_threads parameter
94 Details : Accept the old ost_num_threads parameter but warn that it
95 is deprecated, and fix an off-by-one error that caused an LBUG.
100 Description: Transient SCSI error results in persistent IO issue
101 Details : iobuf->dr_error is not reinitialized to 0 between two
105 Frequency : sometimes when underlying device returns I/O errors
107 Description: OSTs not going read-only during write failures
108 Details : OSTs are not remounted read-only when the journal commit threads
109 get I/O errors because fsfilt_ext3 calls journal_start/stop()
110 instead of the ext3 wrappers.
114 Description: poor connect scaling with increasing client count
115 Details : Don't run filter_grant_sanity_check for more than 100 exports
116 to improve scaling for large numbers of clients.
119 Frequency : SLES10 only
121 Description: sanity-quota.sh quotacheck failed: rc = -22
122 Details : Quotas cannot be enabled on SLES10.
125 Frequency : liblustre clients only
127 Description: getdirentries does not give error when run on compute nodes
128 Details : getdirentries does not fail when the size specified as an argument
129 is too small to contain at least one entry
131 Severity : enhancement
133 Description: Add LNET router traceability for debug purposes
134 Details : If a checksum failure occurs with a router as part of the
135 IO path, the NID of the last router that forwarded the bulk data
136 is printed so it can be identified.
141 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
142 Details : Due to a race condition, liblustre clients were occasionally
145 Severity : enhancement
147 Description: lfs setstripe use optional parameters instead of postional
150 Severity : enhancement
152 Description: Nanosecond timestamp support for ldiskfs
153 Details : The on-disk ldiskfs filesystem has added support for nanosecond
154 resolution timestamps. There is not yet support for this at
155 the Lustre filesystem level.
158 Frequency : during server recovery
160 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
161 Details : request with rq_no_resend flag not awake l_wait_event if they get a
165 Frequency : nfs export on patchless client
167 Description: connectathon hang when test nfs export over patchless client
168 Details : Disconnected dentry cannot be found with lookup, so we do not need
169 to unhash it or make it invalid
172 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
173 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
174 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
179 Description: open req refcounting wrong on reconnect
180 Details : If reconnect happened between getting open reply from server and
181 call to mdc_set_replay_data in ll_file_open, we will schedule
182 replay for unreferenced request that we are about to free.
183 Subsequent close will crash in variety of ways.
184 Check that request is still eligible for replay in
185 mdc_set_replay_data().
190 Description: disable writes to filesystem when reading health_check file
191 Details : the default for reading the health_check proc file has changed
192 to NOT do a journal transaction and write to disk, because this
193 can cause reads of the /proc file to hang and block HA state
194 checking on a healthy but otherwise heavily loaded system. It
195 is possible to return to the previous behaviour during configure
196 with --enable-health-write.
198 Severity : enhancement
200 Description: 64-bit inode version
201 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
202 to the inode. This will be required for version-based recovery.
207 Description: MDS fails to start if a duplicate client export is detected
208 Details : in some rare cases it was possible for a client to connect to
209 an MDS multiple times. Upon recovery the MDS would detect this
210 and fail during startup. Handle this more gracefully.
212 Severity : enhancement
214 Description: Add -o localflock option to simulate old noflock
216 Details : This will achieve local-only flock/fcntl locks
222 Description: log_commit_thread vs filter_destroy race leads to crash
223 Details : Take import reference before releasing llog record semaphore
228 Description: Wrong request locking in request set processing
229 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
230 imp_delayed_list, in this place should be used imp_lock.
233 Frequency : when reconnection
235 Description: Grant Leak when osc reconnect to OST
236 Details : When osc reconnect ost, OST(filter) should check whether it
237 should grant more space to client by comparing fed_grant and
238 cl_avail_grant, and return the granted space to client instead
239 of "new granted" space, because client will call osc_init_grant
240 to update the client grant space info.
243 Frequency : when client reconnect to OST
245 Description: Grant Leak when osc do resend and replay bulk write
246 Details : When osc reconnect to OST, OST(filter)should clear grant info of
247 bulk write request, because the grant info while be sync between
248 OSC and OST when reconnect, and we should ignore the grant info
249 these of resend/replay write req.
254 Description: Grant space more than avaiable left space sometimes.
255 Details : When then OST is about to be full, if two bulk writing from
256 different clients came to OST. Accord the avaliable space of the
257 OST, the first req should be permitted, and the second one
258 should be denied by ENOSPC. But if the seconde arrived before
259 the first one is commited. The OST might wrongly permit second
260 writing, which will cause grant space > avaiable space.
263 Frequency : when client is evicted
265 Description: Grant might be wrongly erased when osc is evicted by OST
266 Details : when the import is evicted by server, it will fork another
267 thread ptlrpc_invalidate_import_thread to invalidate the
268 import, where the grant will be set to 0. While the original
269 thread will update the grant it got when connecting. So if
270 the former happened latter, the grant will be wrongly errased
271 because of this race.
276 Description: Checking Stale with correct fid
277 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
278 to check whether it is stale, because sometimes, we want the
279 enqueue happened anyway, and op_data.fid2 will not be initialized.
281 Severity : enhancement
283 Description: update patchless client
284 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
287 Frequency : only with 2.4 kernel
289 Description: random memory corruption
290 Details : size of struct ll_inode_info is to big for union inode.u and this
291 can be cause of random memory corruption.
296 Description: Memory leak in recovery
297 Details : Lov_mds_md was not free in an error handler in mds_create_object.
298 It should also check obd_fail before fsfilt_start, otherwise if
299 fsfilt_start return -EROFS,(failover mds during mds recovery).
300 then the req will return with repmsg->transno = 0 and rc = EROFS.
301 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
302 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
303 be freed no matter whether fsfilt_commit success or not.
306 Frequency : only with huge count clients
308 Description: Prevents from taking the superblock lock in llap_from_page for
310 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
311 for llap_from_page call in ll_removepage prevents from taking the
312 superblock lock for a soon died page.
317 Description: Not check open intent error before release open handle
318 Details : in some rare cases, the open intent error is not checked before
319 release open handle, which may cause
320 ASSERTION(open_req->rq_transno != 0), because it tries to release
321 the failed open handle.
326 Description: Set cat log bitmap only after create log success.
327 Details : in some rare cases, the cat log bitmap is set too early. and it
328 should be set only after create log success.
332 Description: Accessing a block bevice can re-enable I/O when Lustre is
333 tearing down a device.
334 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
339 Description: service threads may hog cpus when there are a lot of requests
341 Details : Insert cond_resched to give other threads a chance to use some of
347 Description: the cat log was not initialized in recovery
348 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
349 unlink log on mds will not be initialized until mds post
350 recovery. And also in mds post recovery, the unlink log will
351 initialization will be done asynchronausly, so there will be race
352 between add unlink log and unlink log initialization.
356 Description: brw_stats were being printed incorrectly
357 Details : brw_stats were being printed as log2 but all of them were not
358 recorded as log2. Also remove some code duplication arising from
359 filter_tally_{read,write}.
363 Frequency : rare, only in recovery.
364 Description: ASSERTION(req->rq_type != LI_POISON) failed
365 Details : imp_lock should be held while iterating over imp_sending_list for
366 prevent destroy request after get timeout in ptlrpc_queue_wait.
370 Description: replay-single.sh test 52 fails
371 Details : A lock's skiplist need to be cleanup when it being unlinked
372 from its resource list.
376 Description: Short directio read returns full requested size rather than
378 Details : Direct I/O operations should return actual amount of bytes
379 transferred rather than requested size.
381 Severity : enhancement
383 Description: metadata RPC reduction (e.g. for rm performance)
384 Details : decrease the amount of synchronous RPC between clients and servers
385 by canceling conflicing lock before the operation on the client side
386 and packing thier handles into the main operation RPC to server.
388 --------------------------------------------------------------------------------
390 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
395 Frequency : on some architectures
397 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
398 Details : Uninitialized flags sometimes cause configuration commands to
402 Frequency : patchless clients only
404 Description: missing __iget() symbol export
405 Details : The __iget() symbol export is missing. To avoid the need for
406 this on patchless clients the deathrow inode reaper is turned
407 off, and we depend on the VM to clean up old inodes. This
408 dependency was during via the fix for bug 12181.
410 --------------------------------------------------------------------------------
412 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
414 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
415 INTEROPERATE with older versions automatically. In many cases a
416 special upgrade step is needed. Please read the
417 user documentation before upgrading any part of a live system.
418 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
419 version will not interoperate with 1.6 betas before beta5 (1.5.95).
420 * WARNING: Lustre configuration and startup changes are required with
421 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
426 Severity : enhancement
428 Description: MountConf
429 Details : Lustre configuration is now managed via mkfs and mount
430 commands instead of lmc and lconf. New obd types (MGS, MGC)
431 are added for dynamic configuration management. See
432 https://mail.clusterfs.com/wikis/lustre/MountConf for
435 Severity : enhancement
437 Description: dynamic OST addition
438 Details : OSTs can now be added to a live filesystem
440 Severity : enhancement
442 Description: startup order invariance
443 Details : MDTs and OSTs can be started in any order. Clients only
444 require the MDT to complete startup.
446 Severity : enhancement
448 Description: parallel, asynchronous orphan cleanup
449 Details : orphan cleanup is now performed in separate threads for each
450 OST, allowing parallel non-blocking operation.
452 Severity : enhancement
454 Description: optimized stripe assignment
455 Details : stripe assignments are now made based on ost space available,
456 ost previous usage, and OSS previous usage, in order to try
457 to optimize storage space and networking resources.
459 Severity : enhancement
461 Description: Permanently set tunables
462 Details : All writable /proc/fs/lustre tunables can now be permanently
463 set on a per-server basis, at mkfs time or on a live
466 Severity : enhancement
468 Description: Lustre message v2
469 Details : Add lustre message format v2.
471 Severity : enhancement
473 Description: client OST exclusion list
474 Details : Clients can be started with a list of OSTs that should be
475 declared "inactive" for known non-responsive OSTs.
479 Description: SPEC SFS validation failure on NFS v2 over lustre.
480 Details : Changes the blocksize for regular files to be 2x RPC size,
481 and not depend on stripe size.
483 Severity : enhancement
485 Description: Multiple MD RPCs in flight.
486 Details : Further unserialise some read-only MDS RPCs - learn about intents.
487 To avoid overly-overloading MDS, introduce a limit on number of
488 MDS RPCs in flight for a single client and add /proc controls
489 to adjust this limit.
491 Severity : enhancement
493 Description: client read/write statistics
494 Details : Add client read/write call usage stats for performance
495 analysis of user processes.
496 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
497 file access. extents_stats shows chunk size distribution.
498 extents_stats_per_process show chunk size distribution per
501 Severity : enhancement
503 Description: mds statistics
504 Details : Add detailed mds operations statistics in
505 /proc/fs/lustre/mds/*/stats.
509 Description: Failure of copying files with lustre special EAs.
510 Details : Client side always return success for setxattr call for lustre
511 special xattr (currently only "trusted.lov").
516 Description: Refcount LNET uuids
517 Details : The global LNET uuid list grew linearly with every startup;
518 refcount repeated list entries instead of always adding to
521 Severity : enhancement
523 Description: Dynamic service threads
524 Details : Within a small range, start extra service threads
525 automatically when the request queue builds up.
528 Frequency : mixed-endian client/server environments
530 Description: mixed-endian crashes
531 Details : The new msg_v2 system had some failures in mixed-endian
534 Severity : enhancement
536 Description: Easy OST removal
537 Details : OSTs can be permanently deactivated with e.g. 'lctl
538 conf_param lustre-OST0001.osc.active=0'
540 Severity : enhancement
542 Description: MGS proc entries
543 Details : Added basic proc entries for the MGS showing what filesystems
546 Severity : enhancement
548 Description: provide MGS failover
549 Details : Added config lock reacquisition after MGS server failover.
551 Severity : enhancement
553 Description: add Linux 2.4 support
554 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
558 Description: a large application tries to do I/O to the same resource and dies
560 Details : Check the req->rq_arrival time after the call to
561 ost_brw_lock_get(), but before we do anything about
562 processing it & sending the BULK transfer request. This
563 should help move old stale pending locks off the queue as
564 quickly as obd_timeout.
567 Frequency : when an incorrect nid is specified during startup
569 Description: ptlrpc connect to non-existant node causes kernel crash
570 Details : LNET can't be re-entered from an event callback, which
571 happened when we expire a message after the export has been
572 cleaned up. Instead, hand the zombie cleanup off to another
575 Severity : enhancement
577 Description: plain/inodebits lock performance improvement
578 Details : Grouping plain/inodebits in granted list by their request modes
579 and bits policy, thus improving the performance of search through
583 Frequency : only if OST filesystem is corrupted
585 Description: client incorrectly hits assertion in ptlrpc_replay_req()
586 Details : for a short time RPCs with bulk IO are in the replay list,
587 but replay of bulk IOs is unimplemented. If the OST filesystem
588 is corrupted due to disk cache incoherency and then replay is
589 started it is possible to trip an assertion. Avoid putting
590 committed RPCs into the replay list at all to avoid this issue.
593 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
595 Description: System hang on startup
596 Details : This bug allowed the liblustre (e.g. catamount) client to
597 return to the app before handling all startup RPCs. This
598 could leave the node unresponsive to lustre network traffic
599 and manifested as a server ptllnd timeout.
601 Severity : enhancement
603 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
604 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
605 boolean to control whether to print peer debug info when a
606 client's RPC times out.
609 Frequency : only for kernels with patches from Lustre below 1.4.3
611 Description: Remove old rdonly API
612 Details : Remove old rdonly API which unused from at least lustre 1.4.3
615 Frequency : only for devices with external journals
617 Description: Set external device read-only also
618 Details : During a commanded failover stop, we set the disk device
619 read-only while the server shuts down. We now also set any
620 external journal device read-only at the same time.
623 Frequency : when upgrading from 1.4 while trying to change parameters
625 Description: The wrong (new) MDC name was used when setting parameters for
626 upgraded MDT's. Also allows changing of OSC (and MDC)
627 parameters if --writeconf is specified at tunefs upgrade time.
630 Frequency : when setting specific ost indicies
632 Description: QOS code breaks on skipped indicies
633 Details : Add checks for missing OST indicies in the QOS code, so OSTs
634 created with --index need not be sequential.
636 Severity : enhancement
638 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
639 Details : The uninit_groups feature works in conjunction with the kernel
640 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
641 up the pass1 processing of e2fsck. This is a read-only feature
642 in ldiskfs2 only, so older kernels and current ldiskfs cannot
643 mount filesystems that have had this feature enabled.
645 Severity : enhancement
647 Description: Improve multi-block allocation algorithm to avoid fragmentation
648 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
649 allocation locality and avoid filesystem fragmentation.
651 ------------------------------------------------------------------------------
653 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
655 * Support for kernels:
656 2.4.21-47.0.1.EL (RHEL 3)
658 2.6.9-42.0.10.EL (RHEL 4)
659 2.6.12.6 vanilla (kernel.org)
660 2.6.16.27-0.9 (SLES 10)
661 * Recommended e2fsprogs version: 1.39.cfs5
663 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
667 Frequency : occasional, depends on client load and configuration
668 Bugzilla : 12181, 12203
669 Description: data loss for recently-modified files
671 Details : In some cases it is possible that recently written or created
672 files may not be written to disk in a timely manner (this should
673 normally be within 30s unless client IO load is very high).
674 The problem appears as zero-length files or files that are a
675 multiple of 1MB in size after a client crash or client eviction
676 that are missing data at the end of the file.
678 This problem is more likely to be hit on clients where files are
679 repeatedly created and unlinked in the same directory, clients
680 have a large amount of RAM, have many CPUs, the filesystem has
681 many OSTs, the clients are rebooted frequently, and/or the files
682 are not accessed by other nodes after being written.
684 The presence of the problem can be detected by looking at
685 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
686 smaller than the second (nr_unused) then dirty files will not
687 be flushed automatically to disk. "sync; sleep 10" should be
688 run several times on the node before unmounting it to update
689 Lustre (this is also safe to run on nodes without this problem).
691 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
692 that can cause this same problem, so customers using that kernel
693 also need to update the kernel in addition to Lustre. In order
694 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
696 It is normal that files written just before a client crash (less
697 than 30s) may not yet have been flushed to disk, even for local
701 Frequency : frequent on thin XT3 nodes
703 Description: UUID collision on thin XT3 Linux nodes
704 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
705 randomly, since we relied on an insufficiently-seeded PRNG.
710 Description: OSS hangs after "All ost request buffers busy"
711 Details : A deadlock between quota and journal operations caused OSS
712 hangs after printing "All ost request buffers busy."
715 Frequency : always on liblustre builds
717 Description: Cleanup compiler warnings on liblustre
720 Frequency : always on liblustre builds on XT3
722 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
726 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
727 > 32000 subdirectories
728 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
729 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
730 subdirectory count crosses 32000. This will aid e2fsck to
731 correctly handle more than 32000 subdirectories.
734 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
736 Description: System hang on startup
737 Details : This bug allowed the liblustre (e.g. catamount) client to
738 return to the app before handling all startup RPCs. This
739 could leave the node unresponsive to lustre network traffic
740 and manifested as a server ptllnd timeout.
742 Severity : enhancement
744 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
745 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
746 boolean to control whether to print peer debug info when a
747 client's RPC times out.
752 Description: make O_SYNC working on 2.6 kernels
753 Details : 2.6 kernels use different method for mark pages for write,
754 so need add a code to lustre for O_SYNC work.
759 Description: Failure to close file and release space on NFS
760 Details : Put inode details into lock acquired in ll_intent_file_open.
761 Use mdc_intent_lock in ll_intent_open to properly
762 detect all kind of errors unhandled by mdc_enqueue.
767 Description: proc file read during shutdown sometimes raced obd removal,
769 Details : Add lock to prevent obd access after proc file removal.
772 Frequency : Only for files larger than 4GB on 32-bit clients.
774 Description: improperly doing page alignment of locks
775 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
776 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
781 Description: RPCs being resent when they shouldn't be
782 Details : Some RPCs that should not be resent are being resent. This
783 can cause inconsistencies in the RPC state machine. Do not
784 resend such requests.
787 Frequency : rare, only with NFS export
789 Description: Crash on NFS re-export node
790 Details : under very unusual load conditions an assertion is hit in
791 ll_intent_file_open()
794 Frequency : only if OST filesystem is corrupted
796 Description: client incorrectly hits assertion in ptlrpc_replay_req()
797 Details : for a short time RPCs with bulk IO are in the replay list,
798 but replay of bulk IOs is unimplemented. If the OST filesystem
799 is corrupted due to disk cache incoherency and then replay is
800 started it is possible to trip an assertion. Avoid putting
801 committed RPCs into the replay list at all to avoid this issue.
806 Description: large O_DIRECT requests fail under memory pressure/fragmentation
807 Details : Large single O_DIRECT read and write calls can fail to allocate
808 a sufficiently large buffer to process the request. In case of
809 allocation failure the allocation is retried with a smaller
810 buffer and broken into smaller requests.
812 Severity : enhancement
814 Description: Add -o localflock option to simulate old noflock behaviour.
815 Details : This will achieve local-only flock/fcntl locks coherentness.
820 Description: versioning check is incomplete
821 Details : Checking the version difference of client vs. server, report
822 error if the gap is too big.
827 Description: add support PG_writeback bit
828 Details : add support for PG_writeback bit for Lustre, for more carefull
829 work with page cache in 2.6 kernel. This also fix some deadlocks
830 and remove hack for work O_SYNC with 2.6 kernel.
832 Severity : enhancement
834 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
835 Details : The uninit_groups feature works in conjunction with the kernel
836 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
837 up the pass1 processing of e2fsck. This is a read-only feature
838 in ldiskfs2 only, so older kernels and current ldiskfs cannot
839 mount filesystems that have had this feature enabled.
841 Severity : enhancement
843 Description: Improve multi-block allocation algorithm to avoid fragmentation
844 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
845 allocation locality and avoid filesystem fragmentation.
847 ------------------------------------------------------------------------------
849 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
851 * Support for kernels:
852 2.6.9-42.0.3.EL (RHEL 4)
854 2.4.21-47.0.1.EL (RHEL 3)
855 2.6.12.6 vanilla (kernel.org)
856 2.6.16.21-0.8 (SLES10)
857 * Recommended e2fsprogs version: 1.39.cfs2-0
859 * The backwards-compatible /proc/sys/portals symlink has been removed
860 in this release. Before upgrading, please ensure that you change
861 any configuration scripts or /etc/sysctl.conf files that access
862 /proc/sys/portals/* or sysctl portals.* to use the corresponding
863 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
864 in advance of the upgrade on any system running Lustre 1.4.6 or
865 newer, since /proc/sys/lnet was added in that version.
866 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
870 Frequency : only when quota is used
872 Description: avoid scanning export list for quota master
873 Details : Change the algorithms to avoid scanning export list in order
874 to improve the efficiency.
877 Frequency : MDS failover only, very rarely
879 Description: "went back in time" messages on mds failover
880 Details : The greatest transno may be lost when the current operation
881 finishes with an error (transno==0) and the client's last_rcvd
882 record is over-written. Save the greatest transno in the
883 mds_last_transno for this case.
886 Frequency : always for specific kernels and striping counts
888 Description: client may get "Matching packet too big" without ACL support
889 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
890 when trying to access files in certain configurations. The
891 clients should in fact be denied when mounting because they do
895 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
897 Description: many clients connecting with IO in progress causes connect timeouts
898 Details : Avoid synchronous journal commits to avoid delays caused by many
899 clients connecting/disconnecting when bulk IO is in progress.
900 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
901 OST_IO_PORTAL to avoid delays behind potentially many pending
905 Frequency : occasionally with multiple writers to a single file
907 Description: shared writes to file may result in wrong size reported by stat()
908 Details : Allow growing of kms when extent lock is cancelled
911 Frequency : always with random mmap IO to multi-striped file
913 Description: mmap write might be lost if we are writing to a 'hole' in stripe
914 Details : Only if the hole is at the end of OST object so that kms is too
915 small. Fix is to increase kms accordingly in ll_nopage.
918 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
920 Description: writes to a missing object would leak memory on the OST
921 Details : If there is an inconsistency between the MDS and OST filesystems,
922 such that the MDS references an object that doesn't exist, writes
923 to that object will leak memory due to incorrect cleanup in the
924 error handling path, eventually running out of memory on the OST.
929 Description: Creating too long symlink causes lustre errors
930 Details : Check symlink and name lengths before sending requests to MDS.
933 Frequency : only if flock is enabled (not on by default)
935 Description: posix locks not released on fd closure on 2.6.9+
936 Details : We failed to add posix locks to list of inode locks on 2.6.9+
937 kernels, this caused such locks not to be released on fd close and
938 then assertions on fs unmount about still used locks.
941 Frequency : MDS failover only, very rarely
943 Description: clients may get ASSERTION(granted_lock != NULL)
944 Details : When request was taking a long time, and a client was resending
945 a getattr by name lock request. The were multiple lock requests
946 with the same client lock handle and
947 mds_getattr_name->fixup_handle_for_resent_request found one of the
948 lock handles but later failed with ASSERTION(granted_lock != NULL).
953 Description: handle->h_buffer_credits > 0, assertion failure
954 Details : h_buffer_credits is zero after truncate, causing assertion
955 failure. This patch extends the transaction or creates a new
959 Frequency : NFS re-export or patchless client
960 Bugzilla : 11179, 10796
961 Description: Crash on NFS re-export node (__d_move)
962 Details : We do not want to hash the dentry if we don't have a lock.
963 But if this dentry is later used in d_move, we'd hit uninitialised
964 list head d_hash, so we just do this to init d_hash field but
965 leave dentry unhashed.
968 Frequency : NFS re-export or patchless client
970 Description: NFS exports has problem with symbolic link
971 Details : lustre client didn't properly install dentry when re-exported
972 to NFS or running patchless client.
975 Frequency : NFS re-export or patchless client
977 Description: Various nfs/patchless fixes.
978 Details : fixes reuse disconected alias for lookup process - this fixes
979 warning "find_exported_dentry: npd != pd",
980 fix permission error with open files at nfs.
981 fix apply umask when do revalidate.
984 Frequency : occasional
986 Description: Crash on NFS re-export node
987 Details : calling clear_page() on the wrong pointer triggered oops in
988 generic_mapping_read().
991 Frequency : rarely, using O_DIRECT IO
993 Description: unaligned directio crashes client with LASSERT
994 Details : check for unaligned buffers before trying any requests.
997 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
999 Description: stale data returned from RAID cache
1000 Details : If only a small amount of IO is done to the RAID device before
1001 reading it again it is possible to get stale data from the RAID
1002 cache instead of reading it from disk.
1005 Frequency : always for sles10 kernel
1007 Description: sles10 support
1008 Details : ll_follow_link: compile fixes and using of nd_set_link
1009 under newer kernels.
1012 Frequency : depends on arch, kernel and compiler version, always on sles10
1015 Description: recursive or deep enough symlinks cause stack overflow
1016 Details : getting rid of large stack-allocated variable in
1020 Frequency : depends on hardware
1022 Description: lustre write performance loss in the SLES10 kernel
1023 Details : the performance loss is caused by using of write barriers in the
1024 ext3 code. The SLES10 kernel turns barrier support on by
1025 default. The fix is to undo that change for ldiskfs.
1027 ------------------------------------------------------------------------------
1029 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1031 * Support for kernels:
1032 2.6.9-42.0.3EL (RHEL 4)
1033 2.6.5-7.276 (SLES 9)
1034 2.4.21-47.0.1.EL (RHEL 3)
1035 2.6.12.6 vanilla (kernel.org)
1039 Frequency : quota enabled and large files being deleted
1041 Description: releasing more than 4GB of quota at once hangs OST
1042 Details : If a user deletes more than 4GB of files on a single OST it
1043 will cause the OST to spin in an infinite loop. Release
1044 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1049 Description: statfs data retrieved from /proc may be stale or zero
1050 Details : When reading per-device statfs data from /proc, in the
1051 {kbytes,files}_{total,free,avail} files, it may appear
1052 as zero or be out of date.
1055 Frequency : systems with MD RAID1 external journal devices
1057 Description: lconf's call to blkid is confused by RAID1 journal devices
1058 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1059 of returning all block devices that match the journal UUID.
1062 Frequency : always, for aggregate stripe size over 4GB
1064 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1065 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1066 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1069 Frequency : always if "lfs find" used on a local file/directory
1071 Description: "lfs find" segfaults if used on a local file/directory
1072 Details : The case where a directory component was not specified wasn't
1073 handled correctly. Handle this properly.
1076 Frequency : always on ppc64
1078 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1079 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1080 perform endianness conversion.
1083 Frequency : rarely (truncate to non-zero file size after write under load)
1084 Bugzilla : 10730, 10687
1085 Description: Files padded with zeros to next 4K multiple
1086 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1087 it is possible that files that are truncated to a non-zero size
1088 immediately after being written are filled with zero bytes beyond
1089 the truncated size. No file data is lost.
1091 Severity : enhancement
1093 Description: Allow recovery/failover for liblustre clients.
1094 Details : liblustre clients were unaware of failover configurations until
1097 Severity : enhancement
1099 Description: user file locks should fail when not mounting with flock option
1100 Details : Set up an error-returning stub in ll_file_operations.lock field
1101 to prevent incorrect behaviour when client is mounted without
1102 flock option. Also, set up properly f_op->flock field for
1106 Frequency : always on ia64
1108 Description: "lfs df" loops on printing out MDS statfs information
1109 Details : The obd_ioctl_data was not initialized and in some systems
1110 this caused a failure during the ioctl that did not return
1111 an error. Initialize the struct and return an error on failure.
1114 Frequency : SLES 9 only
1116 Description: Error of copying files with lustre special EAs as root
1117 Details : Client side always return success for setxattr call for lustre
1118 special xattr (currently only "trusted.lov").
1121 Frequency : rarely on clusters with both ia64+i386 clients
1123 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1124 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1125 clients are doing concurrent writes to the same file it is
1126 possible that the ia64 clients may LASSERT because the OST
1127 extent locks are not PAGE_SIZE aligned. Ensure that grown
1128 locks are always aligned on the request boundary.
1131 Frequency : specific use, occasional
1133 Description: Overwriting in use executable truncates on-disk binary image
1134 Details : If one node attempts to overwrite an executable in use by
1135 another node, we now correctly return ETXTBSY instead of
1136 truncating the file.
1138 Severity : enhancement
1140 Description: Async OSC create to avoid the blocking unnecessarily.
1141 Details : If a OST has no remain object, system will block on the creating
1142 when need to create a new object on this OST. Now, ways use
1143 pre-created objects when available, instead of blocking on an
1144 empty osc while others are not empty. If we must block, we block
1145 for the shortest possible period of time.
1150 Description: chmod on Lustre root is propagated to other clients
1151 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1152 invalid by the follow_mount time.
1155 Frequency : liblustre clients only
1157 Description: Race in 'instant cancel' lock handling could lead to such locks
1158 never to be granted in case of SMP MDS
1159 Details : Do not destroy not yet granted but cbpending locks in
1163 Frequency : replay/resend of open
1165 Description: non null lock assetion failure in mds_intent_policy
1166 Details : Trying to replay/resend lockless open requests resulted in
1167 mds_open() returning 0 with no lock. Now it sets a flag if
1168 a lock is going to be returned.
1170 Severity : enhancement
1172 Description: Checksum enhancements
1173 Details : New checksum enhancements allow for resending RPCs that failed
1176 Severity : enhancement
1178 Description: Tunables on number of dirty pages in cacche
1179 Details : Allow to set limit on number of dirty pages cached.
1184 Description: client crash on unmount - lock still has references
1185 Details : In some error handling cases it was possible to leak a lock
1186 reference on a client while accessing a file. This was not
1187 harmful to the client during operation, but would cause the
1188 client to crash when the filesystem is unmounted.
1191 Frequency : specific case, rare
1193 Description: ETXTBSY on mds though file not in use by client
1194 Details : ETXTBSY is no longer incorrectly returned when attempting to
1195 chmod or chown a directory that the user previously tried to
1196 execute or a currently-executing binary.
1199 Frequency : extremely rare except on liblustre-based clients
1201 Description: Lustre space not freed when files are deleted
1202 Details : Clean up open-unlinked files after client eviction. Previously
1203 the unlink was skipped and the files remained as orphans.
1208 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1209 Details : In some cases it was possible to send a blocking callback to
1210 a client doing a glimpse, even though that client didn't get
1211 a lock granted. When the glimpse lock is cancelled on the OST
1212 the freed lock is left on the waiting list and corrupted the list.
1215 Frequency : all core dumps
1217 Description: Broke core dumps to lustre
1218 Details : Negative dentry may be unhashed if parent does not have UPDATE
1219 lock, but some callers, e.g. do_coredump, expect dentry to be
1220 hashed after successful create, hash it in ll_create_it.
1222 ------------------------------------------------------------------------------
1224 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1226 * Support for kernels:
1227 2.6.9-42.0.2.EL (RHEL 4)
1228 2.6.5-7.276 (SLES 9)
1229 2.4.21-40.EL (RHEL 3)
1230 2.6.12.6 vanilla (kernel.org)
1234 Frequency : always on RHEL 3
1236 Description: Number of open files grows over time
1237 Details : The number of open files grows over time, whether or not
1238 Lustre is started. This was due to a filp leak introduced
1239 by one of our kernel patches.
1241 ------------------------------------------------------------------------------
1243 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1245 * Support for kernels:
1246 2.6.9-42.EL (RHEL 4)
1247 2.6.5-7.276 (SLES 9)
1248 2.4.21-40.EL (RHEL 3)
1249 2.6.12.6 vanilla (kernel.org)
1254 Bugzilla : 5719, 9635, 9792, 9684
1255 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1256 Details : If a server is under heavy load and cannot reply to new
1257 connection requests before the client resends the (re)connect,
1258 the connection handling code can behave badly if two service
1259 threads are concurrently handing separate (re)connections from
1260 the same client. Add better locking to the connection handling
1261 code, and ensure that only a single connection will be processed
1262 for a given client UUID, even if the lock is dropped.
1264 Severity : enhancement
1266 Description: add TCP zero-copy support to kernel
1267 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1268 sends if the hardware supports scatter-gather and checksumming.
1269 This allows socklnd to do client-write and server-read more
1270 efficiently and reduce CPU utilization from skbuf copying.
1273 Frequency : only if NFS exporting from client
1275 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1276 Details : If setting attributes on a file created under NFS that had
1277 never been opened it would be possible to oops the client
1278 if the file had no objects.
1281 Frequency : always for liblustre
1283 Description: liblustre client does MDS+OSTs setattr RPC for each write
1284 Details : When doing a write from a liblustre client, the client
1285 incorrectly issued an RPC to the MDS and each OST the file was
1286 striped over in order to update the timestamps. When writing
1287 with small chunks and many clients this could overwhelm the MDS
1288 with RPCs. In all cases it would slow down the write because
1289 these RPCs are unnecessary.
1291 Severity : enhancement
1293 Description: allow number of MDS service threads to be changed at module load
1294 Details : It is now possible to change the number of MDS service threads
1295 running. Adding "options mds mds_num_threads={N}" to the MDS's
1296 /etc/modprobe.conf will set the number of threads for the next
1297 time Lustre is restarted (assuming the "mds" module is also
1298 reloaded at that time). The default number of threads will
1299 stay the same, 32 for most systems.
1304 Description: OST crash if filesystem is unformatted or corrupt
1305 Details : If an OST is started on a device that has never been formatted
1306 or if the filesystem is corrupt and cannot even mount then the
1307 error handling cleanup routines would dereference a NULL pointer.
1312 Description: NULL pointer deref in llap_from_page.
1313 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1314 incorrect) mapping if the page was truncated/reclaimed while it was
1315 searched for. Check for this condition and skip such pages when
1316 doing readahead. Introduce extra check to llap_from_page() to
1317 verify page->mapping->host is non-NULL (so page is not anonymous).
1320 Frequency : Sometimes when using sys_sendfile
1322 Description: "page not covered by a lock" warnings from ll_readpage
1323 Details : sendfile called ll_readpage without right page locks present.
1324 Now we introduced ll_file_sendfile that does necessary locking
1325 around call to generic_file_sendfile() much like we do in
1329 Frequency : with certain MDS communication failures at client mount time
1331 Description: NULL pointer deref after failed client mount
1332 Details : a client connection request may delayed by the network layer
1333 and not be sent until after the PTLRPC layer has timed out the
1334 request. If the client fails the mount immediately it will try
1335 to clean up before the network times out the request. Add a
1336 reference from the request import to the obd device and delay
1337 the cleanup until the network drops the request.
1340 Frequency : occasionally during client (re)connect
1342 Description: assertion failure during client (re)connect
1343 Details : processing a client connection request may be delayed by the
1344 client or server longer than the client connect timeout. This
1345 causes the client to resend the connection request. If the
1346 original connection request is replied in this interval, the
1347 client may trip an assertion failure in ptlrpc_connect_interpret()
1348 which thought it would be the only running connect process.
1351 Frequency : only with obd_echo servers and clients that are rebooted
1353 Description: kernel BUG accessing uninitialized data structure
1354 Details : When running an obd_echo server it did not start the ping_evictor
1355 thread, and when a client was evicted an uninitialized data
1356 structure was accessed. Start the ping_evictor in the RPC
1357 service startup instead of the OBD startup.
1359 Severity : enhancement
1360 Bugzilla : 10193 (patchless)
1361 Description: Remove dependency on various unexported kernel interfaces.
1362 Details : No longer need reparent_to_init, exit_mm, exit_files,
1363 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1366 Frequency : rare (only users of deprecated and unsupported LDAP config)
1368 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1369 Details : LDAP apparently contains 'lustreName' attributes instead of
1370 'name'. A simple remapping of the name is sufficient.
1373 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1375 Description: waiting_locks_callback trips kernel BUG if client is evicted
1376 Details : Running with the dump_on_timeout debug flag turned on makes
1377 it possible that the waiting_locks_callback() can try to dump
1378 the Lustre kernel debug logs from an interrupt handler. Defer
1379 this log dumping to the expired_lock_main() thread.
1381 Severity : enhancement
1383 Description: Support NFS exporting on 2.6 kernels.
1384 Details : Implement non-rawops metadata methods for NFS server to use without
1385 changing NFS server code.
1388 Frequency : very rare (synthetic metadata workload only)
1390 Description: two racing renames might cause an MDS thread to deadlock
1391 Details : Running the "racer" program may cause one MDS thread to rename
1392 a file from being the source of a rename to being the target of
1393 a rename at exactly the same time that another thread is doing
1394 so, and the second thread has already enqueued these locks after
1395 doing a lookup of the target and is trying to relock them in
1396 order. Ensure that we don't try to re-lock the same resource.
1399 Frequency : only very large systems with liblustre clients
1401 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1402 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1403 all OSTs in parallel. This allows the eviction of stale liblustre
1404 clients to proceed much faster than if they were done in series,
1405 and also offers similar improvements for other set_info RPCs.
1410 Description: excessive CPU usage during initial read phase on client
1411 Details : During the initial read phase on a client, it would agressively
1412 retry readahead on the file, consuming too much CPU and impacting
1413 performance (since 1.4.5.8). Improve the readahead algorithm
1414 to avoid this, and also improve some other common cases (read
1415 of small files in particular, where "small" is files smaller than
1416 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1421 Description: MDS crash when receiving packet with unknown intent.
1422 Details : Do not LBUG in unknown intent case, just return -EFAULT
1424 Severity : enhancement
1425 Bugzilla : 9293, 9385
1426 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1427 Details : Do not serialize getattr (non-intent version) and statfs.
1430 Frequency : occasional, when OST network is overloaded/intermittent
1432 Description: client evicted by OST after bulk IO timeout
1433 Details : If a client sends a bulk IO request (read or write) the OST
1434 may evict the client if it is unresposive to its data GET/PUT
1435 request. This is incorrect if the network is overloaded (takes
1436 too long to transfer the RPC data) or dropped the OST GET/PUT
1437 request. There is no need to evict the client at all, since
1438 the pinger and/or lock callbacks will handle this, and the
1439 client can restart the bulk request.
1442 Frequency : Always when mmapping file with no objects
1444 Description: client crashes when mmapping file with no objects
1445 Details : Check that we actually have objects in a file before doing any
1446 operations on objects in ll_vm_open, ll_vm_close and
1452 Description: Request leak when working with deleted CWD
1453 Details : Introduce advanced request refcount tracking for requests
1454 referenced from lustre intent.
1456 Severity : Enhancement
1458 Description: Cache open file handles on client.
1459 Details : MDS now will return special lock along with openhandle, if
1460 requested and client is allowed to hold openhandle, even if unused,
1461 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1462 closing files for every read/write openration.
1464 Severity : Enhancement
1466 Description: Cache open negative dentries on client when possible.
1467 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1468 negative dentries on lock revocation.
1473 Description: Remounting a client read-only wasn't possible with a zconf mount
1474 Details : It wasn't possible to remount a client read-only with llmount.
1476 Severity : enhancement
1477 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1478 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1479 MPICH is included to add Lustre-specific ADIO interfaces.
1480 This is based closely on the UFS ADIO layer and only differs
1481 in file creation, in order to allow the OST striping to be set.
1482 This is user-contributed code and not supported by CFS.
1487 Description: extended inode attributes (immutable, append-only) work improperly
1488 when 2.4 and 2.6 kernels are used on client/server or vice versa
1489 Details : Introduce kernel-independent values for these flags.
1491 Severity : enhancement
1494 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1495 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1496 granularity. Now, Fractional MB granularity is be supported,
1497 this is very useful for low memory system.
1499 Severity : enhancement
1501 Description: Getattr by fid
1502 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1503 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1504 have locks and avoids getattr rpc requests for it on every stat.
1507 Frequency : Always, for filesystems larger than 2TB
1509 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1510 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1511 filesystems larger than 2TB to be mounted with mballoc enabled.
1514 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1516 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1517 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1518 filesystems larger than 2TB is mounted then the kernel will
1519 silently corrupt the start of the filesystem. CONFIG_LBD is
1520 enabled for all CFS-supported kernels, but the possibility of
1521 this happening with a modified kernel config exists.
1523 Severity : enhancement
1525 Description: add client O_DIRECT support for 2.6 kernels
1526 Details : It is now possible to do O_DIRECT reads and writes to files
1527 in the Lustre client mountpoint on 2.6 kernel clients.
1529 Severity : enhancement
1531 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1532 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1533 parallel, not waiting for response from every OST before sending
1534 a rpc to the next OST.
1539 Description: setattr vs write race when updating file timestamps
1540 Details : Client processes that update a file timestamp into the past
1541 right after writing to the file (e.g. tar) it is possible that
1542 the updated file modification time can be reset to the current
1543 time due to a race between processing the setattr and write RPC.
1545 Severity : enhancement
1547 Description: Bring 'lfs find' closer in line with regular Linux find.
1548 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1549 -print0 options and obtains all the needed info through the lustre
1552 Severity : enhancement
1554 Description: support up to 1024 configured devices on one node
1555 Details : change obd_dev array from statically allocated to dynamically
1556 allocated structs as they are first used to reduce memory usage
1561 Description: Flush dirty partially truncated pages during truncate
1562 Details : Immediatelly flush partially truncated pages in filter_setattr,
1563 this way we completely avoid having any pages in page cache on OST
1564 and can retire ugly workarounds during writes to flush such pages.
1569 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1570 Details : protect inode from truncation within vfs_unlink() context
1571 just take a reference before calling vfs_unlink() and release it
1572 when parent's i_sem is free.
1577 Description: last_id value checked outside lock on OST caused LASSERT failure
1578 Details : If there were multiple MDS->OST object precreate requests in
1579 flight, it was possible that the OST's last object id was checked
1580 outside a lock and incorrectly tripped an assertion. Move checks
1581 inside locks, and discard old precreate requests.
1584 Frequency : always, if extents are used on OSTs
1586 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1587 Details : OSTs using the extents format would not zero the high 16 bits of
1588 the index physical block number. This is not a problem for any
1589 OST filesystems smaller than 16TB, and no kernels support ext3
1590 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1591 new/modified files) and can be fixed for existing filesystems
1592 with e2fsprogs-1.39-cfs1.
1597 Description: import connection selection may be incorrect if timer wraps
1598 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1599 import connections to be ignored if the 32-bit jiffies counter
1600 wraps. Use a 64-bit jiffies counter.
1603 Frequency : very large clusters immediately after boot
1605 Description: LNET request buffers exhausted under heavy short-term load
1606 Details : If a large number of client requests are generated on a service
1607 that has previously never seen so many requests it is possible
1608 that the request buffer growth cannot keep up with the spike in
1609 demand. Instead of dropping incoming requests, they are held in
1610 the LND until the RPC service can accept more requests.
1613 Frequency : Sometimes during replay
1615 Description: Assertion failure in ll_local_open after replay.
1616 Details : If replay happened on an open request reply before we were able
1617 to set replay handler, reply will become not swabbed tripping the
1618 assertion in ll_local_open. Now we set the handler right after
1619 recognising of open request
1622 Frequency : very rare
1624 Description: kernel reports "badness in vsnprintf"
1625 Details : Reading from the "recovery_status" /proc file in small chunks
1626 may cause a negative length in lprocfs_obd_rd_recovery_status()
1627 call to vsnprintf() (which is otherwise harmless). Exit early
1628 if there is no more space in the output buffer.
1630 Severity : enhancement
1632 Description: clear OBD RPC statistics by writing to them
1633 Details : It is now possible to clear the OBD RPC statistics by writing
1634 to the "stats" file.
1639 Description: Client mtime is not the same on different clients after utimes
1640 Details : In some cases, the client was using the utimes() syscall on
1641 a file cached on another node. The clients now validate the
1642 ctime from the MDS + OSTs to determine which one is right.
1647 Description: Inability to activate failout mode
1648 Details : lconf script incorrectly assumed that in pythong string's numeric
1649 value is used in comparisons.
1652 Frequency : always with multiple stripes per file
1654 Description: Inefficient object allocation for mutli-stripe files
1655 Details : When selecting which OSTs to stripe files over, for files with
1656 a stripe count that divides evenly into the number of OSTs,
1657 the MDS is always picking the same starting OST for each file.
1658 Return the OST selection heuristic to the original design.
1663 Description: mount failures may take full timeout to return an error
1664 Details : Under some heavy load conditions it is possible that a
1665 failed mount can wait for the full obd_timeout interval,
1666 possibly several minutes, before reporting an error.
1667 Instead return an error as soon as the status is known.
1669 Frequency : quota enabled and large files being deleted
1671 Description: releasing more than 4GB of quota at once hangs OST
1672 Details : If a user deletes more than 4GB of files on a single OST it
1673 will cause the OST to spin in an infinite loop. Release
1674 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1679 Description: statfs data retrieved from /proc may be stale or zero
1680 Details : When reading per-device statfs data from /proc, in the
1681 {kbytes,files}_{total,free,avail} files, it may appear
1682 as zero or be out of date.
1685 Frequency : systems with MD RAID1 external journal devices
1687 Description: lconf's call to blkid is confused by RAID1 journal devices
1688 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1689 of returning all block devices that match the journal UUID.
1692 Frequency : always, for aggregate stripe size over 4GB
1694 Description: assertion fails when trying to use 4GB stripe size
1695 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
1696 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
1699 Frequency : always on ppc64
1701 Description: the first write on an ext3 filesystem with mballoc got stuck
1702 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1703 perform endianness conversion.
1705 ------------------------------------------------------------------------------
1707 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1709 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1710 INTEROPERATE with older versions automatically. Please read the
1711 user documentation before upgrading any part of a live system.
1712 * WARNING: Lustre networking configuration changes are required with
1713 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1716 * Support for kernels:
1717 2.6.9-22.0.2.EL (RHEL 4)
1718 2.6.5-7.244 (SLES 9)
1719 2.6.12.6 vanilla (kernel.org)
1722 Severity : enhancement
1723 Bugzilla : 7981/8208
1724 Description: Introduced Lustre Networking (LNET)
1725 Details : LNET is new networking infrastructure for Lustre, it includes
1726 a reorganized network configuration mode (see the user
1727 documentation for full details) as well as support for routing
1728 between different network fabrics. Lustre Networking Devices
1729 (LNDs) for the supported network fabrics have also been
1730 created for this new infrastructure.
1732 Severity : enhancement
1733 Description: Introduced Access control lists
1734 Details : clients can set ACLs on files and directories in order to have
1735 more fine-grained permissions than the standard Unix UGO+RWX.
1736 The MDS must be started with the "-o acl" mount option.
1738 Severity : enhancement
1739 Description: Introduced filesystem quotas
1740 Details : Administrators may now establish per-user quotas on the
1743 Severity : enhancement
1745 Description: Configuration change for the XT3
1746 The PTLLND is now used to run Lustre over Portals on the XT3
1747 The configure option(s) --with-cray-portals are no longer used.
1748 Rather --with-portals=<path-to-portals-includes> is used to
1749 enable building on the XT3. In addition to enable XT3 specific
1750 features the option --enable-cray-xt3 must be used.
1755 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1756 Details : A race between allocating a new llcd and re-getting the llcd_lock
1757 allowed another thread to grab newly-allocated llcd.
1759 Severity : enhancement
1761 Description: 2.6 OST async journal commit and locking fix to improve performance
1762 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1763 2.6 kernels are now async as they already were in 2.4 kernels so
1764 that they can commit concurrently with the network bulk transfer.
1765 For block-allocated files the filter allocation semaphore is held
1766 to avoid filesystem fragmentation during allocation. BKL lock
1767 removed for 2.6 xattr operations where it is no longer needed.
1772 Description: lconf incorrectly determined whether two IP networks could talk
1773 Details : In some more complicated routing and multiple-network
1774 configurations, lconf will avoid trying to make a network
1775 connection to a disjoint part of the IP space. It was doing the
1776 math incorrectly for one set of cases.
1781 Description: Fix for potential infinite loop processing records in an llog.
1782 Details : If an llog record is corrupted/zeroed, it is possible to loop
1783 forever in llog_process(). Validate the llog record length
1784 and skip the remainder of the block on error.
1787 Frequency : occasional (liblustre only)
1789 Description: liblustre could not open files whose last component is a symlink
1790 Details : sysio_path_walk() would incorrectly pass the open intent to
1791 intermediate path components.
1794 Frequency : rare (liblustre only with non-standard tuning)
1795 Bugzilla : 7201 (7350)
1796 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1797 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1798 after it has been released, only do this for referenced locks
1800 Severity : enhancement
1802 Description: specifying an (invalid) directory default stripe_size of -1
1803 would reset the directory default striping
1804 Details : stripe_size -1 was used internally to signal directory stripe
1805 removal, now use "all default" to signal dir stripe removal
1806 as a directory striping of "all default" is not useful
1809 Frequency : common for large clusters running liblustre clients
1811 Description: doing an ls when liblustre clients are running is slow
1812 Details : sending a glimpse AST to a liblustre client waits for every AST
1813 to time out, as liblustre clients will not respond. Since they
1814 cannot cache data we refresh the OST lock LVB from disk instead.
1816 Severity : enhancement
1818 Description: doing an ls at the same time as file IO can be slow
1819 Details : enqueue and other "small" requests can be blocked behind many
1820 large IO requests. Create a new OST IO portal for non-IO
1821 requests so they can be processed faster.
1824 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1826 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1827 created file with mode 000
1828 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1829 allow an ftruncate() as a non-root user to a file with mode 000.
1830 Lustre now respects this flag to disable mode checks when
1831 truncating a file owned by the user
1834 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1837 Description: Revoking locks from clients that went dead or catatonic might take
1839 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1840 cancellation of such locks instant on servers without waiting for
1841 any reply from clients. Clients drops these locks when cancel
1842 notification from server is received without replying.
1845 Frequency : liblustre-only, when liblustre client dies or becomes busy
1847 Description: Doing ls on Linux clients can take a long time with active
1849 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1850 granting such locks to it in the first place if possible. Locks
1851 are taken by proxy on the OST during the read or write and
1852 dropped immediately afterward. Add connect flags handling, do
1853 not grant locks to liblustre clients for glimpse ASTs.
1855 Severity : enhancement
1857 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1858 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1859 sequential reads and behaves badly with applications doing random
1860 reads. Improve it by reading ahead at least read region, and
1861 avoiding excessive large RPC for small reads.
1863 Severity : enhancement
1865 Description: Creating more than 1000 files for a single job may cause a load
1866 imbalance on the OSTs if there are also a large number of OSTs.
1867 Details : qos_prep_create() uses an OST index reseed value that is an
1868 even multiple of the number of available OSTs so that if the
1869 reseed happens in the middle of the object allocation it will
1870 still utilize the OSTs as uniformly as possible.
1875 Description: OST or MDS may oops in ping_evictor_main()
1876 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1877 but doesn't restart at beginning of obd_exports_timed list
1880 Severity : enhancement
1882 Description: improve by-nid export eviction on the MDS and OST
1883 Details : allow multiple exports with the same NID to be evicted at one
1884 time without re-searching the exports list.
1887 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1889 Description: MDS may oops in groups_free()
1890 Details : in rare race conditions a newly allocated group_info struct is
1891 freed again, and this can be NULL. The 2.4 compatibility code
1892 for groups_free() checked for a NULL pointer, but 2.6 did not.
1895 Frequency : common for liblustre clients doing little filesystem IO
1896 Bugzilla : 9352, 7313
1897 Description: server may evict liblustre clients accessing contended locks
1898 Details : if a client is granted a lock or receives a completion AST
1899 with a blocking AST already set it would not reply to the AST
1900 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1903 Frequency : lfs setstripe, only systems with more than 160 OSTs
1905 Description: unable to set striping with a starting offset beyond OST 160
1906 Details : llapi_create_file() incorrectly limited the starting stripe
1907 index to the maximum single-file stripe count.
1910 Frequency : LDAP users only
1912 Description: lconf did not handle in-kernel recovery with LDAP properly
1913 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1915 Severity : enhancement
1917 Description: bind OST threads to NUMA nodes to improve performance
1918 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1919 node and do their allocations there to localize memory access
1921 Severity : enhancement
1923 Description: llmount can determine client NID directly from Myrinet (GM)
1924 Details : the client NID code from gmnalnid was moved directly into
1925 llmount, removing the need to use this or specifying the
1926 client NID explicitly when mounting GM clients with zeroconf
1929 Frequency : if client is started with down MDS
1931 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1932 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1933 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1934 ptlrpc_import_delayed_req() because we didn't check if the
1935 request was interrupted, and we also didn't break out of the
1936 event loop if there was no timeout
1941 Description: data loss during non-page-aligned writes to a single file from
1942 both multiple nodes and multiple threads on one node at same time
1943 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1944 inconsistency led to false short-reads, that were cached and later
1945 used by ->prepare_write() to fill in partially written page,
1946 leading to data loss.
1949 Frequency : always, if lconf --abort_recovery used
1951 Description: lconf --abort_recovery fails with 'Operation not supported'
1952 Details : lconf was attempting to abort recovery on the MDT device and not
1955 Severity : enhancement
1957 Description: remove cleanup logs
1958 Details : replace lconf-generated cleanup logs with lustre internal
1959 cleanup routines. Eliminates the need for client-cleanup and
1962 Severity : enhancement
1964 Description: add support for EAs (user and system) on lustre filesystems
1965 Details : it is now possible to store extended attributes in the Lustre
1966 client filesystem, and with the user_xattr mount option it
1967 is possible to allow users to store EAs on their files also
1969 Severity : enhancement
1971 Description: Add possibility (config option) to show minimal available OST free
1973 Details : When compiled with --enable-mindf configure option, statfs(2)
1974 (and so, df) will return least minimal free space available from
1975 all OSTs as amount of free space on FS, instead of summary of
1976 free spaces of all OSTs.
1978 Severity : enhancement
1980 Description: do not expand extent locks acquired on OST-side
1981 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1982 by server: they are not cached anyway.
1985 Frequency : when mmap is used/binaries executed from Lustre
1987 Description: Unmmap pages before throwing them away from read cache.
1988 Details : llap_shrink cache now attempts to unmap pages before discarding
1989 them (if unmapping failed - do not discard). SLES9 kernel has
1990 extra checks that trigger if this unmapping is not done first.
1995 Description: lconf didn't resolve symlinks before checking to see whether a
1996 given mountpoint was already in use
1999 Frequency : when migrating failover services
2000 Bugzilla : 6395, 9514
2001 Description: When migrating a subset of services from a node (e.g. failback
2002 from a failover service node) the remaining services would
2003 time out and evict clients.
2004 Details : lconf --force (implied by --failover) sets the global obd_timeout
2005 to 5 seconds in order to quickly disconnect, but this caused
2006 other RPCs to time out too quickly. Do not change the global
2007 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2009 Severity : enhancement
2010 Frequency : if MDS is started with down OST
2011 Bugzilla : 9439,5706
2012 Description: Allow startup/shutdown of an MDS without depending on the
2013 availability of the OSTs.
2014 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2015 Add appropriate locking and lov-osc refcounts for safe
2016 cleaning. Add osc abort_inflight calls in case the
2017 synchronize never started.
2020 Frequency : occasional (Cray XT3 only)
2022 Description: root not authorized to access files in CRAY_PORTALS environment
2023 Details : The client process capabilities were not honoured on the MDS in
2024 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2025 been accessed by an authorized user then root was able to access
2026 the file on the local client also. The root user capabilities
2027 are now allowed on the MDS, as this environment has secure UID.
2030 Frequency : occasional
2032 Description: ldiskfs "too long searching" message happens too often
2033 Details : A debugging message (otherwise harmless) prints too often on
2034 the OST console. This has been reduced to only happen when
2035 there are fragmentation problems on the filesystem.
2040 Description: Division by zero in statfs when all OSCs are inactive
2041 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2042 lov_statfs divides by value returned by lov_get_stripecnt().
2046 Bugzilla : 9489, 3273
2047 Description: First write from each client to each OST was only 4kB in size,
2048 to initialize client writeback cache, which caused sub-optimal
2049 RPCs and poor layout on disk for the first writen file.
2050 Details : Clients now request an initial cache grant at (re)connect time
2051 and so that they can start streaming writes to the cache right
2052 away and always do full-sized RPCs if there is enough data.
2053 If the OST is rebooted the client also re-establishes its grant
2054 so that client cached writes will be honoured under the grant.
2059 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2060 Details : Now I/O RPCs go to different portal number and (presumably) fast
2061 lock requests (and glimses) and other RPCs get their own service
2062 threads pool that should be able to service those RPCs
2065 Severity : enhancement
2067 Description: Ability to exchange lustre version between client and servers and
2068 issue warnings at client side if client is too old. Also for
2069 liblustre clients there is ability to refuse connection of too old
2071 Details : New 'version' field is added to connect data structure that is
2072 filled with version info. That info is later checked by server and
2076 Frequency : rare, liblustre only.
2077 Bugzilla : 9296, 9581
2078 Description: Two simultaneous writes from liblustre at offset within same page
2079 might proceed at the same time overwriting eachother with stale
2081 Details : I/O lock withing llu_file_prwv was released too early, before data
2082 actually was hitting the wire. Extended lock-holding time until
2083 server acknowledges receiving data.
2086 Frequency : extremely rare. Never observed in practice.
2088 Description: avoid generating lustre_handle cookie of 0.
2089 Details : class_handle_hash() generates handle cookies by incrementing
2090 global counter, and can hit 0 occasionaly (this is unlikely, but
2091 not impossible, because initial value of cookie counter is
2092 selected randonly). Value of 0 is used as a sentinel meaning
2093 "unassigned handle" --- avoid it. Also coalesce two critical
2094 sections in this function into one.
2096 Severity : enhancement
2098 Description: allow liblustre clients to delegate truncate locking to OST
2099 Details : To avoid overhead of locking, liblustre client instructs OST to
2100 take extent lock in ost_punch() on client's behalf. New connection
2101 flag is added to handle backward compatibility.
2103 Severity : enhancement
2104 Bugzilla : 4928, 7341, 9758
2105 Description: allow number of OST service threads to be specified
2106 Details : a module parameter allows the number of OST service threads
2107 to be specified via "options ost ost_num_threads={N}" in the
2108 OSS's /etc/modules.conf or /etc/modprobe.conf.
2112 Bugzilla : 6146, 9635, 9895
2113 Description: servers crash with bad pointer in target_handle_connect()
2114 Details : In rare cases when a client is reconnecting it was possible that
2115 the connection request was the last reference for that export.
2116 We would temporarily drop the export reference and get a new
2117 one, but this may have been the last reference and the export
2118 was just destroyed. Get new reference before dropping old one.
2120 Severity : enhancement
2121 Frequency : if client is started with failover MDS
2123 Description: Allow multiple MDS hostnames in the mount command
2124 Details : Try to read the configuration from all specified MDS
2125 hostnames during a client mount in case the "primary"
2128 Severity : enhancement
2130 Description: Stop sending data to evicted clients as soon as possible.
2131 Details : Check if the client we are about to send or are sending data to
2132 was evicted already. (Check is done every second of waiting,
2133 for which l_wait_event interface was extended to allow checking
2134 of exit condition at specified intervals).
2137 Frequency : rare, normally only when NFS exporting is done from client
2139 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2141 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2144 Frequency : always, if srand() is called before liblustre initialization
2146 Description: Liblustre uses system PRNG disturbing its usage by user application
2147 Details : Introduce internal to lustre fast and high-quality PRNG for
2148 lustre usage and make liblustre and some other places in generic
2149 lustre code to use it.
2151 Severity : enhancement
2152 Bugzilla : 9477, 9557, 9870
2153 Description: Verify that the MDS configuration logs are updated when xml is
2154 Details : Check if the .xml configuration logs are newer than the config
2155 logs stored on the MDS and report an error if this is the case.
2156 Request --write-conf, or allow starting with --old_conf.
2158 Severity : enhancement
2160 Description: Handle symlinks in the path when checking if Lustre is mounted.
2161 Details : Resolve intermediate symlinks when checking if a client has
2162 mounted a filesystem to avoid duplicate client mounts.
2167 Description: lconf can hit an error exception but still return success.
2168 Details : The lconf command catches the Command error exception at the top
2169 level script context and will exit with the associated exit
2170 status, but doesn't ensure that this exit status is non-zero.
2175 Description: failure of ptlrpc thread startup can cause oops
2176 Details : Starting a ptlrpc service thread can fail if there are a large
2177 number of threads or the server memory is very fragmented.
2178 Handle this without oopsing.
2181 Frequency : always, only if liblustre and non-default acceptor port was used
2183 Description: liblustre cannot connect to servers with non-default acceptor port
2184 Details : tcpnal_set_default_params() was not called and was therefore
2185 ignoring the environment varaible TCPNAL_PORT, as well as other
2186 TCPNAL_ environment variables
2191 Description: two objects could be created on the same OST for a single file
2192 Details : If an OST is down, in some cases it was possible to create two
2193 objects on a single OST for a single file. No problems other
2194 than potential performance impact and spurious error messages.
2198 Bugzilla : 5681, 9562
2199 Description: Client may oops in ll_unhash_aliases
2200 Details : Client dcache may become inconsistent in race condition.
2201 In some cases "getcwd" can fail if the current directory is
2207 Description: Inode refcounting problems in NFS export code
2208 Details : link_raw functions used to call d_instantiate without obtaining
2209 extra inode reference first.
2213 Bugzilla : 9942, 9903
2214 Description: Referencing freed requests leading to crash, memleaks with NFS.
2215 Details : We used to require that call to ll_revalidate_it was always
2216 followed by ll_lookup_it. Also with revalidate_special() it is
2217 possible to call ll_revalidate_it() twice for the same dentry
2218 even if first occurence returned success. This fix changes semantic
2219 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2220 reference on a request referred from the intent.
2221 ll_intent_release() then releases such a request.
2224 Frequency : rare, normally benchmark loads only
2226 Description: unlinked inodes were kept in memory on the client
2227 Details : If a client is repeatedly creating and unlinking files it
2228 can accumulate a lot of stale inodes in the inode slab cache.
2229 If there is no other client load running this can cause the
2230 client node to run out of memory. Instead flush old inodes
2231 from client cache that have the same inode number as a new inode.
2234 Frequency : SLES9 2.6.5 kernel and long filenames only
2235 Bugzilla : 9969, 10379
2236 Description: utime reports stale NFS file handle
2237 Details : SLES9 uses out-of-dentry names in some cases, which confused
2238 the lustre dentry revalidation. Change it to always use the
2242 Frequency : rare, unless heavy write-truncate concurrency is continuous
2243 Bugzilla : 4180, 6984, 7171, 9963, 9331
2244 Description: OST becomes very slow and/or deadlocked during object unlink
2245 Details : filter_destroy() was holding onto the parent directory lock
2246 while truncating+unlinking objects. For very large objects this
2247 may block other threads for a long time and slow overall OST
2248 responsiveness. It may also be possible to get a lock ordering
2249 deadlock in this case, or run out of journal credits because of
2250 the combined truncate+unlink. Solution is to do object truncate
2251 first in one transaction without parent lock, and then do the
2252 final unlink in a new transaction with the parent lock. This
2253 reduces the lock hold time dramatically.
2256 Frequency : rare, 2.4 kernels only
2258 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2259 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2260 before unmounting their filesystem in order to avoid deadlock.
2261 The kernel_locked() function in 2.4 kernels only checks whether
2262 the kernel lock is held, not whether it is this process that is
2263 holding it as 2.6 kernels do.
2268 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2269 Details : The client ptlrpc code may be trying to reconnect to a down
2270 server before a previous connection attempt has timed out.
2271 Increase the reconnect interval to be longer than the connection
2272 timeout interval to avoid sending duplicate connections to
2276 Frequency : echo_client brw_test command
2278 Description: fix echo_client to work with OST preallocated code
2279 Details : OST preallocation code (5137) didn't take echo_client IO path
2280 into account: echo_client calls filter methods outside of any
2281 OST thread and, hence, there is no per-thread preallocated
2282 pages and buffers to use. Solution: hijack pga pages for IO. As
2283 a byproduct, this avoids unnecessary data copying.
2287 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2288 Description: Client can oops in mdc_commit_close() after open replay
2289 Details : It was possible for the MDS to return an open request with no
2290 transaction number in mds_finish_transno() if the client was
2291 evicted, but without actually returning an error. Clients
2292 would later try to replay that open and may trip an assertion
2293 Simplify the client close codepath, and always return an error
2294 from the MDS in case the open is not successful.
2297 Frequency : rare, 2.6 OSTs only
2299 Description: OST may deadlock under high load on fragmented files
2300 Details : If there was a heavy load and highly-fragmented OST filesystems
2301 it was possible to have all the OST threads deadlock waiting on
2302 allocation of biovecs, because the biovecs were not released
2303 until the entire RPC IO was completed. Instead, release biovecs
2304 as soon as they are complete to ensure forward IO progress.
2306 Severity : enhancement
2308 Description: Support for specifying external journal device at mount
2309 Details : If an OST or MDS device is formatted with an external journal
2310 device, this device major/minor is stored in the ext3 superblock
2311 and may not be valid for failover. Allow detecting and
2312 specifying the external journal at mount time.
2317 Description: Mounting an MDS with pending unlinked files may cause oops
2318 Details : target_finish_recovery() calls mds_postrecov() which returned
2319 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2320 considers this an error and immediately begins cleaning up the
2321 lov, just after starting the mds_lov process
2323 Severity : enhancement
2325 Description: Implement 'lfs df' to report actual free space on per-OST basis
2326 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2327 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2328 usage of objects; '-h' to report in human readable format.
2330 ------------------------------------------------------------------------------
2332 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2339 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2340 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2341 at startup during orphan unlinking. Instead, skip buddy update
2342 before setup, as it will be regenerated after recovery is complete.
2347 Description: If an OST is inactive, its locks might reference stale inodes.
2348 Details : lov_change_cbdata() must iterate over all namespaces, even if
2349 they are inactive to clear inode references from the lock.
2351 Severity : enhancement
2352 Frequency : occasional, if non-standard max_dirty_mb used
2354 Description: Client will block write RPCs if not enough grant
2355 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2356 then the client will block writes while waiting for another RPC
2357 to complete instead of consuming its dirty limit. With change
2358 we get improved performance when max_dirty_mb is small.
2360 Severity : enhancement
2361 Bugzilla : 3389, 6253
2362 Description: Add support for supplementary groups on the MDS.
2363 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2364 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2365 lookups for user supplementary groups into a cache.
2369 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2370 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2371 for liblustre O_EXCL is enforced by MDS.
2375 Description: readdir never returns NULL in liblustre.
2376 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2377 dentry in struct dirent.
2381 Frequency : liblustre only, depends on application IO pattern
2382 Description: liblustre clients evicted if not contacting servers
2383 Details : Don't put liblustre clients into the ping_evictor list, so
2384 they will not be evicted by the pinger ever.
2386 Severity : enhancement
2388 Description: Add ability to evict clients by NID from MDS.
2389 Details : By echoing "nid:$NID" string into
2390 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2391 $NID would be instantly evicted from this MDS and from all active
2392 OSTs connected to it.
2396 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2397 Details : lookup_it_finish() used to query file size from OSTs that was not
2402 Description: service threads change working directory to that of init
2403 Details : Starting lustre service threads may pin the working directory
2404 of the parent thread, making that filesystem busy. Threads
2405 now change to the working directory of init to avoid this.
2409 Frequency : during shutdown only
2410 Description: shutdown with a failed MDS or OST can cause unmount to hang
2411 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2416 Frequency : 2.6 only
2417 Description: chmod/chown may include an extra supplementary group
2418 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2419 supplementary group and if none is specified this is used.
2422 Bugzilla : 5479 (6816)
2424 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2425 Details : If lookup is in progress on a file that is unlinked we might try
2426 to revalidate the inode and fail in revalidate after lookup is
2427 complete and ll_file_open() enqueues the open again but
2428 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2431 Frequency : always, if lconf --abort_recovery used
2433 Description: lconf --abort_recovery fails with 'Operation not supported'
2434 Details : lconf was attempting to abort recovery on the MDT device and not
2437 ------------------------------------------------------------------------------
2439 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2444 Frequency : rare (only unsupported configurations with a node running as an
2446 Bugzilla : 6514, 5137
2447 Description: Mounting a Lustre file system on a node running as an OST could
2449 Details : OSTs now preallocates memory needed to write out data at
2450 startup, instead of when needed, to avoid having to
2451 allocate memory in possibly low memory situations.
2452 Specifically, if the file system is mounted on on OST,
2453 memory pressure could force it to try to write out data,
2454 which it needed to allocate memory to do. Due to the low
2455 memory, it would be unable to do so and the node would
2456 become unresponsive.
2458 Severity : enhancement
2460 Description: Addition of lconf --service command line option
2461 Details : lconf now accepts a '--service <arg>' option, which is
2462 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2464 Severity : enhancement
2466 Description: Failover mode is now the default for OSTs.
2467 Details : By default, OSTs will now run in failover mode. To return to
2468 the old behaviour, add '--failout' to the lmc line for OSTs.
2470 Severity : enhancement
2472 Description: Health checks are now provided for MDS and OSTs
2473 Details : Additional detailed health check information on MSD and OSTs
2474 is now provided through the procfs health_check value.
2477 Frequency : occasional, depends on IO load
2479 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2480 after numerous create/delete cycles
2481 Details : The ext3 inode allocation policy would not allocate new inodes
2482 very well on the OSTs because there are no new directories
2483 being created. Instead we look for groups with free space if
2484 the parent directories are nearly full.
2489 Description: Network or server problems during mount may cause partially
2490 mounted clients instead of returning an error.
2491 Details : The config llog parsing code may overwrite the error return
2492 code during mount error handling, returning success instead
2498 Description: MDS can fail to allocate large reply buffers
2499 Details : After long uptimes the MDS can fail to allocate large reply
2500 buffers (e.g. zconf client mount config records) due to memory
2501 fragmentation or consumption by the buffer cache. Preallocate
2502 some large reply buffers so that these replies can be sent even
2503 under memory pressure.
2507 Frequency : rare (liblustre)
2508 Description: fsx running with liblustre complained that using truncate() to
2509 extend the file doesn't work. This patch corrects that issue.
2510 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2511 ATTR_SIZE and lsm use in llu_setattr_raw.
2515 Frequency : rare, only 2.6 kernels
2516 Description: Unusual file access patterns on the MDS may result in inode
2517 data being lost in very rare circumstances.
2518 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2519 code in the 2.6 kernel caused the inode and/or EA data not to
2520 be read from disk, causing single-file corruption.
2524 Frequency : rare, only 2.6 filesystems using extents
2525 Description: Heavy concurrent write and delete load may cause data corruption.
2526 Details : It was possible under high-load situations to have an extent
2527 metadata block in the block device cache from a just-unlinked
2528 file overwrite a newly-allocated data block. We now unmap any
2529 metadata buffers that alias just-allocated data blocks.
2533 Frequency : filesystems with default stripe_count larger than 77
2534 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2535 Details : lconf specifies an inode size of 4096 bytes when the default
2536 stripe_count is larger than 77. This conflicts with the default
2537 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2538 this case to avoid pinning too much memory for large EAs.
2540 ------------------------------------------------------------------------------
2542 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2547 Frequency : rare (extremely heavy IO load with hundreds of clients)
2549 Description: Client is evicted, gets IO error writing to file
2550 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2551 caused clients to be evicted for AST timeouts. The fixes in
2552 bug 5192 mean we no longer need to have such short AST timeouts
2553 so ldlm_timeout has been increased.
2556 Frequency : occasional during --force or --failover shutdown under load
2557 Bugzilla : 5949, 4834
2558 Description: Server oops/LBUG if stopped with --force or --failover under load
2559 Details : a collection of import/export refcount and cleanup ordering
2560 issues fixed for safer force cleanup
2563 Frequency : only filesystems larger than 120 OSTs
2564 Bugzilla : 5990, 6223
2565 Description: lfs getstripe would oops on a very large filesystem
2566 Details : lov_getconfig used kfree on vmalloc'd memory
2569 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2570 Bugzilla : 6242, 6243
2571 Description: reading from files that had been truncated to a non-zero size
2572 but never opened returned no data
2573 Details : ll_file_read() reads zeros from no-object files to EOF
2578 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2579 Details : There's little harm in aborting MDS/OSS recovery and letting it
2580 try again, so I removed the LASSERT and return an error instead.
2582 Severity : enhancement
2584 Description: New debugging infrastructure for tracking down data corruption
2585 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2586 (b) cover more of the client-side code path, and (c) try to narrow
2587 down where problems occurred
2591 Bugzilla : 3819, 4364, 4397, 6313
2592 Description: Racing close and eviction MDS could cause assertion in mds_close
2593 Details : It was possible to get multiple mfd references during close and
2594 client eviction, leading to one thread referencing a freed mfd.
2596 Severity: : enhancement
2597 Bugzilla : 3262, 6359
2598 Description: Attempts to reconnect to servers are now more aggressive.
2599 Details : This builds on the enhanced upcall-less recovery that was added
2600 in 1.4.2. When trying to reconnect to servers, clients will
2601 now try each server in the failover group every 10 seconds. By
2602 default, clients would previously try one server every 25 seconds.
2607 Description: After recovery, certain operations trigger a failed
2608 assertion on a client.
2609 Details : Failing over an mds, using lconf -d --failover, while a
2610 client was doing a readdir() call would cause the client to
2611 LBUG after recovery completed and the readdir() was resent.
2613 Severity : enhancement
2615 Description: Default groups are now added by lconf
2616 Details : You can now run lconf --group <servicename> without having to
2617 manually add groups with lmc.
2620 Frequency : occasional
2622 Description: Nodes with an elan id of 0 trigger a failed assertion
2625 Frequency : always when accessing e.g. tty/console device nodes
2627 Description: tty and some other devices nodes cannot be used on lustre
2628 Details : file's private_data field is used by device data and lustre
2629 values in there got lost. New field was added to struct file to
2630 store fs-specific private data.
2633 Frequency : when exporting Lustre via NFS
2635 Description: NFSD failed occasionally when looking up a path component
2636 Details : NFSD is looking up ".." which was broken in ext3 directories
2637 that had grown large enough to become hashed.
2640 Frequency : Clusters with multiple interfaces not on the same subnet
2642 Description: Nodes will repeatedly try to reconnect to an interface which it
2643 cannot reach and report an error to the log.
2644 Details : Extra peer list entries will be created by lconf with some peers
2645 unreachable. lconf now validates the peer before adding it.
2648 Frequency : Only if a default stripe is set on the filesystem root.
2650 Description: Setting a default stripe on the filesystem root prevented the
2651 filesystem from being remounted.
2652 Details : The client was sending extra request flags in the root getattr
2653 request and did not allocate a reply buffer for the dir EA.
2656 Frequency : occasional, higher if lots of files are accessed by one client
2657 Bugzilla : 6159, 6097
2658 Description: Client trips assertion regarding lsm mismatch/magic
2659 Details : While revalidating inodes the VFS looks up inodes with ifind()
2660 and in rare cases can find an inode that is being freed.
2661 The ll_test_inode() code will free the lsm during ifind()
2662 when it finds an existing inode and then the VFS later attaches
2663 this free lsm to a new inode.
2667 Bugzilla : 6422, 7030
2668 Description: MDS deadlock between mkdir and client eviction
2669 Details : Creating a new file via mkdir or mknod (starting a transaction
2670 and getting the ns lock) can deadlock with client eviction
2671 (gets ns lock and trying to finish a synchronous transaction).
2674 Frequency : occasional
2675 Description: While starting a server, the fsfilt_ext3 module could not be
2677 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2678 kernels. Previously, lconf would still use the ext3 name
2679 when trying to load modules. Now, it will correctly use
2680 ext3 on 2.4 and ldiskfs on 2.6.
2682 Severity : enhancement
2683 Description: The default stripe count has been changed to 1
2684 Details : The interpretation of the default stripe count (0, to lfs
2685 or lmc) has been changed to mean striping across a single
2686 OST, rather than all available. For general usage we have
2687 found a stripe count of 1 or 2 works best.
2689 Severity : enhancement
2690 Description: Add support for compiling against Cray portals.
2691 Details : Conditional compiling for some areas that are different
2695 Frequency : occasional
2696 Bugzilla : 6409, 6834
2697 Description: Creating files with an explicit stripe count may lead to
2698 a failed assertion on the MDS
2699 Details : If some OSTs are full or unavailable, creating files may
2700 trigger a failed assertion on the MDS. Now, Lustre will
2701 try to use other servers or return an error to the
2705 Frequency : occasional
2707 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2708 caused lock timeout during readahead (since 1.4.2).
2709 Details : Processes doing readahead might match a lock that hasn't been
2710 granted yet if there are overlapping and conflicting lock
2711 requests. The readahead process waits on ungranted lock
2712 (original lock is CBPENDING), while OST waits for that process
2713 to cancel CBPENDING read lock and eventually evicts client.
2715 Severity : enhancement
2717 Description: Initial enabling of flock support for clients
2718 Details : Implements fcntl advisory locking and file status functions.
2719 This feature is provided as an optional mount flag (default
2720 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2721 locking are implemented yet, and those that are are not guaranteed
2722 to be completely correct in production environments.
2723 mount -t lustre -o [flock|noflock] ...
2726 Frequency : occasional
2728 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2729 assertion in the ext3 JBD (journaling) layer.
2730 Details : The b_committed_data struct is protected by the big kernel lock
2731 in 2.4 kernels, serializing journal_commit_transaction() and
2732 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2733 access to this struct. In 2.6 kernels there is finer grained
2734 locking to improve SMP performance of the JBD layer.
2738 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2740 -----------------------------------------------------------------------------
2742 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2744 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2745 versions of Lustre. Please update all servers and clients to
2746 version 1.4.2 or later at the same time. You must also run
2747 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2748 to update the configuration logs.
2750 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2751 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2752 - swab llog records of type '0' so we get proper header size/idx (5861)
2753 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2754 - fix rename of one directory over another leaking an inode (5953)
2755 - avoid SetPageDirty on 2.6 (5981)
2756 - don't re-add just-being-destroyed locks to the waiting list (5653)
2757 - when creating new directories, inherit the parent's custom
2758 striping settings if present parent (3048)
2759 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2760 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2761 - don't hold client locks on temporary worklist from l_lru (5666)
2762 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2763 - automatically evict dead clients (5921)
2764 - Update file size properly in create+truncate+fstat case (6196)
2765 - Do not unhash mountpoint dentries, do not allow removal of
2767 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2768 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2769 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2770 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2771 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2772 - Correctly handle reads of files with no objects (6243)
2773 - lctl recover will also mark a device active if deactivate used (5933)
2775 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2776 - allow --write-conf on an MDS with different nettype than client (5619)
2777 - don't write config llogs to MDS for mounts not from that MDS (5617)
2778 - lconf should create multiple TCP connections from a client (5201)
2779 - init scripts are now turned off by default; run chkconfig --on
2780 lustre and chkconfig --on lustrefs to use them
2781 - upcalls are no longer needed for clients to recover to failover
2783 - add --abort-recovery option to lconf to abort recovery on device
2785 - add support for an arbitrary number of OSTs (3026)
2786 - Quota support protocol changes.
2787 - forward compatibility changes to wire structs (6007)
2788 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2789 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2790 - improved "lustre status" script
2791 - initialize blocksize for non-regular files (6062)
2792 - added --disable-server and --disable-client configure options (5782)
2793 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2794 - Vanilla 2.4.29 support
2795 - increase maximum number of obd devices to 520 (6242)
2796 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2797 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2798 SLES 9 2.6.5 kernels
2799 - limit stripes per file to 160 (the maximum EA size) (6093)
2801 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2804 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2805 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2806 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2807 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2808 - small ext3 extents cleanups and fixes (5733)
2809 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2810 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2811 - avoid needless client->OST connect, fix handle mismatch (5317)
2812 - fix DLM error path that led to out-of-sync client, long delays (5779)
2813 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2814 - fix several locking issues related to i_size (5492,5624,5654,5672)
2815 - don't move pending lock onto export if it is already evicted (5683)
2816 - fix kernel oops when creating .foo in unlinked directory (5548)
2817 - fix deadlock in obdfilter statistics vs. object create (5811)
2818 - use time_{before,after} to avoid timer jiffies wrap (5882)
2819 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2820 - Do not leak request if server was not able to process it (5154)
2821 - If mds_open unable to find parent dir, make that negative lookup(5154)
2822 - don't create new directories with extent-mapping (5909, 5936)
2824 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2825 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2826 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2827 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2828 - support for RHEL4 kernel on i686 (5773)
2829 - provide error messages when incompatible logs are encountered (5898)
2831 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2832 * version 1.4.0.10 (1.4.1 release candidate 1)
2834 - don't keep a lock reference when lock is not granted (4238)
2835 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2836 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2837 - drop import inflight refcount on signal_completed_replay error (5255)
2838 - unlock page after async write error during send (3677)
2839 - handle missing objects in filter_preprw_read properly (5265)
2840 - no transno return for symlink open, don't save no-trasno open (3440)
2841 - don't try to complete elan receive that already failed (4012)
2842 - free RPC server reply state on error (5406)
2843 - clean up thread from ptlrpc_start_thread() on error (5160)
2844 - readahead could read extra page into cache that wasn't ejected (5388)
2845 - prevent races in class_attach/setup/cleanup/detach (5260)
2846 - don't dereference de->d_inode after l_dput of de (5458)
2847 - use "int" for stripe value returned from lock_to_stripe (5544)
2848 - mballoc allocation and error-checking fixes in 2.6 (5504)
2849 - block device patches to fix I/O request sizes in 2.6 (5482)
2850 - look up hostnames for IB nals (5602)
2851 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2852 - don't start multiple acceptors for the same port (5277)
2853 - fix incorrect LASSERT in mds_getattr_name (5635)
2854 - export a proc file for general "ping" checking (5628)
2855 - fix "lfs check" to not block when the MDS is down (5628)
2857 - service request history (4965)
2858 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2859 - create an "evict_client" /proc entry on OSTs, like the MDS has
2860 - fix mount usage message, return errors per mount(8) (5168)
2861 - change grep [] to grep "[]" in tests so they work in more UMLs
2862 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2863 - remove ancient LOV_MAGIC_V0 EA support (5047)
2864 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2865 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2866 - allow readahead allocations to fail when low on memory (5383)
2867 - mmap locking landed again, after considerable improvement (2828)
2868 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2870 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2873 - send OST transaction number in read/write reply to free req (4966)
2874 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2875 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2876 - the watchdog thread now runs as interruptible (5246)
2877 - flock/lockf fixes (but it's still disabled, pending 5135)
2878 - don't use EXT3 constants in llite code (5094)
2879 - memory shortage at startup could cause assertion (5176)
2881 - reorganization of lov code
2882 - single portals codebase
2884 - add extents/mballoc support (5025)
2885 - direct I/O reads in the obdfilter (4048)
2886 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2888 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2891 - send OST transaction number in read/write reply to free req (4966)
2892 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2893 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2894 - the watchdog thread now runs as interruptible (5246)
2895 - handle missing objects in filter_preprw_read properly (5265)
2896 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2897 - drop import inflight refcount on signal_completed_replay error (5255)
2898 - unlock page after async write error during send (3677)
2899 - return original error code on reconstructed replies (3761)
2900 - no transno return for symlink open, don't save no-trasno open (3440)
2902 - add pid to ldlm debugging output (4922)
2903 - bump the watchdog timeouts -- we can't handle 30sec yet
2904 - extra debugging for orphan dentry/inode bug (5259)
2906 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2909 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2910 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2911 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2912 - fix NULL dereference in /proc/sys/portals/routes (4827)
2913 - allow failed mdc_close() operations to be interrupted (4561)
2914 - stop precreate on OST before MDS would time out on it (4778)
2915 - don't send partial-page writes before EOF from client (4410)
2916 - discard client grant for sub-page writes on large-page clients (4520)
2917 - don't free dentries not owned by NFS code, check generation (4806)
2918 - fix lsm leak if mds_create_objects() fails (4801)
2919 - limit debug_daemon file size, always print CERROR messages (4789)
2920 - use transno after validating reply (3892)
2921 - process timed out requests if import state changes (3754)
2922 - update mtime on OST during writes, return in glimpse (4829)
2923 - add mkfsoptions to LDAP (4679)
2924 - use ->max_readahead method instead of zapping global ra (5039)
2925 - don't interrupt __l_wait_event() during strace
2927 - add software watchdogs to catch hung threads quickly (4941)
2928 - make lustrefs init script start after nfs is mounted
2929 - fix CWARN/ERROR duplication (4930)
2930 - return async write errors to application if possible (2248)
2931 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2932 - print NAL number in %x format (4645)
2933 - update barely-supported suse-2.4.21-171 series (4842)
2934 - support for sles 9 %post scripts
2935 - support for building 2.6 kernel-source packages
2936 - support for sles km_* packages
2938 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2941 - ignore -ENOENT errors in osc_destroy (3639)
2942 - notify osc create thread that OSC is being cleaned up (4600)
2943 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2944 - reconstruct ost_handle() like mds_handle() (4657)
2945 - create a new thread to do import eviction to avoid deadlock (3969)
2946 - let lconf resolve symlinked-to devices (4629)
2947 - don't unlink "objects" from directory with default EA (4554)
2948 - hold socknal file ref over connect in case target is down (4394)
2949 - allow more than 32000 subdirectories in a single directory (3244)
2950 - fix blocks count for O_DIRECT writes (3751)
2951 - OST returns ENOSPC from object create when no space left (4539)
2952 - don't send truncate RPC if file size isn't changing (4410)
2953 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2954 - bind to privileged port in socknal and tcpnal (3689)
2956 - rate limit CERROR/CWARN console message to avoid overload (4519)
2957 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2958 - basic mmap support (3918)
2959 - kernel patch series update from b1_4 (4711)
2961 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2964 - avoid crash during MDS cleanup with OST shut down (2775)
2965 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2966 - don't use bad inodes on the MDS (3744)
2967 - dynamic object preallocation to improve recovery speed (4236)
2968 - don't hold spinlock over lock dumping or change debug flags (4401)
2969 - don't zero obd_dev when it is force cleaned (3651)
2970 - print grants to console if they go negative (4431)
2971 - "lctl deactivate" will stop automatic recovery attempts (3406)
2972 - look for existing locks in ldlm_handle_enqueue() (3764)
2973 - don't resolve lock handle twice in recovery avoiding race (4401)
2974 - revalidate should check working dir is a directory (4134)
2976 - don't always mark "slow" obdfilter messages as errors (4418)
2978 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2981 - don't close LustreDB during write_conf until it is done (3860)
2982 - fix typo in lconf for_each_profile (3821)
2983 - allow dumping logs from multiple threads at one time (3820)
2984 - don't allow multiple threads in OSC recovery (3812)
2985 - fix debug_size parameters (3864)
2986 - fix mds_postrecov to initialize import for llog ctxt (3121)
2987 - replace config semaphore with spinlock (3306)
2988 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2989 - don't allow enqueue to complete on a destroyed export (3822)
2990 - down write_lock before checking llog header bitmap (3825)
2991 - recover from lock replay timeout (3764)
2992 - up llog sem before sending rpc (3652)
2993 - reduce ns lock hold times when setting kms (3267)
2994 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2995 - fix NULL deref and obd_dev leak on setup error (3312)
2996 - replace some LBUG about llog ops with error handling (3841)
2997 - don't match INVALID dentries from d_lookup and spin (3784)
2998 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2999 - fix invalid assertion in ptlrpc_set_wait (3880)
3001 - add libwrap support for the TCP acceptor (3996)
3002 - add /proc/sys/portals/routes for non-root route listing (3994)
3003 - allow setting MDS UUID in .xml (2580)
3004 - print the stack of a process that LBUGs (4228)
3006 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3009 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3010 - make sure to unset replay flag from failed open requests (3440)
3011 - if default stripe count is 0, use OST count for inode size (3636)
3012 - update parent mtime/ctime on client for create/unlink (2611)
3013 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3014 - free recovery state on server during a forced cleanup (3571)
3015 - unregister_reply for resent reqs (3063)
3016 - loop back devices mounting and status check on 2.6 (3563)
3017 - fix resource-creation race that can provoke i_size == 0 (3513)
3018 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3019 - more debugging for page-accounting assertion (3746)
3020 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3021 - avoid deadlock after precreation failure (3758)
3022 - fix race and lock order deadlock in orphan handling (3450, 3750)
3023 - add validity checks when grabbing inodes from l_ast_data (3599)
3025 - add /proc/.../recovery_status to obdfilter (3428)
3026 - lightweight CDEBUG infrastructure, debug daemon (3668)
3027 - change default OSC RPC parameters to be better on small clusters
3028 - turn off OST read cache for files smaller than 32MB
3029 - install man pages and include them in rpms (3100)
3030 - add new init script for (un)mounting lustre filesystems (2593)
3031 - run chkconfig in %post for init scripts (3701)
3032 - drop scimac NAL (unmaintained)
3034 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3037 - clean kiobufs before and after use (3485)
3038 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3039 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3040 - change init script to fail more clearly if not run as root (1528)
3041 - allow clients to reconnect during replay (1742)
3042 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3043 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3044 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3045 - fix cond_resched (3554)
3046 - only evict unfinished clients after recovery (3515)
3047 - allow bulk resend, prevent data loss (3570)
3048 - dynamic ptlrpc request buffer allocation (2102)
3049 - don't allow unlinking open directory if it isn't empty (2904)
3050 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3051 - remove extraneous obd dereference causing LASSERT failure (3334)
3052 - don't use get_cycles() when creating temp. files on the mds (3156)
3053 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3054 - handle EEXIST for set-stripe, set proper directory name (3336)
3056 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3057 - fix ksocknal_fmb_callback() error messages (2918)
3059 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3062 - don't copy lvb into (possibly NULL) reply on error (2983)
3063 - don't deref dentry after dput, don't free lvb on error (2922)
3064 - use the kms to determine writeback rpc length (2947)
3065 - increment oti_logcookies when osc is inactive (2948)
3066 - update client's i_blocks count via lvb messages (2543)
3067 - handle intent open/close of special files properly (1557)
3068 - mount MDS with errors=remount-ro, like obdfilter (2009)
3069 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3070 - don't use cancelling-locks' kms values (2947)
3071 - use highest lock extent for kms, not last one (2925)
3072 - don't dereference ERR_PTR() dentry in error handling path (3107)
3073 - fix thread race in portals_debug_dumplog() (3122)
3074 - create lprocfs device entries at setup instead of at attach (1519)
3075 - common AST error handler, don't evict client on completion race (3145)
3076 - zero nameidata in detach_mnt in 2.6 (3118)
3077 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3078 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3079 - initialize RPC timeout timer earlier for 2.6 (3219)
3080 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3081 - print nal/nid for unknown nid (3258)
3082 - additional checks for oscc recovery before doing precreate (3284)
3083 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3084 - don't crash in mdc_close for bad permissions on open (3285)
3085 - zero i_rdev for non-device files (3147)
3086 - clear page->private before handing to FS, better assertion (3119)
3087 - tune the read pipeline (3236)
3088 - fix incorrect decref of invalidated dentry (2350)
3089 - provide read-ahead stats and refine rpc in flight stats (3328)
3090 - don't hold journal transaction open across create RPC (3313)
3091 - update atime on MDS at close time (3265)
3092 - close LDAP connection when recovering to avoid server load (3315)
3093 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3094 - don't leak open file on MDS after open resend (3325)
3095 - serialize filter_precreate and filter_destroy_precreated (3329)
3096 - loop device shouldn't call sync_dev() for nul device (3092)
3097 - clear page cache after eviction (2766)
3098 - resynchronize MDS->OST in background (2824)
3099 - refuse to mount the same filesystem twice on same mountpoint (3394)
3100 - allow llmount to create routes for mounting behind routers (3320)
3101 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3102 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3103 - fix rare problem with rename on htree directories (3417)
3105 - allow default OST striping configuration per directory (1414)
3106 - fix compilation for qswnal for 2.6 kernels (3125)
3107 - increase maximum number of MDS request buffers for large systems
3108 - change liblustreapi to be useful for external progs like lfsck (3098)
3109 - increase local configuration timeout for slow disks (3353)
3110 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3112 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3115 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3116 - don't overwrite extent policy data in reply if lock was blocked (2901)
3117 - drop filter export grants atomically with removal from device (2663)
3118 - del obd_self_export from work_list in class_disconnect_exports (2908)
3119 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3120 - swab reply message in mdc_close, other PPC fixes (2464)
3121 - fix destroying of named logs (2325)
3122 - overwrite old logs when running lconf --write_conf (2264)
3123 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3124 - fix race in target_handle_connect (2898)
3125 - mds_reint_create() should take same inode create lock (2926)
3126 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3127 - don't close files for self_export to avoid uninitialized obd (2936)
3128 - allow MDS with the same name as client node (2939)
3129 - hold dentry reference for closed log files for unlink (2325)
3130 - reserve space for all logs during transactions (2059)
3131 - don't evict page beyond end of stripe extent (2925)
3132 - don't oops on a deleted current working directory (2399)
3133 - handle hard links to targets without a parent properly (2517)
3134 - don't dereference NULL lock when racing during eviction (2867)
3135 - don't grow lock extents when lots of conflicting locks (2919)
3137 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3140 - account for cache space usage on clients to avoid data loss (974)
3141 - lfsck support in lustre kernel code (2349)
3142 - reduce journal credits needed for BRW writes (2370)
3143 - orphan handling to avoid losing space on client/server crashes
3144 - ptlrpcd can be blocked, stopping ALL progress (2477)
3145 - use lock value blocks to assist in proper KMS, faster stat (1021)
3146 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3147 - recovery for initial connections (2355)
3148 - fixes for mds_cleanup_orphans (1934)
3149 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3150 - block all file creations until orphan recovery completes (1901)
3151 - client remove rq_connection from request struct (2423)
3152 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3153 - recovery timer race (2670)
3154 - mdc_close recovey bug (2532)
3155 - ptlrpc cleanup bug (2710)
3156 - mds timeout on local locks (2588)
3157 - namespace lock held during RPCs (2431)
3158 - handle interrupted sync write properly (2503)
3159 - don't try to handle a message that hasn't been replied to (2699)
3160 - client assert failure during cleanup after abort recovery (2701)
3161 - leak mdc device after failed mount (2712)
3162 - ptlrpc_check_set allows timedout requests to complete (2714)
3163 - wait for inflight reqs when ptlrpcd finishes (2710)
3164 - make sure unregistered services are removed from the srv_list
3165 - reset bulk XID's when resending them (caught by 1138 test)
3166 - unregister_bulk after timeout
3167 - fix lconf error (2694)
3168 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3169 - readahead locks pages, leaves pending causing memory pressure (2673)
3170 - increase OST request buffers to 4096 on large machines (2729)
3171 - fix up permission of existing directories in simple_mkdir (2661)
3172 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3173 - don't assign transno to errored transactions (2742)
3174 - don't delete objects on OST if given a bogus objid from MDS (2751)
3175 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3176 - if rq_no_resend, then timeout request after recovery (2432)
3177 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3178 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3179 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3180 - MDS recovery shouldn't delete objects that it has given out (2730)
3181 - if enqueue arrives after completion, don't clobber LVB (2819)
3182 - don't unlock pages twice when trigger_group_io returns error (2814)
3183 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3184 - don't write pages to disk if there was an error (1450)
3185 - don't ping imports that have recovery disabled (2676)
3186 - take buffered bytes into account when balancing socknal conn (2817)
3187 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3188 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3189 - remove little-endian swabbing of llog records (1987)
3190 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3191 - retry reposting request buffers if they fail (1191)
3192 - grow extent at grant time to avoid granting a revoked lock (2809)
3193 - lock revoke doesn't evict page if covered by a second lock (2765)
3194 - disable VM readahead to avoid reading outside lock extents (2805)
3196 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3197 - updated kernel patches for hp-2.4.20 kernel (2681)
3199 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3202 - fix truncated write corruption (2366)
3203 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3205 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3206 - ensure len(uuid) < 37 in lmc (1171)
3207 - fix ia64 OOPS in llog_test (2255)
3208 - zero end of page at obdfilter for partial page writes (2648)
3209 - don't leave stale dentries around after renames (bug 2428)
3210 - fix timeouts when evicting a client with a single lock held (2642)
3211 - set deadline for the initial HELLO message to drain (2634)
3212 - print out dotted-quad IP addresses in the socknal (2302)
3214 - additional debugging for MDS client eviction problem (2443)
3215 - fix mkfsoptions support for osts (2603, 2604)
3217 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3220 - add series for the vanilla 2.6.0 kernel
3221 - add series for the vanilla 2.4.24 kernel
3222 - add series for a cray x86/64 UL kernel drop
3223 - fix xattr patches for the vanilla 2.4.19 series
3225 - generate true UUIDs in lmc (1171)
3226 - have portals stack dumping break in UML (2466)
3227 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3228 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3229 - wake callback waiting threads on client eviction (2460)
3230 - Add --ptldebug and --subsystem to lmc (1719)
3231 - update assertion to allow safe interrupt allocation
3232 - set rq_no_resend for cancel requests (2432)
3233 - recalculate ptlrpcd timeout after resend (2494)
3234 - call vfs_rmdir when removing pending directories (2368)
3235 - fix renaming a file to itself (2429)
3236 - lmc creates a default one-stripe lov (2454)
3237 - expand procfs space to handle large clusters (2326)
3238 - increase UML stack to avoid overflow
3239 - update lconf's list of debug and subsystem masks
3240 - fix lfs find --obd (2510)
3241 - /proc tunable for disabling filter read caching (2591)
3242 - stop rpm packages from altering slapd.conf (2301)
3243 - disable nagle in the socknal under 0conf (2578)
3244 - choose mds inode size based on stripe count (2572)
3245 - fix kernel-source rpm problems (2516)
3247 - add --disable-doc to avoid pdf generation (2421)
3248 - update documentation, tests, type-os, comments
3249 - avoid format warnings on ia64
3250 - remove the TOE NAL
3251 - tiny code cleanups by removing unused fields
3253 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3256 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3257 - remove the most blatant lies from BUILDING file (2371)
3258 - change default debug level to reasonable production setting
3259 - reduce client side cache size to reduce cache flush time
3260 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3261 - make TCP zerocopy and pinger support enabled by default (2476)
3262 - sync writes completed after process exits caused crashes (2319)
3263 - maintain correct mount count on the MDS (2356)
3264 - backout 1557, because 2316 wasn't really fixed
3265 - better file I/O statistics gathering in /proc
3266 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3267 - another kernel patch to fix zero-copy TCP function export
3268 - don't take duplicate lock when processing re-sent getattr (2420)
3269 - lctl uses obd_self_export instead of creating new conn (2353)
3270 - MDS/OST recovery case which requires object creation asserted (2425)
3271 - move lfs from /usr/sbin to /usr/bin in packages
3272 - fix race between mds_client_add and mds_client_free (2417)
3273 - use kmalloc instead of slabs in portals (2430)
3274 - don't create duplicate records when a failover MDS is present (2442)
3275 - remove unnecessary mount age check (2332)
3276 - don't remove directory inodes from locks prematurely (2451)
3277 - don't break if MDS service name is the same as hostname (2103)
3278 - fix races in client write RPC generation when cache full (2482)
3280 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3283 - remove now-unused request->rq_obd (278)
3284 - if an allocation fails, print out how much memory we've used (1933)
3285 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3286 - add the "configurable stack size" patch to most series files (1256)
3287 - ability to write large log records, for 100+ OST configs (2306)
3288 - fix NULL deref when filter_prep fails (2314)
3289 - fix operator precedence error in filter_sync
3290 - dynamic allocation of socknal TX descriptors (2315)
3291 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3292 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3293 - ensure that CWARN messages in Portals always get to the syslog
3294 - __init/__exit are not for prototype decls (ldlm_init/exit)
3295 - x86-64 compile warning fixes
3296 - fix gateway LMC keyword conflict (2318)
3297 - fix MDS lock inversions in getattr/reint paths (1844)
3298 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3299 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3300 - DLM race condition prevented some lock evictions (2328)
3301 - ENOMEM detection and retry on socknal sends (2230)
3302 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3303 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3304 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3305 - better error messages when a client is rejected during recovery (1505)
3306 - avoid cancelling locks which were never granted, after failure (2330)
3307 - fix i_sem/journal inversion in mds_client_add (2333)
3308 - fix truncate/getattr lock cycle deadlock (2334)
3309 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3310 - fix two rare exit paths which could leak an l_lock() ref (2321)
3311 - fencepost error in MDS/OST orphan recovery (2226)
3312 - make log record alignment 8 bytes (1988)
3313 - lstripe now fails when requested offset > ost_count (2237)
3314 - ensure that all kernel series have a complete list.h (1607)
3315 - fix crashes in special-file operations (2316)
3316 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3318 - allow configurable automake binary, for testing new versions
3319 - small update to the lfs documentation
3321 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3323 * fix negative export reference count in fsfilt_sync (2312)
3325 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3326 * release candidate 0.9.1
3328 - orphans are moved into the PENDING directory for possible recovery
3329 - replayed opens now open by fid for orphan/rename safety (1042)
3330 - last close of an orphan inode generates a transno (683)
3331 - chdir() and mount() now pin the directory entry (1020)
3332 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3333 - discard very old requests without processing them (1502)
3334 - remove some common, well-understood CERRORs (1505)
3335 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3336 - clear "grant" flags in OST replies until OST grant code lands (1644)
3337 - fix read performance by not clobbering i_blksize on client (1598)
3338 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3339 - make LRU size a /proc tunable, clears locks when reduced (707)
3340 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3341 - server locks take references on exports now (1558)
3342 - build fixes for 2.4.20-rh trees (1663)
3343 - return an error from lov_create if all OSCs are inactive (1751)
3344 - fix import levels when a reconnect happens without a timeout (1597)
3345 - exit early from mds_open if we get a lookup error (1749)
3346 - partial page read at EOF wouldn't wait for disk before sending (1642)
3347 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3348 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3349 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3350 - write/truncate lock inversion (1639)
3351 - Don't auto-load obdclass, portals modules during cleanup (1495)
3352 - fix timestamps from jumping to "now" (1763)
3353 - extra journal assertions (1648)
3354 - add an extra multiunlink test (1771)
3355 - fix read_record/write_record API (1776)
3356 - fix leak of offset_extent, possible incorrect i_size later (1772)
3357 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3358 - Debugging for the kqswnal_get_idle_tx problems (1820)
3359 - Allow recovery to be attempted multiple times (1536)
3360 - Write out MDS last_rcvd file after it is first created (1600)
3361 - Fix tx_descriptor leak in failed transmit situations (1827)
3362 - ext3 journaling fixes for assertion failure after IO error (1871)
3363 - class_export_put() on freed export after completion AST error (1896)
3364 - Fix revalidate looping in VFS (1322)
3365 - Don't access a freed export during MDS_REINT timeout (1521)
3366 - Add open-unlink recovery support on the MDS (1673,1764)
3367 - Return an error if no MDS data was read from last_rcvd (1946)
3368 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3369 - Don't setup a disk device that doesn't match exported UUID (317)
3370 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3371 - avoid committing NULL handle in force close
3372 - local.sh is now a one-stripe LOV configuration
3373 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3374 - fix ext3 htree duplicate directory entry corruption (1516)
3375 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3376 - update to new LOV EA format (2097)
3377 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3379 2003-06-15 Phil Schwan <phil@clusterfs.com>
3382 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3383 - per-import/export recovery handling (958, 931, 959)
3384 - multiple last-rcvd slots, for serving multiple FSes (949)
3385 - connections are again shared between multiple imp/exports (963, 964)
3386 - "umount -f" would hang if any requests needed to be sent (393, 978)
3387 - avoid pinning large req buffer by copying for queued messages (989)
3388 - add "uuid" to "lctl device" command to help upcalls (991)
3389 - "open" RPCs with transnos would confuse recovery counters (1037)
3390 - do proper endian conversion of all wire messages (288, 340, 891)
3391 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3392 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3393 - fix ptlrpc_connection leak in target_handle_connect (1174)
3394 - fix import refcounting bug in OST and MDS cleanup (1134)
3395 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3396 - fix very unlikely obd_types race condition (501)
3397 - remove osc_open hack for echo_client (1187)
3398 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3399 - a failure in read_inode2 leads to deadlock (1139)
3400 - cancel ack-locks as soon as transaction is committed (1072)
3401 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3402 - make sure to commitrw after any preprw to avoid deadlock (1162)
3403 - failing to execute a file in a lustre FS would lock inode (1203)
3404 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3405 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3406 - obd_uuid2tgt would walk off the end of the list (1255)
3407 - on IA64 the osc would give portals incorrect bulk size (1258)
3408 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3409 - fix lock inversion caused by new llite matching code (1282)
3410 - limit the number of dirty pages on a client to 10MB (1286)
3411 - timed out locks were not being corrected cancelled (1289)
3412 - fix O_DIRECT above 4GB on IA-32 (1292)
3413 * major user-visible changes
3414 - fail out/fail over policy now controlled by the upcall (993)
3416 - add OBD_PING to check server availability and failure (954)
3417 - lustre messages are now sent in sending host order (288, 340, 891)
3418 - add eadatalen to MDS getattr reply (340)
3419 - OST read replies may contain second buffer, with per-page status (593)
3421 2003-03-11 Phil Schwan <phil@clusterfs.com>
3424 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3425 - failed open()s could cause deadlock; fixed (867, 869)
3426 - stop cancelling OST locks when files are closed (481)
3427 - overlapping XID spaces caused network corruption (851, 853)
3428 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3429 - setattr_raw added, to do single-RPC, server-side setattrs
3430 - lmc/lconf syntax change for OST UUIDs
3431 - fix crashy race condition between ptlrpc_free_req and osc_close
3432 - don't use request in mdc_enqueue if we hit a timeout (889)
3433 - don't set the inode i_size for regular files from the MDS (896)
3434 - handle out of order completion AST (842)
3435 - don't LBUG if a lock request times out after receiving AST (913)
3436 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3437 - if a bad lock AST arrives, send an error instead of dropping entirely
3438 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3439 - fix leak in bulk IO when only partially completed (899, 900, 926)
3440 - fix O_DIRECT for ia64 (55)
3441 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3442 - C-z after timeout could hang a process forever; fixed (977)
3444 - client-side I/O cache (678, 924, 929, 941, 970)
3446 - READPAGE and SETATTRs which don't take server-side locks get
3449 2003-02-11 Phil Schwan <phil@clusterfs.com>
3452 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3453 - Fix fsfilt_extN_readpage() to read a full page of directory
3454 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3455 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3456 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3457 - Most intent operations were removed, in favour of a new RPC mode
3458 that does a single RPC to the server and bypasses most of the VFS
3459 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3460 - Aggressively cancel local locks on DLM servers
3461 - mds_reint_unlink sends EA to the client if it's the last nlink.
3462 client uses that EA to unlink OST objects.
3463 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3464 - recursive symlinks were fixed (439)
3465 - fixed NULL deref in DEBUG_REQ
3466 - filter_update_lastobjid no longer calls sync, which annoyed extN
3467 - fixed multi-client small-writes to a single file problem (445)
3468 - fixed mtime updates during file writes (607)
3469 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3470 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3471 - fixed Linux OST size reporting problem (444, 656)
3472 - OST now updates object mtime with writes or setattr (607, 619)
3473 - client verifies file size before zeroing page past EOF (445)
3474 - OST now writes last allocated objid to disk with allocation (108)
3475 - LOV on echo now works (409)
3477 - mds_reint_unlink sends a new buffer, with the EA included. this
3478 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3479 set if a regular file was being unlinked, and it was the last link
3480 - use PtlGet from the target for bulk writes (315)
3481 - OST now updates object mtime with writes or setattr (607, 619)
3482 - LDLM now has a grant-time callback to revalidate locked items, if
3484 - Many MDS operations were reorganized to combat race conditions
3486 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3487 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3489 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3492 - Fully reactivate OST imports after reconnection (512, others)
3493 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3494 - More graceful error handling for truncating on dead OST (515)
3495 - Don't error out unless we're actually accessing dead stripes (474)
3496 - Fix garbage sizes when stripes are missing (410)
3497 - LRU counters were broken, causing constant lock purge (433, 432)
3498 - garbage on read from stripes with failed OSTs (441)
3499 - mark OSCs as active before reconnecting during recovery (438)
3500 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3501 - lfind did not preserve OST order in output (443)
3502 - symlinks cause hung clients, incorrect data (439)
3503 - stop dereferencing request after dropping refcount (457)
3504 - don't LASSERT(spin_is_locked) on non-SMP (455)
3505 - fixes for many rename() bugs
3506 - fstat didn't correctly synchronize attributes (399)
3507 - server must handle lock cancellation during blocking AST prep (487)
3508 - bulk descriptors were free()d too soon (511)
3509 - fix paths in lconf, which would load incorrect modules (451, 507)
3510 - fix confusing lconf 'host not found' error message (386)
3511 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3512 - fix race condition in mdc_blocking_ast() for inode access (526)
3513 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3514 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3515 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3516 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3517 - fix lstripe in several ways (kernel side) (527)
3518 - fix request leak in ldlm_cli_enqueue (262)
3519 - incorrect OSC was marked inactive after OST failure
3520 - call mds_fs_cleanup before unmounting filesystem (524)
3521 - fix races between taking ns_lock and ldlm_lock_change_resource
3522 - fix races updating LOV export open file list
3523 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3524 - fix recovery NULL deref in ldlm_cli_cancel_unused
3525 - fix some DLM races by using new hash table for lock handles (419)
3526 - permit the client to specify desired inodes, at replay
3527 - duplicate requests when we queue them for replay reintegration
3528 - fix last_rcvd offset calculation
3529 - sync after each recovered transaction, so we always make progress
3530 - never, not always, ERESTART requests without transnos
3531 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3533 - skip replay if the MDS says that the client is already connected
3534 - don't check for a recovery-enabled export to match lctl's UUID
3535 - don't INC_USE_COUNT for phantom exports
3536 - don't crash when cleaning up phantom exports (567)
3537 - don't double-finish or set replay data for errored mdc_open requests
3538 - abort requests when they time out, so we don't get old replies
3539 - send/receive replies for AST messages again
3540 - if the client says that it doesn't have the lock, cancel it on the
3542 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3543 mark it as destroyed, it will all work out when decref is called
3544 - fix module use counts (22, 581)
3546 - ASTs now expect a reply (server cancels lock on error reply)
3548 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3551 - fix many simultaneous client startup (392)
3552 - fix dentry->d_it clobbering
3553 - credentials weren't being shipped for readdir/getattr operations
3554 - remove invalid assertions triggered during some concurrent MD
3556 - proper Lustre versions added (336, 389)
3557 - fix memory leak for create error case (398)
3558 - fix LOV locking bug that would get cli/srv out of sync
3559 - fix echo client over LOV (409)
3560 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3561 - fix double-O_EXCL intent crash (424)
3562 - avoid sending multiple lock CANCELs (352)
3564 - MDS can do multi-client recovery (modulo bugs in new code)
3566 - many updates, edits, cleanups
3568 2002-11-18 Phil Schwan <phil@clusterfs.com>
3571 - fix null d_it dereference (346)
3572 - fix full OST/dbench hang (333)
3573 - fix permission problem with file removal (286)
3574 - fix removal of OSCs from LOV when they fail
3575 - fix NULL deref during bulk timeout (214)
3576 - fix problems related to multiple filesystems on one MDS (241)
3577 - fixed serious subtle metadata locking bugs
3578 - free locks on clients when inodes are removed due to memory
3580 - fix inode pointer in lock data (285)
3581 - partial support for multiple MDS on a single host (241)
3582 - data locks weren't cancelled at clear_inode time (290, 311)
3583 - intent locks could lead to unbounded lock growth (205)
3584 - added a maximum lock count, an LRU list, and a flusher
3585 - fix multiple rename (365)
3586 - properly abstracted the echo client
3587 - OSC locked 1 byte too many; fixed
3588 - rewrote brw callback code:
3589 - fixed recovery bugs related to LOVs (306)
3590 - fixed too-many-pages-in-one-write crash (191)
3591 - fixed (again) crash in sync_io_timeout (214)
3592 - probably fixed callback-related race (385)
3594 - Add capability to MDS protocol
3595 - LDLM cancellations and callbacks on different portals
3597 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3600 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3601 - defer open object destruction to close time (601981, 138)
3602 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3603 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3604 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3605 - don't call obd_setattr in ll_file_release for destroyed objects
3606 * protocol change to lustre_msg: move |version| and add |flags|
3607 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3608 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3609 * added replay of create, unlink, link and rename operations during
3610 MDS failover; recovery should be much more robust now
3611 * remove failed OSCs from LOVs (only lov_create uses this so far)
3612 * the lustre-HOWTO was brought (more) up to date (582544)
3614 2002-10-23 Phil Schwan <phil@clusterfs.com>
3617 - in-use dentries weren't being reused properly (617851)
3618 - prevent multiple LDLM setup (599178)
3619 - fix LOV size calculations for truncate (617853)
3620 - fix client handling of MDS intent errors (POSIX)
3621 - fix permission bug in lovstripe.c test (624321)
3622 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3623 - truncate past end of file could corrupt data
3624 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3625 - a race in recovery could return ETIMEDOUT to apps (623947)
3626 - building outside the source directory was fixed
3627 * the lustre-HOWTO was brought (more) up to date (582544)
3628 * major progress was made on recovery functionality
3630 2002-10-10 Phil Schwan <phil@clusterfs.com>
3633 - recovery deadlock fix
3634 - rm -rf causes LBUG fix (617817)
3635 - file open by multiple tasks fix (618962)
3636 - directory permissions bugs (602707 and 620007)
3637 - journal_stop fixed with locking (611313)
3638 - O_APPEND failures resolved (618273, perhaps 614459)
3639 - lconf PATH fix (619770)
3640 - IA64 build fix (621450)
3641 - RPC buffer sizes scale with amount of memory
3643 2002-10-01 Phil Schwan <phil@clusterfs.com>
3646 - locks would be cancelled without throwing away data pages,
3647 resulting in inconsistent data (605627)
3648 - inode attributes were not always being refreshed (605627, 612449)
3649 - lconf now continues to cleanup after lctl reports an error
3650 - MDS now enforces user permissions (602707)
3651 - lprocfs cleanup fixed, but not yet enabled (614157)
3652 - fixed infinite server hang, should a client not respond to an AST
3653 - avoid going into recovery if user calls readlink() with a buffer
3654 that's too small (613941)
3655 - AST RPCs no longer require replies (614867) -- this may be changed
3656 - don't crash server if client sends an IOV that's too big (611336)
3657 - fixed lock conversion deadlock (611892)
3658 - fixed the following of symlinks (614622)
3659 * recovery: the server can remove locks from a client that dies, other
3660 clients can make progress
3661 * more extN patch fixes
3662 * compile-time configurable ptlrpc buffer allocations
3664 - collaborative read cache document
3665 - Lustre Lite Performance CDR document-in-progress
3667 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3670 - fix typo in patch-2.4.18
3672 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3675 - clear ptlrpc request each time in handle_incoming_request()
3676 - unlink of files now destroys the object on the OST
3678 2002-09-19 Peter Braam <braam@clusterfs.com>
3680 * add hard link support
3681 * change obdfile creation method
3682 * kernel patch changed
3684 2002-09-19 Peter Braam <braam@clusterfs.com>
3687 - stack overflow bug in extN fixed
3689 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3691 * documentation updates
3692 - add man pages for config tools
3693 - update tests/README to describe testing with new config tools
3694 - finish metadata API descriptions
3695 * bug fixes and cleanups
3696 - statfs workaround for 16TB limit
3697 - LOV stripe allocation improved, can stripe on subset of OSTs
3698 - LOV file size/IO offset was wrong for files > 4GB in size
3699 - object EA data was being dropped, caused files to be unreadable
3700 - memory overflow with non-LOV OST caused memory corruption
3701 - fixed regression tests to work with new config tools, obdfilter
3702 - fixed bug when directory size became larger than 1 block
3703 - fixed bug (for single client case) when PWD was deleted
3704 - invalidate local directory pages when doing intent-based ops
3705 - avoid LDLM oops when lock callback contained bad data
3707 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3709 * documentation updates
3710 * bug fixes and cleanups
3711 - configuration tools
3714 - 64-bit compile warnings
3715 - 64-bit internal statfs data
3717 * test_brw on persistent OST devices
3719 * lprocfs (disabled)
3721 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3723 * documentation updates
3724 * bug fixes and cleanups
3725 * configuration tools
3727 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3730 * many small fixes to 0_5_4
3731 * io/network handling
3732 * thinkos in MDS operations
3734 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3737 * crucial basic fixes to 0.5.3
3738 * IOR, Iozone work over Elan
3741 2002-08-07 Phil Schwan <phil@clusterfs.com>
3742 * version 0_5_3, our first alpha
3743 * we use the new Portals iovs
3744 * documentation updates
3745 * bug fixes and cleanups
3746 * small changes in the DLM wire protocol
3748 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3749 * version 0_5_1 with some initial stability,
3750 * locking on MD and file I/O.
3751 * documentation updates
3752 * several bug fixes since 0.5.0
3753 * small changes in wire protocol
3755 2002-07-18 Phil Schwan <phil@clusterfs.com>
3757 * delivered as Lustre Light Alpha
3758 * fixed a crash after handling invalid MDS requests
3759 * fixed directory pages for architectures with non-4k pages sizes
3761 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3762 * release version v0_4_4
3763 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3765 * Fixes a number of interruption problems with OST operations.
3766 * Update documentation for portals header changes
3767 * Move all wire protocol structs/defines to lustre_idl.h
3768 * Fixes symlink length bug.
3769 * Add tcpdump to repository.
3771 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3772 * release version v0_4_3
3773 * Fixes statfs for inodes on extN.
3774 * Fixes bug in runtests which would delete /etc/hosts.
3775 * Use 64-bit object IDs wherever possible (not into VFS though)
3776 Remove ost_get_info, which is unused by lustre, and out of date.
3778 2002-07-03 Peter Braam <braam@clusterfs.com>
3779 * release version v0_4_2 Fixes a lookup error (type not passed)
3780 * move forward to head of Portals
3781 * move forward to latest Lustre kernel
3783 2002-06-25 Peter Braam <braam@clusterfs.com>
3784 * release version v0_4_1. Hopefully stable on single node use.