1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
6 2.6.16.46-0.14 (SLES 10),
7 2.6.18-8.1.8.EL5 (RHEL 5).
8 2.6.18.8 vanilla (kernel.org)
9 * Client support for unpatched kernels:
10 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
11 2.6.9-42.0.10.EL (RHEL 4),
12 2.6.16 - 2.6.21 vanilla (kernel.org)
13 * Recommended e2fsprogs version: 1.40.2-cfs1
14 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
18 Description: don't use GFP_* in generic Lustre code.
19 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
23 Description: obdclass is limited by single OBD_ALLOC(idarray)
24 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
28 Description: warnings with build patchless client with vanila 2.6.19 and up
29 Details : change old ctl_table style and replace ctl_table/ctl_table_header
30 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
34 Frequency : Only for SLES
35 Description: sanity_quota fail test_1
36 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
37 file, which makes TSTID[2] inunique.
41 Frequency : Only for SLES9
42 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
43 Details : Update target/ChangeLog/which_patch .
47 Frequency : Only for test_10 in sanity-quota.sh
48 Description: a bug in quota test code
49 Details : checking a test flag in a wrong place
53 Frequency : Only for test_13 in sanity-quota.sh
54 Description: a bug in quota test script
55 Details : shouldn't check overfull allocated quota
59 Description: jbd statistics
60 Details : Port older jbd statistics patch for sles10
62 Severity : enhancement
64 Description: Add printing inode info into message about error in writepage.
69 Description: reply_lock_interpret crash due to race with it and lock cancel.
70 Details : Do not replay locks that are being cancelled. Do not reference
71 locks by their address during replay, just by their handle.
75 Frequency : When flocks are used.
76 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
77 Details : Flock locks might destroy just granted lock if it could be merged
78 with another existing flock, this is done in completion handler,
79 so teach ldlm_cli_enquque_fini that this is a valid case for
85 Description: Oops in read and write path when failing to allocate lock.
86 Details : Check if lock allocation failed and return error back.
90 Description: lstripe command fails for valid OST index
91 Details : The stripe offset is compared to lov->desc.ld_tgt_count
92 instead of lov->desc.ld_active_tgt_count.
96 Description: sanity.sh failed test 103
97 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
102 Description: df doesn't work properly if diskfs blocksize != 4K
103 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
107 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
108 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
112 Description: obd_health_check_timeout too short
113 Details : set obd_health_check_timeout as 1.5x of obd_timeout
117 Description: llapi_file_create() does not allow some changes
118 Details : add llapi_file_open() that allows specifying the mode and
119 open flags, and also returns an open file handle.
123 Description: merge and cleanup kernel patches.
124 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
128 Description: Add journal checksum support.(Kernel part)
129 Details : The journal checksum feature adds two new flags i.e
130 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
131 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
132 indicates that the commit block contains the checksum for
133 the blocks described by the descriptor blocks. Now commit
134 record can be sent to disk without waiting for descriptor
135 blocks to be written to disk. This behavior is controlled
136 using JBD2_FEATURE_ASYNC_COMMIT flag.
140 Description: OSS needs mutliple precreate threads
141 Details : Add ability to start more than one create thread per OSS.
145 Description: Sanity.sh test_65k failed: setstripe should have succeeded
146 Details : execute lfs setstripe on client
150 Description: mds_obd_create error creating tmp object
151 Details : When the user sets quota on root, llog will be affected and can't
152 create files and write files.
155 Frequency : Always on ia64 patchless client, and possibly others.
157 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
158 Details : This allows the patchless client to be loaded on architectures
163 Description: RedHat Update kernel for RHEL5
164 Details : Add the kernel config file for RHEL5.
168 Description: RedHat Update kernel for RHEL5
169 Details : Modify the kernel config file more closer RHEL5.
173 Description: Build failure against Centos5 (RHEL5)
174 Details : Define PAGE_SIZE when it isn't present.
178 Description: test_7 on sanity-quota.sh can't be used on separate machines
179 Details : add facet to handle it
183 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
184 Details : with NFS, the anon dentry's parent was set to itself in
185 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
186 corresponding dentry other than use rec->ur_name.
188 Severity : enhancement
190 Description: lfs setstripe enhancement
191 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
195 Description: testing performance impact of enabling checksumming
196 Details : enable checksum by default, allow --disable-checksum
197 configure option and "-o nochecksum" mount option
199 --------------------------------------------------------------------------------
201 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
203 * Support for kernels:
204 2.6.5-7.283 (SLES 9),
205 2.6.9-55.EL (RHEL 4),
206 2.6.16.46-0.14 (SLES 10),
207 2.6.18.8 vanilla (kernel.org)
208 * Client support for unpatched kernels:
209 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
210 2.6.16 - 2.6.22 vanilla (kernel.org)
211 * Due to recently discovered recovery problems, we do not recommend
212 using patchless RHEL 4 clients with this or any earlier release.
213 * Recommended e2fsprogs version: 1.39.cfs8
214 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
215 * Starting with this release, the ldiskfs backing filesystem required
216 by Lustre is now in its own package, lustre-ldiskfs. This package
217 should be installed. It is versioned separately from Lustre and
218 may be released separately in future.
222 Description: block reactivating mgc import until all deactivates complete
223 Details : Fix race when failing back MDT/MGS to itself (testing)
225 Severity : enhancement
227 Description: add optional extra BUILD_VERSION info
228 Details : add a new environment variable (namely LUSTRE_VERS) which allows
229 to override the lustre version.
232 Frequency : 2.6.18 servers only
234 Description: ll_kern_mount() doesn't release the module reference
235 Details : The ldiskfs module reference count never drops down to 0
236 because ll_kern_mount() doesn't release the module reference.
241 Description: server LBUG when using old ost_num_threads parameter
242 Details : Accept the old ost_num_threads parameter but warn that it
243 is deprecated, and fix an off-by-one error that caused an LBUG.
248 Description: Transient SCSI error results in persistent IO issue
249 Details : iobuf->dr_error is not reinitialized to 0 between two
253 Frequency : sometimes when underlying device returns I/O errors
255 Description: OSTs not going read-only during write failures
256 Details : OSTs are not remounted read-only when the journal commit threads
257 get I/O errors because fsfilt_ext3 calls journal_start/stop()
258 instead of the ext3 wrappers.
262 Description: poor connect scaling with increasing client count
263 Details : Don't run filter_grant_sanity_check for more than 100 exports
264 to improve scaling for large numbers of clients.
267 Frequency : SLES10 only
269 Description: sanity-quota.sh quotacheck failed: rc = -22
270 Details : Quotas cannot be enabled on SLES10.
273 Frequency : liblustre clients only
275 Description: getdirentries does not give error when run on compute nodes
276 Details : getdirentries does not fail when the size specified as an argument
277 is too small to contain at least one entry
279 Severity : enhancement
281 Description: Add LNET router traceability for debug purposes
282 Details : If a checksum failure occurs with a router as part of the
283 IO path, the NID of the last router that forwarded the bulk data
284 is printed so it can be identified.
289 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
290 Details : Due to a race condition, liblustre clients were occasionally
293 Severity : enhancement
295 Description: lfs setstripe use optional parameters instead of postional
298 Severity : enhancement
300 Description: Nanosecond timestamp support for ldiskfs
301 Details : The on-disk ldiskfs filesystem has added support for nanosecond
302 resolution timestamps. There is not yet support for this at
303 the Lustre filesystem level.
306 Frequency : during server recovery
308 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
309 Details : request with rq_no_resend flag not awake l_wait_event if they get a
313 Frequency : nfs export on patchless client
315 Description: connectathon hang when test nfs export over patchless client
316 Details : Disconnected dentry cannot be found with lookup, so we do not need
317 to unhash it or make it invalid
320 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
321 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
322 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
327 Description: open req refcounting wrong on reconnect
328 Details : If reconnect happened between getting open reply from server and
329 call to mdc_set_replay_data in ll_file_open, we will schedule
330 replay for unreferenced request that we are about to free.
331 Subsequent close will crash in variety of ways.
332 Check that request is still eligible for replay in
333 mdc_set_replay_data().
338 Description: disable writes to filesystem when reading health_check file
339 Details : the default for reading the health_check proc file has changed
340 to NOT do a journal transaction and write to disk, because this
341 can cause reads of the /proc file to hang and block HA state
342 checking on a healthy but otherwise heavily loaded system. It
343 is possible to return to the previous behaviour during configure
344 with --enable-health-write.
346 Severity : enhancement
348 Description: 64-bit inode version
349 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
350 to the inode. This will be required for version-based recovery.
355 Description: MDS fails to start if a duplicate client export is detected
356 Details : in some rare cases it was possible for a client to connect to
357 an MDS multiple times. Upon recovery the MDS would detect this
358 and fail during startup. Handle this more gracefully.
360 Severity : enhancement
362 Description: Add -o localflock option to simulate old noflock
364 Details : This will achieve local-only flock/fcntl locks
370 Description: log_commit_thread vs filter_destroy race leads to crash
371 Details : Take import reference before releasing llog record semaphore
376 Description: Wrong request locking in request set processing
377 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
378 imp_delayed_list, in this place should be used imp_lock.
381 Frequency : when reconnection
383 Description: Grant Leak when osc reconnect to OST
384 Details : When osc reconnect ost, OST(filter) should check whether it
385 should grant more space to client by comparing fed_grant and
386 cl_avail_grant, and return the granted space to client instead
387 of "new granted" space, because client will call osc_init_grant
388 to update the client grant space info.
391 Frequency : when client reconnect to OST
393 Description: Grant Leak when osc do resend and replay bulk write
394 Details : When osc reconnect to OST, OST(filter)should clear grant info of
395 bulk write request, because the grant info while be sync between
396 OSC and OST when reconnect, and we should ignore the grant info
397 these of resend/replay write req.
402 Description: Grant space more than avaiable left space sometimes.
403 Details : When then OST is about to be full, if two bulk writing from
404 different clients came to OST. Accord the avaliable space of the
405 OST, the first req should be permitted, and the second one
406 should be denied by ENOSPC. But if the seconde arrived before
407 the first one is commited. The OST might wrongly permit second
408 writing, which will cause grant space > avaiable space.
411 Frequency : when client is evicted
413 Description: Grant might be wrongly erased when osc is evicted by OST
414 Details : when the import is evicted by server, it will fork another
415 thread ptlrpc_invalidate_import_thread to invalidate the
416 import, where the grant will be set to 0. While the original
417 thread will update the grant it got when connecting. So if
418 the former happened latter, the grant will be wrongly errased
419 because of this race.
424 Description: Checking Stale with correct fid
425 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
426 to check whether it is stale, because sometimes, we want the
427 enqueue happened anyway, and op_data.fid2 will not be initialized.
429 Severity : enhancement
431 Description: update patchless client
432 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
435 Frequency : only with 2.4 kernel
437 Description: random memory corruption
438 Details : size of struct ll_inode_info is to big for union inode.u and this
439 can be cause of random memory corruption.
444 Description: Memory leak in recovery
445 Details : Lov_mds_md was not free in an error handler in mds_create_object.
446 It should also check obd_fail before fsfilt_start, otherwise if
447 fsfilt_start return -EROFS,(failover mds during mds recovery).
448 then the req will return with repmsg->transno = 0 and rc = EROFS.
449 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
450 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
451 be freed no matter whether fsfilt_commit success or not.
454 Frequency : only with huge count clients
456 Description: Prevents from taking the superblock lock in llap_from_page for
458 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
459 for llap_from_page call in ll_removepage prevents from taking the
460 superblock lock for a soon died page.
465 Description: Not check open intent error before release open handle
466 Details : in some rare cases, the open intent error is not checked before
467 release open handle, which may cause
468 ASSERTION(open_req->rq_transno != 0), because it tries to release
469 the failed open handle.
474 Description: Set cat log bitmap only after create log success.
475 Details : in some rare cases, the cat log bitmap is set too early. and it
476 should be set only after create log success.
480 Description: Accessing a block bevice can re-enable I/O when Lustre is
481 tearing down a device.
482 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
487 Description: service threads may hog cpus when there are a lot of requests
489 Details : Insert cond_resched to give other threads a chance to use some of
495 Description: the cat log was not initialized in recovery
496 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
497 unlink log on mds will not be initialized until mds post
498 recovery. And also in mds post recovery, the unlink log will
499 initialization will be done asynchronausly, so there will be race
500 between add unlink log and unlink log initialization.
504 Description: brw_stats were being printed incorrectly
505 Details : brw_stats were being printed as log2 but all of them were not
506 recorded as log2. Also remove some code duplication arising from
507 filter_tally_{read,write}.
511 Frequency : rare, only in recovery.
512 Description: ASSERTION(req->rq_type != LI_POISON) failed
513 Details : imp_lock should be held while iterating over imp_sending_list for
514 prevent destroy request after get timeout in ptlrpc_queue_wait.
518 Description: replay-single.sh test 52 fails
519 Details : A lock's skiplist need to be cleanup when it being unlinked
520 from its resource list.
524 Description: Short directio read returns full requested size rather than
526 Details : Direct I/O operations should return actual amount of bytes
527 transferred rather than requested size.
529 Severity : enhancement
531 Description: metadata RPC reduction (e.g. for rm performance)
532 Details : decrease the amount of synchronous RPC between clients and servers
533 by canceling conflicing lock before the operation on the client side
534 and packing thier handles into the main operation RPC to server.
536 Severity : enhancement
538 Description: Async OSC create to avoid the blocking unnecessarily.
539 Details : If a OST has no remain object, system will block on the creating
540 when need to create a new object on this OST. Now, ways use
541 pre-created objects when available, instead of blocking on an
542 empty osc while others are not empty. If we must block, we block
543 for the shortest possible period of time.
545 --------------------------------------------------------------------------------
547 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
552 Frequency : on some architectures
554 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
555 Details : Uninitialized flags sometimes cause configuration commands to
559 Frequency : patchless clients only
561 Description: missing __iget() symbol export
562 Details : The __iget() symbol export is missing. To avoid the need for
563 this on patchless clients the deathrow inode reaper is turned
564 off, and we depend on the VM to clean up old inodes. This
565 dependency was during via the fix for bug 12181.
567 --------------------------------------------------------------------------------
569 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
571 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
572 INTEROPERATE with older versions automatically. In many cases a
573 special upgrade step is needed. Please read the
574 user documentation before upgrading any part of a live system.
575 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
576 version will not interoperate with 1.6 betas before beta5 (1.5.95).
577 * WARNING: Lustre configuration and startup changes are required with
578 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
584 Severity : enhancement
586 Description: MountConf
587 Details : Lustre configuration is now managed via mkfs and mount
588 commands instead of lmc and lconf. New obd types (MGS, MGC)
589 are added for dynamic configuration management. See
590 https://mail.clusterfs.com/wikis/lustre/MountConf for
593 Severity : enhancement
595 Description: dynamic OST addition
596 Details : OSTs can now be added to a live filesystem
598 Severity : enhancement
600 Description: startup order invariance
601 Details : MDTs and OSTs can be started in any order. Clients only
602 require the MDT to complete startup.
604 Severity : enhancement
606 Description: parallel, asynchronous orphan cleanup
607 Details : orphan cleanup is now performed in separate threads for each
608 OST, allowing parallel non-blocking operation.
610 Severity : enhancement
612 Description: optimized stripe assignment
613 Details : stripe assignments are now made based on ost space available,
614 ost previous usage, and OSS previous usage, in order to try
615 to optimize storage space and networking resources.
617 Severity : enhancement
619 Description: Permanently set tunables
620 Details : All writable /proc/fs/lustre tunables can now be permanently
621 set on a per-server basis, at mkfs time or on a live
624 Severity : enhancement
626 Description: Lustre message v2
627 Details : Add lustre message format v2.
629 Severity : enhancement
631 Description: client OST exclusion list
632 Details : Clients can be started with a list of OSTs that should be
633 declared "inactive" for known non-responsive OSTs.
637 Description: SPEC SFS validation failure on NFS v2 over lustre.
638 Details : Changes the blocksize for regular files to be 2x RPC size,
639 and not depend on stripe size.
641 Severity : enhancement
643 Description: Multiple MD RPCs in flight.
644 Details : Further unserialise some read-only MDS RPCs - learn about intents.
645 To avoid overly-overloading MDS, introduce a limit on number of
646 MDS RPCs in flight for a single client and add /proc controls
647 to adjust this limit.
649 Severity : enhancement
651 Description: client read/write statistics
652 Details : Add client read/write call usage stats for performance
653 analysis of user processes.
654 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
655 file access. extents_stats shows chunk size distribution.
656 extents_stats_per_process show chunk size distribution per
659 Severity : enhancement
661 Description: mds statistics
662 Details : Add detailed mds operations statistics in
663 /proc/fs/lustre/mds/*/stats.
667 Description: Failure of copying files with lustre special EAs.
668 Details : Client side always return success for setxattr call for lustre
669 special xattr (currently only "trusted.lov").
674 Description: Refcount LNET uuids
675 Details : The global LNET uuid list grew linearly with every startup;
676 refcount repeated list entries instead of always adding to
679 Severity : enhancement
681 Description: Dynamic service threads
682 Details : Within a small range, start extra service threads
683 automatically when the request queue builds up.
686 Frequency : mixed-endian client/server environments
688 Description: mixed-endian crashes
689 Details : The new msg_v2 system had some failures in mixed-endian
692 Severity : enhancement
694 Description: Easy OST removal
695 Details : OSTs can be permanently deactivated with e.g. 'lctl
696 conf_param lustre-OST0001.osc.active=0'
698 Severity : enhancement
700 Description: MGS proc entries
701 Details : Added basic proc entries for the MGS showing what filesystems
704 Severity : enhancement
706 Description: provide MGS failover
707 Details : Added config lock reacquisition after MGS server failover.
709 Severity : enhancement
711 Description: add Linux 2.4 support
712 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
716 Description: a large application tries to do I/O to the same resource and dies
718 Details : Check the req->rq_arrival time after the call to
719 ost_brw_lock_get(), but before we do anything about
720 processing it & sending the BULK transfer request. This
721 should help move old stale pending locks off the queue as
722 quickly as obd_timeout.
725 Frequency : when an incorrect nid is specified during startup
727 Description: ptlrpc connect to non-existant node causes kernel crash
728 Details : LNET can't be re-entered from an event callback, which
729 happened when we expire a message after the export has been
730 cleaned up. Instead, hand the zombie cleanup off to another
733 Severity : enhancement
735 Description: plain/inodebits lock performance improvement
736 Details : Grouping plain/inodebits in granted list by their request modes
737 and bits policy, thus improving the performance of search through
741 Frequency : only if OST filesystem is corrupted
743 Description: client incorrectly hits assertion in ptlrpc_replay_req()
744 Details : for a short time RPCs with bulk IO are in the replay list,
745 but replay of bulk IOs is unimplemented. If the OST filesystem
746 is corrupted due to disk cache incoherency and then replay is
747 started it is possible to trip an assertion. Avoid putting
748 committed RPCs into the replay list at all to avoid this issue.
751 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
753 Description: System hang on startup
754 Details : This bug allowed the liblustre (e.g. catamount) client to
755 return to the app before handling all startup RPCs. This
756 could leave the node unresponsive to lustre network traffic
757 and manifested as a server ptllnd timeout.
759 Severity : enhancement
761 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
762 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
763 boolean to control whether to print peer debug info when a
764 client's RPC times out.
767 Frequency : only for kernels with patches from Lustre below 1.4.3
769 Description: Remove old rdonly API
770 Details : Remove old rdonly API which unused from at least lustre 1.4.3
773 Frequency : only for devices with external journals
775 Description: Set external device read-only also
776 Details : During a commanded failover stop, we set the disk device
777 read-only while the server shuts down. We now also set any
778 external journal device read-only at the same time.
781 Frequency : when upgrading from 1.4 while trying to change parameters
783 Description: The wrong (new) MDC name was used when setting parameters for
784 upgraded MDT's. Also allows changing of OSC (and MDC)
785 parameters if --writeconf is specified at tunefs upgrade time.
788 Frequency : when setting specific ost indicies
790 Description: QOS code breaks on skipped indicies
791 Details : Add checks for missing OST indicies in the QOS code, so OSTs
792 created with --index need not be sequential.
794 Severity : enhancement
796 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
797 Details : The uninit_groups feature works in conjunction with the kernel
798 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
799 up the pass1 processing of e2fsck. This is a read-only feature
800 in ldiskfs2 only, so older kernels and current ldiskfs cannot
801 mount filesystems that have had this feature enabled.
803 Severity : enhancement
805 Description: Improve multi-block allocation algorithm to avoid fragmentation
806 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
807 allocation locality and avoid filesystem fragmentation.
809 ------------------------------------------------------------------------------
811 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
813 * Support for kernels:
814 2.4.21-47.0.1.EL (RHEL 3)
816 2.6.9-42.0.10.EL (RHEL 4)
817 2.6.12.6 vanilla (kernel.org)
818 2.6.16.27-0.9 (SLES 10)
819 * Recommended e2fsprogs version: 1.39.cfs5
821 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
825 Frequency : occasional, depends on client load and configuration
826 Bugzilla : 12181, 12203
827 Description: data loss for recently-modified files
829 Details : In some cases it is possible that recently written or created
830 files may not be written to disk in a timely manner (this should
831 normally be within 30s unless client IO load is very high).
832 The problem appears as zero-length files or files that are a
833 multiple of 1MB in size after a client crash or client eviction
834 that are missing data at the end of the file.
836 This problem is more likely to be hit on clients where files are
837 repeatedly created and unlinked in the same directory, clients
838 have a large amount of RAM, have many CPUs, the filesystem has
839 many OSTs, the clients are rebooted frequently, and/or the files
840 are not accessed by other nodes after being written.
842 The presence of the problem can be detected by looking at
843 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
844 smaller than the second (nr_unused) then dirty files will not
845 be flushed automatically to disk. "sync; sleep 10" should be
846 run several times on the node before unmounting it to update
847 Lustre (this is also safe to run on nodes without this problem).
849 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
850 that can cause this same problem, so customers using that kernel
851 also need to update the kernel in addition to Lustre. In order
852 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
854 It is normal that files written just before a client crash (less
855 than 30s) may not yet have been flushed to disk, even for local
859 Frequency : frequent on thin XT3 nodes
861 Description: UUID collision on thin XT3 Linux nodes
862 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
863 randomly, since we relied on an insufficiently-seeded PRNG.
868 Description: OSS hangs after "All ost request buffers busy"
869 Details : A deadlock between quota and journal operations caused OSS
870 hangs after printing "All ost request buffers busy."
873 Frequency : always on liblustre builds
875 Description: Cleanup compiler warnings on liblustre
878 Frequency : always on liblustre builds on XT3
880 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
884 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
885 > 32000 subdirectories
886 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
887 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
888 subdirectory count crosses 32000. This will aid e2fsck to
889 correctly handle more than 32000 subdirectories.
892 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
894 Description: System hang on startup
895 Details : This bug allowed the liblustre (e.g. catamount) client to
896 return to the app before handling all startup RPCs. This
897 could leave the node unresponsive to lustre network traffic
898 and manifested as a server ptllnd timeout.
900 Severity : enhancement
902 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
903 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
904 boolean to control whether to print peer debug info when a
905 client's RPC times out.
910 Description: make O_SYNC working on 2.6 kernels
911 Details : 2.6 kernels use different method for mark pages for write,
912 so need add a code to lustre for O_SYNC work.
917 Description: Failure to close file and release space on NFS
918 Details : Put inode details into lock acquired in ll_intent_file_open.
919 Use mdc_intent_lock in ll_intent_open to properly
920 detect all kind of errors unhandled by mdc_enqueue.
925 Description: proc file read during shutdown sometimes raced obd removal,
927 Details : Add lock to prevent obd access after proc file removal.
930 Frequency : Only for files larger than 4GB on 32-bit clients.
932 Description: improperly doing page alignment of locks
933 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
934 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
939 Description: RPCs being resent when they shouldn't be
940 Details : Some RPCs that should not be resent are being resent. This
941 can cause inconsistencies in the RPC state machine. Do not
942 resend such requests.
945 Frequency : rare, only with NFS export
947 Description: Crash on NFS re-export node
948 Details : under very unusual load conditions an assertion is hit in
949 ll_intent_file_open()
952 Frequency : only if OST filesystem is corrupted
954 Description: client incorrectly hits assertion in ptlrpc_replay_req()
955 Details : for a short time RPCs with bulk IO are in the replay list,
956 but replay of bulk IOs is unimplemented. If the OST filesystem
957 is corrupted due to disk cache incoherency and then replay is
958 started it is possible to trip an assertion. Avoid putting
959 committed RPCs into the replay list at all to avoid this issue.
964 Description: large O_DIRECT requests fail under memory pressure/fragmentation
965 Details : Large single O_DIRECT read and write calls can fail to allocate
966 a sufficiently large buffer to process the request. In case of
967 allocation failure the allocation is retried with a smaller
968 buffer and broken into smaller requests.
970 Severity : enhancement
972 Description: Add -o localflock option to simulate old noflock behaviour.
973 Details : This will achieve local-only flock/fcntl locks coherentness.
978 Description: versioning check is incomplete
979 Details : Checking the version difference of client vs. server, report
980 error if the gap is too big.
985 Description: add support PG_writeback bit
986 Details : add support for PG_writeback bit for Lustre, for more carefull
987 work with page cache in 2.6 kernel. This also fix some deadlocks
988 and remove hack for work O_SYNC with 2.6 kernel.
990 Severity : enhancement
992 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
993 Details : The uninit_groups feature works in conjunction with the kernel
994 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
995 up the pass1 processing of e2fsck. This is a read-only feature
996 in ldiskfs2 only, so older kernels and current ldiskfs cannot
997 mount filesystems that have had this feature enabled.
999 Severity : enhancement
1001 Description: Improve multi-block allocation algorithm to avoid fragmentation
1002 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1003 allocation locality and avoid filesystem fragmentation.
1005 ------------------------------------------------------------------------------
1007 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1009 * Support for kernels:
1010 2.6.9-42.0.3.EL (RHEL 4)
1011 2.6.5-7.276 (SLES 9)
1012 2.4.21-47.0.1.EL (RHEL 3)
1013 2.6.12.6 vanilla (kernel.org)
1014 2.6.16.21-0.8 (SLES10)
1015 * Recommended e2fsprogs version: 1.39.cfs2-0
1017 * The backwards-compatible /proc/sys/portals symlink has been removed
1018 in this release. Before upgrading, please ensure that you change
1019 any configuration scripts or /etc/sysctl.conf files that access
1020 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1021 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1022 in advance of the upgrade on any system running Lustre 1.4.6 or
1023 newer, since /proc/sys/lnet was added in that version.
1024 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1028 Frequency : only when quota is used
1030 Description: avoid scanning export list for quota master
1031 Details : Change the algorithms to avoid scanning export list in order
1032 to improve the efficiency.
1035 Frequency : MDS failover only, very rarely
1037 Description: "went back in time" messages on mds failover
1038 Details : The greatest transno may be lost when the current operation
1039 finishes with an error (transno==0) and the client's last_rcvd
1040 record is over-written. Save the greatest transno in the
1041 mds_last_transno for this case.
1044 Frequency : always for specific kernels and striping counts
1046 Description: client may get "Matching packet too big" without ACL support
1047 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1048 when trying to access files in certain configurations. The
1049 clients should in fact be denied when mounting because they do
1050 not understand ACLs.
1053 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1055 Description: many clients connecting with IO in progress causes connect timeouts
1056 Details : Avoid synchronous journal commits to avoid delays caused by many
1057 clients connecting/disconnecting when bulk IO is in progress.
1058 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1059 OST_IO_PORTAL to avoid delays behind potentially many pending
1063 Frequency : occasionally with multiple writers to a single file
1065 Description: shared writes to file may result in wrong size reported by stat()
1066 Details : Allow growing of kms when extent lock is cancelled
1069 Frequency : always with random mmap IO to multi-striped file
1071 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1072 Details : Only if the hole is at the end of OST object so that kms is too
1073 small. Fix is to increase kms accordingly in ll_nopage.
1076 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1078 Description: writes to a missing object would leak memory on the OST
1079 Details : If there is an inconsistency between the MDS and OST filesystems,
1080 such that the MDS references an object that doesn't exist, writes
1081 to that object will leak memory due to incorrect cleanup in the
1082 error handling path, eventually running out of memory on the OST.
1087 Description: Creating too long symlink causes lustre errors
1088 Details : Check symlink and name lengths before sending requests to MDS.
1091 Frequency : only if flock is enabled (not on by default)
1093 Description: posix locks not released on fd closure on 2.6.9+
1094 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1095 kernels, this caused such locks not to be released on fd close and
1096 then assertions on fs unmount about still used locks.
1099 Frequency : MDS failover only, very rarely
1101 Description: clients may get ASSERTION(granted_lock != NULL)
1102 Details : When request was taking a long time, and a client was resending
1103 a getattr by name lock request. The were multiple lock requests
1104 with the same client lock handle and
1105 mds_getattr_name->fixup_handle_for_resent_request found one of the
1106 lock handles but later failed with ASSERTION(granted_lock != NULL).
1111 Description: handle->h_buffer_credits > 0, assertion failure
1112 Details : h_buffer_credits is zero after truncate, causing assertion
1113 failure. This patch extends the transaction or creates a new
1117 Frequency : NFS re-export or patchless client
1118 Bugzilla : 11179, 10796
1119 Description: Crash on NFS re-export node (__d_move)
1120 Details : We do not want to hash the dentry if we don't have a lock.
1121 But if this dentry is later used in d_move, we'd hit uninitialised
1122 list head d_hash, so we just do this to init d_hash field but
1123 leave dentry unhashed.
1126 Frequency : NFS re-export or patchless client
1128 Description: NFS exports has problem with symbolic link
1129 Details : lustre client didn't properly install dentry when re-exported
1130 to NFS or running patchless client.
1133 Frequency : NFS re-export or patchless client
1135 Description: Various nfs/patchless fixes.
1136 Details : fixes reuse disconected alias for lookup process - this fixes
1137 warning "find_exported_dentry: npd != pd",
1138 fix permission error with open files at nfs.
1139 fix apply umask when do revalidate.
1142 Frequency : occasional
1144 Description: Crash on NFS re-export node
1145 Details : calling clear_page() on the wrong pointer triggered oops in
1146 generic_mapping_read().
1149 Frequency : rarely, using O_DIRECT IO
1151 Description: unaligned directio crashes client with LASSERT
1152 Details : check for unaligned buffers before trying any requests.
1155 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1157 Description: stale data returned from RAID cache
1158 Details : If only a small amount of IO is done to the RAID device before
1159 reading it again it is possible to get stale data from the RAID
1160 cache instead of reading it from disk.
1163 Frequency : always for sles10 kernel
1165 Description: sles10 support
1166 Details : ll_follow_link: compile fixes and using of nd_set_link
1167 under newer kernels.
1170 Frequency : depends on arch, kernel and compiler version, always on sles10
1173 Description: recursive or deep enough symlinks cause stack overflow
1174 Details : getting rid of large stack-allocated variable in
1178 Frequency : depends on hardware
1180 Description: lustre write performance loss in the SLES10 kernel
1181 Details : the performance loss is caused by using of write barriers in the
1182 ext3 code. The SLES10 kernel turns barrier support on by
1183 default. The fix is to undo that change for ldiskfs.
1185 ------------------------------------------------------------------------------
1187 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1189 * Support for kernels:
1190 2.6.9-42.0.3EL (RHEL 4)
1191 2.6.5-7.276 (SLES 9)
1192 2.4.21-47.0.1.EL (RHEL 3)
1193 2.6.12.6 vanilla (kernel.org)
1197 Frequency : quota enabled and large files being deleted
1199 Description: releasing more than 4GB of quota at once hangs OST
1200 Details : If a user deletes more than 4GB of files on a single OST it
1201 will cause the OST to spin in an infinite loop. Release
1202 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1207 Description: statfs data retrieved from /proc may be stale or zero
1208 Details : When reading per-device statfs data from /proc, in the
1209 {kbytes,files}_{total,free,avail} files, it may appear
1210 as zero or be out of date.
1213 Frequency : systems with MD RAID1 external journal devices
1215 Description: lconf's call to blkid is confused by RAID1 journal devices
1216 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1217 of returning all block devices that match the journal UUID.
1220 Frequency : always, for aggregate stripe size over 4GB
1222 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1223 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1224 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1227 Frequency : always if "lfs find" used on a local file/directory
1229 Description: "lfs find" segfaults if used on a local file/directory
1230 Details : The case where a directory component was not specified wasn't
1231 handled correctly. Handle this properly.
1234 Frequency : always on ppc64
1236 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1237 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1238 perform endianness conversion.
1241 Frequency : rarely (truncate to non-zero file size after write under load)
1242 Bugzilla : 10730, 10687
1243 Description: Files padded with zeros to next 4K multiple
1244 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1245 it is possible that files that are truncated to a non-zero size
1246 immediately after being written are filled with zero bytes beyond
1247 the truncated size. No file data is lost.
1249 Severity : enhancement
1251 Description: Allow recovery/failover for liblustre clients.
1252 Details : liblustre clients were unaware of failover configurations until
1255 Severity : enhancement
1257 Description: user file locks should fail when not mounting with flock option
1258 Details : Set up an error-returning stub in ll_file_operations.lock field
1259 to prevent incorrect behaviour when client is mounted without
1260 flock option. Also, set up properly f_op->flock field for
1264 Frequency : always on ia64
1266 Description: "lfs df" loops on printing out MDS statfs information
1267 Details : The obd_ioctl_data was not initialized and in some systems
1268 this caused a failure during the ioctl that did not return
1269 an error. Initialize the struct and return an error on failure.
1272 Frequency : SLES 9 only
1274 Description: Error of copying files with lustre special EAs as root
1275 Details : Client side always return success for setxattr call for lustre
1276 special xattr (currently only "trusted.lov").
1279 Frequency : rarely on clusters with both ia64+i386 clients
1281 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1282 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1283 clients are doing concurrent writes to the same file it is
1284 possible that the ia64 clients may LASSERT because the OST
1285 extent locks are not PAGE_SIZE aligned. Ensure that grown
1286 locks are always aligned on the request boundary.
1289 Frequency : specific use, occasional
1291 Description: Overwriting in use executable truncates on-disk binary image
1292 Details : If one node attempts to overwrite an executable in use by
1293 another node, we now correctly return ETXTBSY instead of
1294 truncating the file.
1296 Severity : enhancement
1298 Description: Async OSC create to avoid the blocking unnecessarily.
1299 Details : If a OST has no remain object, system will block on the creating
1300 when need to create a new object on this OST. Now, ways use
1301 pre-created objects when available, instead of blocking on an
1302 empty osc while others are not empty. If we must block, we block
1303 for the shortest possible period of time.
1308 Description: chmod on Lustre root is propagated to other clients
1309 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1310 invalid by the follow_mount time.
1313 Frequency : liblustre clients only
1315 Description: Race in 'instant cancel' lock handling could lead to such locks
1316 never to be granted in case of SMP MDS
1317 Details : Do not destroy not yet granted but cbpending locks in
1321 Frequency : replay/resend of open
1323 Description: non null lock assetion failure in mds_intent_policy
1324 Details : Trying to replay/resend lockless open requests resulted in
1325 mds_open() returning 0 with no lock. Now it sets a flag if
1326 a lock is going to be returned.
1328 Severity : enhancement
1330 Description: Checksum enhancements
1331 Details : New checksum enhancements allow for resending RPCs that failed
1334 Severity : enhancement
1336 Description: Tunables on number of dirty pages in cacche
1337 Details : Allow to set limit on number of dirty pages cached.
1342 Description: client crash on unmount - lock still has references
1343 Details : In some error handling cases it was possible to leak a lock
1344 reference on a client while accessing a file. This was not
1345 harmful to the client during operation, but would cause the
1346 client to crash when the filesystem is unmounted.
1349 Frequency : specific case, rare
1351 Description: ETXTBSY on mds though file not in use by client
1352 Details : ETXTBSY is no longer incorrectly returned when attempting to
1353 chmod or chown a directory that the user previously tried to
1354 execute or a currently-executing binary.
1357 Frequency : extremely rare except on liblustre-based clients
1359 Description: Lustre space not freed when files are deleted
1360 Details : Clean up open-unlinked files after client eviction. Previously
1361 the unlink was skipped and the files remained as orphans.
1366 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1367 Details : In some cases it was possible to send a blocking callback to
1368 a client doing a glimpse, even though that client didn't get
1369 a lock granted. When the glimpse lock is cancelled on the OST
1370 the freed lock is left on the waiting list and corrupted the list.
1373 Frequency : all core dumps
1375 Description: Broke core dumps to lustre
1376 Details : Negative dentry may be unhashed if parent does not have UPDATE
1377 lock, but some callers, e.g. do_coredump, expect dentry to be
1378 hashed after successful create, hash it in ll_create_it.
1380 ------------------------------------------------------------------------------
1382 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1384 * Support for kernels:
1385 2.6.9-42.0.2.EL (RHEL 4)
1386 2.6.5-7.276 (SLES 9)
1387 2.4.21-40.EL (RHEL 3)
1388 2.6.12.6 vanilla (kernel.org)
1392 Frequency : always on RHEL 3
1394 Description: Number of open files grows over time
1395 Details : The number of open files grows over time, whether or not
1396 Lustre is started. This was due to a filp leak introduced
1397 by one of our kernel patches.
1399 ------------------------------------------------------------------------------
1401 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1403 * Support for kernels:
1404 2.6.9-42.EL (RHEL 4)
1405 2.6.5-7.276 (SLES 9)
1406 2.4.21-40.EL (RHEL 3)
1407 2.6.12.6 vanilla (kernel.org)
1412 Bugzilla : 5719, 9635, 9792, 9684
1413 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1414 Details : If a server is under heavy load and cannot reply to new
1415 connection requests before the client resends the (re)connect,
1416 the connection handling code can behave badly if two service
1417 threads are concurrently handing separate (re)connections from
1418 the same client. Add better locking to the connection handling
1419 code, and ensure that only a single connection will be processed
1420 for a given client UUID, even if the lock is dropped.
1422 Severity : enhancement
1424 Description: add TCP zero-copy support to kernel
1425 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1426 sends if the hardware supports scatter-gather and checksumming.
1427 This allows socklnd to do client-write and server-read more
1428 efficiently and reduce CPU utilization from skbuf copying.
1431 Frequency : only if NFS exporting from client
1433 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1434 Details : If setting attributes on a file created under NFS that had
1435 never been opened it would be possible to oops the client
1436 if the file had no objects.
1439 Frequency : always for liblustre
1441 Description: liblustre client does MDS+OSTs setattr RPC for each write
1442 Details : When doing a write from a liblustre client, the client
1443 incorrectly issued an RPC to the MDS and each OST the file was
1444 striped over in order to update the timestamps. When writing
1445 with small chunks and many clients this could overwhelm the MDS
1446 with RPCs. In all cases it would slow down the write because
1447 these RPCs are unnecessary.
1449 Severity : enhancement
1451 Description: allow number of MDS service threads to be changed at module load
1452 Details : It is now possible to change the number of MDS service threads
1453 running. Adding "options mds mds_num_threads={N}" to the MDS's
1454 /etc/modprobe.conf will set the number of threads for the next
1455 time Lustre is restarted (assuming the "mds" module is also
1456 reloaded at that time). The default number of threads will
1457 stay the same, 32 for most systems.
1462 Description: OST crash if filesystem is unformatted or corrupt
1463 Details : If an OST is started on a device that has never been formatted
1464 or if the filesystem is corrupt and cannot even mount then the
1465 error handling cleanup routines would dereference a NULL pointer.
1470 Description: NULL pointer deref in llap_from_page.
1471 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1472 incorrect) mapping if the page was truncated/reclaimed while it was
1473 searched for. Check for this condition and skip such pages when
1474 doing readahead. Introduce extra check to llap_from_page() to
1475 verify page->mapping->host is non-NULL (so page is not anonymous).
1478 Frequency : Sometimes when using sys_sendfile
1480 Description: "page not covered by a lock" warnings from ll_readpage
1481 Details : sendfile called ll_readpage without right page locks present.
1482 Now we introduced ll_file_sendfile that does necessary locking
1483 around call to generic_file_sendfile() much like we do in
1487 Frequency : with certain MDS communication failures at client mount time
1489 Description: NULL pointer deref after failed client mount
1490 Details : a client connection request may delayed by the network layer
1491 and not be sent until after the PTLRPC layer has timed out the
1492 request. If the client fails the mount immediately it will try
1493 to clean up before the network times out the request. Add a
1494 reference from the request import to the obd device and delay
1495 the cleanup until the network drops the request.
1498 Frequency : occasionally during client (re)connect
1500 Description: assertion failure during client (re)connect
1501 Details : processing a client connection request may be delayed by the
1502 client or server longer than the client connect timeout. This
1503 causes the client to resend the connection request. If the
1504 original connection request is replied in this interval, the
1505 client may trip an assertion failure in ptlrpc_connect_interpret()
1506 which thought it would be the only running connect process.
1509 Frequency : only with obd_echo servers and clients that are rebooted
1511 Description: kernel BUG accessing uninitialized data structure
1512 Details : When running an obd_echo server it did not start the ping_evictor
1513 thread, and when a client was evicted an uninitialized data
1514 structure was accessed. Start the ping_evictor in the RPC
1515 service startup instead of the OBD startup.
1517 Severity : enhancement
1518 Bugzilla : 10193 (patchless)
1519 Description: Remove dependency on various unexported kernel interfaces.
1520 Details : No longer need reparent_to_init, exit_mm, exit_files,
1521 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1524 Frequency : rare (only users of deprecated and unsupported LDAP config)
1526 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1527 Details : LDAP apparently contains 'lustreName' attributes instead of
1528 'name'. A simple remapping of the name is sufficient.
1531 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1533 Description: waiting_locks_callback trips kernel BUG if client is evicted
1534 Details : Running with the dump_on_timeout debug flag turned on makes
1535 it possible that the waiting_locks_callback() can try to dump
1536 the Lustre kernel debug logs from an interrupt handler. Defer
1537 this log dumping to the expired_lock_main() thread.
1539 Severity : enhancement
1541 Description: Support NFS exporting on 2.6 kernels.
1542 Details : Implement non-rawops metadata methods for NFS server to use without
1543 changing NFS server code.
1546 Frequency : very rare (synthetic metadata workload only)
1548 Description: two racing renames might cause an MDS thread to deadlock
1549 Details : Running the "racer" program may cause one MDS thread to rename
1550 a file from being the source of a rename to being the target of
1551 a rename at exactly the same time that another thread is doing
1552 so, and the second thread has already enqueued these locks after
1553 doing a lookup of the target and is trying to relock them in
1554 order. Ensure that we don't try to re-lock the same resource.
1557 Frequency : only very large systems with liblustre clients
1559 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1560 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1561 all OSTs in parallel. This allows the eviction of stale liblustre
1562 clients to proceed much faster than if they were done in series,
1563 and also offers similar improvements for other set_info RPCs.
1568 Description: excessive CPU usage during initial read phase on client
1569 Details : During the initial read phase on a client, it would agressively
1570 retry readahead on the file, consuming too much CPU and impacting
1571 performance (since 1.4.5.8). Improve the readahead algorithm
1572 to avoid this, and also improve some other common cases (read
1573 of small files in particular, where "small" is files smaller than
1574 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1579 Description: MDS crash when receiving packet with unknown intent.
1580 Details : Do not LBUG in unknown intent case, just return -EFAULT
1582 Severity : enhancement
1583 Bugzilla : 9293, 9385
1584 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1585 Details : Do not serialize getattr (non-intent version) and statfs.
1588 Frequency : occasional, when OST network is overloaded/intermittent
1590 Description: client evicted by OST after bulk IO timeout
1591 Details : If a client sends a bulk IO request (read or write) the OST
1592 may evict the client if it is unresposive to its data GET/PUT
1593 request. This is incorrect if the network is overloaded (takes
1594 too long to transfer the RPC data) or dropped the OST GET/PUT
1595 request. There is no need to evict the client at all, since
1596 the pinger and/or lock callbacks will handle this, and the
1597 client can restart the bulk request.
1600 Frequency : Always when mmapping file with no objects
1602 Description: client crashes when mmapping file with no objects
1603 Details : Check that we actually have objects in a file before doing any
1604 operations on objects in ll_vm_open, ll_vm_close and
1610 Description: Request leak when working with deleted CWD
1611 Details : Introduce advanced request refcount tracking for requests
1612 referenced from lustre intent.
1614 Severity : Enhancement
1616 Description: Cache open file handles on client.
1617 Details : MDS now will return special lock along with openhandle, if
1618 requested and client is allowed to hold openhandle, even if unused,
1619 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1620 closing files for every read/write openration.
1622 Severity : Enhancement
1624 Description: Cache open negative dentries on client when possible.
1625 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1626 negative dentries on lock revocation.
1631 Description: Remounting a client read-only wasn't possible with a zconf mount
1632 Details : It wasn't possible to remount a client read-only with llmount.
1634 Severity : enhancement
1635 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1636 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1637 MPICH is included to add Lustre-specific ADIO interfaces.
1638 This is based closely on the UFS ADIO layer and only differs
1639 in file creation, in order to allow the OST striping to be set.
1640 This is user-contributed code and not supported by CFS.
1645 Description: extended inode attributes (immutable, append-only) work improperly
1646 when 2.4 and 2.6 kernels are used on client/server or vice versa
1647 Details : Introduce kernel-independent values for these flags.
1649 Severity : enhancement
1652 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1653 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1654 granularity. Now, Fractional MB granularity is be supported,
1655 this is very useful for low memory system.
1657 Severity : enhancement
1659 Description: Getattr by fid
1660 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1661 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1662 have locks and avoids getattr rpc requests for it on every stat.
1665 Frequency : Always, for filesystems larger than 2TB
1667 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1668 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1669 filesystems larger than 2TB to be mounted with mballoc enabled.
1672 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1674 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1675 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1676 filesystems larger than 2TB is mounted then the kernel will
1677 silently corrupt the start of the filesystem. CONFIG_LBD is
1678 enabled for all CFS-supported kernels, but the possibility of
1679 this happening with a modified kernel config exists.
1681 Severity : enhancement
1683 Description: add client O_DIRECT support for 2.6 kernels
1684 Details : It is now possible to do O_DIRECT reads and writes to files
1685 in the Lustre client mountpoint on 2.6 kernel clients.
1687 Severity : enhancement
1689 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1690 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1691 parallel, not waiting for response from every OST before sending
1692 a rpc to the next OST.
1697 Description: setattr vs write race when updating file timestamps
1698 Details : Client processes that update a file timestamp into the past
1699 right after writing to the file (e.g. tar) it is possible that
1700 the updated file modification time can be reset to the current
1701 time due to a race between processing the setattr and write RPC.
1703 Severity : enhancement
1705 Description: Bring 'lfs find' closer in line with regular Linux find.
1706 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1707 -print0 options and obtains all the needed info through the lustre
1710 Severity : enhancement
1712 Description: support up to 1024 configured devices on one node
1713 Details : change obd_dev array from statically allocated to dynamically
1714 allocated structs as they are first used to reduce memory usage
1719 Description: Flush dirty partially truncated pages during truncate
1720 Details : Immediatelly flush partially truncated pages in filter_setattr,
1721 this way we completely avoid having any pages in page cache on OST
1722 and can retire ugly workarounds during writes to flush such pages.
1727 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1728 Details : protect inode from truncation within vfs_unlink() context
1729 just take a reference before calling vfs_unlink() and release it
1730 when parent's i_sem is free.
1735 Description: last_id value checked outside lock on OST caused LASSERT failure
1736 Details : If there were multiple MDS->OST object precreate requests in
1737 flight, it was possible that the OST's last object id was checked
1738 outside a lock and incorrectly tripped an assertion. Move checks
1739 inside locks, and discard old precreate requests.
1742 Frequency : always, if extents are used on OSTs
1744 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1745 Details : OSTs using the extents format would not zero the high 16 bits of
1746 the index physical block number. This is not a problem for any
1747 OST filesystems smaller than 16TB, and no kernels support ext3
1748 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1749 new/modified files) and can be fixed for existing filesystems
1750 with e2fsprogs-1.39-cfs1.
1755 Description: import connection selection may be incorrect if timer wraps
1756 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1757 import connections to be ignored if the 32-bit jiffies counter
1758 wraps. Use a 64-bit jiffies counter.
1761 Frequency : very large clusters immediately after boot
1763 Description: LNET request buffers exhausted under heavy short-term load
1764 Details : If a large number of client requests are generated on a service
1765 that has previously never seen so many requests it is possible
1766 that the request buffer growth cannot keep up with the spike in
1767 demand. Instead of dropping incoming requests, they are held in
1768 the LND until the RPC service can accept more requests.
1771 Frequency : Sometimes during replay
1773 Description: Assertion failure in ll_local_open after replay.
1774 Details : If replay happened on an open request reply before we were able
1775 to set replay handler, reply will become not swabbed tripping the
1776 assertion in ll_local_open. Now we set the handler right after
1777 recognising of open request
1780 Frequency : very rare
1782 Description: kernel reports "badness in vsnprintf"
1783 Details : Reading from the "recovery_status" /proc file in small chunks
1784 may cause a negative length in lprocfs_obd_rd_recovery_status()
1785 call to vsnprintf() (which is otherwise harmless). Exit early
1786 if there is no more space in the output buffer.
1788 Severity : enhancement
1790 Description: clear OBD RPC statistics by writing to them
1791 Details : It is now possible to clear the OBD RPC statistics by writing
1792 to the "stats" file.
1797 Description: Client mtime is not the same on different clients after utimes
1798 Details : In some cases, the client was using the utimes() syscall on
1799 a file cached on another node. The clients now validate the
1800 ctime from the MDS + OSTs to determine which one is right.
1805 Description: Inability to activate failout mode
1806 Details : lconf script incorrectly assumed that in pythong string's numeric
1807 value is used in comparisons.
1810 Frequency : always with multiple stripes per file
1812 Description: Inefficient object allocation for mutli-stripe files
1813 Details : When selecting which OSTs to stripe files over, for files with
1814 a stripe count that divides evenly into the number of OSTs,
1815 the MDS is always picking the same starting OST for each file.
1816 Return the OST selection heuristic to the original design.
1821 Description: mount failures may take full timeout to return an error
1822 Details : Under some heavy load conditions it is possible that a
1823 failed mount can wait for the full obd_timeout interval,
1824 possibly several minutes, before reporting an error.
1825 Instead return an error as soon as the status is known.
1827 Frequency : quota enabled and large files being deleted
1829 Description: releasing more than 4GB of quota at once hangs OST
1830 Details : If a user deletes more than 4GB of files on a single OST it
1831 will cause the OST to spin in an infinite loop. Release
1832 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1837 Description: statfs data retrieved from /proc may be stale or zero
1838 Details : When reading per-device statfs data from /proc, in the
1839 {kbytes,files}_{total,free,avail} files, it may appear
1840 as zero or be out of date.
1843 Frequency : systems with MD RAID1 external journal devices
1845 Description: lconf's call to blkid is confused by RAID1 journal devices
1846 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1847 of returning all block devices that match the journal UUID.
1850 Frequency : always, for aggregate stripe size over 4GB
1852 Description: assertion fails when trying to use 4GB stripe size
1853 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
1854 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
1857 Frequency : always on ppc64
1859 Description: the first write on an ext3 filesystem with mballoc got stuck
1860 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1861 perform endianness conversion.
1863 ------------------------------------------------------------------------------
1865 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1867 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1868 INTEROPERATE with older versions automatically. Please read the
1869 user documentation before upgrading any part of a live system.
1870 * WARNING: Lustre networking configuration changes are required with
1871 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1874 * Support for kernels:
1875 2.6.9-22.0.2.EL (RHEL 4)
1876 2.6.5-7.244 (SLES 9)
1877 2.6.12.6 vanilla (kernel.org)
1880 Severity : enhancement
1881 Bugzilla : 7981/8208
1882 Description: Introduced Lustre Networking (LNET)
1883 Details : LNET is new networking infrastructure for Lustre, it includes
1884 a reorganized network configuration mode (see the user
1885 documentation for full details) as well as support for routing
1886 between different network fabrics. Lustre Networking Devices
1887 (LNDs) for the supported network fabrics have also been
1888 created for this new infrastructure.
1890 Severity : enhancement
1891 Description: Introduced Access control lists
1892 Details : clients can set ACLs on files and directories in order to have
1893 more fine-grained permissions than the standard Unix UGO+RWX.
1894 The MDS must be started with the "-o acl" mount option.
1896 Severity : enhancement
1897 Description: Introduced filesystem quotas
1898 Details : Administrators may now establish per-user quotas on the
1901 Severity : enhancement
1903 Description: Configuration change for the XT3
1904 The PTLLND is now used to run Lustre over Portals on the XT3
1905 The configure option(s) --with-cray-portals are no longer used.
1906 Rather --with-portals=<path-to-portals-includes> is used to
1907 enable building on the XT3. In addition to enable XT3 specific
1908 features the option --enable-cray-xt3 must be used.
1913 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1914 Details : A race between allocating a new llcd and re-getting the llcd_lock
1915 allowed another thread to grab newly-allocated llcd.
1917 Severity : enhancement
1919 Description: 2.6 OST async journal commit and locking fix to improve performance
1920 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1921 2.6 kernels are now async as they already were in 2.4 kernels so
1922 that they can commit concurrently with the network bulk transfer.
1923 For block-allocated files the filter allocation semaphore is held
1924 to avoid filesystem fragmentation during allocation. BKL lock
1925 removed for 2.6 xattr operations where it is no longer needed.
1930 Description: lconf incorrectly determined whether two IP networks could talk
1931 Details : In some more complicated routing and multiple-network
1932 configurations, lconf will avoid trying to make a network
1933 connection to a disjoint part of the IP space. It was doing the
1934 math incorrectly for one set of cases.
1939 Description: Fix for potential infinite loop processing records in an llog.
1940 Details : If an llog record is corrupted/zeroed, it is possible to loop
1941 forever in llog_process(). Validate the llog record length
1942 and skip the remainder of the block on error.
1945 Frequency : occasional (liblustre only)
1947 Description: liblustre could not open files whose last component is a symlink
1948 Details : sysio_path_walk() would incorrectly pass the open intent to
1949 intermediate path components.
1952 Frequency : rare (liblustre only with non-standard tuning)
1953 Bugzilla : 7201 (7350)
1954 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1955 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1956 after it has been released, only do this for referenced locks
1958 Severity : enhancement
1960 Description: specifying an (invalid) directory default stripe_size of -1
1961 would reset the directory default striping
1962 Details : stripe_size -1 was used internally to signal directory stripe
1963 removal, now use "all default" to signal dir stripe removal
1964 as a directory striping of "all default" is not useful
1967 Frequency : common for large clusters running liblustre clients
1969 Description: doing an ls when liblustre clients are running is slow
1970 Details : sending a glimpse AST to a liblustre client waits for every AST
1971 to time out, as liblustre clients will not respond. Since they
1972 cannot cache data we refresh the OST lock LVB from disk instead.
1974 Severity : enhancement
1976 Description: doing an ls at the same time as file IO can be slow
1977 Details : enqueue and other "small" requests can be blocked behind many
1978 large IO requests. Create a new OST IO portal for non-IO
1979 requests so they can be processed faster.
1982 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1984 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1985 created file with mode 000
1986 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1987 allow an ftruncate() as a non-root user to a file with mode 000.
1988 Lustre now respects this flag to disable mode checks when
1989 truncating a file owned by the user
1992 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1995 Description: Revoking locks from clients that went dead or catatonic might take
1997 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1998 cancellation of such locks instant on servers without waiting for
1999 any reply from clients. Clients drops these locks when cancel
2000 notification from server is received without replying.
2003 Frequency : liblustre-only, when liblustre client dies or becomes busy
2005 Description: Doing ls on Linux clients can take a long time with active
2007 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2008 granting such locks to it in the first place if possible. Locks
2009 are taken by proxy on the OST during the read or write and
2010 dropped immediately afterward. Add connect flags handling, do
2011 not grant locks to liblustre clients for glimpse ASTs.
2013 Severity : enhancement
2015 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2016 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2017 sequential reads and behaves badly with applications doing random
2018 reads. Improve it by reading ahead at least read region, and
2019 avoiding excessive large RPC for small reads.
2021 Severity : enhancement
2023 Description: Creating more than 1000 files for a single job may cause a load
2024 imbalance on the OSTs if there are also a large number of OSTs.
2025 Details : qos_prep_create() uses an OST index reseed value that is an
2026 even multiple of the number of available OSTs so that if the
2027 reseed happens in the middle of the object allocation it will
2028 still utilize the OSTs as uniformly as possible.
2033 Description: OST or MDS may oops in ping_evictor_main()
2034 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2035 but doesn't restart at beginning of obd_exports_timed list
2038 Severity : enhancement
2040 Description: improve by-nid export eviction on the MDS and OST
2041 Details : allow multiple exports with the same NID to be evicted at one
2042 time without re-searching the exports list.
2045 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2047 Description: MDS may oops in groups_free()
2048 Details : in rare race conditions a newly allocated group_info struct is
2049 freed again, and this can be NULL. The 2.4 compatibility code
2050 for groups_free() checked for a NULL pointer, but 2.6 did not.
2053 Frequency : common for liblustre clients doing little filesystem IO
2054 Bugzilla : 9352, 7313
2055 Description: server may evict liblustre clients accessing contended locks
2056 Details : if a client is granted a lock or receives a completion AST
2057 with a blocking AST already set it would not reply to the AST
2058 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2061 Frequency : lfs setstripe, only systems with more than 160 OSTs
2063 Description: unable to set striping with a starting offset beyond OST 160
2064 Details : llapi_create_file() incorrectly limited the starting stripe
2065 index to the maximum single-file stripe count.
2068 Frequency : LDAP users only
2070 Description: lconf did not handle in-kernel recovery with LDAP properly
2071 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2073 Severity : enhancement
2075 Description: bind OST threads to NUMA nodes to improve performance
2076 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2077 node and do their allocations there to localize memory access
2079 Severity : enhancement
2081 Description: llmount can determine client NID directly from Myrinet (GM)
2082 Details : the client NID code from gmnalnid was moved directly into
2083 llmount, removing the need to use this or specifying the
2084 client NID explicitly when mounting GM clients with zeroconf
2087 Frequency : if client is started with down MDS
2089 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2090 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2091 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2092 ptlrpc_import_delayed_req() because we didn't check if the
2093 request was interrupted, and we also didn't break out of the
2094 event loop if there was no timeout
2099 Description: data loss during non-page-aligned writes to a single file from
2100 both multiple nodes and multiple threads on one node at same time
2101 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2102 inconsistency led to false short-reads, that were cached and later
2103 used by ->prepare_write() to fill in partially written page,
2104 leading to data loss.
2107 Frequency : always, if lconf --abort_recovery used
2109 Description: lconf --abort_recovery fails with 'Operation not supported'
2110 Details : lconf was attempting to abort recovery on the MDT device and not
2113 Severity : enhancement
2115 Description: remove cleanup logs
2116 Details : replace lconf-generated cleanup logs with lustre internal
2117 cleanup routines. Eliminates the need for client-cleanup and
2120 Severity : enhancement
2122 Description: add support for EAs (user and system) on lustre filesystems
2123 Details : it is now possible to store extended attributes in the Lustre
2124 client filesystem, and with the user_xattr mount option it
2125 is possible to allow users to store EAs on their files also
2127 Severity : enhancement
2129 Description: Add possibility (config option) to show minimal available OST free
2131 Details : When compiled with --enable-mindf configure option, statfs(2)
2132 (and so, df) will return least minimal free space available from
2133 all OSTs as amount of free space on FS, instead of summary of
2134 free spaces of all OSTs.
2136 Severity : enhancement
2138 Description: do not expand extent locks acquired on OST-side
2139 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2140 by server: they are not cached anyway.
2143 Frequency : when mmap is used/binaries executed from Lustre
2145 Description: Unmmap pages before throwing them away from read cache.
2146 Details : llap_shrink cache now attempts to unmap pages before discarding
2147 them (if unmapping failed - do not discard). SLES9 kernel has
2148 extra checks that trigger if this unmapping is not done first.
2153 Description: lconf didn't resolve symlinks before checking to see whether a
2154 given mountpoint was already in use
2157 Frequency : when migrating failover services
2158 Bugzilla : 6395, 9514
2159 Description: When migrating a subset of services from a node (e.g. failback
2160 from a failover service node) the remaining services would
2161 time out and evict clients.
2162 Details : lconf --force (implied by --failover) sets the global obd_timeout
2163 to 5 seconds in order to quickly disconnect, but this caused
2164 other RPCs to time out too quickly. Do not change the global
2165 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2167 Severity : enhancement
2168 Frequency : if MDS is started with down OST
2169 Bugzilla : 9439,5706
2170 Description: Allow startup/shutdown of an MDS without depending on the
2171 availability of the OSTs.
2172 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2173 Add appropriate locking and lov-osc refcounts for safe
2174 cleaning. Add osc abort_inflight calls in case the
2175 synchronize never started.
2178 Frequency : occasional (Cray XT3 only)
2180 Description: root not authorized to access files in CRAY_PORTALS environment
2181 Details : The client process capabilities were not honoured on the MDS in
2182 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2183 been accessed by an authorized user then root was able to access
2184 the file on the local client also. The root user capabilities
2185 are now allowed on the MDS, as this environment has secure UID.
2188 Frequency : occasional
2190 Description: ldiskfs "too long searching" message happens too often
2191 Details : A debugging message (otherwise harmless) prints too often on
2192 the OST console. This has been reduced to only happen when
2193 there are fragmentation problems on the filesystem.
2198 Description: Division by zero in statfs when all OSCs are inactive
2199 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2200 lov_statfs divides by value returned by lov_get_stripecnt().
2204 Bugzilla : 9489, 3273
2205 Description: First write from each client to each OST was only 4kB in size,
2206 to initialize client writeback cache, which caused sub-optimal
2207 RPCs and poor layout on disk for the first writen file.
2208 Details : Clients now request an initial cache grant at (re)connect time
2209 and so that they can start streaming writes to the cache right
2210 away and always do full-sized RPCs if there is enough data.
2211 If the OST is rebooted the client also re-establishes its grant
2212 so that client cached writes will be honoured under the grant.
2217 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2218 Details : Now I/O RPCs go to different portal number and (presumably) fast
2219 lock requests (and glimses) and other RPCs get their own service
2220 threads pool that should be able to service those RPCs
2223 Severity : enhancement
2225 Description: Ability to exchange lustre version between client and servers and
2226 issue warnings at client side if client is too old. Also for
2227 liblustre clients there is ability to refuse connection of too old
2229 Details : New 'version' field is added to connect data structure that is
2230 filled with version info. That info is later checked by server and
2234 Frequency : rare, liblustre only.
2235 Bugzilla : 9296, 9581
2236 Description: Two simultaneous writes from liblustre at offset within same page
2237 might proceed at the same time overwriting eachother with stale
2239 Details : I/O lock withing llu_file_prwv was released too early, before data
2240 actually was hitting the wire. Extended lock-holding time until
2241 server acknowledges receiving data.
2244 Frequency : extremely rare. Never observed in practice.
2246 Description: avoid generating lustre_handle cookie of 0.
2247 Details : class_handle_hash() generates handle cookies by incrementing
2248 global counter, and can hit 0 occasionaly (this is unlikely, but
2249 not impossible, because initial value of cookie counter is
2250 selected randonly). Value of 0 is used as a sentinel meaning
2251 "unassigned handle" --- avoid it. Also coalesce two critical
2252 sections in this function into one.
2254 Severity : enhancement
2256 Description: allow liblustre clients to delegate truncate locking to OST
2257 Details : To avoid overhead of locking, liblustre client instructs OST to
2258 take extent lock in ost_punch() on client's behalf. New connection
2259 flag is added to handle backward compatibility.
2261 Severity : enhancement
2262 Bugzilla : 4928, 7341, 9758
2263 Description: allow number of OST service threads to be specified
2264 Details : a module parameter allows the number of OST service threads
2265 to be specified via "options ost ost_num_threads={N}" in the
2266 OSS's /etc/modules.conf or /etc/modprobe.conf.
2270 Bugzilla : 6146, 9635, 9895
2271 Description: servers crash with bad pointer in target_handle_connect()
2272 Details : In rare cases when a client is reconnecting it was possible that
2273 the connection request was the last reference for that export.
2274 We would temporarily drop the export reference and get a new
2275 one, but this may have been the last reference and the export
2276 was just destroyed. Get new reference before dropping old one.
2278 Severity : enhancement
2279 Frequency : if client is started with failover MDS
2281 Description: Allow multiple MDS hostnames in the mount command
2282 Details : Try to read the configuration from all specified MDS
2283 hostnames during a client mount in case the "primary"
2286 Severity : enhancement
2288 Description: Stop sending data to evicted clients as soon as possible.
2289 Details : Check if the client we are about to send or are sending data to
2290 was evicted already. (Check is done every second of waiting,
2291 for which l_wait_event interface was extended to allow checking
2292 of exit condition at specified intervals).
2295 Frequency : rare, normally only when NFS exporting is done from client
2297 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2299 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2302 Frequency : always, if srand() is called before liblustre initialization
2304 Description: Liblustre uses system PRNG disturbing its usage by user application
2305 Details : Introduce internal to lustre fast and high-quality PRNG for
2306 lustre usage and make liblustre and some other places in generic
2307 lustre code to use it.
2309 Severity : enhancement
2310 Bugzilla : 9477, 9557, 9870
2311 Description: Verify that the MDS configuration logs are updated when xml is
2312 Details : Check if the .xml configuration logs are newer than the config
2313 logs stored on the MDS and report an error if this is the case.
2314 Request --write-conf, or allow starting with --old_conf.
2316 Severity : enhancement
2318 Description: Handle symlinks in the path when checking if Lustre is mounted.
2319 Details : Resolve intermediate symlinks when checking if a client has
2320 mounted a filesystem to avoid duplicate client mounts.
2325 Description: lconf can hit an error exception but still return success.
2326 Details : The lconf command catches the Command error exception at the top
2327 level script context and will exit with the associated exit
2328 status, but doesn't ensure that this exit status is non-zero.
2333 Description: failure of ptlrpc thread startup can cause oops
2334 Details : Starting a ptlrpc service thread can fail if there are a large
2335 number of threads or the server memory is very fragmented.
2336 Handle this without oopsing.
2339 Frequency : always, only if liblustre and non-default acceptor port was used
2341 Description: liblustre cannot connect to servers with non-default acceptor port
2342 Details : tcpnal_set_default_params() was not called and was therefore
2343 ignoring the environment varaible TCPNAL_PORT, as well as other
2344 TCPNAL_ environment variables
2349 Description: two objects could be created on the same OST for a single file
2350 Details : If an OST is down, in some cases it was possible to create two
2351 objects on a single OST for a single file. No problems other
2352 than potential performance impact and spurious error messages.
2356 Bugzilla : 5681, 9562
2357 Description: Client may oops in ll_unhash_aliases
2358 Details : Client dcache may become inconsistent in race condition.
2359 In some cases "getcwd" can fail if the current directory is
2365 Description: Inode refcounting problems in NFS export code
2366 Details : link_raw functions used to call d_instantiate without obtaining
2367 extra inode reference first.
2371 Bugzilla : 9942, 9903
2372 Description: Referencing freed requests leading to crash, memleaks with NFS.
2373 Details : We used to require that call to ll_revalidate_it was always
2374 followed by ll_lookup_it. Also with revalidate_special() it is
2375 possible to call ll_revalidate_it() twice for the same dentry
2376 even if first occurence returned success. This fix changes semantic
2377 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2378 reference on a request referred from the intent.
2379 ll_intent_release() then releases such a request.
2382 Frequency : rare, normally benchmark loads only
2384 Description: unlinked inodes were kept in memory on the client
2385 Details : If a client is repeatedly creating and unlinking files it
2386 can accumulate a lot of stale inodes in the inode slab cache.
2387 If there is no other client load running this can cause the
2388 client node to run out of memory. Instead flush old inodes
2389 from client cache that have the same inode number as a new inode.
2392 Frequency : SLES9 2.6.5 kernel and long filenames only
2393 Bugzilla : 9969, 10379
2394 Description: utime reports stale NFS file handle
2395 Details : SLES9 uses out-of-dentry names in some cases, which confused
2396 the lustre dentry revalidation. Change it to always use the
2400 Frequency : rare, unless heavy write-truncate concurrency is continuous
2401 Bugzilla : 4180, 6984, 7171, 9963, 9331
2402 Description: OST becomes very slow and/or deadlocked during object unlink
2403 Details : filter_destroy() was holding onto the parent directory lock
2404 while truncating+unlinking objects. For very large objects this
2405 may block other threads for a long time and slow overall OST
2406 responsiveness. It may also be possible to get a lock ordering
2407 deadlock in this case, or run out of journal credits because of
2408 the combined truncate+unlink. Solution is to do object truncate
2409 first in one transaction without parent lock, and then do the
2410 final unlink in a new transaction with the parent lock. This
2411 reduces the lock hold time dramatically.
2414 Frequency : rare, 2.4 kernels only
2416 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2417 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2418 before unmounting their filesystem in order to avoid deadlock.
2419 The kernel_locked() function in 2.4 kernels only checks whether
2420 the kernel lock is held, not whether it is this process that is
2421 holding it as 2.6 kernels do.
2426 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2427 Details : The client ptlrpc code may be trying to reconnect to a down
2428 server before a previous connection attempt has timed out.
2429 Increase the reconnect interval to be longer than the connection
2430 timeout interval to avoid sending duplicate connections to
2434 Frequency : echo_client brw_test command
2436 Description: fix echo_client to work with OST preallocated code
2437 Details : OST preallocation code (5137) didn't take echo_client IO path
2438 into account: echo_client calls filter methods outside of any
2439 OST thread and, hence, there is no per-thread preallocated
2440 pages and buffers to use. Solution: hijack pga pages for IO. As
2441 a byproduct, this avoids unnecessary data copying.
2445 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2446 Description: Client can oops in mdc_commit_close() after open replay
2447 Details : It was possible for the MDS to return an open request with no
2448 transaction number in mds_finish_transno() if the client was
2449 evicted, but without actually returning an error. Clients
2450 would later try to replay that open and may trip an assertion
2451 Simplify the client close codepath, and always return an error
2452 from the MDS in case the open is not successful.
2455 Frequency : rare, 2.6 OSTs only
2457 Description: OST may deadlock under high load on fragmented files
2458 Details : If there was a heavy load and highly-fragmented OST filesystems
2459 it was possible to have all the OST threads deadlock waiting on
2460 allocation of biovecs, because the biovecs were not released
2461 until the entire RPC IO was completed. Instead, release biovecs
2462 as soon as they are complete to ensure forward IO progress.
2464 Severity : enhancement
2466 Description: Support for specifying external journal device at mount
2467 Details : If an OST or MDS device is formatted with an external journal
2468 device, this device major/minor is stored in the ext3 superblock
2469 and may not be valid for failover. Allow detecting and
2470 specifying the external journal at mount time.
2475 Description: Mounting an MDS with pending unlinked files may cause oops
2476 Details : target_finish_recovery() calls mds_postrecov() which returned
2477 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2478 considers this an error and immediately begins cleaning up the
2479 lov, just after starting the mds_lov process
2481 Severity : enhancement
2483 Description: Implement 'lfs df' to report actual free space on per-OST basis
2484 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2485 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2486 usage of objects; '-h' to report in human readable format.
2488 ------------------------------------------------------------------------------
2490 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2497 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2498 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2499 at startup during orphan unlinking. Instead, skip buddy update
2500 before setup, as it will be regenerated after recovery is complete.
2505 Description: If an OST is inactive, its locks might reference stale inodes.
2506 Details : lov_change_cbdata() must iterate over all namespaces, even if
2507 they are inactive to clear inode references from the lock.
2509 Severity : enhancement
2510 Frequency : occasional, if non-standard max_dirty_mb used
2512 Description: Client will block write RPCs if not enough grant
2513 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2514 then the client will block writes while waiting for another RPC
2515 to complete instead of consuming its dirty limit. With change
2516 we get improved performance when max_dirty_mb is small.
2518 Severity : enhancement
2519 Bugzilla : 3389, 6253
2520 Description: Add support for supplementary groups on the MDS.
2521 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2522 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2523 lookups for user supplementary groups into a cache.
2527 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2528 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2529 for liblustre O_EXCL is enforced by MDS.
2533 Description: readdir never returns NULL in liblustre.
2534 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2535 dentry in struct dirent.
2539 Frequency : liblustre only, depends on application IO pattern
2540 Description: liblustre clients evicted if not contacting servers
2541 Details : Don't put liblustre clients into the ping_evictor list, so
2542 they will not be evicted by the pinger ever.
2544 Severity : enhancement
2546 Description: Add ability to evict clients by NID from MDS.
2547 Details : By echoing "nid:$NID" string into
2548 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2549 $NID would be instantly evicted from this MDS and from all active
2550 OSTs connected to it.
2554 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2555 Details : lookup_it_finish() used to query file size from OSTs that was not
2560 Description: service threads change working directory to that of init
2561 Details : Starting lustre service threads may pin the working directory
2562 of the parent thread, making that filesystem busy. Threads
2563 now change to the working directory of init to avoid this.
2567 Frequency : during shutdown only
2568 Description: shutdown with a failed MDS or OST can cause unmount to hang
2569 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2574 Frequency : 2.6 only
2575 Description: chmod/chown may include an extra supplementary group
2576 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2577 supplementary group and if none is specified this is used.
2580 Bugzilla : 5479 (6816)
2582 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2583 Details : If lookup is in progress on a file that is unlinked we might try
2584 to revalidate the inode and fail in revalidate after lookup is
2585 complete and ll_file_open() enqueues the open again but
2586 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2589 Frequency : always, if lconf --abort_recovery used
2591 Description: lconf --abort_recovery fails with 'Operation not supported'
2592 Details : lconf was attempting to abort recovery on the MDT device and not
2595 ------------------------------------------------------------------------------
2597 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2602 Frequency : rare (only unsupported configurations with a node running as an
2604 Bugzilla : 6514, 5137
2605 Description: Mounting a Lustre file system on a node running as an OST could
2607 Details : OSTs now preallocates memory needed to write out data at
2608 startup, instead of when needed, to avoid having to
2609 allocate memory in possibly low memory situations.
2610 Specifically, if the file system is mounted on on OST,
2611 memory pressure could force it to try to write out data,
2612 which it needed to allocate memory to do. Due to the low
2613 memory, it would be unable to do so and the node would
2614 become unresponsive.
2616 Severity : enhancement
2618 Description: Addition of lconf --service command line option
2619 Details : lconf now accepts a '--service <arg>' option, which is
2620 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2622 Severity : enhancement
2624 Description: Failover mode is now the default for OSTs.
2625 Details : By default, OSTs will now run in failover mode. To return to
2626 the old behaviour, add '--failout' to the lmc line for OSTs.
2628 Severity : enhancement
2630 Description: Health checks are now provided for MDS and OSTs
2631 Details : Additional detailed health check information on MSD and OSTs
2632 is now provided through the procfs health_check value.
2635 Frequency : occasional, depends on IO load
2637 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2638 after numerous create/delete cycles
2639 Details : The ext3 inode allocation policy would not allocate new inodes
2640 very well on the OSTs because there are no new directories
2641 being created. Instead we look for groups with free space if
2642 the parent directories are nearly full.
2647 Description: Network or server problems during mount may cause partially
2648 mounted clients instead of returning an error.
2649 Details : The config llog parsing code may overwrite the error return
2650 code during mount error handling, returning success instead
2656 Description: MDS can fail to allocate large reply buffers
2657 Details : After long uptimes the MDS can fail to allocate large reply
2658 buffers (e.g. zconf client mount config records) due to memory
2659 fragmentation or consumption by the buffer cache. Preallocate
2660 some large reply buffers so that these replies can be sent even
2661 under memory pressure.
2665 Frequency : rare (liblustre)
2666 Description: fsx running with liblustre complained that using truncate() to
2667 extend the file doesn't work. This patch corrects that issue.
2668 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2669 ATTR_SIZE and lsm use in llu_setattr_raw.
2673 Frequency : rare, only 2.6 kernels
2674 Description: Unusual file access patterns on the MDS may result in inode
2675 data being lost in very rare circumstances.
2676 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2677 code in the 2.6 kernel caused the inode and/or EA data not to
2678 be read from disk, causing single-file corruption.
2682 Frequency : rare, only 2.6 filesystems using extents
2683 Description: Heavy concurrent write and delete load may cause data corruption.
2684 Details : It was possible under high-load situations to have an extent
2685 metadata block in the block device cache from a just-unlinked
2686 file overwrite a newly-allocated data block. We now unmap any
2687 metadata buffers that alias just-allocated data blocks.
2691 Frequency : filesystems with default stripe_count larger than 77
2692 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2693 Details : lconf specifies an inode size of 4096 bytes when the default
2694 stripe_count is larger than 77. This conflicts with the default
2695 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2696 this case to avoid pinning too much memory for large EAs.
2698 ------------------------------------------------------------------------------
2700 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2705 Frequency : rare (extremely heavy IO load with hundreds of clients)
2707 Description: Client is evicted, gets IO error writing to file
2708 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2709 caused clients to be evicted for AST timeouts. The fixes in
2710 bug 5192 mean we no longer need to have such short AST timeouts
2711 so ldlm_timeout has been increased.
2714 Frequency : occasional during --force or --failover shutdown under load
2715 Bugzilla : 5949, 4834
2716 Description: Server oops/LBUG if stopped with --force or --failover under load
2717 Details : a collection of import/export refcount and cleanup ordering
2718 issues fixed for safer force cleanup
2721 Frequency : only filesystems larger than 120 OSTs
2722 Bugzilla : 5990, 6223
2723 Description: lfs getstripe would oops on a very large filesystem
2724 Details : lov_getconfig used kfree on vmalloc'd memory
2727 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2728 Bugzilla : 6242, 6243
2729 Description: reading from files that had been truncated to a non-zero size
2730 but never opened returned no data
2731 Details : ll_file_read() reads zeros from no-object files to EOF
2736 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2737 Details : There's little harm in aborting MDS/OSS recovery and letting it
2738 try again, so I removed the LASSERT and return an error instead.
2740 Severity : enhancement
2742 Description: New debugging infrastructure for tracking down data corruption
2743 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2744 (b) cover more of the client-side code path, and (c) try to narrow
2745 down where problems occurred
2749 Bugzilla : 3819, 4364, 4397, 6313
2750 Description: Racing close and eviction MDS could cause assertion in mds_close
2751 Details : It was possible to get multiple mfd references during close and
2752 client eviction, leading to one thread referencing a freed mfd.
2754 Severity: : enhancement
2755 Bugzilla : 3262, 6359
2756 Description: Attempts to reconnect to servers are now more aggressive.
2757 Details : This builds on the enhanced upcall-less recovery that was added
2758 in 1.4.2. When trying to reconnect to servers, clients will
2759 now try each server in the failover group every 10 seconds. By
2760 default, clients would previously try one server every 25 seconds.
2765 Description: After recovery, certain operations trigger a failed
2766 assertion on a client.
2767 Details : Failing over an mds, using lconf -d --failover, while a
2768 client was doing a readdir() call would cause the client to
2769 LBUG after recovery completed and the readdir() was resent.
2771 Severity : enhancement
2773 Description: Default groups are now added by lconf
2774 Details : You can now run lconf --group <servicename> without having to
2775 manually add groups with lmc.
2778 Frequency : occasional
2780 Description: Nodes with an elan id of 0 trigger a failed assertion
2783 Frequency : always when accessing e.g. tty/console device nodes
2785 Description: tty and some other devices nodes cannot be used on lustre
2786 Details : file's private_data field is used by device data and lustre
2787 values in there got lost. New field was added to struct file to
2788 store fs-specific private data.
2791 Frequency : when exporting Lustre via NFS
2793 Description: NFSD failed occasionally when looking up a path component
2794 Details : NFSD is looking up ".." which was broken in ext3 directories
2795 that had grown large enough to become hashed.
2798 Frequency : Clusters with multiple interfaces not on the same subnet
2800 Description: Nodes will repeatedly try to reconnect to an interface which it
2801 cannot reach and report an error to the log.
2802 Details : Extra peer list entries will be created by lconf with some peers
2803 unreachable. lconf now validates the peer before adding it.
2806 Frequency : Only if a default stripe is set on the filesystem root.
2808 Description: Setting a default stripe on the filesystem root prevented the
2809 filesystem from being remounted.
2810 Details : The client was sending extra request flags in the root getattr
2811 request and did not allocate a reply buffer for the dir EA.
2814 Frequency : occasional, higher if lots of files are accessed by one client
2815 Bugzilla : 6159, 6097
2816 Description: Client trips assertion regarding lsm mismatch/magic
2817 Details : While revalidating inodes the VFS looks up inodes with ifind()
2818 and in rare cases can find an inode that is being freed.
2819 The ll_test_inode() code will free the lsm during ifind()
2820 when it finds an existing inode and then the VFS later attaches
2821 this free lsm to a new inode.
2825 Bugzilla : 6422, 7030
2826 Description: MDS deadlock between mkdir and client eviction
2827 Details : Creating a new file via mkdir or mknod (starting a transaction
2828 and getting the ns lock) can deadlock with client eviction
2829 (gets ns lock and trying to finish a synchronous transaction).
2832 Frequency : occasional
2833 Description: While starting a server, the fsfilt_ext3 module could not be
2835 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2836 kernels. Previously, lconf would still use the ext3 name
2837 when trying to load modules. Now, it will correctly use
2838 ext3 on 2.4 and ldiskfs on 2.6.
2840 Severity : enhancement
2841 Description: The default stripe count has been changed to 1
2842 Details : The interpretation of the default stripe count (0, to lfs
2843 or lmc) has been changed to mean striping across a single
2844 OST, rather than all available. For general usage we have
2845 found a stripe count of 1 or 2 works best.
2847 Severity : enhancement
2848 Description: Add support for compiling against Cray portals.
2849 Details : Conditional compiling for some areas that are different
2853 Frequency : occasional
2854 Bugzilla : 6409, 6834
2855 Description: Creating files with an explicit stripe count may lead to
2856 a failed assertion on the MDS
2857 Details : If some OSTs are full or unavailable, creating files may
2858 trigger a failed assertion on the MDS. Now, Lustre will
2859 try to use other servers or return an error to the
2863 Frequency : occasional
2865 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2866 caused lock timeout during readahead (since 1.4.2).
2867 Details : Processes doing readahead might match a lock that hasn't been
2868 granted yet if there are overlapping and conflicting lock
2869 requests. The readahead process waits on ungranted lock
2870 (original lock is CBPENDING), while OST waits for that process
2871 to cancel CBPENDING read lock and eventually evicts client.
2873 Severity : enhancement
2875 Description: Initial enabling of flock support for clients
2876 Details : Implements fcntl advisory locking and file status functions.
2877 This feature is provided as an optional mount flag (default
2878 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2879 locking are implemented yet, and those that are are not guaranteed
2880 to be completely correct in production environments.
2881 mount -t lustre -o [flock|noflock] ...
2884 Frequency : occasional
2886 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2887 assertion in the ext3 JBD (journaling) layer.
2888 Details : The b_committed_data struct is protected by the big kernel lock
2889 in 2.4 kernels, serializing journal_commit_transaction() and
2890 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2891 access to this struct. In 2.6 kernels there is finer grained
2892 locking to improve SMP performance of the JBD layer.
2896 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2901 Description: merge and cleanup kernel patches.
2904 -----------------------------------------------------------------------------
2906 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2908 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2909 versions of Lustre. Please update all servers and clients to
2910 version 1.4.2 or later at the same time. You must also run
2911 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2912 to update the configuration logs.
2914 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2915 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2916 - swab llog records of type '0' so we get proper header size/idx (5861)
2917 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2918 - fix rename of one directory over another leaking an inode (5953)
2919 - avoid SetPageDirty on 2.6 (5981)
2920 - don't re-add just-being-destroyed locks to the waiting list (5653)
2921 - when creating new directories, inherit the parent's custom
2922 striping settings if present parent (3048)
2923 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2924 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2925 - don't hold client locks on temporary worklist from l_lru (5666)
2926 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2927 - automatically evict dead clients (5921)
2928 - Update file size properly in create+truncate+fstat case (6196)
2929 - Do not unhash mountpoint dentries, do not allow removal of
2931 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2932 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2933 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2934 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2935 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2936 - Correctly handle reads of files with no objects (6243)
2937 - lctl recover will also mark a device active if deactivate used (5933)
2939 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2940 - allow --write-conf on an MDS with different nettype than client (5619)
2941 - don't write config llogs to MDS for mounts not from that MDS (5617)
2942 - lconf should create multiple TCP connections from a client (5201)
2943 - init scripts are now turned off by default; run chkconfig --on
2944 lustre and chkconfig --on lustrefs to use them
2945 - upcalls are no longer needed for clients to recover to failover
2947 - add --abort-recovery option to lconf to abort recovery on device
2949 - add support for an arbitrary number of OSTs (3026)
2950 - Quota support protocol changes.
2951 - forward compatibility changes to wire structs (6007)
2952 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2953 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2954 - improved "lustre status" script
2955 - initialize blocksize for non-regular files (6062)
2956 - added --disable-server and --disable-client configure options (5782)
2957 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2958 - Vanilla 2.4.29 support
2959 - increase maximum number of obd devices to 520 (6242)
2960 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2961 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2962 SLES 9 2.6.5 kernels
2963 - limit stripes per file to 160 (the maximum EA size) (6093)
2965 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2968 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2969 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2970 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2971 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2972 - small ext3 extents cleanups and fixes (5733)
2973 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2974 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2975 - avoid needless client->OST connect, fix handle mismatch (5317)
2976 - fix DLM error path that led to out-of-sync client, long delays (5779)
2977 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2978 - fix several locking issues related to i_size (5492,5624,5654,5672)
2979 - don't move pending lock onto export if it is already evicted (5683)
2980 - fix kernel oops when creating .foo in unlinked directory (5548)
2981 - fix deadlock in obdfilter statistics vs. object create (5811)
2982 - use time_{before,after} to avoid timer jiffies wrap (5882)
2983 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2984 - Do not leak request if server was not able to process it (5154)
2985 - If mds_open unable to find parent dir, make that negative lookup(5154)
2986 - don't create new directories with extent-mapping (5909, 5936)
2988 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2989 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2990 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2991 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2992 - support for RHEL4 kernel on i686 (5773)
2993 - provide error messages when incompatible logs are encountered (5898)
2995 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2996 * version 1.4.0.10 (1.4.1 release candidate 1)
2998 - don't keep a lock reference when lock is not granted (4238)
2999 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3000 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3001 - drop import inflight refcount on signal_completed_replay error (5255)
3002 - unlock page after async write error during send (3677)
3003 - handle missing objects in filter_preprw_read properly (5265)
3004 - no transno return for symlink open, don't save no-trasno open (3440)
3005 - don't try to complete elan receive that already failed (4012)
3006 - free RPC server reply state on error (5406)
3007 - clean up thread from ptlrpc_start_thread() on error (5160)
3008 - readahead could read extra page into cache that wasn't ejected (5388)
3009 - prevent races in class_attach/setup/cleanup/detach (5260)
3010 - don't dereference de->d_inode after l_dput of de (5458)
3011 - use "int" for stripe value returned from lock_to_stripe (5544)
3012 - mballoc allocation and error-checking fixes in 2.6 (5504)
3013 - block device patches to fix I/O request sizes in 2.6 (5482)
3014 - look up hostnames for IB nals (5602)
3015 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3016 - don't start multiple acceptors for the same port (5277)
3017 - fix incorrect LASSERT in mds_getattr_name (5635)
3018 - export a proc file for general "ping" checking (5628)
3019 - fix "lfs check" to not block when the MDS is down (5628)
3021 - service request history (4965)
3022 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3023 - create an "evict_client" /proc entry on OSTs, like the MDS has
3024 - fix mount usage message, return errors per mount(8) (5168)
3025 - change grep [] to grep "[]" in tests so they work in more UMLs
3026 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3027 - remove ancient LOV_MAGIC_V0 EA support (5047)
3028 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3029 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3030 - allow readahead allocations to fail when low on memory (5383)
3031 - mmap locking landed again, after considerable improvement (2828)
3032 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3034 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3037 - send OST transaction number in read/write reply to free req (4966)
3038 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3039 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3040 - the watchdog thread now runs as interruptible (5246)
3041 - flock/lockf fixes (but it's still disabled, pending 5135)
3042 - don't use EXT3 constants in llite code (5094)
3043 - memory shortage at startup could cause assertion (5176)
3045 - reorganization of lov code
3046 - single portals codebase
3048 - add extents/mballoc support (5025)
3049 - direct I/O reads in the obdfilter (4048)
3050 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3052 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3055 - send OST transaction number in read/write reply to free req (4966)
3056 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3057 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3058 - the watchdog thread now runs as interruptible (5246)
3059 - handle missing objects in filter_preprw_read properly (5265)
3060 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3061 - drop import inflight refcount on signal_completed_replay error (5255)
3062 - unlock page after async write error during send (3677)
3063 - return original error code on reconstructed replies (3761)
3064 - no transno return for symlink open, don't save no-trasno open (3440)
3066 - add pid to ldlm debugging output (4922)
3067 - bump the watchdog timeouts -- we can't handle 30sec yet
3068 - extra debugging for orphan dentry/inode bug (5259)
3070 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3073 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3074 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3075 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3076 - fix NULL dereference in /proc/sys/portals/routes (4827)
3077 - allow failed mdc_close() operations to be interrupted (4561)
3078 - stop precreate on OST before MDS would time out on it (4778)
3079 - don't send partial-page writes before EOF from client (4410)
3080 - discard client grant for sub-page writes on large-page clients (4520)
3081 - don't free dentries not owned by NFS code, check generation (4806)
3082 - fix lsm leak if mds_create_objects() fails (4801)
3083 - limit debug_daemon file size, always print CERROR messages (4789)
3084 - use transno after validating reply (3892)
3085 - process timed out requests if import state changes (3754)
3086 - update mtime on OST during writes, return in glimpse (4829)
3087 - add mkfsoptions to LDAP (4679)
3088 - use ->max_readahead method instead of zapping global ra (5039)
3089 - don't interrupt __l_wait_event() during strace
3091 - add software watchdogs to catch hung threads quickly (4941)
3092 - make lustrefs init script start after nfs is mounted
3093 - fix CWARN/ERROR duplication (4930)
3094 - return async write errors to application if possible (2248)
3095 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3096 - print NAL number in %x format (4645)
3097 - update barely-supported suse-2.4.21-171 series (4842)
3098 - support for sles 9 %post scripts
3099 - support for building 2.6 kernel-source packages
3100 - support for sles km_* packages
3102 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3105 - ignore -ENOENT errors in osc_destroy (3639)
3106 - notify osc create thread that OSC is being cleaned up (4600)
3107 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3108 - reconstruct ost_handle() like mds_handle() (4657)
3109 - create a new thread to do import eviction to avoid deadlock (3969)
3110 - let lconf resolve symlinked-to devices (4629)
3111 - don't unlink "objects" from directory with default EA (4554)
3112 - hold socknal file ref over connect in case target is down (4394)
3113 - allow more than 32000 subdirectories in a single directory (3244)
3114 - fix blocks count for O_DIRECT writes (3751)
3115 - OST returns ENOSPC from object create when no space left (4539)
3116 - don't send truncate RPC if file size isn't changing (4410)
3117 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3118 - bind to privileged port in socknal and tcpnal (3689)
3120 - rate limit CERROR/CWARN console message to avoid overload (4519)
3121 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3122 - basic mmap support (3918)
3123 - kernel patch series update from b1_4 (4711)
3125 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3128 - avoid crash during MDS cleanup with OST shut down (2775)
3129 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3130 - don't use bad inodes on the MDS (3744)
3131 - dynamic object preallocation to improve recovery speed (4236)
3132 - don't hold spinlock over lock dumping or change debug flags (4401)
3133 - don't zero obd_dev when it is force cleaned (3651)
3134 - print grants to console if they go negative (4431)
3135 - "lctl deactivate" will stop automatic recovery attempts (3406)
3136 - look for existing locks in ldlm_handle_enqueue() (3764)
3137 - don't resolve lock handle twice in recovery avoiding race (4401)
3138 - revalidate should check working dir is a directory (4134)
3140 - don't always mark "slow" obdfilter messages as errors (4418)
3142 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3145 - don't close LustreDB during write_conf until it is done (3860)
3146 - fix typo in lconf for_each_profile (3821)
3147 - allow dumping logs from multiple threads at one time (3820)
3148 - don't allow multiple threads in OSC recovery (3812)
3149 - fix debug_size parameters (3864)
3150 - fix mds_postrecov to initialize import for llog ctxt (3121)
3151 - replace config semaphore with spinlock (3306)
3152 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3153 - don't allow enqueue to complete on a destroyed export (3822)
3154 - down write_lock before checking llog header bitmap (3825)
3155 - recover from lock replay timeout (3764)
3156 - up llog sem before sending rpc (3652)
3157 - reduce ns lock hold times when setting kms (3267)
3158 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3159 - fix NULL deref and obd_dev leak on setup error (3312)
3160 - replace some LBUG about llog ops with error handling (3841)
3161 - don't match INVALID dentries from d_lookup and spin (3784)
3162 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3163 - fix invalid assertion in ptlrpc_set_wait (3880)
3165 - add libwrap support for the TCP acceptor (3996)
3166 - add /proc/sys/portals/routes for non-root route listing (3994)
3167 - allow setting MDS UUID in .xml (2580)
3168 - print the stack of a process that LBUGs (4228)
3170 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3173 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3174 - make sure to unset replay flag from failed open requests (3440)
3175 - if default stripe count is 0, use OST count for inode size (3636)
3176 - update parent mtime/ctime on client for create/unlink (2611)
3177 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3178 - free recovery state on server during a forced cleanup (3571)
3179 - unregister_reply for resent reqs (3063)
3180 - loop back devices mounting and status check on 2.6 (3563)
3181 - fix resource-creation race that can provoke i_size == 0 (3513)
3182 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3183 - more debugging for page-accounting assertion (3746)
3184 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3185 - avoid deadlock after precreation failure (3758)
3186 - fix race and lock order deadlock in orphan handling (3450, 3750)
3187 - add validity checks when grabbing inodes from l_ast_data (3599)
3189 - add /proc/.../recovery_status to obdfilter (3428)
3190 - lightweight CDEBUG infrastructure, debug daemon (3668)
3191 - change default OSC RPC parameters to be better on small clusters
3192 - turn off OST read cache for files smaller than 32MB
3193 - install man pages and include them in rpms (3100)
3194 - add new init script for (un)mounting lustre filesystems (2593)
3195 - run chkconfig in %post for init scripts (3701)
3196 - drop scimac NAL (unmaintained)
3198 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3201 - clean kiobufs before and after use (3485)
3202 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3203 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3204 - change init script to fail more clearly if not run as root (1528)
3205 - allow clients to reconnect during replay (1742)
3206 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3207 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3208 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3209 - fix cond_resched (3554)
3210 - only evict unfinished clients after recovery (3515)
3211 - allow bulk resend, prevent data loss (3570)
3212 - dynamic ptlrpc request buffer allocation (2102)
3213 - don't allow unlinking open directory if it isn't empty (2904)
3214 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3215 - remove extraneous obd dereference causing LASSERT failure (3334)
3216 - don't use get_cycles() when creating temp. files on the mds (3156)
3217 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3218 - handle EEXIST for set-stripe, set proper directory name (3336)
3220 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3221 - fix ksocknal_fmb_callback() error messages (2918)
3223 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3226 - don't copy lvb into (possibly NULL) reply on error (2983)
3227 - don't deref dentry after dput, don't free lvb on error (2922)
3228 - use the kms to determine writeback rpc length (2947)
3229 - increment oti_logcookies when osc is inactive (2948)
3230 - update client's i_blocks count via lvb messages (2543)
3231 - handle intent open/close of special files properly (1557)
3232 - mount MDS with errors=remount-ro, like obdfilter (2009)
3233 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3234 - don't use cancelling-locks' kms values (2947)
3235 - use highest lock extent for kms, not last one (2925)
3236 - don't dereference ERR_PTR() dentry in error handling path (3107)
3237 - fix thread race in portals_debug_dumplog() (3122)
3238 - create lprocfs device entries at setup instead of at attach (1519)
3239 - common AST error handler, don't evict client on completion race (3145)
3240 - zero nameidata in detach_mnt in 2.6 (3118)
3241 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3242 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3243 - initialize RPC timeout timer earlier for 2.6 (3219)
3244 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3245 - print nal/nid for unknown nid (3258)
3246 - additional checks for oscc recovery before doing precreate (3284)
3247 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3248 - don't crash in mdc_close for bad permissions on open (3285)
3249 - zero i_rdev for non-device files (3147)
3250 - clear page->private before handing to FS, better assertion (3119)
3251 - tune the read pipeline (3236)
3252 - fix incorrect decref of invalidated dentry (2350)
3253 - provide read-ahead stats and refine rpc in flight stats (3328)
3254 - don't hold journal transaction open across create RPC (3313)
3255 - update atime on MDS at close time (3265)
3256 - close LDAP connection when recovering to avoid server load (3315)
3257 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3258 - don't leak open file on MDS after open resend (3325)
3259 - serialize filter_precreate and filter_destroy_precreated (3329)
3260 - loop device shouldn't call sync_dev() for nul device (3092)
3261 - clear page cache after eviction (2766)
3262 - resynchronize MDS->OST in background (2824)
3263 - refuse to mount the same filesystem twice on same mountpoint (3394)
3264 - allow llmount to create routes for mounting behind routers (3320)
3265 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3266 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3267 - fix rare problem with rename on htree directories (3417)
3269 - allow default OST striping configuration per directory (1414)
3270 - fix compilation for qswnal for 2.6 kernels (3125)
3271 - increase maximum number of MDS request buffers for large systems
3272 - change liblustreapi to be useful for external progs like lfsck (3098)
3273 - increase local configuration timeout for slow disks (3353)
3274 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3276 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3279 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3280 - don't overwrite extent policy data in reply if lock was blocked (2901)
3281 - drop filter export grants atomically with removal from device (2663)
3282 - del obd_self_export from work_list in class_disconnect_exports (2908)
3283 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3284 - swab reply message in mdc_close, other PPC fixes (2464)
3285 - fix destroying of named logs (2325)
3286 - overwrite old logs when running lconf --write_conf (2264)
3287 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3288 - fix race in target_handle_connect (2898)
3289 - mds_reint_create() should take same inode create lock (2926)
3290 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3291 - don't close files for self_export to avoid uninitialized obd (2936)
3292 - allow MDS with the same name as client node (2939)
3293 - hold dentry reference for closed log files for unlink (2325)
3294 - reserve space for all logs during transactions (2059)
3295 - don't evict page beyond end of stripe extent (2925)
3296 - don't oops on a deleted current working directory (2399)
3297 - handle hard links to targets without a parent properly (2517)
3298 - don't dereference NULL lock when racing during eviction (2867)
3299 - don't grow lock extents when lots of conflicting locks (2919)
3301 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3304 - account for cache space usage on clients to avoid data loss (974)
3305 - lfsck support in lustre kernel code (2349)
3306 - reduce journal credits needed for BRW writes (2370)
3307 - orphan handling to avoid losing space on client/server crashes
3308 - ptlrpcd can be blocked, stopping ALL progress (2477)
3309 - use lock value blocks to assist in proper KMS, faster stat (1021)
3310 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3311 - recovery for initial connections (2355)
3312 - fixes for mds_cleanup_orphans (1934)
3313 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3314 - block all file creations until orphan recovery completes (1901)
3315 - client remove rq_connection from request struct (2423)
3316 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3317 - recovery timer race (2670)
3318 - mdc_close recovey bug (2532)
3319 - ptlrpc cleanup bug (2710)
3320 - mds timeout on local locks (2588)
3321 - namespace lock held during RPCs (2431)
3322 - handle interrupted sync write properly (2503)
3323 - don't try to handle a message that hasn't been replied to (2699)
3324 - client assert failure during cleanup after abort recovery (2701)
3325 - leak mdc device after failed mount (2712)
3326 - ptlrpc_check_set allows timedout requests to complete (2714)
3327 - wait for inflight reqs when ptlrpcd finishes (2710)
3328 - make sure unregistered services are removed from the srv_list
3329 - reset bulk XID's when resending them (caught by 1138 test)
3330 - unregister_bulk after timeout
3331 - fix lconf error (2694)
3332 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3333 - readahead locks pages, leaves pending causing memory pressure (2673)
3334 - increase OST request buffers to 4096 on large machines (2729)
3335 - fix up permission of existing directories in simple_mkdir (2661)
3336 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3337 - don't assign transno to errored transactions (2742)
3338 - don't delete objects on OST if given a bogus objid from MDS (2751)
3339 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3340 - if rq_no_resend, then timeout request after recovery (2432)
3341 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3342 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3343 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3344 - MDS recovery shouldn't delete objects that it has given out (2730)
3345 - if enqueue arrives after completion, don't clobber LVB (2819)
3346 - don't unlock pages twice when trigger_group_io returns error (2814)
3347 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3348 - don't write pages to disk if there was an error (1450)
3349 - don't ping imports that have recovery disabled (2676)
3350 - take buffered bytes into account when balancing socknal conn (2817)
3351 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3352 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3353 - remove little-endian swabbing of llog records (1987)
3354 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3355 - retry reposting request buffers if they fail (1191)
3356 - grow extent at grant time to avoid granting a revoked lock (2809)
3357 - lock revoke doesn't evict page if covered by a second lock (2765)
3358 - disable VM readahead to avoid reading outside lock extents (2805)
3360 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3361 - updated kernel patches for hp-2.4.20 kernel (2681)
3363 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3366 - fix truncated write corruption (2366)
3367 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3369 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3370 - ensure len(uuid) < 37 in lmc (1171)
3371 - fix ia64 OOPS in llog_test (2255)
3372 - zero end of page at obdfilter for partial page writes (2648)
3373 - don't leave stale dentries around after renames (bug 2428)
3374 - fix timeouts when evicting a client with a single lock held (2642)
3375 - set deadline for the initial HELLO message to drain (2634)
3376 - print out dotted-quad IP addresses in the socknal (2302)
3378 - additional debugging for MDS client eviction problem (2443)
3379 - fix mkfsoptions support for osts (2603, 2604)
3381 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3384 - add series for the vanilla 2.6.0 kernel
3385 - add series for the vanilla 2.4.24 kernel
3386 - add series for a cray x86/64 UL kernel drop
3387 - fix xattr patches for the vanilla 2.4.19 series
3389 - generate true UUIDs in lmc (1171)
3390 - have portals stack dumping break in UML (2466)
3391 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3392 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3393 - wake callback waiting threads on client eviction (2460)
3394 - Add --ptldebug and --subsystem to lmc (1719)
3395 - update assertion to allow safe interrupt allocation
3396 - set rq_no_resend for cancel requests (2432)
3397 - recalculate ptlrpcd timeout after resend (2494)
3398 - call vfs_rmdir when removing pending directories (2368)
3399 - fix renaming a file to itself (2429)
3400 - lmc creates a default one-stripe lov (2454)
3401 - expand procfs space to handle large clusters (2326)
3402 - increase UML stack to avoid overflow
3403 - update lconf's list of debug and subsystem masks
3404 - fix lfs find --obd (2510)
3405 - /proc tunable for disabling filter read caching (2591)
3406 - stop rpm packages from altering slapd.conf (2301)
3407 - disable nagle in the socknal under 0conf (2578)
3408 - choose mds inode size based on stripe count (2572)
3409 - fix kernel-source rpm problems (2516)
3411 - add --disable-doc to avoid pdf generation (2421)
3412 - update documentation, tests, type-os, comments
3413 - avoid format warnings on ia64
3414 - remove the TOE NAL
3415 - tiny code cleanups by removing unused fields
3417 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3420 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3421 - remove the most blatant lies from BUILDING file (2371)
3422 - change default debug level to reasonable production setting
3423 - reduce client side cache size to reduce cache flush time
3424 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3425 - make TCP zerocopy and pinger support enabled by default (2476)
3426 - sync writes completed after process exits caused crashes (2319)
3427 - maintain correct mount count on the MDS (2356)
3428 - backout 1557, because 2316 wasn't really fixed
3429 - better file I/O statistics gathering in /proc
3430 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3431 - another kernel patch to fix zero-copy TCP function export
3432 - don't take duplicate lock when processing re-sent getattr (2420)
3433 - lctl uses obd_self_export instead of creating new conn (2353)
3434 - MDS/OST recovery case which requires object creation asserted (2425)
3435 - move lfs from /usr/sbin to /usr/bin in packages
3436 - fix race between mds_client_add and mds_client_free (2417)
3437 - use kmalloc instead of slabs in portals (2430)
3438 - don't create duplicate records when a failover MDS is present (2442)
3439 - remove unnecessary mount age check (2332)
3440 - don't remove directory inodes from locks prematurely (2451)
3441 - don't break if MDS service name is the same as hostname (2103)
3442 - fix races in client write RPC generation when cache full (2482)
3444 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3447 - remove now-unused request->rq_obd (278)
3448 - if an allocation fails, print out how much memory we've used (1933)
3449 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3450 - add the "configurable stack size" patch to most series files (1256)
3451 - ability to write large log records, for 100+ OST configs (2306)
3452 - fix NULL deref when filter_prep fails (2314)
3453 - fix operator precedence error in filter_sync
3454 - dynamic allocation of socknal TX descriptors (2315)
3455 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3456 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3457 - ensure that CWARN messages in Portals always get to the syslog
3458 - __init/__exit are not for prototype decls (ldlm_init/exit)
3459 - x86-64 compile warning fixes
3460 - fix gateway LMC keyword conflict (2318)
3461 - fix MDS lock inversions in getattr/reint paths (1844)
3462 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3463 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3464 - DLM race condition prevented some lock evictions (2328)
3465 - ENOMEM detection and retry on socknal sends (2230)
3466 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3467 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3468 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3469 - better error messages when a client is rejected during recovery (1505)
3470 - avoid cancelling locks which were never granted, after failure (2330)
3471 - fix i_sem/journal inversion in mds_client_add (2333)
3472 - fix truncate/getattr lock cycle deadlock (2334)
3473 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3474 - fix two rare exit paths which could leak an l_lock() ref (2321)
3475 - fencepost error in MDS/OST orphan recovery (2226)
3476 - make log record alignment 8 bytes (1988)
3477 - lstripe now fails when requested offset > ost_count (2237)
3478 - ensure that all kernel series have a complete list.h (1607)
3479 - fix crashes in special-file operations (2316)
3480 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3482 - allow configurable automake binary, for testing new versions
3483 - small update to the lfs documentation
3485 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3487 * fix negative export reference count in fsfilt_sync (2312)
3489 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3490 * release candidate 0.9.1
3492 - orphans are moved into the PENDING directory for possible recovery
3493 - replayed opens now open by fid for orphan/rename safety (1042)
3494 - last close of an orphan inode generates a transno (683)
3495 - chdir() and mount() now pin the directory entry (1020)
3496 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3497 - discard very old requests without processing them (1502)
3498 - remove some common, well-understood CERRORs (1505)
3499 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3500 - clear "grant" flags in OST replies until OST grant code lands (1644)
3501 - fix read performance by not clobbering i_blksize on client (1598)
3502 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3503 - make LRU size a /proc tunable, clears locks when reduced (707)
3504 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3505 - server locks take references on exports now (1558)
3506 - build fixes for 2.4.20-rh trees (1663)
3507 - return an error from lov_create if all OSCs are inactive (1751)
3508 - fix import levels when a reconnect happens without a timeout (1597)
3509 - exit early from mds_open if we get a lookup error (1749)
3510 - partial page read at EOF wouldn't wait for disk before sending (1642)
3511 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3512 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3513 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3514 - write/truncate lock inversion (1639)
3515 - Don't auto-load obdclass, portals modules during cleanup (1495)
3516 - fix timestamps from jumping to "now" (1763)
3517 - extra journal assertions (1648)
3518 - add an extra multiunlink test (1771)
3519 - fix read_record/write_record API (1776)
3520 - fix leak of offset_extent, possible incorrect i_size later (1772)
3521 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3522 - Debugging for the kqswnal_get_idle_tx problems (1820)
3523 - Allow recovery to be attempted multiple times (1536)
3524 - Write out MDS last_rcvd file after it is first created (1600)
3525 - Fix tx_descriptor leak in failed transmit situations (1827)
3526 - ext3 journaling fixes for assertion failure after IO error (1871)
3527 - class_export_put() on freed export after completion AST error (1896)
3528 - Fix revalidate looping in VFS (1322)
3529 - Don't access a freed export during MDS_REINT timeout (1521)
3530 - Add open-unlink recovery support on the MDS (1673,1764)
3531 - Return an error if no MDS data was read from last_rcvd (1946)
3532 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3533 - Don't setup a disk device that doesn't match exported UUID (317)
3534 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3535 - avoid committing NULL handle in force close
3536 - local.sh is now a one-stripe LOV configuration
3537 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3538 - fix ext3 htree duplicate directory entry corruption (1516)
3539 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3540 - update to new LOV EA format (2097)
3541 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3543 2003-06-15 Phil Schwan <phil@clusterfs.com>
3546 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3547 - per-import/export recovery handling (958, 931, 959)
3548 - multiple last-rcvd slots, for serving multiple FSes (949)
3549 - connections are again shared between multiple imp/exports (963, 964)
3550 - "umount -f" would hang if any requests needed to be sent (393, 978)
3551 - avoid pinning large req buffer by copying for queued messages (989)
3552 - add "uuid" to "lctl device" command to help upcalls (991)
3553 - "open" RPCs with transnos would confuse recovery counters (1037)
3554 - do proper endian conversion of all wire messages (288, 340, 891)
3555 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3556 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3557 - fix ptlrpc_connection leak in target_handle_connect (1174)
3558 - fix import refcounting bug in OST and MDS cleanup (1134)
3559 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3560 - fix very unlikely obd_types race condition (501)
3561 - remove osc_open hack for echo_client (1187)
3562 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3563 - a failure in read_inode2 leads to deadlock (1139)
3564 - cancel ack-locks as soon as transaction is committed (1072)
3565 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3566 - make sure to commitrw after any preprw to avoid deadlock (1162)
3567 - failing to execute a file in a lustre FS would lock inode (1203)
3568 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3569 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3570 - obd_uuid2tgt would walk off the end of the list (1255)
3571 - on IA64 the osc would give portals incorrect bulk size (1258)
3572 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3573 - fix lock inversion caused by new llite matching code (1282)
3574 - limit the number of dirty pages on a client to 10MB (1286)
3575 - timed out locks were not being corrected cancelled (1289)
3576 - fix O_DIRECT above 4GB on IA-32 (1292)
3577 * major user-visible changes
3578 - fail out/fail over policy now controlled by the upcall (993)
3580 - add OBD_PING to check server availability and failure (954)
3581 - lustre messages are now sent in sending host order (288, 340, 891)
3582 - add eadatalen to MDS getattr reply (340)
3583 - OST read replies may contain second buffer, with per-page status (593)
3585 2003-03-11 Phil Schwan <phil@clusterfs.com>
3588 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3589 - failed open()s could cause deadlock; fixed (867, 869)
3590 - stop cancelling OST locks when files are closed (481)
3591 - overlapping XID spaces caused network corruption (851, 853)
3592 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3593 - setattr_raw added, to do single-RPC, server-side setattrs
3594 - lmc/lconf syntax change for OST UUIDs
3595 - fix crashy race condition between ptlrpc_free_req and osc_close
3596 - don't use request in mdc_enqueue if we hit a timeout (889)
3597 - don't set the inode i_size for regular files from the MDS (896)
3598 - handle out of order completion AST (842)
3599 - don't LBUG if a lock request times out after receiving AST (913)
3600 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3601 - if a bad lock AST arrives, send an error instead of dropping entirely
3602 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3603 - fix leak in bulk IO when only partially completed (899, 900, 926)
3604 - fix O_DIRECT for ia64 (55)
3605 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3606 - C-z after timeout could hang a process forever; fixed (977)
3608 - client-side I/O cache (678, 924, 929, 941, 970)
3610 - READPAGE and SETATTRs which don't take server-side locks get
3613 2003-02-11 Phil Schwan <phil@clusterfs.com>
3616 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3617 - Fix fsfilt_extN_readpage() to read a full page of directory
3618 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3619 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3620 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3621 - Most intent operations were removed, in favour of a new RPC mode
3622 that does a single RPC to the server and bypasses most of the VFS
3623 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3624 - Aggressively cancel local locks on DLM servers
3625 - mds_reint_unlink sends EA to the client if it's the last nlink.
3626 client uses that EA to unlink OST objects.
3627 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3628 - recursive symlinks were fixed (439)
3629 - fixed NULL deref in DEBUG_REQ
3630 - filter_update_lastobjid no longer calls sync, which annoyed extN
3631 - fixed multi-client small-writes to a single file problem (445)
3632 - fixed mtime updates during file writes (607)
3633 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3634 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3635 - fixed Linux OST size reporting problem (444, 656)
3636 - OST now updates object mtime with writes or setattr (607, 619)
3637 - client verifies file size before zeroing page past EOF (445)
3638 - OST now writes last allocated objid to disk with allocation (108)
3639 - LOV on echo now works (409)
3641 - mds_reint_unlink sends a new buffer, with the EA included. this
3642 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3643 set if a regular file was being unlinked, and it was the last link
3644 - use PtlGet from the target for bulk writes (315)
3645 - OST now updates object mtime with writes or setattr (607, 619)
3646 - LDLM now has a grant-time callback to revalidate locked items, if
3648 - Many MDS operations were reorganized to combat race conditions
3650 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3651 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3653 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3656 - Fully reactivate OST imports after reconnection (512, others)
3657 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3658 - More graceful error handling for truncating on dead OST (515)
3659 - Don't error out unless we're actually accessing dead stripes (474)
3660 - Fix garbage sizes when stripes are missing (410)
3661 - LRU counters were broken, causing constant lock purge (433, 432)
3662 - garbage on read from stripes with failed OSTs (441)
3663 - mark OSCs as active before reconnecting during recovery (438)
3664 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3665 - lfind did not preserve OST order in output (443)
3666 - symlinks cause hung clients, incorrect data (439)
3667 - stop dereferencing request after dropping refcount (457)
3668 - don't LASSERT(spin_is_locked) on non-SMP (455)
3669 - fixes for many rename() bugs
3670 - fstat didn't correctly synchronize attributes (399)
3671 - server must handle lock cancellation during blocking AST prep (487)
3672 - bulk descriptors were free()d too soon (511)
3673 - fix paths in lconf, which would load incorrect modules (451, 507)
3674 - fix confusing lconf 'host not found' error message (386)
3675 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3676 - fix race condition in mdc_blocking_ast() for inode access (526)
3677 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3678 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3679 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3680 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3681 - fix lstripe in several ways (kernel side) (527)
3682 - fix request leak in ldlm_cli_enqueue (262)
3683 - incorrect OSC was marked inactive after OST failure
3684 - call mds_fs_cleanup before unmounting filesystem (524)
3685 - fix races between taking ns_lock and ldlm_lock_change_resource
3686 - fix races updating LOV export open file list
3687 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3688 - fix recovery NULL deref in ldlm_cli_cancel_unused
3689 - fix some DLM races by using new hash table for lock handles (419)
3690 - permit the client to specify desired inodes, at replay
3691 - duplicate requests when we queue them for replay reintegration
3692 - fix last_rcvd offset calculation
3693 - sync after each recovered transaction, so we always make progress
3694 - never, not always, ERESTART requests without transnos
3695 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3697 - skip replay if the MDS says that the client is already connected
3698 - don't check for a recovery-enabled export to match lctl's UUID
3699 - don't INC_USE_COUNT for phantom exports
3700 - don't crash when cleaning up phantom exports (567)
3701 - don't double-finish or set replay data for errored mdc_open requests
3702 - abort requests when they time out, so we don't get old replies
3703 - send/receive replies for AST messages again
3704 - if the client says that it doesn't have the lock, cancel it on the
3706 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3707 mark it as destroyed, it will all work out when decref is called
3708 - fix module use counts (22, 581)
3710 - ASTs now expect a reply (server cancels lock on error reply)
3712 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3715 - fix many simultaneous client startup (392)
3716 - fix dentry->d_it clobbering
3717 - credentials weren't being shipped for readdir/getattr operations
3718 - remove invalid assertions triggered during some concurrent MD
3720 - proper Lustre versions added (336, 389)
3721 - fix memory leak for create error case (398)
3722 - fix LOV locking bug that would get cli/srv out of sync
3723 - fix echo client over LOV (409)
3724 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3725 - fix double-O_EXCL intent crash (424)
3726 - avoid sending multiple lock CANCELs (352)
3728 - MDS can do multi-client recovery (modulo bugs in new code)
3730 - many updates, edits, cleanups
3732 2002-11-18 Phil Schwan <phil@clusterfs.com>
3735 - fix null d_it dereference (346)
3736 - fix full OST/dbench hang (333)
3737 - fix permission problem with file removal (286)
3738 - fix removal of OSCs from LOV when they fail
3739 - fix NULL deref during bulk timeout (214)
3740 - fix problems related to multiple filesystems on one MDS (241)
3741 - fixed serious subtle metadata locking bugs
3742 - free locks on clients when inodes are removed due to memory
3744 - fix inode pointer in lock data (285)
3745 - partial support for multiple MDS on a single host (241)
3746 - data locks weren't cancelled at clear_inode time (290, 311)
3747 - intent locks could lead to unbounded lock growth (205)
3748 - added a maximum lock count, an LRU list, and a flusher
3749 - fix multiple rename (365)
3750 - properly abstracted the echo client
3751 - OSC locked 1 byte too many; fixed
3752 - rewrote brw callback code:
3753 - fixed recovery bugs related to LOVs (306)
3754 - fixed too-many-pages-in-one-write crash (191)
3755 - fixed (again) crash in sync_io_timeout (214)
3756 - probably fixed callback-related race (385)
3758 - Add capability to MDS protocol
3759 - LDLM cancellations and callbacks on different portals
3761 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3764 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3765 - defer open object destruction to close time (601981, 138)
3766 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3767 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3768 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3769 - don't call obd_setattr in ll_file_release for destroyed objects
3770 * protocol change to lustre_msg: move |version| and add |flags|
3771 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3772 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3773 * added replay of create, unlink, link and rename operations during
3774 MDS failover; recovery should be much more robust now
3775 * remove failed OSCs from LOVs (only lov_create uses this so far)
3776 * the lustre-HOWTO was brought (more) up to date (582544)
3778 2002-10-23 Phil Schwan <phil@clusterfs.com>
3781 - in-use dentries weren't being reused properly (617851)
3782 - prevent multiple LDLM setup (599178)
3783 - fix LOV size calculations for truncate (617853)
3784 - fix client handling of MDS intent errors (POSIX)
3785 - fix permission bug in lovstripe.c test (624321)
3786 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3787 - truncate past end of file could corrupt data
3788 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3789 - a race in recovery could return ETIMEDOUT to apps (623947)
3790 - building outside the source directory was fixed
3791 * the lustre-HOWTO was brought (more) up to date (582544)
3792 * major progress was made on recovery functionality
3794 2002-10-10 Phil Schwan <phil@clusterfs.com>
3797 - recovery deadlock fix
3798 - rm -rf causes LBUG fix (617817)
3799 - file open by multiple tasks fix (618962)
3800 - directory permissions bugs (602707 and 620007)
3801 - journal_stop fixed with locking (611313)
3802 - O_APPEND failures resolved (618273, perhaps 614459)
3803 - lconf PATH fix (619770)
3804 - IA64 build fix (621450)
3805 - RPC buffer sizes scale with amount of memory
3807 2002-10-01 Phil Schwan <phil@clusterfs.com>
3810 - locks would be cancelled without throwing away data pages,
3811 resulting in inconsistent data (605627)
3812 - inode attributes were not always being refreshed (605627, 612449)
3813 - lconf now continues to cleanup after lctl reports an error
3814 - MDS now enforces user permissions (602707)
3815 - lprocfs cleanup fixed, but not yet enabled (614157)
3816 - fixed infinite server hang, should a client not respond to an AST
3817 - avoid going into recovery if user calls readlink() with a buffer
3818 that's too small (613941)
3819 - AST RPCs no longer require replies (614867) -- this may be changed
3820 - don't crash server if client sends an IOV that's too big (611336)
3821 - fixed lock conversion deadlock (611892)
3822 - fixed the following of symlinks (614622)
3823 * recovery: the server can remove locks from a client that dies, other
3824 clients can make progress
3825 * more extN patch fixes
3826 * compile-time configurable ptlrpc buffer allocations
3828 - collaborative read cache document
3829 - Lustre Lite Performance CDR document-in-progress
3831 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3834 - fix typo in patch-2.4.18
3836 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3839 - clear ptlrpc request each time in handle_incoming_request()
3840 - unlink of files now destroys the object on the OST
3842 2002-09-19 Peter Braam <braam@clusterfs.com>
3844 * add hard link support
3845 * change obdfile creation method
3846 * kernel patch changed
3848 2002-09-19 Peter Braam <braam@clusterfs.com>
3851 - stack overflow bug in extN fixed
3853 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3855 * documentation updates
3856 - add man pages for config tools
3857 - update tests/README to describe testing with new config tools
3858 - finish metadata API descriptions
3859 * bug fixes and cleanups
3860 - statfs workaround for 16TB limit
3861 - LOV stripe allocation improved, can stripe on subset of OSTs
3862 - LOV file size/IO offset was wrong for files > 4GB in size
3863 - object EA data was being dropped, caused files to be unreadable
3864 - memory overflow with non-LOV OST caused memory corruption
3865 - fixed regression tests to work with new config tools, obdfilter
3866 - fixed bug when directory size became larger than 1 block
3867 - fixed bug (for single client case) when PWD was deleted
3868 - invalidate local directory pages when doing intent-based ops
3869 - avoid LDLM oops when lock callback contained bad data
3871 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3873 * documentation updates
3874 * bug fixes and cleanups
3875 - configuration tools
3878 - 64-bit compile warnings
3879 - 64-bit internal statfs data
3881 * test_brw on persistent OST devices
3883 * lprocfs (disabled)
3885 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3887 * documentation updates
3888 * bug fixes and cleanups
3889 * configuration tools
3891 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3894 * many small fixes to 0_5_4
3895 * io/network handling
3896 * thinkos in MDS operations
3898 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3901 * crucial basic fixes to 0.5.3
3902 * IOR, Iozone work over Elan
3905 2002-08-07 Phil Schwan <phil@clusterfs.com>
3906 * version 0_5_3, our first alpha
3907 * we use the new Portals iovs
3908 * documentation updates
3909 * bug fixes and cleanups
3910 * small changes in the DLM wire protocol
3912 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3913 * version 0_5_1 with some initial stability,
3914 * locking on MD and file I/O.
3915 * documentation updates
3916 * several bug fixes since 0.5.0
3917 * small changes in wire protocol
3919 2002-07-18 Phil Schwan <phil@clusterfs.com>
3921 * delivered as Lustre Light Alpha
3922 * fixed a crash after handling invalid MDS requests
3923 * fixed directory pages for architectures with non-4k pages sizes
3925 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3926 * release version v0_4_4
3927 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3929 * Fixes a number of interruption problems with OST operations.
3930 * Update documentation for portals header changes
3931 * Move all wire protocol structs/defines to lustre_idl.h
3932 * Fixes symlink length bug.
3933 * Add tcpdump to repository.
3935 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3936 * release version v0_4_3
3937 * Fixes statfs for inodes on extN.
3938 * Fixes bug in runtests which would delete /etc/hosts.
3939 * Use 64-bit object IDs wherever possible (not into VFS though)
3940 Remove ost_get_info, which is unused by lustre, and out of date.
3942 2002-07-03 Peter Braam <braam@clusterfs.com>
3943 * release version v0_4_2 Fixes a lookup error (type not passed)
3944 * move forward to head of Portals
3945 * move forward to latest Lustre kernel
3947 2002-06-25 Peter Braam <braam@clusterfs.com>
3948 * release version v0_4_1. Hopefully stable on single node use.