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: 1.40.2-cfs1
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
58 Description: jbd statistics
59 Details : Port older jbd statistics patch for sles10
61 Severity : enhancement
63 Description: Add printing inode info into message about error in writepage.
68 Description: reply_lock_interpret crash due to race with it and lock cancel.
69 Details : Do not replay locks that are being cancelled. Do not reference
70 locks by their address during replay, just by their handle.
74 Frequency : When flocks are used.
75 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
76 Details : Flock locks might destroy just granted lock if it could be merged
77 with another existing flock, this is done in completion handler,
78 so teach ldlm_cli_enquque_fini that this is a valid case for
84 Description: Oops in read and write path when failing to allocate lock.
85 Details : Check if lock allocation failed and return error back.
89 Description: lstripe command fails for valid OST index
90 Details : The stripe offset is compared to lov->desc.ld_tgt_count
91 instead of lov->desc.ld_active_tgt_count.
95 Description: sanity.sh failed test 103
96 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
101 Description: df doesn't work properly if diskfs blocksize != 4K
102 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
106 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
107 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
111 Description: obd_health_check_timeout too short
112 Details : set obd_health_check_timeout as 1.5x of obd_timeout
116 Description: llapi_file_create() does not allow some changes
117 Details : add llapi_file_open() that allows specifying the mode and
118 open flags, and also returns an open file handle.
122 Description: merge and cleanup kernel patches.
123 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
127 Description: Add journal checksum support.(Kernel part)
128 Details : The journal checksum feature adds two new flags i.e
129 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
130 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
131 indicates that the commit block contains the checksum for
132 the blocks described by the descriptor blocks. Now commit
133 record can be sent to disk without waiting for descriptor
134 blocks to be written to disk. This behavior is controlled
135 using JBD2_FEATURE_ASYNC_COMMIT flag.
139 Description: OSS needs mutliple precreate threads
140 Details : Add ability to start more than one create thread per OSS.
144 Description: Sanity.sh test_65k failed: setstripe should have succeeded
145 Details : execute lfs setstripe on client
149 Description: mds_obd_create error creating tmp object
150 Details : When the user sets quota on root, llog will be affected and can't
151 create files and write files.
154 Frequency : Always on ia64 patchless client, and possibly others.
156 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
157 Details : This allows the patchless client to be loaded on architectures
160 --------------------------------------------------------------------------------
162 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
164 * Support for kernels:
165 2.6.5-7.283 (SLES 9),
166 2.6.9-55.EL (RHEL 4),
167 2.6.16.46-0.14 (SLES 10),
168 2.6.18.8 vanilla (kernel.org)
169 * Client support for unpatched kernels:
170 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
171 2.6.16 - 2.6.22 vanilla (kernel.org)
172 * Due to recently discovered recovery problems, we do not recommend
173 using patchless RHEL 4 clients with this or any earlier release.
174 * Recommended e2fsprogs version: 1.39.cfs8
175 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
176 * Starting with this release, the ldiskfs backing filesystem required
177 by Lustre is now in its own package, lustre-ldiskfs. This package
178 should be installed. It is versioned separately from Lustre and
179 may be released separately in future.
183 Description: block reactivating mgc import until all deactivates complete
184 Details : Fix race when failing back MDT/MGS to itself (testing)
186 Severity : enhancement
188 Description: add optional extra BUILD_VERSION info
189 Details : add a new environment variable (namely LUSTRE_VERS) which allows
190 to override the lustre version.
193 Frequency : 2.6.18 servers only
195 Description: ll_kern_mount() doesn't release the module reference
196 Details : The ldiskfs module reference count never drops down to 0
197 because ll_kern_mount() doesn't release the module reference.
202 Description: server LBUG when using old ost_num_threads parameter
203 Details : Accept the old ost_num_threads parameter but warn that it
204 is deprecated, and fix an off-by-one error that caused an LBUG.
209 Description: Transient SCSI error results in persistent IO issue
210 Details : iobuf->dr_error is not reinitialized to 0 between two
214 Frequency : sometimes when underlying device returns I/O errors
216 Description: OSTs not going read-only during write failures
217 Details : OSTs are not remounted read-only when the journal commit threads
218 get I/O errors because fsfilt_ext3 calls journal_start/stop()
219 instead of the ext3 wrappers.
223 Description: poor connect scaling with increasing client count
224 Details : Don't run filter_grant_sanity_check for more than 100 exports
225 to improve scaling for large numbers of clients.
228 Frequency : SLES10 only
230 Description: sanity-quota.sh quotacheck failed: rc = -22
231 Details : Quotas cannot be enabled on SLES10.
234 Frequency : liblustre clients only
236 Description: getdirentries does not give error when run on compute nodes
237 Details : getdirentries does not fail when the size specified as an argument
238 is too small to contain at least one entry
240 Severity : enhancement
242 Description: Add LNET router traceability for debug purposes
243 Details : If a checksum failure occurs with a router as part of the
244 IO path, the NID of the last router that forwarded the bulk data
245 is printed so it can be identified.
250 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
251 Details : Due to a race condition, liblustre clients were occasionally
254 Severity : enhancement
256 Description: lfs setstripe use optional parameters instead of postional
259 Severity : enhancement
261 Description: Nanosecond timestamp support for ldiskfs
262 Details : The on-disk ldiskfs filesystem has added support for nanosecond
263 resolution timestamps. There is not yet support for this at
264 the Lustre filesystem level.
267 Frequency : during server recovery
269 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
270 Details : request with rq_no_resend flag not awake l_wait_event if they get a
274 Frequency : nfs export on patchless client
276 Description: connectathon hang when test nfs export over patchless client
277 Details : Disconnected dentry cannot be found with lookup, so we do not need
278 to unhash it or make it invalid
281 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
282 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
283 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
288 Description: open req refcounting wrong on reconnect
289 Details : If reconnect happened between getting open reply from server and
290 call to mdc_set_replay_data in ll_file_open, we will schedule
291 replay for unreferenced request that we are about to free.
292 Subsequent close will crash in variety of ways.
293 Check that request is still eligible for replay in
294 mdc_set_replay_data().
299 Description: disable writes to filesystem when reading health_check file
300 Details : the default for reading the health_check proc file has changed
301 to NOT do a journal transaction and write to disk, because this
302 can cause reads of the /proc file to hang and block HA state
303 checking on a healthy but otherwise heavily loaded system. It
304 is possible to return to the previous behaviour during configure
305 with --enable-health-write.
307 Severity : enhancement
309 Description: 64-bit inode version
310 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
311 to the inode. This will be required for version-based recovery.
316 Description: MDS fails to start if a duplicate client export is detected
317 Details : in some rare cases it was possible for a client to connect to
318 an MDS multiple times. Upon recovery the MDS would detect this
319 and fail during startup. Handle this more gracefully.
321 Severity : enhancement
323 Description: Add -o localflock option to simulate old noflock
325 Details : This will achieve local-only flock/fcntl locks
331 Description: log_commit_thread vs filter_destroy race leads to crash
332 Details : Take import reference before releasing llog record semaphore
337 Description: Wrong request locking in request set processing
338 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
339 imp_delayed_list, in this place should be used imp_lock.
342 Frequency : when reconnection
344 Description: Grant Leak when osc reconnect to OST
345 Details : When osc reconnect ost, OST(filter) should check whether it
346 should grant more space to client by comparing fed_grant and
347 cl_avail_grant, and return the granted space to client instead
348 of "new granted" space, because client will call osc_init_grant
349 to update the client grant space info.
352 Frequency : when client reconnect to OST
354 Description: Grant Leak when osc do resend and replay bulk write
355 Details : When osc reconnect to OST, OST(filter)should clear grant info of
356 bulk write request, because the grant info while be sync between
357 OSC and OST when reconnect, and we should ignore the grant info
358 these of resend/replay write req.
363 Description: Grant space more than avaiable left space sometimes.
364 Details : When then OST is about to be full, if two bulk writing from
365 different clients came to OST. Accord the avaliable space of the
366 OST, the first req should be permitted, and the second one
367 should be denied by ENOSPC. But if the seconde arrived before
368 the first one is commited. The OST might wrongly permit second
369 writing, which will cause grant space > avaiable space.
372 Frequency : when client is evicted
374 Description: Grant might be wrongly erased when osc is evicted by OST
375 Details : when the import is evicted by server, it will fork another
376 thread ptlrpc_invalidate_import_thread to invalidate the
377 import, where the grant will be set to 0. While the original
378 thread will update the grant it got when connecting. So if
379 the former happened latter, the grant will be wrongly errased
380 because of this race.
385 Description: Checking Stale with correct fid
386 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
387 to check whether it is stale, because sometimes, we want the
388 enqueue happened anyway, and op_data.fid2 will not be initialized.
390 Severity : enhancement
392 Description: update patchless client
393 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
396 Frequency : only with 2.4 kernel
398 Description: random memory corruption
399 Details : size of struct ll_inode_info is to big for union inode.u and this
400 can be cause of random memory corruption.
405 Description: Memory leak in recovery
406 Details : Lov_mds_md was not free in an error handler in mds_create_object.
407 It should also check obd_fail before fsfilt_start, otherwise if
408 fsfilt_start return -EROFS,(failover mds during mds recovery).
409 then the req will return with repmsg->transno = 0 and rc = EROFS.
410 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
411 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
412 be freed no matter whether fsfilt_commit success or not.
415 Frequency : only with huge count clients
417 Description: Prevents from taking the superblock lock in llap_from_page for
419 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
420 for llap_from_page call in ll_removepage prevents from taking the
421 superblock lock for a soon died page.
426 Description: Not check open intent error before release open handle
427 Details : in some rare cases, the open intent error is not checked before
428 release open handle, which may cause
429 ASSERTION(open_req->rq_transno != 0), because it tries to release
430 the failed open handle.
435 Description: Set cat log bitmap only after create log success.
436 Details : in some rare cases, the cat log bitmap is set too early. and it
437 should be set only after create log success.
441 Description: Accessing a block bevice can re-enable I/O when Lustre is
442 tearing down a device.
443 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
448 Description: service threads may hog cpus when there are a lot of requests
450 Details : Insert cond_resched to give other threads a chance to use some of
456 Description: the cat log was not initialized in recovery
457 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
458 unlink log on mds will not be initialized until mds post
459 recovery. And also in mds post recovery, the unlink log will
460 initialization will be done asynchronausly, so there will be race
461 between add unlink log and unlink log initialization.
465 Description: brw_stats were being printed incorrectly
466 Details : brw_stats were being printed as log2 but all of them were not
467 recorded as log2. Also remove some code duplication arising from
468 filter_tally_{read,write}.
472 Frequency : rare, only in recovery.
473 Description: ASSERTION(req->rq_type != LI_POISON) failed
474 Details : imp_lock should be held while iterating over imp_sending_list for
475 prevent destroy request after get timeout in ptlrpc_queue_wait.
479 Description: replay-single.sh test 52 fails
480 Details : A lock's skiplist need to be cleanup when it being unlinked
481 from its resource list.
485 Description: Short directio read returns full requested size rather than
487 Details : Direct I/O operations should return actual amount of bytes
488 transferred rather than requested size.
490 Severity : enhancement
492 Description: metadata RPC reduction (e.g. for rm performance)
493 Details : decrease the amount of synchronous RPC between clients and servers
494 by canceling conflicing lock before the operation on the client side
495 and packing thier handles into the main operation RPC to server.
497 Severity : enhancement
499 Description: Async OSC create to avoid the blocking unnecessarily.
500 Details : If a OST has no remain object, system will block on the creating
501 when need to create a new object on this OST. Now, ways use
502 pre-created objects when available, instead of blocking on an
503 empty osc while others are not empty. If we must block, we block
504 for the shortest possible period of time.
506 --------------------------------------------------------------------------------
508 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
513 Frequency : on some architectures
515 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
516 Details : Uninitialized flags sometimes cause configuration commands to
520 Frequency : patchless clients only
522 Description: missing __iget() symbol export
523 Details : The __iget() symbol export is missing. To avoid the need for
524 this on patchless clients the deathrow inode reaper is turned
525 off, and we depend on the VM to clean up old inodes. This
526 dependency was during via the fix for bug 12181.
528 --------------------------------------------------------------------------------
530 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
532 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
533 INTEROPERATE with older versions automatically. In many cases a
534 special upgrade step is needed. Please read the
535 user documentation before upgrading any part of a live system.
536 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
537 version will not interoperate with 1.6 betas before beta5 (1.5.95).
538 * WARNING: Lustre configuration and startup changes are required with
539 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
545 Severity : enhancement
547 Description: MountConf
548 Details : Lustre configuration is now managed via mkfs and mount
549 commands instead of lmc and lconf. New obd types (MGS, MGC)
550 are added for dynamic configuration management. See
551 https://mail.clusterfs.com/wikis/lustre/MountConf for
554 Severity : enhancement
556 Description: dynamic OST addition
557 Details : OSTs can now be added to a live filesystem
559 Severity : enhancement
561 Description: startup order invariance
562 Details : MDTs and OSTs can be started in any order. Clients only
563 require the MDT to complete startup.
565 Severity : enhancement
567 Description: parallel, asynchronous orphan cleanup
568 Details : orphan cleanup is now performed in separate threads for each
569 OST, allowing parallel non-blocking operation.
571 Severity : enhancement
573 Description: optimized stripe assignment
574 Details : stripe assignments are now made based on ost space available,
575 ost previous usage, and OSS previous usage, in order to try
576 to optimize storage space and networking resources.
578 Severity : enhancement
580 Description: Permanently set tunables
581 Details : All writable /proc/fs/lustre tunables can now be permanently
582 set on a per-server basis, at mkfs time or on a live
585 Severity : enhancement
587 Description: Lustre message v2
588 Details : Add lustre message format v2.
590 Severity : enhancement
592 Description: client OST exclusion list
593 Details : Clients can be started with a list of OSTs that should be
594 declared "inactive" for known non-responsive OSTs.
598 Description: SPEC SFS validation failure on NFS v2 over lustre.
599 Details : Changes the blocksize for regular files to be 2x RPC size,
600 and not depend on stripe size.
602 Severity : enhancement
604 Description: Multiple MD RPCs in flight.
605 Details : Further unserialise some read-only MDS RPCs - learn about intents.
606 To avoid overly-overloading MDS, introduce a limit on number of
607 MDS RPCs in flight for a single client and add /proc controls
608 to adjust this limit.
610 Severity : enhancement
612 Description: client read/write statistics
613 Details : Add client read/write call usage stats for performance
614 analysis of user processes.
615 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
616 file access. extents_stats shows chunk size distribution.
617 extents_stats_per_process show chunk size distribution per
620 Severity : enhancement
622 Description: mds statistics
623 Details : Add detailed mds operations statistics in
624 /proc/fs/lustre/mds/*/stats.
628 Description: Failure of copying files with lustre special EAs.
629 Details : Client side always return success for setxattr call for lustre
630 special xattr (currently only "trusted.lov").
635 Description: Refcount LNET uuids
636 Details : The global LNET uuid list grew linearly with every startup;
637 refcount repeated list entries instead of always adding to
640 Severity : enhancement
642 Description: Dynamic service threads
643 Details : Within a small range, start extra service threads
644 automatically when the request queue builds up.
647 Frequency : mixed-endian client/server environments
649 Description: mixed-endian crashes
650 Details : The new msg_v2 system had some failures in mixed-endian
653 Severity : enhancement
655 Description: Easy OST removal
656 Details : OSTs can be permanently deactivated with e.g. 'lctl
657 conf_param lustre-OST0001.osc.active=0'
659 Severity : enhancement
661 Description: MGS proc entries
662 Details : Added basic proc entries for the MGS showing what filesystems
665 Severity : enhancement
667 Description: provide MGS failover
668 Details : Added config lock reacquisition after MGS server failover.
670 Severity : enhancement
672 Description: add Linux 2.4 support
673 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
677 Description: a large application tries to do I/O to the same resource and dies
679 Details : Check the req->rq_arrival time after the call to
680 ost_brw_lock_get(), but before we do anything about
681 processing it & sending the BULK transfer request. This
682 should help move old stale pending locks off the queue as
683 quickly as obd_timeout.
686 Frequency : when an incorrect nid is specified during startup
688 Description: ptlrpc connect to non-existant node causes kernel crash
689 Details : LNET can't be re-entered from an event callback, which
690 happened when we expire a message after the export has been
691 cleaned up. Instead, hand the zombie cleanup off to another
694 Severity : enhancement
696 Description: plain/inodebits lock performance improvement
697 Details : Grouping plain/inodebits in granted list by their request modes
698 and bits policy, thus improving the performance of search through
702 Frequency : only if OST filesystem is corrupted
704 Description: client incorrectly hits assertion in ptlrpc_replay_req()
705 Details : for a short time RPCs with bulk IO are in the replay list,
706 but replay of bulk IOs is unimplemented. If the OST filesystem
707 is corrupted due to disk cache incoherency and then replay is
708 started it is possible to trip an assertion. Avoid putting
709 committed RPCs into the replay list at all to avoid this issue.
712 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
714 Description: System hang on startup
715 Details : This bug allowed the liblustre (e.g. catamount) client to
716 return to the app before handling all startup RPCs. This
717 could leave the node unresponsive to lustre network traffic
718 and manifested as a server ptllnd timeout.
720 Severity : enhancement
722 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
723 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
724 boolean to control whether to print peer debug info when a
725 client's RPC times out.
728 Frequency : only for kernels with patches from Lustre below 1.4.3
730 Description: Remove old rdonly API
731 Details : Remove old rdonly API which unused from at least lustre 1.4.3
734 Frequency : only for devices with external journals
736 Description: Set external device read-only also
737 Details : During a commanded failover stop, we set the disk device
738 read-only while the server shuts down. We now also set any
739 external journal device read-only at the same time.
742 Frequency : when upgrading from 1.4 while trying to change parameters
744 Description: The wrong (new) MDC name was used when setting parameters for
745 upgraded MDT's. Also allows changing of OSC (and MDC)
746 parameters if --writeconf is specified at tunefs upgrade time.
749 Frequency : when setting specific ost indicies
751 Description: QOS code breaks on skipped indicies
752 Details : Add checks for missing OST indicies in the QOS code, so OSTs
753 created with --index need not be sequential.
755 Severity : enhancement
757 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
758 Details : The uninit_groups feature works in conjunction with the kernel
759 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
760 up the pass1 processing of e2fsck. This is a read-only feature
761 in ldiskfs2 only, so older kernels and current ldiskfs cannot
762 mount filesystems that have had this feature enabled.
764 Severity : enhancement
766 Description: Improve multi-block allocation algorithm to avoid fragmentation
767 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
768 allocation locality and avoid filesystem fragmentation.
770 ------------------------------------------------------------------------------
772 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
774 * Support for kernels:
775 2.4.21-47.0.1.EL (RHEL 3)
777 2.6.9-42.0.10.EL (RHEL 4)
778 2.6.12.6 vanilla (kernel.org)
779 2.6.16.27-0.9 (SLES 10)
780 * Recommended e2fsprogs version: 1.39.cfs5
782 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
786 Frequency : occasional, depends on client load and configuration
787 Bugzilla : 12181, 12203
788 Description: data loss for recently-modified files
790 Details : In some cases it is possible that recently written or created
791 files may not be written to disk in a timely manner (this should
792 normally be within 30s unless client IO load is very high).
793 The problem appears as zero-length files or files that are a
794 multiple of 1MB in size after a client crash or client eviction
795 that are missing data at the end of the file.
797 This problem is more likely to be hit on clients where files are
798 repeatedly created and unlinked in the same directory, clients
799 have a large amount of RAM, have many CPUs, the filesystem has
800 many OSTs, the clients are rebooted frequently, and/or the files
801 are not accessed by other nodes after being written.
803 The presence of the problem can be detected by looking at
804 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
805 smaller than the second (nr_unused) then dirty files will not
806 be flushed automatically to disk. "sync; sleep 10" should be
807 run several times on the node before unmounting it to update
808 Lustre (this is also safe to run on nodes without this problem).
810 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
811 that can cause this same problem, so customers using that kernel
812 also need to update the kernel in addition to Lustre. In order
813 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
815 It is normal that files written just before a client crash (less
816 than 30s) may not yet have been flushed to disk, even for local
820 Frequency : frequent on thin XT3 nodes
822 Description: UUID collision on thin XT3 Linux nodes
823 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
824 randomly, since we relied on an insufficiently-seeded PRNG.
829 Description: OSS hangs after "All ost request buffers busy"
830 Details : A deadlock between quota and journal operations caused OSS
831 hangs after printing "All ost request buffers busy."
834 Frequency : always on liblustre builds
836 Description: Cleanup compiler warnings on liblustre
839 Frequency : always on liblustre builds on XT3
841 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
845 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
846 > 32000 subdirectories
847 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
848 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
849 subdirectory count crosses 32000. This will aid e2fsck to
850 correctly handle more than 32000 subdirectories.
853 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
855 Description: System hang on startup
856 Details : This bug allowed the liblustre (e.g. catamount) client to
857 return to the app before handling all startup RPCs. This
858 could leave the node unresponsive to lustre network traffic
859 and manifested as a server ptllnd timeout.
861 Severity : enhancement
863 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
864 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
865 boolean to control whether to print peer debug info when a
866 client's RPC times out.
871 Description: make O_SYNC working on 2.6 kernels
872 Details : 2.6 kernels use different method for mark pages for write,
873 so need add a code to lustre for O_SYNC work.
878 Description: Failure to close file and release space on NFS
879 Details : Put inode details into lock acquired in ll_intent_file_open.
880 Use mdc_intent_lock in ll_intent_open to properly
881 detect all kind of errors unhandled by mdc_enqueue.
886 Description: proc file read during shutdown sometimes raced obd removal,
888 Details : Add lock to prevent obd access after proc file removal.
891 Frequency : Only for files larger than 4GB on 32-bit clients.
893 Description: improperly doing page alignment of locks
894 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
895 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
900 Description: RPCs being resent when they shouldn't be
901 Details : Some RPCs that should not be resent are being resent. This
902 can cause inconsistencies in the RPC state machine. Do not
903 resend such requests.
906 Frequency : rare, only with NFS export
908 Description: Crash on NFS re-export node
909 Details : under very unusual load conditions an assertion is hit in
910 ll_intent_file_open()
913 Frequency : only if OST filesystem is corrupted
915 Description: client incorrectly hits assertion in ptlrpc_replay_req()
916 Details : for a short time RPCs with bulk IO are in the replay list,
917 but replay of bulk IOs is unimplemented. If the OST filesystem
918 is corrupted due to disk cache incoherency and then replay is
919 started it is possible to trip an assertion. Avoid putting
920 committed RPCs into the replay list at all to avoid this issue.
925 Description: large O_DIRECT requests fail under memory pressure/fragmentation
926 Details : Large single O_DIRECT read and write calls can fail to allocate
927 a sufficiently large buffer to process the request. In case of
928 allocation failure the allocation is retried with a smaller
929 buffer and broken into smaller requests.
931 Severity : enhancement
933 Description: Add -o localflock option to simulate old noflock behaviour.
934 Details : This will achieve local-only flock/fcntl locks coherentness.
939 Description: versioning check is incomplete
940 Details : Checking the version difference of client vs. server, report
941 error if the gap is too big.
946 Description: add support PG_writeback bit
947 Details : add support for PG_writeback bit for Lustre, for more carefull
948 work with page cache in 2.6 kernel. This also fix some deadlocks
949 and remove hack for work O_SYNC with 2.6 kernel.
951 Severity : enhancement
953 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
954 Details : The uninit_groups feature works in conjunction with the kernel
955 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
956 up the pass1 processing of e2fsck. This is a read-only feature
957 in ldiskfs2 only, so older kernels and current ldiskfs cannot
958 mount filesystems that have had this feature enabled.
960 Severity : enhancement
962 Description: Improve multi-block allocation algorithm to avoid fragmentation
963 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
964 allocation locality and avoid filesystem fragmentation.
966 ------------------------------------------------------------------------------
968 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
970 * Support for kernels:
971 2.6.9-42.0.3.EL (RHEL 4)
973 2.4.21-47.0.1.EL (RHEL 3)
974 2.6.12.6 vanilla (kernel.org)
975 2.6.16.21-0.8 (SLES10)
976 * Recommended e2fsprogs version: 1.39.cfs2-0
978 * The backwards-compatible /proc/sys/portals symlink has been removed
979 in this release. Before upgrading, please ensure that you change
980 any configuration scripts or /etc/sysctl.conf files that access
981 /proc/sys/portals/* or sysctl portals.* to use the corresponding
982 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
983 in advance of the upgrade on any system running Lustre 1.4.6 or
984 newer, since /proc/sys/lnet was added in that version.
985 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
989 Frequency : only when quota is used
991 Description: avoid scanning export list for quota master
992 Details : Change the algorithms to avoid scanning export list in order
993 to improve the efficiency.
996 Frequency : MDS failover only, very rarely
998 Description: "went back in time" messages on mds failover
999 Details : The greatest transno may be lost when the current operation
1000 finishes with an error (transno==0) and the client's last_rcvd
1001 record is over-written. Save the greatest transno in the
1002 mds_last_transno for this case.
1005 Frequency : always for specific kernels and striping counts
1007 Description: client may get "Matching packet too big" without ACL support
1008 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1009 when trying to access files in certain configurations. The
1010 clients should in fact be denied when mounting because they do
1011 not understand ACLs.
1014 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1016 Description: many clients connecting with IO in progress causes connect timeouts
1017 Details : Avoid synchronous journal commits to avoid delays caused by many
1018 clients connecting/disconnecting when bulk IO is in progress.
1019 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1020 OST_IO_PORTAL to avoid delays behind potentially many pending
1024 Frequency : occasionally with multiple writers to a single file
1026 Description: shared writes to file may result in wrong size reported by stat()
1027 Details : Allow growing of kms when extent lock is cancelled
1030 Frequency : always with random mmap IO to multi-striped file
1032 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1033 Details : Only if the hole is at the end of OST object so that kms is too
1034 small. Fix is to increase kms accordingly in ll_nopage.
1037 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1039 Description: writes to a missing object would leak memory on the OST
1040 Details : If there is an inconsistency between the MDS and OST filesystems,
1041 such that the MDS references an object that doesn't exist, writes
1042 to that object will leak memory due to incorrect cleanup in the
1043 error handling path, eventually running out of memory on the OST.
1048 Description: Creating too long symlink causes lustre errors
1049 Details : Check symlink and name lengths before sending requests to MDS.
1052 Frequency : only if flock is enabled (not on by default)
1054 Description: posix locks not released on fd closure on 2.6.9+
1055 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1056 kernels, this caused such locks not to be released on fd close and
1057 then assertions on fs unmount about still used locks.
1060 Frequency : MDS failover only, very rarely
1062 Description: clients may get ASSERTION(granted_lock != NULL)
1063 Details : When request was taking a long time, and a client was resending
1064 a getattr by name lock request. The were multiple lock requests
1065 with the same client lock handle and
1066 mds_getattr_name->fixup_handle_for_resent_request found one of the
1067 lock handles but later failed with ASSERTION(granted_lock != NULL).
1072 Description: handle->h_buffer_credits > 0, assertion failure
1073 Details : h_buffer_credits is zero after truncate, causing assertion
1074 failure. This patch extends the transaction or creates a new
1078 Frequency : NFS re-export or patchless client
1079 Bugzilla : 11179, 10796
1080 Description: Crash on NFS re-export node (__d_move)
1081 Details : We do not want to hash the dentry if we don't have a lock.
1082 But if this dentry is later used in d_move, we'd hit uninitialised
1083 list head d_hash, so we just do this to init d_hash field but
1084 leave dentry unhashed.
1087 Frequency : NFS re-export or patchless client
1089 Description: NFS exports has problem with symbolic link
1090 Details : lustre client didn't properly install dentry when re-exported
1091 to NFS or running patchless client.
1094 Frequency : NFS re-export or patchless client
1096 Description: Various nfs/patchless fixes.
1097 Details : fixes reuse disconected alias for lookup process - this fixes
1098 warning "find_exported_dentry: npd != pd",
1099 fix permission error with open files at nfs.
1100 fix apply umask when do revalidate.
1103 Frequency : occasional
1105 Description: Crash on NFS re-export node
1106 Details : calling clear_page() on the wrong pointer triggered oops in
1107 generic_mapping_read().
1110 Frequency : rarely, using O_DIRECT IO
1112 Description: unaligned directio crashes client with LASSERT
1113 Details : check for unaligned buffers before trying any requests.
1116 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1118 Description: stale data returned from RAID cache
1119 Details : If only a small amount of IO is done to the RAID device before
1120 reading it again it is possible to get stale data from the RAID
1121 cache instead of reading it from disk.
1124 Frequency : always for sles10 kernel
1126 Description: sles10 support
1127 Details : ll_follow_link: compile fixes and using of nd_set_link
1128 under newer kernels.
1131 Frequency : depends on arch, kernel and compiler version, always on sles10
1134 Description: recursive or deep enough symlinks cause stack overflow
1135 Details : getting rid of large stack-allocated variable in
1139 Frequency : depends on hardware
1141 Description: lustre write performance loss in the SLES10 kernel
1142 Details : the performance loss is caused by using of write barriers in the
1143 ext3 code. The SLES10 kernel turns barrier support on by
1144 default. The fix is to undo that change for ldiskfs.
1146 ------------------------------------------------------------------------------
1148 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1150 * Support for kernels:
1151 2.6.9-42.0.3EL (RHEL 4)
1152 2.6.5-7.276 (SLES 9)
1153 2.4.21-47.0.1.EL (RHEL 3)
1154 2.6.12.6 vanilla (kernel.org)
1158 Frequency : quota enabled and large files being deleted
1160 Description: releasing more than 4GB of quota at once hangs OST
1161 Details : If a user deletes more than 4GB of files on a single OST it
1162 will cause the OST to spin in an infinite loop. Release
1163 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1168 Description: statfs data retrieved from /proc may be stale or zero
1169 Details : When reading per-device statfs data from /proc, in the
1170 {kbytes,files}_{total,free,avail} files, it may appear
1171 as zero or be out of date.
1174 Frequency : systems with MD RAID1 external journal devices
1176 Description: lconf's call to blkid is confused by RAID1 journal devices
1177 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1178 of returning all block devices that match the journal UUID.
1181 Frequency : always, for aggregate stripe size over 4GB
1183 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1184 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1185 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1188 Frequency : always if "lfs find" used on a local file/directory
1190 Description: "lfs find" segfaults if used on a local file/directory
1191 Details : The case where a directory component was not specified wasn't
1192 handled correctly. Handle this properly.
1195 Frequency : always on ppc64
1197 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1198 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1199 perform endianness conversion.
1202 Frequency : rarely (truncate to non-zero file size after write under load)
1203 Bugzilla : 10730, 10687
1204 Description: Files padded with zeros to next 4K multiple
1205 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1206 it is possible that files that are truncated to a non-zero size
1207 immediately after being written are filled with zero bytes beyond
1208 the truncated size. No file data is lost.
1210 Severity : enhancement
1212 Description: Allow recovery/failover for liblustre clients.
1213 Details : liblustre clients were unaware of failover configurations until
1216 Severity : enhancement
1218 Description: user file locks should fail when not mounting with flock option
1219 Details : Set up an error-returning stub in ll_file_operations.lock field
1220 to prevent incorrect behaviour when client is mounted without
1221 flock option. Also, set up properly f_op->flock field for
1225 Frequency : always on ia64
1227 Description: "lfs df" loops on printing out MDS statfs information
1228 Details : The obd_ioctl_data was not initialized and in some systems
1229 this caused a failure during the ioctl that did not return
1230 an error. Initialize the struct and return an error on failure.
1233 Frequency : SLES 9 only
1235 Description: Error of copying files with lustre special EAs as root
1236 Details : Client side always return success for setxattr call for lustre
1237 special xattr (currently only "trusted.lov").
1240 Frequency : rarely on clusters with both ia64+i386 clients
1242 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1243 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1244 clients are doing concurrent writes to the same file it is
1245 possible that the ia64 clients may LASSERT because the OST
1246 extent locks are not PAGE_SIZE aligned. Ensure that grown
1247 locks are always aligned on the request boundary.
1250 Frequency : specific use, occasional
1252 Description: Overwriting in use executable truncates on-disk binary image
1253 Details : If one node attempts to overwrite an executable in use by
1254 another node, we now correctly return ETXTBSY instead of
1255 truncating the file.
1257 Severity : enhancement
1259 Description: Async OSC create to avoid the blocking unnecessarily.
1260 Details : If a OST has no remain object, system will block on the creating
1261 when need to create a new object on this OST. Now, ways use
1262 pre-created objects when available, instead of blocking on an
1263 empty osc while others are not empty. If we must block, we block
1264 for the shortest possible period of time.
1269 Description: chmod on Lustre root is propagated to other clients
1270 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1271 invalid by the follow_mount time.
1274 Frequency : liblustre clients only
1276 Description: Race in 'instant cancel' lock handling could lead to such locks
1277 never to be granted in case of SMP MDS
1278 Details : Do not destroy not yet granted but cbpending locks in
1282 Frequency : replay/resend of open
1284 Description: non null lock assetion failure in mds_intent_policy
1285 Details : Trying to replay/resend lockless open requests resulted in
1286 mds_open() returning 0 with no lock. Now it sets a flag if
1287 a lock is going to be returned.
1289 Severity : enhancement
1291 Description: Checksum enhancements
1292 Details : New checksum enhancements allow for resending RPCs that failed
1295 Severity : enhancement
1297 Description: Tunables on number of dirty pages in cacche
1298 Details : Allow to set limit on number of dirty pages cached.
1303 Description: client crash on unmount - lock still has references
1304 Details : In some error handling cases it was possible to leak a lock
1305 reference on a client while accessing a file. This was not
1306 harmful to the client during operation, but would cause the
1307 client to crash when the filesystem is unmounted.
1310 Frequency : specific case, rare
1312 Description: ETXTBSY on mds though file not in use by client
1313 Details : ETXTBSY is no longer incorrectly returned when attempting to
1314 chmod or chown a directory that the user previously tried to
1315 execute or a currently-executing binary.
1318 Frequency : extremely rare except on liblustre-based clients
1320 Description: Lustre space not freed when files are deleted
1321 Details : Clean up open-unlinked files after client eviction. Previously
1322 the unlink was skipped and the files remained as orphans.
1327 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1328 Details : In some cases it was possible to send a blocking callback to
1329 a client doing a glimpse, even though that client didn't get
1330 a lock granted. When the glimpse lock is cancelled on the OST
1331 the freed lock is left on the waiting list and corrupted the list.
1334 Frequency : all core dumps
1336 Description: Broke core dumps to lustre
1337 Details : Negative dentry may be unhashed if parent does not have UPDATE
1338 lock, but some callers, e.g. do_coredump, expect dentry to be
1339 hashed after successful create, hash it in ll_create_it.
1341 ------------------------------------------------------------------------------
1343 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1345 * Support for kernels:
1346 2.6.9-42.0.2.EL (RHEL 4)
1347 2.6.5-7.276 (SLES 9)
1348 2.4.21-40.EL (RHEL 3)
1349 2.6.12.6 vanilla (kernel.org)
1353 Frequency : always on RHEL 3
1355 Description: Number of open files grows over time
1356 Details : The number of open files grows over time, whether or not
1357 Lustre is started. This was due to a filp leak introduced
1358 by one of our kernel patches.
1360 ------------------------------------------------------------------------------
1362 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1364 * Support for kernels:
1365 2.6.9-42.EL (RHEL 4)
1366 2.6.5-7.276 (SLES 9)
1367 2.4.21-40.EL (RHEL 3)
1368 2.6.12.6 vanilla (kernel.org)
1373 Bugzilla : 5719, 9635, 9792, 9684
1374 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1375 Details : If a server is under heavy load and cannot reply to new
1376 connection requests before the client resends the (re)connect,
1377 the connection handling code can behave badly if two service
1378 threads are concurrently handing separate (re)connections from
1379 the same client. Add better locking to the connection handling
1380 code, and ensure that only a single connection will be processed
1381 for a given client UUID, even if the lock is dropped.
1383 Severity : enhancement
1385 Description: add TCP zero-copy support to kernel
1386 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1387 sends if the hardware supports scatter-gather and checksumming.
1388 This allows socklnd to do client-write and server-read more
1389 efficiently and reduce CPU utilization from skbuf copying.
1392 Frequency : only if NFS exporting from client
1394 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1395 Details : If setting attributes on a file created under NFS that had
1396 never been opened it would be possible to oops the client
1397 if the file had no objects.
1400 Frequency : always for liblustre
1402 Description: liblustre client does MDS+OSTs setattr RPC for each write
1403 Details : When doing a write from a liblustre client, the client
1404 incorrectly issued an RPC to the MDS and each OST the file was
1405 striped over in order to update the timestamps. When writing
1406 with small chunks and many clients this could overwhelm the MDS
1407 with RPCs. In all cases it would slow down the write because
1408 these RPCs are unnecessary.
1410 Severity : enhancement
1412 Description: allow number of MDS service threads to be changed at module load
1413 Details : It is now possible to change the number of MDS service threads
1414 running. Adding "options mds mds_num_threads={N}" to the MDS's
1415 /etc/modprobe.conf will set the number of threads for the next
1416 time Lustre is restarted (assuming the "mds" module is also
1417 reloaded at that time). The default number of threads will
1418 stay the same, 32 for most systems.
1423 Description: OST crash if filesystem is unformatted or corrupt
1424 Details : If an OST is started on a device that has never been formatted
1425 or if the filesystem is corrupt and cannot even mount then the
1426 error handling cleanup routines would dereference a NULL pointer.
1431 Description: NULL pointer deref in llap_from_page.
1432 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1433 incorrect) mapping if the page was truncated/reclaimed while it was
1434 searched for. Check for this condition and skip such pages when
1435 doing readahead. Introduce extra check to llap_from_page() to
1436 verify page->mapping->host is non-NULL (so page is not anonymous).
1439 Frequency : Sometimes when using sys_sendfile
1441 Description: "page not covered by a lock" warnings from ll_readpage
1442 Details : sendfile called ll_readpage without right page locks present.
1443 Now we introduced ll_file_sendfile that does necessary locking
1444 around call to generic_file_sendfile() much like we do in
1448 Frequency : with certain MDS communication failures at client mount time
1450 Description: NULL pointer deref after failed client mount
1451 Details : a client connection request may delayed by the network layer
1452 and not be sent until after the PTLRPC layer has timed out the
1453 request. If the client fails the mount immediately it will try
1454 to clean up before the network times out the request. Add a
1455 reference from the request import to the obd device and delay
1456 the cleanup until the network drops the request.
1459 Frequency : occasionally during client (re)connect
1461 Description: assertion failure during client (re)connect
1462 Details : processing a client connection request may be delayed by the
1463 client or server longer than the client connect timeout. This
1464 causes the client to resend the connection request. If the
1465 original connection request is replied in this interval, the
1466 client may trip an assertion failure in ptlrpc_connect_interpret()
1467 which thought it would be the only running connect process.
1470 Frequency : only with obd_echo servers and clients that are rebooted
1472 Description: kernel BUG accessing uninitialized data structure
1473 Details : When running an obd_echo server it did not start the ping_evictor
1474 thread, and when a client was evicted an uninitialized data
1475 structure was accessed. Start the ping_evictor in the RPC
1476 service startup instead of the OBD startup.
1478 Severity : enhancement
1479 Bugzilla : 10193 (patchless)
1480 Description: Remove dependency on various unexported kernel interfaces.
1481 Details : No longer need reparent_to_init, exit_mm, exit_files,
1482 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1485 Frequency : rare (only users of deprecated and unsupported LDAP config)
1487 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1488 Details : LDAP apparently contains 'lustreName' attributes instead of
1489 'name'. A simple remapping of the name is sufficient.
1492 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1494 Description: waiting_locks_callback trips kernel BUG if client is evicted
1495 Details : Running with the dump_on_timeout debug flag turned on makes
1496 it possible that the waiting_locks_callback() can try to dump
1497 the Lustre kernel debug logs from an interrupt handler. Defer
1498 this log dumping to the expired_lock_main() thread.
1500 Severity : enhancement
1502 Description: Support NFS exporting on 2.6 kernels.
1503 Details : Implement non-rawops metadata methods for NFS server to use without
1504 changing NFS server code.
1507 Frequency : very rare (synthetic metadata workload only)
1509 Description: two racing renames might cause an MDS thread to deadlock
1510 Details : Running the "racer" program may cause one MDS thread to rename
1511 a file from being the source of a rename to being the target of
1512 a rename at exactly the same time that another thread is doing
1513 so, and the second thread has already enqueued these locks after
1514 doing a lookup of the target and is trying to relock them in
1515 order. Ensure that we don't try to re-lock the same resource.
1518 Frequency : only very large systems with liblustre clients
1520 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1521 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1522 all OSTs in parallel. This allows the eviction of stale liblustre
1523 clients to proceed much faster than if they were done in series,
1524 and also offers similar improvements for other set_info RPCs.
1529 Description: excessive CPU usage during initial read phase on client
1530 Details : During the initial read phase on a client, it would agressively
1531 retry readahead on the file, consuming too much CPU and impacting
1532 performance (since 1.4.5.8). Improve the readahead algorithm
1533 to avoid this, and also improve some other common cases (read
1534 of small files in particular, where "small" is files smaller than
1535 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1540 Description: MDS crash when receiving packet with unknown intent.
1541 Details : Do not LBUG in unknown intent case, just return -EFAULT
1543 Severity : enhancement
1544 Bugzilla : 9293, 9385
1545 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1546 Details : Do not serialize getattr (non-intent version) and statfs.
1549 Frequency : occasional, when OST network is overloaded/intermittent
1551 Description: client evicted by OST after bulk IO timeout
1552 Details : If a client sends a bulk IO request (read or write) the OST
1553 may evict the client if it is unresposive to its data GET/PUT
1554 request. This is incorrect if the network is overloaded (takes
1555 too long to transfer the RPC data) or dropped the OST GET/PUT
1556 request. There is no need to evict the client at all, since
1557 the pinger and/or lock callbacks will handle this, and the
1558 client can restart the bulk request.
1561 Frequency : Always when mmapping file with no objects
1563 Description: client crashes when mmapping file with no objects
1564 Details : Check that we actually have objects in a file before doing any
1565 operations on objects in ll_vm_open, ll_vm_close and
1571 Description: Request leak when working with deleted CWD
1572 Details : Introduce advanced request refcount tracking for requests
1573 referenced from lustre intent.
1575 Severity : Enhancement
1577 Description: Cache open file handles on client.
1578 Details : MDS now will return special lock along with openhandle, if
1579 requested and client is allowed to hold openhandle, even if unused,
1580 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1581 closing files for every read/write openration.
1583 Severity : Enhancement
1585 Description: Cache open negative dentries on client when possible.
1586 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1587 negative dentries on lock revocation.
1592 Description: Remounting a client read-only wasn't possible with a zconf mount
1593 Details : It wasn't possible to remount a client read-only with llmount.
1595 Severity : enhancement
1596 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1597 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1598 MPICH is included to add Lustre-specific ADIO interfaces.
1599 This is based closely on the UFS ADIO layer and only differs
1600 in file creation, in order to allow the OST striping to be set.
1601 This is user-contributed code and not supported by CFS.
1606 Description: extended inode attributes (immutable, append-only) work improperly
1607 when 2.4 and 2.6 kernels are used on client/server or vice versa
1608 Details : Introduce kernel-independent values for these flags.
1610 Severity : enhancement
1613 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1614 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1615 granularity. Now, Fractional MB granularity is be supported,
1616 this is very useful for low memory system.
1618 Severity : enhancement
1620 Description: Getattr by fid
1621 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1622 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1623 have locks and avoids getattr rpc requests for it on every stat.
1626 Frequency : Always, for filesystems larger than 2TB
1628 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1629 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1630 filesystems larger than 2TB to be mounted with mballoc enabled.
1633 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1635 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1636 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1637 filesystems larger than 2TB is mounted then the kernel will
1638 silently corrupt the start of the filesystem. CONFIG_LBD is
1639 enabled for all CFS-supported kernels, but the possibility of
1640 this happening with a modified kernel config exists.
1642 Severity : enhancement
1644 Description: add client O_DIRECT support for 2.6 kernels
1645 Details : It is now possible to do O_DIRECT reads and writes to files
1646 in the Lustre client mountpoint on 2.6 kernel clients.
1648 Severity : enhancement
1650 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1651 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1652 parallel, not waiting for response from every OST before sending
1653 a rpc to the next OST.
1658 Description: setattr vs write race when updating file timestamps
1659 Details : Client processes that update a file timestamp into the past
1660 right after writing to the file (e.g. tar) it is possible that
1661 the updated file modification time can be reset to the current
1662 time due to a race between processing the setattr and write RPC.
1664 Severity : enhancement
1666 Description: Bring 'lfs find' closer in line with regular Linux find.
1667 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1668 -print0 options and obtains all the needed info through the lustre
1671 Severity : enhancement
1673 Description: support up to 1024 configured devices on one node
1674 Details : change obd_dev array from statically allocated to dynamically
1675 allocated structs as they are first used to reduce memory usage
1680 Description: Flush dirty partially truncated pages during truncate
1681 Details : Immediatelly flush partially truncated pages in filter_setattr,
1682 this way we completely avoid having any pages in page cache on OST
1683 and can retire ugly workarounds during writes to flush such pages.
1688 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1689 Details : protect inode from truncation within vfs_unlink() context
1690 just take a reference before calling vfs_unlink() and release it
1691 when parent's i_sem is free.
1696 Description: last_id value checked outside lock on OST caused LASSERT failure
1697 Details : If there were multiple MDS->OST object precreate requests in
1698 flight, it was possible that the OST's last object id was checked
1699 outside a lock and incorrectly tripped an assertion. Move checks
1700 inside locks, and discard old precreate requests.
1703 Frequency : always, if extents are used on OSTs
1705 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1706 Details : OSTs using the extents format would not zero the high 16 bits of
1707 the index physical block number. This is not a problem for any
1708 OST filesystems smaller than 16TB, and no kernels support ext3
1709 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1710 new/modified files) and can be fixed for existing filesystems
1711 with e2fsprogs-1.39-cfs1.
1716 Description: import connection selection may be incorrect if timer wraps
1717 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1718 import connections to be ignored if the 32-bit jiffies counter
1719 wraps. Use a 64-bit jiffies counter.
1722 Frequency : very large clusters immediately after boot
1724 Description: LNET request buffers exhausted under heavy short-term load
1725 Details : If a large number of client requests are generated on a service
1726 that has previously never seen so many requests it is possible
1727 that the request buffer growth cannot keep up with the spike in
1728 demand. Instead of dropping incoming requests, they are held in
1729 the LND until the RPC service can accept more requests.
1732 Frequency : Sometimes during replay
1734 Description: Assertion failure in ll_local_open after replay.
1735 Details : If replay happened on an open request reply before we were able
1736 to set replay handler, reply will become not swabbed tripping the
1737 assertion in ll_local_open. Now we set the handler right after
1738 recognising of open request
1741 Frequency : very rare
1743 Description: kernel reports "badness in vsnprintf"
1744 Details : Reading from the "recovery_status" /proc file in small chunks
1745 may cause a negative length in lprocfs_obd_rd_recovery_status()
1746 call to vsnprintf() (which is otherwise harmless). Exit early
1747 if there is no more space in the output buffer.
1749 Severity : enhancement
1751 Description: clear OBD RPC statistics by writing to them
1752 Details : It is now possible to clear the OBD RPC statistics by writing
1753 to the "stats" file.
1758 Description: Client mtime is not the same on different clients after utimes
1759 Details : In some cases, the client was using the utimes() syscall on
1760 a file cached on another node. The clients now validate the
1761 ctime from the MDS + OSTs to determine which one is right.
1766 Description: Inability to activate failout mode
1767 Details : lconf script incorrectly assumed that in pythong string's numeric
1768 value is used in comparisons.
1771 Frequency : always with multiple stripes per file
1773 Description: Inefficient object allocation for mutli-stripe files
1774 Details : When selecting which OSTs to stripe files over, for files with
1775 a stripe count that divides evenly into the number of OSTs,
1776 the MDS is always picking the same starting OST for each file.
1777 Return the OST selection heuristic to the original design.
1782 Description: mount failures may take full timeout to return an error
1783 Details : Under some heavy load conditions it is possible that a
1784 failed mount can wait for the full obd_timeout interval,
1785 possibly several minutes, before reporting an error.
1786 Instead return an error as soon as the status is known.
1788 Frequency : quota enabled and large files being deleted
1790 Description: releasing more than 4GB of quota at once hangs OST
1791 Details : If a user deletes more than 4GB of files on a single OST it
1792 will cause the OST to spin in an infinite loop. Release
1793 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1798 Description: statfs data retrieved from /proc may be stale or zero
1799 Details : When reading per-device statfs data from /proc, in the
1800 {kbytes,files}_{total,free,avail} files, it may appear
1801 as zero or be out of date.
1804 Frequency : systems with MD RAID1 external journal devices
1806 Description: lconf's call to blkid is confused by RAID1 journal devices
1807 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1808 of returning all block devices that match the journal UUID.
1811 Frequency : always, for aggregate stripe size over 4GB
1813 Description: assertion fails when trying to use 4GB stripe size
1814 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
1815 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
1818 Frequency : always on ppc64
1820 Description: the first write on an ext3 filesystem with mballoc got stuck
1821 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1822 perform endianness conversion.
1824 ------------------------------------------------------------------------------
1826 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1828 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1829 INTEROPERATE with older versions automatically. Please read the
1830 user documentation before upgrading any part of a live system.
1831 * WARNING: Lustre networking configuration changes are required with
1832 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1835 * Support for kernels:
1836 2.6.9-22.0.2.EL (RHEL 4)
1837 2.6.5-7.244 (SLES 9)
1838 2.6.12.6 vanilla (kernel.org)
1841 Severity : enhancement
1842 Bugzilla : 7981/8208
1843 Description: Introduced Lustre Networking (LNET)
1844 Details : LNET is new networking infrastructure for Lustre, it includes
1845 a reorganized network configuration mode (see the user
1846 documentation for full details) as well as support for routing
1847 between different network fabrics. Lustre Networking Devices
1848 (LNDs) for the supported network fabrics have also been
1849 created for this new infrastructure.
1851 Severity : enhancement
1852 Description: Introduced Access control lists
1853 Details : clients can set ACLs on files and directories in order to have
1854 more fine-grained permissions than the standard Unix UGO+RWX.
1855 The MDS must be started with the "-o acl" mount option.
1857 Severity : enhancement
1858 Description: Introduced filesystem quotas
1859 Details : Administrators may now establish per-user quotas on the
1862 Severity : enhancement
1864 Description: Configuration change for the XT3
1865 The PTLLND is now used to run Lustre over Portals on the XT3
1866 The configure option(s) --with-cray-portals are no longer used.
1867 Rather --with-portals=<path-to-portals-includes> is used to
1868 enable building on the XT3. In addition to enable XT3 specific
1869 features the option --enable-cray-xt3 must be used.
1874 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1875 Details : A race between allocating a new llcd and re-getting the llcd_lock
1876 allowed another thread to grab newly-allocated llcd.
1878 Severity : enhancement
1880 Description: 2.6 OST async journal commit and locking fix to improve performance
1881 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1882 2.6 kernels are now async as they already were in 2.4 kernels so
1883 that they can commit concurrently with the network bulk transfer.
1884 For block-allocated files the filter allocation semaphore is held
1885 to avoid filesystem fragmentation during allocation. BKL lock
1886 removed for 2.6 xattr operations where it is no longer needed.
1891 Description: lconf incorrectly determined whether two IP networks could talk
1892 Details : In some more complicated routing and multiple-network
1893 configurations, lconf will avoid trying to make a network
1894 connection to a disjoint part of the IP space. It was doing the
1895 math incorrectly for one set of cases.
1900 Description: Fix for potential infinite loop processing records in an llog.
1901 Details : If an llog record is corrupted/zeroed, it is possible to loop
1902 forever in llog_process(). Validate the llog record length
1903 and skip the remainder of the block on error.
1906 Frequency : occasional (liblustre only)
1908 Description: liblustre could not open files whose last component is a symlink
1909 Details : sysio_path_walk() would incorrectly pass the open intent to
1910 intermediate path components.
1913 Frequency : rare (liblustre only with non-standard tuning)
1914 Bugzilla : 7201 (7350)
1915 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1916 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1917 after it has been released, only do this for referenced locks
1919 Severity : enhancement
1921 Description: specifying an (invalid) directory default stripe_size of -1
1922 would reset the directory default striping
1923 Details : stripe_size -1 was used internally to signal directory stripe
1924 removal, now use "all default" to signal dir stripe removal
1925 as a directory striping of "all default" is not useful
1928 Frequency : common for large clusters running liblustre clients
1930 Description: doing an ls when liblustre clients are running is slow
1931 Details : sending a glimpse AST to a liblustre client waits for every AST
1932 to time out, as liblustre clients will not respond. Since they
1933 cannot cache data we refresh the OST lock LVB from disk instead.
1935 Severity : enhancement
1937 Description: doing an ls at the same time as file IO can be slow
1938 Details : enqueue and other "small" requests can be blocked behind many
1939 large IO requests. Create a new OST IO portal for non-IO
1940 requests so they can be processed faster.
1943 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1945 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1946 created file with mode 000
1947 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1948 allow an ftruncate() as a non-root user to a file with mode 000.
1949 Lustre now respects this flag to disable mode checks when
1950 truncating a file owned by the user
1953 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1956 Description: Revoking locks from clients that went dead or catatonic might take
1958 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1959 cancellation of such locks instant on servers without waiting for
1960 any reply from clients. Clients drops these locks when cancel
1961 notification from server is received without replying.
1964 Frequency : liblustre-only, when liblustre client dies or becomes busy
1966 Description: Doing ls on Linux clients can take a long time with active
1968 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1969 granting such locks to it in the first place if possible. Locks
1970 are taken by proxy on the OST during the read or write and
1971 dropped immediately afterward. Add connect flags handling, do
1972 not grant locks to liblustre clients for glimpse ASTs.
1974 Severity : enhancement
1976 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1977 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1978 sequential reads and behaves badly with applications doing random
1979 reads. Improve it by reading ahead at least read region, and
1980 avoiding excessive large RPC for small reads.
1982 Severity : enhancement
1984 Description: Creating more than 1000 files for a single job may cause a load
1985 imbalance on the OSTs if there are also a large number of OSTs.
1986 Details : qos_prep_create() uses an OST index reseed value that is an
1987 even multiple of the number of available OSTs so that if the
1988 reseed happens in the middle of the object allocation it will
1989 still utilize the OSTs as uniformly as possible.
1994 Description: OST or MDS may oops in ping_evictor_main()
1995 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1996 but doesn't restart at beginning of obd_exports_timed list
1999 Severity : enhancement
2001 Description: improve by-nid export eviction on the MDS and OST
2002 Details : allow multiple exports with the same NID to be evicted at one
2003 time without re-searching the exports list.
2006 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2008 Description: MDS may oops in groups_free()
2009 Details : in rare race conditions a newly allocated group_info struct is
2010 freed again, and this can be NULL. The 2.4 compatibility code
2011 for groups_free() checked for a NULL pointer, but 2.6 did not.
2014 Frequency : common for liblustre clients doing little filesystem IO
2015 Bugzilla : 9352, 7313
2016 Description: server may evict liblustre clients accessing contended locks
2017 Details : if a client is granted a lock or receives a completion AST
2018 with a blocking AST already set it would not reply to the AST
2019 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2022 Frequency : lfs setstripe, only systems with more than 160 OSTs
2024 Description: unable to set striping with a starting offset beyond OST 160
2025 Details : llapi_create_file() incorrectly limited the starting stripe
2026 index to the maximum single-file stripe count.
2029 Frequency : LDAP users only
2031 Description: lconf did not handle in-kernel recovery with LDAP properly
2032 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2034 Severity : enhancement
2036 Description: bind OST threads to NUMA nodes to improve performance
2037 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2038 node and do their allocations there to localize memory access
2040 Severity : enhancement
2042 Description: llmount can determine client NID directly from Myrinet (GM)
2043 Details : the client NID code from gmnalnid was moved directly into
2044 llmount, removing the need to use this or specifying the
2045 client NID explicitly when mounting GM clients with zeroconf
2048 Frequency : if client is started with down MDS
2050 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2051 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2052 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2053 ptlrpc_import_delayed_req() because we didn't check if the
2054 request was interrupted, and we also didn't break out of the
2055 event loop if there was no timeout
2060 Description: data loss during non-page-aligned writes to a single file from
2061 both multiple nodes and multiple threads on one node at same time
2062 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2063 inconsistency led to false short-reads, that were cached and later
2064 used by ->prepare_write() to fill in partially written page,
2065 leading to data loss.
2068 Frequency : always, if lconf --abort_recovery used
2070 Description: lconf --abort_recovery fails with 'Operation not supported'
2071 Details : lconf was attempting to abort recovery on the MDT device and not
2074 Severity : enhancement
2076 Description: remove cleanup logs
2077 Details : replace lconf-generated cleanup logs with lustre internal
2078 cleanup routines. Eliminates the need for client-cleanup and
2081 Severity : enhancement
2083 Description: add support for EAs (user and system) on lustre filesystems
2084 Details : it is now possible to store extended attributes in the Lustre
2085 client filesystem, and with the user_xattr mount option it
2086 is possible to allow users to store EAs on their files also
2088 Severity : enhancement
2090 Description: Add possibility (config option) to show minimal available OST free
2092 Details : When compiled with --enable-mindf configure option, statfs(2)
2093 (and so, df) will return least minimal free space available from
2094 all OSTs as amount of free space on FS, instead of summary of
2095 free spaces of all OSTs.
2097 Severity : enhancement
2099 Description: do not expand extent locks acquired on OST-side
2100 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2101 by server: they are not cached anyway.
2104 Frequency : when mmap is used/binaries executed from Lustre
2106 Description: Unmmap pages before throwing them away from read cache.
2107 Details : llap_shrink cache now attempts to unmap pages before discarding
2108 them (if unmapping failed - do not discard). SLES9 kernel has
2109 extra checks that trigger if this unmapping is not done first.
2114 Description: lconf didn't resolve symlinks before checking to see whether a
2115 given mountpoint was already in use
2118 Frequency : when migrating failover services
2119 Bugzilla : 6395, 9514
2120 Description: When migrating a subset of services from a node (e.g. failback
2121 from a failover service node) the remaining services would
2122 time out and evict clients.
2123 Details : lconf --force (implied by --failover) sets the global obd_timeout
2124 to 5 seconds in order to quickly disconnect, but this caused
2125 other RPCs to time out too quickly. Do not change the global
2126 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2128 Severity : enhancement
2129 Frequency : if MDS is started with down OST
2130 Bugzilla : 9439,5706
2131 Description: Allow startup/shutdown of an MDS without depending on the
2132 availability of the OSTs.
2133 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2134 Add appropriate locking and lov-osc refcounts for safe
2135 cleaning. Add osc abort_inflight calls in case the
2136 synchronize never started.
2139 Frequency : occasional (Cray XT3 only)
2141 Description: root not authorized to access files in CRAY_PORTALS environment
2142 Details : The client process capabilities were not honoured on the MDS in
2143 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2144 been accessed by an authorized user then root was able to access
2145 the file on the local client also. The root user capabilities
2146 are now allowed on the MDS, as this environment has secure UID.
2149 Frequency : occasional
2151 Description: ldiskfs "too long searching" message happens too often
2152 Details : A debugging message (otherwise harmless) prints too often on
2153 the OST console. This has been reduced to only happen when
2154 there are fragmentation problems on the filesystem.
2159 Description: Division by zero in statfs when all OSCs are inactive
2160 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2161 lov_statfs divides by value returned by lov_get_stripecnt().
2165 Bugzilla : 9489, 3273
2166 Description: First write from each client to each OST was only 4kB in size,
2167 to initialize client writeback cache, which caused sub-optimal
2168 RPCs and poor layout on disk for the first writen file.
2169 Details : Clients now request an initial cache grant at (re)connect time
2170 and so that they can start streaming writes to the cache right
2171 away and always do full-sized RPCs if there is enough data.
2172 If the OST is rebooted the client also re-establishes its grant
2173 so that client cached writes will be honoured under the grant.
2178 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2179 Details : Now I/O RPCs go to different portal number and (presumably) fast
2180 lock requests (and glimses) and other RPCs get their own service
2181 threads pool that should be able to service those RPCs
2184 Severity : enhancement
2186 Description: Ability to exchange lustre version between client and servers and
2187 issue warnings at client side if client is too old. Also for
2188 liblustre clients there is ability to refuse connection of too old
2190 Details : New 'version' field is added to connect data structure that is
2191 filled with version info. That info is later checked by server and
2195 Frequency : rare, liblustre only.
2196 Bugzilla : 9296, 9581
2197 Description: Two simultaneous writes from liblustre at offset within same page
2198 might proceed at the same time overwriting eachother with stale
2200 Details : I/O lock withing llu_file_prwv was released too early, before data
2201 actually was hitting the wire. Extended lock-holding time until
2202 server acknowledges receiving data.
2205 Frequency : extremely rare. Never observed in practice.
2207 Description: avoid generating lustre_handle cookie of 0.
2208 Details : class_handle_hash() generates handle cookies by incrementing
2209 global counter, and can hit 0 occasionaly (this is unlikely, but
2210 not impossible, because initial value of cookie counter is
2211 selected randonly). Value of 0 is used as a sentinel meaning
2212 "unassigned handle" --- avoid it. Also coalesce two critical
2213 sections in this function into one.
2215 Severity : enhancement
2217 Description: allow liblustre clients to delegate truncate locking to OST
2218 Details : To avoid overhead of locking, liblustre client instructs OST to
2219 take extent lock in ost_punch() on client's behalf. New connection
2220 flag is added to handle backward compatibility.
2222 Severity : enhancement
2223 Bugzilla : 4928, 7341, 9758
2224 Description: allow number of OST service threads to be specified
2225 Details : a module parameter allows the number of OST service threads
2226 to be specified via "options ost ost_num_threads={N}" in the
2227 OSS's /etc/modules.conf or /etc/modprobe.conf.
2231 Bugzilla : 6146, 9635, 9895
2232 Description: servers crash with bad pointer in target_handle_connect()
2233 Details : In rare cases when a client is reconnecting it was possible that
2234 the connection request was the last reference for that export.
2235 We would temporarily drop the export reference and get a new
2236 one, but this may have been the last reference and the export
2237 was just destroyed. Get new reference before dropping old one.
2239 Severity : enhancement
2240 Frequency : if client is started with failover MDS
2242 Description: Allow multiple MDS hostnames in the mount command
2243 Details : Try to read the configuration from all specified MDS
2244 hostnames during a client mount in case the "primary"
2247 Severity : enhancement
2249 Description: Stop sending data to evicted clients as soon as possible.
2250 Details : Check if the client we are about to send or are sending data to
2251 was evicted already. (Check is done every second of waiting,
2252 for which l_wait_event interface was extended to allow checking
2253 of exit condition at specified intervals).
2256 Frequency : rare, normally only when NFS exporting is done from client
2258 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2260 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2263 Frequency : always, if srand() is called before liblustre initialization
2265 Description: Liblustre uses system PRNG disturbing its usage by user application
2266 Details : Introduce internal to lustre fast and high-quality PRNG for
2267 lustre usage and make liblustre and some other places in generic
2268 lustre code to use it.
2270 Severity : enhancement
2271 Bugzilla : 9477, 9557, 9870
2272 Description: Verify that the MDS configuration logs are updated when xml is
2273 Details : Check if the .xml configuration logs are newer than the config
2274 logs stored on the MDS and report an error if this is the case.
2275 Request --write-conf, or allow starting with --old_conf.
2277 Severity : enhancement
2279 Description: Handle symlinks in the path when checking if Lustre is mounted.
2280 Details : Resolve intermediate symlinks when checking if a client has
2281 mounted a filesystem to avoid duplicate client mounts.
2286 Description: lconf can hit an error exception but still return success.
2287 Details : The lconf command catches the Command error exception at the top
2288 level script context and will exit with the associated exit
2289 status, but doesn't ensure that this exit status is non-zero.
2294 Description: failure of ptlrpc thread startup can cause oops
2295 Details : Starting a ptlrpc service thread can fail if there are a large
2296 number of threads or the server memory is very fragmented.
2297 Handle this without oopsing.
2300 Frequency : always, only if liblustre and non-default acceptor port was used
2302 Description: liblustre cannot connect to servers with non-default acceptor port
2303 Details : tcpnal_set_default_params() was not called and was therefore
2304 ignoring the environment varaible TCPNAL_PORT, as well as other
2305 TCPNAL_ environment variables
2310 Description: two objects could be created on the same OST for a single file
2311 Details : If an OST is down, in some cases it was possible to create two
2312 objects on a single OST for a single file. No problems other
2313 than potential performance impact and spurious error messages.
2317 Bugzilla : 5681, 9562
2318 Description: Client may oops in ll_unhash_aliases
2319 Details : Client dcache may become inconsistent in race condition.
2320 In some cases "getcwd" can fail if the current directory is
2326 Description: Inode refcounting problems in NFS export code
2327 Details : link_raw functions used to call d_instantiate without obtaining
2328 extra inode reference first.
2332 Bugzilla : 9942, 9903
2333 Description: Referencing freed requests leading to crash, memleaks with NFS.
2334 Details : We used to require that call to ll_revalidate_it was always
2335 followed by ll_lookup_it. Also with revalidate_special() it is
2336 possible to call ll_revalidate_it() twice for the same dentry
2337 even if first occurence returned success. This fix changes semantic
2338 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2339 reference on a request referred from the intent.
2340 ll_intent_release() then releases such a request.
2343 Frequency : rare, normally benchmark loads only
2345 Description: unlinked inodes were kept in memory on the client
2346 Details : If a client is repeatedly creating and unlinking files it
2347 can accumulate a lot of stale inodes in the inode slab cache.
2348 If there is no other client load running this can cause the
2349 client node to run out of memory. Instead flush old inodes
2350 from client cache that have the same inode number as a new inode.
2353 Frequency : SLES9 2.6.5 kernel and long filenames only
2354 Bugzilla : 9969, 10379
2355 Description: utime reports stale NFS file handle
2356 Details : SLES9 uses out-of-dentry names in some cases, which confused
2357 the lustre dentry revalidation. Change it to always use the
2361 Frequency : rare, unless heavy write-truncate concurrency is continuous
2362 Bugzilla : 4180, 6984, 7171, 9963, 9331
2363 Description: OST becomes very slow and/or deadlocked during object unlink
2364 Details : filter_destroy() was holding onto the parent directory lock
2365 while truncating+unlinking objects. For very large objects this
2366 may block other threads for a long time and slow overall OST
2367 responsiveness. It may also be possible to get a lock ordering
2368 deadlock in this case, or run out of journal credits because of
2369 the combined truncate+unlink. Solution is to do object truncate
2370 first in one transaction without parent lock, and then do the
2371 final unlink in a new transaction with the parent lock. This
2372 reduces the lock hold time dramatically.
2375 Frequency : rare, 2.4 kernels only
2377 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2378 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2379 before unmounting their filesystem in order to avoid deadlock.
2380 The kernel_locked() function in 2.4 kernels only checks whether
2381 the kernel lock is held, not whether it is this process that is
2382 holding it as 2.6 kernels do.
2387 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2388 Details : The client ptlrpc code may be trying to reconnect to a down
2389 server before a previous connection attempt has timed out.
2390 Increase the reconnect interval to be longer than the connection
2391 timeout interval to avoid sending duplicate connections to
2395 Frequency : echo_client brw_test command
2397 Description: fix echo_client to work with OST preallocated code
2398 Details : OST preallocation code (5137) didn't take echo_client IO path
2399 into account: echo_client calls filter methods outside of any
2400 OST thread and, hence, there is no per-thread preallocated
2401 pages and buffers to use. Solution: hijack pga pages for IO. As
2402 a byproduct, this avoids unnecessary data copying.
2406 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2407 Description: Client can oops in mdc_commit_close() after open replay
2408 Details : It was possible for the MDS to return an open request with no
2409 transaction number in mds_finish_transno() if the client was
2410 evicted, but without actually returning an error. Clients
2411 would later try to replay that open and may trip an assertion
2412 Simplify the client close codepath, and always return an error
2413 from the MDS in case the open is not successful.
2416 Frequency : rare, 2.6 OSTs only
2418 Description: OST may deadlock under high load on fragmented files
2419 Details : If there was a heavy load and highly-fragmented OST filesystems
2420 it was possible to have all the OST threads deadlock waiting on
2421 allocation of biovecs, because the biovecs were not released
2422 until the entire RPC IO was completed. Instead, release biovecs
2423 as soon as they are complete to ensure forward IO progress.
2425 Severity : enhancement
2427 Description: Support for specifying external journal device at mount
2428 Details : If an OST or MDS device is formatted with an external journal
2429 device, this device major/minor is stored in the ext3 superblock
2430 and may not be valid for failover. Allow detecting and
2431 specifying the external journal at mount time.
2436 Description: Mounting an MDS with pending unlinked files may cause oops
2437 Details : target_finish_recovery() calls mds_postrecov() which returned
2438 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2439 considers this an error and immediately begins cleaning up the
2440 lov, just after starting the mds_lov process
2442 Severity : enhancement
2444 Description: Implement 'lfs df' to report actual free space on per-OST basis
2445 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2446 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2447 usage of objects; '-h' to report in human readable format.
2449 ------------------------------------------------------------------------------
2451 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2458 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2459 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2460 at startup during orphan unlinking. Instead, skip buddy update
2461 before setup, as it will be regenerated after recovery is complete.
2466 Description: If an OST is inactive, its locks might reference stale inodes.
2467 Details : lov_change_cbdata() must iterate over all namespaces, even if
2468 they are inactive to clear inode references from the lock.
2470 Severity : enhancement
2471 Frequency : occasional, if non-standard max_dirty_mb used
2473 Description: Client will block write RPCs if not enough grant
2474 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2475 then the client will block writes while waiting for another RPC
2476 to complete instead of consuming its dirty limit. With change
2477 we get improved performance when max_dirty_mb is small.
2479 Severity : enhancement
2480 Bugzilla : 3389, 6253
2481 Description: Add support for supplementary groups on the MDS.
2482 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2483 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2484 lookups for user supplementary groups into a cache.
2488 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2489 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2490 for liblustre O_EXCL is enforced by MDS.
2494 Description: readdir never returns NULL in liblustre.
2495 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2496 dentry in struct dirent.
2500 Frequency : liblustre only, depends on application IO pattern
2501 Description: liblustre clients evicted if not contacting servers
2502 Details : Don't put liblustre clients into the ping_evictor list, so
2503 they will not be evicted by the pinger ever.
2505 Severity : enhancement
2507 Description: Add ability to evict clients by NID from MDS.
2508 Details : By echoing "nid:$NID" string into
2509 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2510 $NID would be instantly evicted from this MDS and from all active
2511 OSTs connected to it.
2515 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2516 Details : lookup_it_finish() used to query file size from OSTs that was not
2521 Description: service threads change working directory to that of init
2522 Details : Starting lustre service threads may pin the working directory
2523 of the parent thread, making that filesystem busy. Threads
2524 now change to the working directory of init to avoid this.
2528 Frequency : during shutdown only
2529 Description: shutdown with a failed MDS or OST can cause unmount to hang
2530 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2535 Frequency : 2.6 only
2536 Description: chmod/chown may include an extra supplementary group
2537 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2538 supplementary group and if none is specified this is used.
2541 Bugzilla : 5479 (6816)
2543 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2544 Details : If lookup is in progress on a file that is unlinked we might try
2545 to revalidate the inode and fail in revalidate after lookup is
2546 complete and ll_file_open() enqueues the open again but
2547 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2550 Frequency : always, if lconf --abort_recovery used
2552 Description: lconf --abort_recovery fails with 'Operation not supported'
2553 Details : lconf was attempting to abort recovery on the MDT device and not
2556 ------------------------------------------------------------------------------
2558 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2563 Frequency : rare (only unsupported configurations with a node running as an
2565 Bugzilla : 6514, 5137
2566 Description: Mounting a Lustre file system on a node running as an OST could
2568 Details : OSTs now preallocates memory needed to write out data at
2569 startup, instead of when needed, to avoid having to
2570 allocate memory in possibly low memory situations.
2571 Specifically, if the file system is mounted on on OST,
2572 memory pressure could force it to try to write out data,
2573 which it needed to allocate memory to do. Due to the low
2574 memory, it would be unable to do so and the node would
2575 become unresponsive.
2577 Severity : enhancement
2579 Description: Addition of lconf --service command line option
2580 Details : lconf now accepts a '--service <arg>' option, which is
2581 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2583 Severity : enhancement
2585 Description: Failover mode is now the default for OSTs.
2586 Details : By default, OSTs will now run in failover mode. To return to
2587 the old behaviour, add '--failout' to the lmc line for OSTs.
2589 Severity : enhancement
2591 Description: Health checks are now provided for MDS and OSTs
2592 Details : Additional detailed health check information on MSD and OSTs
2593 is now provided through the procfs health_check value.
2596 Frequency : occasional, depends on IO load
2598 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2599 after numerous create/delete cycles
2600 Details : The ext3 inode allocation policy would not allocate new inodes
2601 very well on the OSTs because there are no new directories
2602 being created. Instead we look for groups with free space if
2603 the parent directories are nearly full.
2608 Description: Network or server problems during mount may cause partially
2609 mounted clients instead of returning an error.
2610 Details : The config llog parsing code may overwrite the error return
2611 code during mount error handling, returning success instead
2617 Description: MDS can fail to allocate large reply buffers
2618 Details : After long uptimes the MDS can fail to allocate large reply
2619 buffers (e.g. zconf client mount config records) due to memory
2620 fragmentation or consumption by the buffer cache. Preallocate
2621 some large reply buffers so that these replies can be sent even
2622 under memory pressure.
2626 Frequency : rare (liblustre)
2627 Description: fsx running with liblustre complained that using truncate() to
2628 extend the file doesn't work. This patch corrects that issue.
2629 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2630 ATTR_SIZE and lsm use in llu_setattr_raw.
2634 Frequency : rare, only 2.6 kernels
2635 Description: Unusual file access patterns on the MDS may result in inode
2636 data being lost in very rare circumstances.
2637 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2638 code in the 2.6 kernel caused the inode and/or EA data not to
2639 be read from disk, causing single-file corruption.
2643 Frequency : rare, only 2.6 filesystems using extents
2644 Description: Heavy concurrent write and delete load may cause data corruption.
2645 Details : It was possible under high-load situations to have an extent
2646 metadata block in the block device cache from a just-unlinked
2647 file overwrite a newly-allocated data block. We now unmap any
2648 metadata buffers that alias just-allocated data blocks.
2652 Frequency : filesystems with default stripe_count larger than 77
2653 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2654 Details : lconf specifies an inode size of 4096 bytes when the default
2655 stripe_count is larger than 77. This conflicts with the default
2656 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2657 this case to avoid pinning too much memory for large EAs.
2659 ------------------------------------------------------------------------------
2661 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2666 Frequency : rare (extremely heavy IO load with hundreds of clients)
2668 Description: Client is evicted, gets IO error writing to file
2669 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2670 caused clients to be evicted for AST timeouts. The fixes in
2671 bug 5192 mean we no longer need to have such short AST timeouts
2672 so ldlm_timeout has been increased.
2675 Frequency : occasional during --force or --failover shutdown under load
2676 Bugzilla : 5949, 4834
2677 Description: Server oops/LBUG if stopped with --force or --failover under load
2678 Details : a collection of import/export refcount and cleanup ordering
2679 issues fixed for safer force cleanup
2682 Frequency : only filesystems larger than 120 OSTs
2683 Bugzilla : 5990, 6223
2684 Description: lfs getstripe would oops on a very large filesystem
2685 Details : lov_getconfig used kfree on vmalloc'd memory
2688 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2689 Bugzilla : 6242, 6243
2690 Description: reading from files that had been truncated to a non-zero size
2691 but never opened returned no data
2692 Details : ll_file_read() reads zeros from no-object files to EOF
2697 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2698 Details : There's little harm in aborting MDS/OSS recovery and letting it
2699 try again, so I removed the LASSERT and return an error instead.
2701 Severity : enhancement
2703 Description: New debugging infrastructure for tracking down data corruption
2704 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2705 (b) cover more of the client-side code path, and (c) try to narrow
2706 down where problems occurred
2710 Bugzilla : 3819, 4364, 4397, 6313
2711 Description: Racing close and eviction MDS could cause assertion in mds_close
2712 Details : It was possible to get multiple mfd references during close and
2713 client eviction, leading to one thread referencing a freed mfd.
2715 Severity: : enhancement
2716 Bugzilla : 3262, 6359
2717 Description: Attempts to reconnect to servers are now more aggressive.
2718 Details : This builds on the enhanced upcall-less recovery that was added
2719 in 1.4.2. When trying to reconnect to servers, clients will
2720 now try each server in the failover group every 10 seconds. By
2721 default, clients would previously try one server every 25 seconds.
2726 Description: After recovery, certain operations trigger a failed
2727 assertion on a client.
2728 Details : Failing over an mds, using lconf -d --failover, while a
2729 client was doing a readdir() call would cause the client to
2730 LBUG after recovery completed and the readdir() was resent.
2732 Severity : enhancement
2734 Description: Default groups are now added by lconf
2735 Details : You can now run lconf --group <servicename> without having to
2736 manually add groups with lmc.
2739 Frequency : occasional
2741 Description: Nodes with an elan id of 0 trigger a failed assertion
2744 Frequency : always when accessing e.g. tty/console device nodes
2746 Description: tty and some other devices nodes cannot be used on lustre
2747 Details : file's private_data field is used by device data and lustre
2748 values in there got lost. New field was added to struct file to
2749 store fs-specific private data.
2752 Frequency : when exporting Lustre via NFS
2754 Description: NFSD failed occasionally when looking up a path component
2755 Details : NFSD is looking up ".." which was broken in ext3 directories
2756 that had grown large enough to become hashed.
2759 Frequency : Clusters with multiple interfaces not on the same subnet
2761 Description: Nodes will repeatedly try to reconnect to an interface which it
2762 cannot reach and report an error to the log.
2763 Details : Extra peer list entries will be created by lconf with some peers
2764 unreachable. lconf now validates the peer before adding it.
2767 Frequency : Only if a default stripe is set on the filesystem root.
2769 Description: Setting a default stripe on the filesystem root prevented the
2770 filesystem from being remounted.
2771 Details : The client was sending extra request flags in the root getattr
2772 request and did not allocate a reply buffer for the dir EA.
2775 Frequency : occasional, higher if lots of files are accessed by one client
2776 Bugzilla : 6159, 6097
2777 Description: Client trips assertion regarding lsm mismatch/magic
2778 Details : While revalidating inodes the VFS looks up inodes with ifind()
2779 and in rare cases can find an inode that is being freed.
2780 The ll_test_inode() code will free the lsm during ifind()
2781 when it finds an existing inode and then the VFS later attaches
2782 this free lsm to a new inode.
2786 Bugzilla : 6422, 7030
2787 Description: MDS deadlock between mkdir and client eviction
2788 Details : Creating a new file via mkdir or mknod (starting a transaction
2789 and getting the ns lock) can deadlock with client eviction
2790 (gets ns lock and trying to finish a synchronous transaction).
2793 Frequency : occasional
2794 Description: While starting a server, the fsfilt_ext3 module could not be
2796 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2797 kernels. Previously, lconf would still use the ext3 name
2798 when trying to load modules. Now, it will correctly use
2799 ext3 on 2.4 and ldiskfs on 2.6.
2801 Severity : enhancement
2802 Description: The default stripe count has been changed to 1
2803 Details : The interpretation of the default stripe count (0, to lfs
2804 or lmc) has been changed to mean striping across a single
2805 OST, rather than all available. For general usage we have
2806 found a stripe count of 1 or 2 works best.
2808 Severity : enhancement
2809 Description: Add support for compiling against Cray portals.
2810 Details : Conditional compiling for some areas that are different
2814 Frequency : occasional
2815 Bugzilla : 6409, 6834
2816 Description: Creating files with an explicit stripe count may lead to
2817 a failed assertion on the MDS
2818 Details : If some OSTs are full or unavailable, creating files may
2819 trigger a failed assertion on the MDS. Now, Lustre will
2820 try to use other servers or return an error to the
2824 Frequency : occasional
2826 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2827 caused lock timeout during readahead (since 1.4.2).
2828 Details : Processes doing readahead might match a lock that hasn't been
2829 granted yet if there are overlapping and conflicting lock
2830 requests. The readahead process waits on ungranted lock
2831 (original lock is CBPENDING), while OST waits for that process
2832 to cancel CBPENDING read lock and eventually evicts client.
2834 Severity : enhancement
2836 Description: Initial enabling of flock support for clients
2837 Details : Implements fcntl advisory locking and file status functions.
2838 This feature is provided as an optional mount flag (default
2839 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2840 locking are implemented yet, and those that are are not guaranteed
2841 to be completely correct in production environments.
2842 mount -t lustre -o [flock|noflock] ...
2845 Frequency : occasional
2847 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2848 assertion in the ext3 JBD (journaling) layer.
2849 Details : The b_committed_data struct is protected by the big kernel lock
2850 in 2.4 kernels, serializing journal_commit_transaction() and
2851 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2852 access to this struct. In 2.6 kernels there is finer grained
2853 locking to improve SMP performance of the JBD layer.
2857 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2862 Description: merge and cleanup kernel patches.
2865 -----------------------------------------------------------------------------
2867 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2869 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2870 versions of Lustre. Please update all servers and clients to
2871 version 1.4.2 or later at the same time. You must also run
2872 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2873 to update the configuration logs.
2875 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2876 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2877 - swab llog records of type '0' so we get proper header size/idx (5861)
2878 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2879 - fix rename of one directory over another leaking an inode (5953)
2880 - avoid SetPageDirty on 2.6 (5981)
2881 - don't re-add just-being-destroyed locks to the waiting list (5653)
2882 - when creating new directories, inherit the parent's custom
2883 striping settings if present parent (3048)
2884 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2885 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2886 - don't hold client locks on temporary worklist from l_lru (5666)
2887 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2888 - automatically evict dead clients (5921)
2889 - Update file size properly in create+truncate+fstat case (6196)
2890 - Do not unhash mountpoint dentries, do not allow removal of
2892 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2893 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2894 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2895 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2896 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2897 - Correctly handle reads of files with no objects (6243)
2898 - lctl recover will also mark a device active if deactivate used (5933)
2900 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2901 - allow --write-conf on an MDS with different nettype than client (5619)
2902 - don't write config llogs to MDS for mounts not from that MDS (5617)
2903 - lconf should create multiple TCP connections from a client (5201)
2904 - init scripts are now turned off by default; run chkconfig --on
2905 lustre and chkconfig --on lustrefs to use them
2906 - upcalls are no longer needed for clients to recover to failover
2908 - add --abort-recovery option to lconf to abort recovery on device
2910 - add support for an arbitrary number of OSTs (3026)
2911 - Quota support protocol changes.
2912 - forward compatibility changes to wire structs (6007)
2913 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2914 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2915 - improved "lustre status" script
2916 - initialize blocksize for non-regular files (6062)
2917 - added --disable-server and --disable-client configure options (5782)
2918 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2919 - Vanilla 2.4.29 support
2920 - increase maximum number of obd devices to 520 (6242)
2921 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2922 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2923 SLES 9 2.6.5 kernels
2924 - limit stripes per file to 160 (the maximum EA size) (6093)
2926 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2929 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2930 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2931 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2932 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2933 - small ext3 extents cleanups and fixes (5733)
2934 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2935 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2936 - avoid needless client->OST connect, fix handle mismatch (5317)
2937 - fix DLM error path that led to out-of-sync client, long delays (5779)
2938 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2939 - fix several locking issues related to i_size (5492,5624,5654,5672)
2940 - don't move pending lock onto export if it is already evicted (5683)
2941 - fix kernel oops when creating .foo in unlinked directory (5548)
2942 - fix deadlock in obdfilter statistics vs. object create (5811)
2943 - use time_{before,after} to avoid timer jiffies wrap (5882)
2944 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2945 - Do not leak request if server was not able to process it (5154)
2946 - If mds_open unable to find parent dir, make that negative lookup(5154)
2947 - don't create new directories with extent-mapping (5909, 5936)
2949 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2950 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2951 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2952 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2953 - support for RHEL4 kernel on i686 (5773)
2954 - provide error messages when incompatible logs are encountered (5898)
2956 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2957 * version 1.4.0.10 (1.4.1 release candidate 1)
2959 - don't keep a lock reference when lock is not granted (4238)
2960 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2961 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2962 - drop import inflight refcount on signal_completed_replay error (5255)
2963 - unlock page after async write error during send (3677)
2964 - handle missing objects in filter_preprw_read properly (5265)
2965 - no transno return for symlink open, don't save no-trasno open (3440)
2966 - don't try to complete elan receive that already failed (4012)
2967 - free RPC server reply state on error (5406)
2968 - clean up thread from ptlrpc_start_thread() on error (5160)
2969 - readahead could read extra page into cache that wasn't ejected (5388)
2970 - prevent races in class_attach/setup/cleanup/detach (5260)
2971 - don't dereference de->d_inode after l_dput of de (5458)
2972 - use "int" for stripe value returned from lock_to_stripe (5544)
2973 - mballoc allocation and error-checking fixes in 2.6 (5504)
2974 - block device patches to fix I/O request sizes in 2.6 (5482)
2975 - look up hostnames for IB nals (5602)
2976 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2977 - don't start multiple acceptors for the same port (5277)
2978 - fix incorrect LASSERT in mds_getattr_name (5635)
2979 - export a proc file for general "ping" checking (5628)
2980 - fix "lfs check" to not block when the MDS is down (5628)
2982 - service request history (4965)
2983 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2984 - create an "evict_client" /proc entry on OSTs, like the MDS has
2985 - fix mount usage message, return errors per mount(8) (5168)
2986 - change grep [] to grep "[]" in tests so they work in more UMLs
2987 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2988 - remove ancient LOV_MAGIC_V0 EA support (5047)
2989 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2990 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2991 - allow readahead allocations to fail when low on memory (5383)
2992 - mmap locking landed again, after considerable improvement (2828)
2993 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2995 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2998 - send OST transaction number in read/write reply to free req (4966)
2999 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3000 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3001 - the watchdog thread now runs as interruptible (5246)
3002 - flock/lockf fixes (but it's still disabled, pending 5135)
3003 - don't use EXT3 constants in llite code (5094)
3004 - memory shortage at startup could cause assertion (5176)
3006 - reorganization of lov code
3007 - single portals codebase
3009 - add extents/mballoc support (5025)
3010 - direct I/O reads in the obdfilter (4048)
3011 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3013 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3016 - send OST transaction number in read/write reply to free req (4966)
3017 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3018 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3019 - the watchdog thread now runs as interruptible (5246)
3020 - handle missing objects in filter_preprw_read properly (5265)
3021 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3022 - drop import inflight refcount on signal_completed_replay error (5255)
3023 - unlock page after async write error during send (3677)
3024 - return original error code on reconstructed replies (3761)
3025 - no transno return for symlink open, don't save no-trasno open (3440)
3027 - add pid to ldlm debugging output (4922)
3028 - bump the watchdog timeouts -- we can't handle 30sec yet
3029 - extra debugging for orphan dentry/inode bug (5259)
3031 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3034 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3035 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3036 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3037 - fix NULL dereference in /proc/sys/portals/routes (4827)
3038 - allow failed mdc_close() operations to be interrupted (4561)
3039 - stop precreate on OST before MDS would time out on it (4778)
3040 - don't send partial-page writes before EOF from client (4410)
3041 - discard client grant for sub-page writes on large-page clients (4520)
3042 - don't free dentries not owned by NFS code, check generation (4806)
3043 - fix lsm leak if mds_create_objects() fails (4801)
3044 - limit debug_daemon file size, always print CERROR messages (4789)
3045 - use transno after validating reply (3892)
3046 - process timed out requests if import state changes (3754)
3047 - update mtime on OST during writes, return in glimpse (4829)
3048 - add mkfsoptions to LDAP (4679)
3049 - use ->max_readahead method instead of zapping global ra (5039)
3050 - don't interrupt __l_wait_event() during strace
3052 - add software watchdogs to catch hung threads quickly (4941)
3053 - make lustrefs init script start after nfs is mounted
3054 - fix CWARN/ERROR duplication (4930)
3055 - return async write errors to application if possible (2248)
3056 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3057 - print NAL number in %x format (4645)
3058 - update barely-supported suse-2.4.21-171 series (4842)
3059 - support for sles 9 %post scripts
3060 - support for building 2.6 kernel-source packages
3061 - support for sles km_* packages
3063 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3066 - ignore -ENOENT errors in osc_destroy (3639)
3067 - notify osc create thread that OSC is being cleaned up (4600)
3068 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3069 - reconstruct ost_handle() like mds_handle() (4657)
3070 - create a new thread to do import eviction to avoid deadlock (3969)
3071 - let lconf resolve symlinked-to devices (4629)
3072 - don't unlink "objects" from directory with default EA (4554)
3073 - hold socknal file ref over connect in case target is down (4394)
3074 - allow more than 32000 subdirectories in a single directory (3244)
3075 - fix blocks count for O_DIRECT writes (3751)
3076 - OST returns ENOSPC from object create when no space left (4539)
3077 - don't send truncate RPC if file size isn't changing (4410)
3078 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3079 - bind to privileged port in socknal and tcpnal (3689)
3081 - rate limit CERROR/CWARN console message to avoid overload (4519)
3082 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3083 - basic mmap support (3918)
3084 - kernel patch series update from b1_4 (4711)
3086 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3089 - avoid crash during MDS cleanup with OST shut down (2775)
3090 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3091 - don't use bad inodes on the MDS (3744)
3092 - dynamic object preallocation to improve recovery speed (4236)
3093 - don't hold spinlock over lock dumping or change debug flags (4401)
3094 - don't zero obd_dev when it is force cleaned (3651)
3095 - print grants to console if they go negative (4431)
3096 - "lctl deactivate" will stop automatic recovery attempts (3406)
3097 - look for existing locks in ldlm_handle_enqueue() (3764)
3098 - don't resolve lock handle twice in recovery avoiding race (4401)
3099 - revalidate should check working dir is a directory (4134)
3101 - don't always mark "slow" obdfilter messages as errors (4418)
3103 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3106 - don't close LustreDB during write_conf until it is done (3860)
3107 - fix typo in lconf for_each_profile (3821)
3108 - allow dumping logs from multiple threads at one time (3820)
3109 - don't allow multiple threads in OSC recovery (3812)
3110 - fix debug_size parameters (3864)
3111 - fix mds_postrecov to initialize import for llog ctxt (3121)
3112 - replace config semaphore with spinlock (3306)
3113 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3114 - don't allow enqueue to complete on a destroyed export (3822)
3115 - down write_lock before checking llog header bitmap (3825)
3116 - recover from lock replay timeout (3764)
3117 - up llog sem before sending rpc (3652)
3118 - reduce ns lock hold times when setting kms (3267)
3119 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3120 - fix NULL deref and obd_dev leak on setup error (3312)
3121 - replace some LBUG about llog ops with error handling (3841)
3122 - don't match INVALID dentries from d_lookup and spin (3784)
3123 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3124 - fix invalid assertion in ptlrpc_set_wait (3880)
3126 - add libwrap support for the TCP acceptor (3996)
3127 - add /proc/sys/portals/routes for non-root route listing (3994)
3128 - allow setting MDS UUID in .xml (2580)
3129 - print the stack of a process that LBUGs (4228)
3131 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3134 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3135 - make sure to unset replay flag from failed open requests (3440)
3136 - if default stripe count is 0, use OST count for inode size (3636)
3137 - update parent mtime/ctime on client for create/unlink (2611)
3138 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3139 - free recovery state on server during a forced cleanup (3571)
3140 - unregister_reply for resent reqs (3063)
3141 - loop back devices mounting and status check on 2.6 (3563)
3142 - fix resource-creation race that can provoke i_size == 0 (3513)
3143 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3144 - more debugging for page-accounting assertion (3746)
3145 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3146 - avoid deadlock after precreation failure (3758)
3147 - fix race and lock order deadlock in orphan handling (3450, 3750)
3148 - add validity checks when grabbing inodes from l_ast_data (3599)
3150 - add /proc/.../recovery_status to obdfilter (3428)
3151 - lightweight CDEBUG infrastructure, debug daemon (3668)
3152 - change default OSC RPC parameters to be better on small clusters
3153 - turn off OST read cache for files smaller than 32MB
3154 - install man pages and include them in rpms (3100)
3155 - add new init script for (un)mounting lustre filesystems (2593)
3156 - run chkconfig in %post for init scripts (3701)
3157 - drop scimac NAL (unmaintained)
3159 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3162 - clean kiobufs before and after use (3485)
3163 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3164 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3165 - change init script to fail more clearly if not run as root (1528)
3166 - allow clients to reconnect during replay (1742)
3167 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3168 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3169 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3170 - fix cond_resched (3554)
3171 - only evict unfinished clients after recovery (3515)
3172 - allow bulk resend, prevent data loss (3570)
3173 - dynamic ptlrpc request buffer allocation (2102)
3174 - don't allow unlinking open directory if it isn't empty (2904)
3175 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3176 - remove extraneous obd dereference causing LASSERT failure (3334)
3177 - don't use get_cycles() when creating temp. files on the mds (3156)
3178 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3179 - handle EEXIST for set-stripe, set proper directory name (3336)
3181 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3182 - fix ksocknal_fmb_callback() error messages (2918)
3184 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3187 - don't copy lvb into (possibly NULL) reply on error (2983)
3188 - don't deref dentry after dput, don't free lvb on error (2922)
3189 - use the kms to determine writeback rpc length (2947)
3190 - increment oti_logcookies when osc is inactive (2948)
3191 - update client's i_blocks count via lvb messages (2543)
3192 - handle intent open/close of special files properly (1557)
3193 - mount MDS with errors=remount-ro, like obdfilter (2009)
3194 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3195 - don't use cancelling-locks' kms values (2947)
3196 - use highest lock extent for kms, not last one (2925)
3197 - don't dereference ERR_PTR() dentry in error handling path (3107)
3198 - fix thread race in portals_debug_dumplog() (3122)
3199 - create lprocfs device entries at setup instead of at attach (1519)
3200 - common AST error handler, don't evict client on completion race (3145)
3201 - zero nameidata in detach_mnt in 2.6 (3118)
3202 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3203 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3204 - initialize RPC timeout timer earlier for 2.6 (3219)
3205 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3206 - print nal/nid for unknown nid (3258)
3207 - additional checks for oscc recovery before doing precreate (3284)
3208 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3209 - don't crash in mdc_close for bad permissions on open (3285)
3210 - zero i_rdev for non-device files (3147)
3211 - clear page->private before handing to FS, better assertion (3119)
3212 - tune the read pipeline (3236)
3213 - fix incorrect decref of invalidated dentry (2350)
3214 - provide read-ahead stats and refine rpc in flight stats (3328)
3215 - don't hold journal transaction open across create RPC (3313)
3216 - update atime on MDS at close time (3265)
3217 - close LDAP connection when recovering to avoid server load (3315)
3218 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3219 - don't leak open file on MDS after open resend (3325)
3220 - serialize filter_precreate and filter_destroy_precreated (3329)
3221 - loop device shouldn't call sync_dev() for nul device (3092)
3222 - clear page cache after eviction (2766)
3223 - resynchronize MDS->OST in background (2824)
3224 - refuse to mount the same filesystem twice on same mountpoint (3394)
3225 - allow llmount to create routes for mounting behind routers (3320)
3226 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3227 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3228 - fix rare problem with rename on htree directories (3417)
3230 - allow default OST striping configuration per directory (1414)
3231 - fix compilation for qswnal for 2.6 kernels (3125)
3232 - increase maximum number of MDS request buffers for large systems
3233 - change liblustreapi to be useful for external progs like lfsck (3098)
3234 - increase local configuration timeout for slow disks (3353)
3235 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3237 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3240 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3241 - don't overwrite extent policy data in reply if lock was blocked (2901)
3242 - drop filter export grants atomically with removal from device (2663)
3243 - del obd_self_export from work_list in class_disconnect_exports (2908)
3244 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3245 - swab reply message in mdc_close, other PPC fixes (2464)
3246 - fix destroying of named logs (2325)
3247 - overwrite old logs when running lconf --write_conf (2264)
3248 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3249 - fix race in target_handle_connect (2898)
3250 - mds_reint_create() should take same inode create lock (2926)
3251 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3252 - don't close files for self_export to avoid uninitialized obd (2936)
3253 - allow MDS with the same name as client node (2939)
3254 - hold dentry reference for closed log files for unlink (2325)
3255 - reserve space for all logs during transactions (2059)
3256 - don't evict page beyond end of stripe extent (2925)
3257 - don't oops on a deleted current working directory (2399)
3258 - handle hard links to targets without a parent properly (2517)
3259 - don't dereference NULL lock when racing during eviction (2867)
3260 - don't grow lock extents when lots of conflicting locks (2919)
3262 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3265 - account for cache space usage on clients to avoid data loss (974)
3266 - lfsck support in lustre kernel code (2349)
3267 - reduce journal credits needed for BRW writes (2370)
3268 - orphan handling to avoid losing space on client/server crashes
3269 - ptlrpcd can be blocked, stopping ALL progress (2477)
3270 - use lock value blocks to assist in proper KMS, faster stat (1021)
3271 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3272 - recovery for initial connections (2355)
3273 - fixes for mds_cleanup_orphans (1934)
3274 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3275 - block all file creations until orphan recovery completes (1901)
3276 - client remove rq_connection from request struct (2423)
3277 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3278 - recovery timer race (2670)
3279 - mdc_close recovey bug (2532)
3280 - ptlrpc cleanup bug (2710)
3281 - mds timeout on local locks (2588)
3282 - namespace lock held during RPCs (2431)
3283 - handle interrupted sync write properly (2503)
3284 - don't try to handle a message that hasn't been replied to (2699)
3285 - client assert failure during cleanup after abort recovery (2701)
3286 - leak mdc device after failed mount (2712)
3287 - ptlrpc_check_set allows timedout requests to complete (2714)
3288 - wait for inflight reqs when ptlrpcd finishes (2710)
3289 - make sure unregistered services are removed from the srv_list
3290 - reset bulk XID's when resending them (caught by 1138 test)
3291 - unregister_bulk after timeout
3292 - fix lconf error (2694)
3293 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3294 - readahead locks pages, leaves pending causing memory pressure (2673)
3295 - increase OST request buffers to 4096 on large machines (2729)
3296 - fix up permission of existing directories in simple_mkdir (2661)
3297 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3298 - don't assign transno to errored transactions (2742)
3299 - don't delete objects on OST if given a bogus objid from MDS (2751)
3300 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3301 - if rq_no_resend, then timeout request after recovery (2432)
3302 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3303 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3304 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3305 - MDS recovery shouldn't delete objects that it has given out (2730)
3306 - if enqueue arrives after completion, don't clobber LVB (2819)
3307 - don't unlock pages twice when trigger_group_io returns error (2814)
3308 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3309 - don't write pages to disk if there was an error (1450)
3310 - don't ping imports that have recovery disabled (2676)
3311 - take buffered bytes into account when balancing socknal conn (2817)
3312 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3313 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3314 - remove little-endian swabbing of llog records (1987)
3315 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3316 - retry reposting request buffers if they fail (1191)
3317 - grow extent at grant time to avoid granting a revoked lock (2809)
3318 - lock revoke doesn't evict page if covered by a second lock (2765)
3319 - disable VM readahead to avoid reading outside lock extents (2805)
3321 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3322 - updated kernel patches for hp-2.4.20 kernel (2681)
3324 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3327 - fix truncated write corruption (2366)
3328 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3330 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3331 - ensure len(uuid) < 37 in lmc (1171)
3332 - fix ia64 OOPS in llog_test (2255)
3333 - zero end of page at obdfilter for partial page writes (2648)
3334 - don't leave stale dentries around after renames (bug 2428)
3335 - fix timeouts when evicting a client with a single lock held (2642)
3336 - set deadline for the initial HELLO message to drain (2634)
3337 - print out dotted-quad IP addresses in the socknal (2302)
3339 - additional debugging for MDS client eviction problem (2443)
3340 - fix mkfsoptions support for osts (2603, 2604)
3342 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3345 - add series for the vanilla 2.6.0 kernel
3346 - add series for the vanilla 2.4.24 kernel
3347 - add series for a cray x86/64 UL kernel drop
3348 - fix xattr patches for the vanilla 2.4.19 series
3350 - generate true UUIDs in lmc (1171)
3351 - have portals stack dumping break in UML (2466)
3352 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3353 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3354 - wake callback waiting threads on client eviction (2460)
3355 - Add --ptldebug and --subsystem to lmc (1719)
3356 - update assertion to allow safe interrupt allocation
3357 - set rq_no_resend for cancel requests (2432)
3358 - recalculate ptlrpcd timeout after resend (2494)
3359 - call vfs_rmdir when removing pending directories (2368)
3360 - fix renaming a file to itself (2429)
3361 - lmc creates a default one-stripe lov (2454)
3362 - expand procfs space to handle large clusters (2326)
3363 - increase UML stack to avoid overflow
3364 - update lconf's list of debug and subsystem masks
3365 - fix lfs find --obd (2510)
3366 - /proc tunable for disabling filter read caching (2591)
3367 - stop rpm packages from altering slapd.conf (2301)
3368 - disable nagle in the socknal under 0conf (2578)
3369 - choose mds inode size based on stripe count (2572)
3370 - fix kernel-source rpm problems (2516)
3372 - add --disable-doc to avoid pdf generation (2421)
3373 - update documentation, tests, type-os, comments
3374 - avoid format warnings on ia64
3375 - remove the TOE NAL
3376 - tiny code cleanups by removing unused fields
3378 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3381 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3382 - remove the most blatant lies from BUILDING file (2371)
3383 - change default debug level to reasonable production setting
3384 - reduce client side cache size to reduce cache flush time
3385 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3386 - make TCP zerocopy and pinger support enabled by default (2476)
3387 - sync writes completed after process exits caused crashes (2319)
3388 - maintain correct mount count on the MDS (2356)
3389 - backout 1557, because 2316 wasn't really fixed
3390 - better file I/O statistics gathering in /proc
3391 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3392 - another kernel patch to fix zero-copy TCP function export
3393 - don't take duplicate lock when processing re-sent getattr (2420)
3394 - lctl uses obd_self_export instead of creating new conn (2353)
3395 - MDS/OST recovery case which requires object creation asserted (2425)
3396 - move lfs from /usr/sbin to /usr/bin in packages
3397 - fix race between mds_client_add and mds_client_free (2417)
3398 - use kmalloc instead of slabs in portals (2430)
3399 - don't create duplicate records when a failover MDS is present (2442)
3400 - remove unnecessary mount age check (2332)
3401 - don't remove directory inodes from locks prematurely (2451)
3402 - don't break if MDS service name is the same as hostname (2103)
3403 - fix races in client write RPC generation when cache full (2482)
3405 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3408 - remove now-unused request->rq_obd (278)
3409 - if an allocation fails, print out how much memory we've used (1933)
3410 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3411 - add the "configurable stack size" patch to most series files (1256)
3412 - ability to write large log records, for 100+ OST configs (2306)
3413 - fix NULL deref when filter_prep fails (2314)
3414 - fix operator precedence error in filter_sync
3415 - dynamic allocation of socknal TX descriptors (2315)
3416 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3417 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3418 - ensure that CWARN messages in Portals always get to the syslog
3419 - __init/__exit are not for prototype decls (ldlm_init/exit)
3420 - x86-64 compile warning fixes
3421 - fix gateway LMC keyword conflict (2318)
3422 - fix MDS lock inversions in getattr/reint paths (1844)
3423 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3424 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3425 - DLM race condition prevented some lock evictions (2328)
3426 - ENOMEM detection and retry on socknal sends (2230)
3427 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3428 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3429 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3430 - better error messages when a client is rejected during recovery (1505)
3431 - avoid cancelling locks which were never granted, after failure (2330)
3432 - fix i_sem/journal inversion in mds_client_add (2333)
3433 - fix truncate/getattr lock cycle deadlock (2334)
3434 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3435 - fix two rare exit paths which could leak an l_lock() ref (2321)
3436 - fencepost error in MDS/OST orphan recovery (2226)
3437 - make log record alignment 8 bytes (1988)
3438 - lstripe now fails when requested offset > ost_count (2237)
3439 - ensure that all kernel series have a complete list.h (1607)
3440 - fix crashes in special-file operations (2316)
3441 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3443 - allow configurable automake binary, for testing new versions
3444 - small update to the lfs documentation
3446 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3448 * fix negative export reference count in fsfilt_sync (2312)
3450 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3451 * release candidate 0.9.1
3453 - orphans are moved into the PENDING directory for possible recovery
3454 - replayed opens now open by fid for orphan/rename safety (1042)
3455 - last close of an orphan inode generates a transno (683)
3456 - chdir() and mount() now pin the directory entry (1020)
3457 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3458 - discard very old requests without processing them (1502)
3459 - remove some common, well-understood CERRORs (1505)
3460 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3461 - clear "grant" flags in OST replies until OST grant code lands (1644)
3462 - fix read performance by not clobbering i_blksize on client (1598)
3463 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3464 - make LRU size a /proc tunable, clears locks when reduced (707)
3465 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3466 - server locks take references on exports now (1558)
3467 - build fixes for 2.4.20-rh trees (1663)
3468 - return an error from lov_create if all OSCs are inactive (1751)
3469 - fix import levels when a reconnect happens without a timeout (1597)
3470 - exit early from mds_open if we get a lookup error (1749)
3471 - partial page read at EOF wouldn't wait for disk before sending (1642)
3472 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3473 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3474 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3475 - write/truncate lock inversion (1639)
3476 - Don't auto-load obdclass, portals modules during cleanup (1495)
3477 - fix timestamps from jumping to "now" (1763)
3478 - extra journal assertions (1648)
3479 - add an extra multiunlink test (1771)
3480 - fix read_record/write_record API (1776)
3481 - fix leak of offset_extent, possible incorrect i_size later (1772)
3482 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3483 - Debugging for the kqswnal_get_idle_tx problems (1820)
3484 - Allow recovery to be attempted multiple times (1536)
3485 - Write out MDS last_rcvd file after it is first created (1600)
3486 - Fix tx_descriptor leak in failed transmit situations (1827)
3487 - ext3 journaling fixes for assertion failure after IO error (1871)
3488 - class_export_put() on freed export after completion AST error (1896)
3489 - Fix revalidate looping in VFS (1322)
3490 - Don't access a freed export during MDS_REINT timeout (1521)
3491 - Add open-unlink recovery support on the MDS (1673,1764)
3492 - Return an error if no MDS data was read from last_rcvd (1946)
3493 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3494 - Don't setup a disk device that doesn't match exported UUID (317)
3495 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3496 - avoid committing NULL handle in force close
3497 - local.sh is now a one-stripe LOV configuration
3498 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3499 - fix ext3 htree duplicate directory entry corruption (1516)
3500 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3501 - update to new LOV EA format (2097)
3502 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3504 2003-06-15 Phil Schwan <phil@clusterfs.com>
3507 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3508 - per-import/export recovery handling (958, 931, 959)
3509 - multiple last-rcvd slots, for serving multiple FSes (949)
3510 - connections are again shared between multiple imp/exports (963, 964)
3511 - "umount -f" would hang if any requests needed to be sent (393, 978)
3512 - avoid pinning large req buffer by copying for queued messages (989)
3513 - add "uuid" to "lctl device" command to help upcalls (991)
3514 - "open" RPCs with transnos would confuse recovery counters (1037)
3515 - do proper endian conversion of all wire messages (288, 340, 891)
3516 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3517 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3518 - fix ptlrpc_connection leak in target_handle_connect (1174)
3519 - fix import refcounting bug in OST and MDS cleanup (1134)
3520 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3521 - fix very unlikely obd_types race condition (501)
3522 - remove osc_open hack for echo_client (1187)
3523 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3524 - a failure in read_inode2 leads to deadlock (1139)
3525 - cancel ack-locks as soon as transaction is committed (1072)
3526 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3527 - make sure to commitrw after any preprw to avoid deadlock (1162)
3528 - failing to execute a file in a lustre FS would lock inode (1203)
3529 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3530 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3531 - obd_uuid2tgt would walk off the end of the list (1255)
3532 - on IA64 the osc would give portals incorrect bulk size (1258)
3533 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3534 - fix lock inversion caused by new llite matching code (1282)
3535 - limit the number of dirty pages on a client to 10MB (1286)
3536 - timed out locks were not being corrected cancelled (1289)
3537 - fix O_DIRECT above 4GB on IA-32 (1292)
3538 * major user-visible changes
3539 - fail out/fail over policy now controlled by the upcall (993)
3541 - add OBD_PING to check server availability and failure (954)
3542 - lustre messages are now sent in sending host order (288, 340, 891)
3543 - add eadatalen to MDS getattr reply (340)
3544 - OST read replies may contain second buffer, with per-page status (593)
3546 2003-03-11 Phil Schwan <phil@clusterfs.com>
3549 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3550 - failed open()s could cause deadlock; fixed (867, 869)
3551 - stop cancelling OST locks when files are closed (481)
3552 - overlapping XID spaces caused network corruption (851, 853)
3553 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3554 - setattr_raw added, to do single-RPC, server-side setattrs
3555 - lmc/lconf syntax change for OST UUIDs
3556 - fix crashy race condition between ptlrpc_free_req and osc_close
3557 - don't use request in mdc_enqueue if we hit a timeout (889)
3558 - don't set the inode i_size for regular files from the MDS (896)
3559 - handle out of order completion AST (842)
3560 - don't LBUG if a lock request times out after receiving AST (913)
3561 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3562 - if a bad lock AST arrives, send an error instead of dropping entirely
3563 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3564 - fix leak in bulk IO when only partially completed (899, 900, 926)
3565 - fix O_DIRECT for ia64 (55)
3566 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3567 - C-z after timeout could hang a process forever; fixed (977)
3569 - client-side I/O cache (678, 924, 929, 941, 970)
3571 - READPAGE and SETATTRs which don't take server-side locks get
3574 2003-02-11 Phil Schwan <phil@clusterfs.com>
3577 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3578 - Fix fsfilt_extN_readpage() to read a full page of directory
3579 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3580 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3581 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3582 - Most intent operations were removed, in favour of a new RPC mode
3583 that does a single RPC to the server and bypasses most of the VFS
3584 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3585 - Aggressively cancel local locks on DLM servers
3586 - mds_reint_unlink sends EA to the client if it's the last nlink.
3587 client uses that EA to unlink OST objects.
3588 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3589 - recursive symlinks were fixed (439)
3590 - fixed NULL deref in DEBUG_REQ
3591 - filter_update_lastobjid no longer calls sync, which annoyed extN
3592 - fixed multi-client small-writes to a single file problem (445)
3593 - fixed mtime updates during file writes (607)
3594 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3595 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3596 - fixed Linux OST size reporting problem (444, 656)
3597 - OST now updates object mtime with writes or setattr (607, 619)
3598 - client verifies file size before zeroing page past EOF (445)
3599 - OST now writes last allocated objid to disk with allocation (108)
3600 - LOV on echo now works (409)
3602 - mds_reint_unlink sends a new buffer, with the EA included. this
3603 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3604 set if a regular file was being unlinked, and it was the last link
3605 - use PtlGet from the target for bulk writes (315)
3606 - OST now updates object mtime with writes or setattr (607, 619)
3607 - LDLM now has a grant-time callback to revalidate locked items, if
3609 - Many MDS operations were reorganized to combat race conditions
3611 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3612 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3614 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3617 - Fully reactivate OST imports after reconnection (512, others)
3618 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3619 - More graceful error handling for truncating on dead OST (515)
3620 - Don't error out unless we're actually accessing dead stripes (474)
3621 - Fix garbage sizes when stripes are missing (410)
3622 - LRU counters were broken, causing constant lock purge (433, 432)
3623 - garbage on read from stripes with failed OSTs (441)
3624 - mark OSCs as active before reconnecting during recovery (438)
3625 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3626 - lfind did not preserve OST order in output (443)
3627 - symlinks cause hung clients, incorrect data (439)
3628 - stop dereferencing request after dropping refcount (457)
3629 - don't LASSERT(spin_is_locked) on non-SMP (455)
3630 - fixes for many rename() bugs
3631 - fstat didn't correctly synchronize attributes (399)
3632 - server must handle lock cancellation during blocking AST prep (487)
3633 - bulk descriptors were free()d too soon (511)
3634 - fix paths in lconf, which would load incorrect modules (451, 507)
3635 - fix confusing lconf 'host not found' error message (386)
3636 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3637 - fix race condition in mdc_blocking_ast() for inode access (526)
3638 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3639 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3640 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3641 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3642 - fix lstripe in several ways (kernel side) (527)
3643 - fix request leak in ldlm_cli_enqueue (262)
3644 - incorrect OSC was marked inactive after OST failure
3645 - call mds_fs_cleanup before unmounting filesystem (524)
3646 - fix races between taking ns_lock and ldlm_lock_change_resource
3647 - fix races updating LOV export open file list
3648 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3649 - fix recovery NULL deref in ldlm_cli_cancel_unused
3650 - fix some DLM races by using new hash table for lock handles (419)
3651 - permit the client to specify desired inodes, at replay
3652 - duplicate requests when we queue them for replay reintegration
3653 - fix last_rcvd offset calculation
3654 - sync after each recovered transaction, so we always make progress
3655 - never, not always, ERESTART requests without transnos
3656 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3658 - skip replay if the MDS says that the client is already connected
3659 - don't check for a recovery-enabled export to match lctl's UUID
3660 - don't INC_USE_COUNT for phantom exports
3661 - don't crash when cleaning up phantom exports (567)
3662 - don't double-finish or set replay data for errored mdc_open requests
3663 - abort requests when they time out, so we don't get old replies
3664 - send/receive replies for AST messages again
3665 - if the client says that it doesn't have the lock, cancel it on the
3667 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3668 mark it as destroyed, it will all work out when decref is called
3669 - fix module use counts (22, 581)
3671 - ASTs now expect a reply (server cancels lock on error reply)
3673 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3676 - fix many simultaneous client startup (392)
3677 - fix dentry->d_it clobbering
3678 - credentials weren't being shipped for readdir/getattr operations
3679 - remove invalid assertions triggered during some concurrent MD
3681 - proper Lustre versions added (336, 389)
3682 - fix memory leak for create error case (398)
3683 - fix LOV locking bug that would get cli/srv out of sync
3684 - fix echo client over LOV (409)
3685 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3686 - fix double-O_EXCL intent crash (424)
3687 - avoid sending multiple lock CANCELs (352)
3689 - MDS can do multi-client recovery (modulo bugs in new code)
3691 - many updates, edits, cleanups
3693 2002-11-18 Phil Schwan <phil@clusterfs.com>
3696 - fix null d_it dereference (346)
3697 - fix full OST/dbench hang (333)
3698 - fix permission problem with file removal (286)
3699 - fix removal of OSCs from LOV when they fail
3700 - fix NULL deref during bulk timeout (214)
3701 - fix problems related to multiple filesystems on one MDS (241)
3702 - fixed serious subtle metadata locking bugs
3703 - free locks on clients when inodes are removed due to memory
3705 - fix inode pointer in lock data (285)
3706 - partial support for multiple MDS on a single host (241)
3707 - data locks weren't cancelled at clear_inode time (290, 311)
3708 - intent locks could lead to unbounded lock growth (205)
3709 - added a maximum lock count, an LRU list, and a flusher
3710 - fix multiple rename (365)
3711 - properly abstracted the echo client
3712 - OSC locked 1 byte too many; fixed
3713 - rewrote brw callback code:
3714 - fixed recovery bugs related to LOVs (306)
3715 - fixed too-many-pages-in-one-write crash (191)
3716 - fixed (again) crash in sync_io_timeout (214)
3717 - probably fixed callback-related race (385)
3719 - Add capability to MDS protocol
3720 - LDLM cancellations and callbacks on different portals
3722 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3725 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3726 - defer open object destruction to close time (601981, 138)
3727 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3728 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3729 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3730 - don't call obd_setattr in ll_file_release for destroyed objects
3731 * protocol change to lustre_msg: move |version| and add |flags|
3732 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3733 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3734 * added replay of create, unlink, link and rename operations during
3735 MDS failover; recovery should be much more robust now
3736 * remove failed OSCs from LOVs (only lov_create uses this so far)
3737 * the lustre-HOWTO was brought (more) up to date (582544)
3739 2002-10-23 Phil Schwan <phil@clusterfs.com>
3742 - in-use dentries weren't being reused properly (617851)
3743 - prevent multiple LDLM setup (599178)
3744 - fix LOV size calculations for truncate (617853)
3745 - fix client handling of MDS intent errors (POSIX)
3746 - fix permission bug in lovstripe.c test (624321)
3747 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3748 - truncate past end of file could corrupt data
3749 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3750 - a race in recovery could return ETIMEDOUT to apps (623947)
3751 - building outside the source directory was fixed
3752 * the lustre-HOWTO was brought (more) up to date (582544)
3753 * major progress was made on recovery functionality
3755 2002-10-10 Phil Schwan <phil@clusterfs.com>
3758 - recovery deadlock fix
3759 - rm -rf causes LBUG fix (617817)
3760 - file open by multiple tasks fix (618962)
3761 - directory permissions bugs (602707 and 620007)
3762 - journal_stop fixed with locking (611313)
3763 - O_APPEND failures resolved (618273, perhaps 614459)
3764 - lconf PATH fix (619770)
3765 - IA64 build fix (621450)
3766 - RPC buffer sizes scale with amount of memory
3768 2002-10-01 Phil Schwan <phil@clusterfs.com>
3771 - locks would be cancelled without throwing away data pages,
3772 resulting in inconsistent data (605627)
3773 - inode attributes were not always being refreshed (605627, 612449)
3774 - lconf now continues to cleanup after lctl reports an error
3775 - MDS now enforces user permissions (602707)
3776 - lprocfs cleanup fixed, but not yet enabled (614157)
3777 - fixed infinite server hang, should a client not respond to an AST
3778 - avoid going into recovery if user calls readlink() with a buffer
3779 that's too small (613941)
3780 - AST RPCs no longer require replies (614867) -- this may be changed
3781 - don't crash server if client sends an IOV that's too big (611336)
3782 - fixed lock conversion deadlock (611892)
3783 - fixed the following of symlinks (614622)
3784 * recovery: the server can remove locks from a client that dies, other
3785 clients can make progress
3786 * more extN patch fixes
3787 * compile-time configurable ptlrpc buffer allocations
3789 - collaborative read cache document
3790 - Lustre Lite Performance CDR document-in-progress
3792 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3795 - fix typo in patch-2.4.18
3797 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3800 - clear ptlrpc request each time in handle_incoming_request()
3801 - unlink of files now destroys the object on the OST
3803 2002-09-19 Peter Braam <braam@clusterfs.com>
3805 * add hard link support
3806 * change obdfile creation method
3807 * kernel patch changed
3809 2002-09-19 Peter Braam <braam@clusterfs.com>
3812 - stack overflow bug in extN fixed
3814 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3816 * documentation updates
3817 - add man pages for config tools
3818 - update tests/README to describe testing with new config tools
3819 - finish metadata API descriptions
3820 * bug fixes and cleanups
3821 - statfs workaround for 16TB limit
3822 - LOV stripe allocation improved, can stripe on subset of OSTs
3823 - LOV file size/IO offset was wrong for files > 4GB in size
3824 - object EA data was being dropped, caused files to be unreadable
3825 - memory overflow with non-LOV OST caused memory corruption
3826 - fixed regression tests to work with new config tools, obdfilter
3827 - fixed bug when directory size became larger than 1 block
3828 - fixed bug (for single client case) when PWD was deleted
3829 - invalidate local directory pages when doing intent-based ops
3830 - avoid LDLM oops when lock callback contained bad data
3832 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3834 * documentation updates
3835 * bug fixes and cleanups
3836 - configuration tools
3839 - 64-bit compile warnings
3840 - 64-bit internal statfs data
3842 * test_brw on persistent OST devices
3844 * lprocfs (disabled)
3846 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3848 * documentation updates
3849 * bug fixes and cleanups
3850 * configuration tools
3852 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3855 * many small fixes to 0_5_4
3856 * io/network handling
3857 * thinkos in MDS operations
3859 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3862 * crucial basic fixes to 0.5.3
3863 * IOR, Iozone work over Elan
3866 2002-08-07 Phil Schwan <phil@clusterfs.com>
3867 * version 0_5_3, our first alpha
3868 * we use the new Portals iovs
3869 * documentation updates
3870 * bug fixes and cleanups
3871 * small changes in the DLM wire protocol
3873 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3874 * version 0_5_1 with some initial stability,
3875 * locking on MD and file I/O.
3876 * documentation updates
3877 * several bug fixes since 0.5.0
3878 * small changes in wire protocol
3880 2002-07-18 Phil Schwan <phil@clusterfs.com>
3882 * delivered as Lustre Light Alpha
3883 * fixed a crash after handling invalid MDS requests
3884 * fixed directory pages for architectures with non-4k pages sizes
3886 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3887 * release version v0_4_4
3888 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3890 * Fixes a number of interruption problems with OST operations.
3891 * Update documentation for portals header changes
3892 * Move all wire protocol structs/defines to lustre_idl.h
3893 * Fixes symlink length bug.
3894 * Add tcpdump to repository.
3896 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3897 * release version v0_4_3
3898 * Fixes statfs for inodes on extN.
3899 * Fixes bug in runtests which would delete /etc/hosts.
3900 * Use 64-bit object IDs wherever possible (not into VFS though)
3901 Remove ost_get_info, which is unused by lustre, and out of date.
3903 2002-07-03 Peter Braam <braam@clusterfs.com>
3904 * release version v0_4_2 Fixes a lookup error (type not passed)
3905 * move forward to head of Portals
3906 * move forward to latest Lustre kernel
3908 2002-06-25 Peter Braam <braam@clusterfs.com>
3909 * release version v0_4_1. Hopefully stable on single node use.