1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5 2.6.16.46-0.14 (SLES 10)
6 2.6.18.8 vanilla (kernel.org)
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.9-42.0.10.EL (RHEL 4)
10 2.6.16 - 2.6.21 vanilla (kernel.org)
11 * Recommended e2fsprogs version: 1.39.cfs8
12 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
14 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
16 * Support for kernels:
19 2.6.16.46-0.14 (SLES 10)
20 2.6.18.8 vanilla (kernel.org)
21 * Client support for unpatched kernels:
22 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
23 2.6.16 - 2.6.22 vanilla (kernel.org)
24 * Due to recently discovered recovery problems, we do not recommend
25 using patchless RHEL 4 clients with this or any earlier release.
26 * Recommended e2fsprogs version: 1.39.cfs8
27 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
28 * Starting with this release, the ldiskfs backing filesystem required
29 by Lustre is now in its own package, lustre-ldiskfs. This package
30 should be installed. It is versioned separately from Lustre and
31 may be released separately in future.
33 Severity : enhancement
35 Description: add optional extra BUILD_VERSION info
36 Details : add a new environment variable (namely LUSTRE_VERS) which allows
37 to override the lustre version.
40 Frequency : 2.6.18 servers only
42 Description: ll_kern_mount() doesn't release the module reference
43 Details : The ldiskfs module reference count never drops down to 0
44 because ll_kern_mount() doesn't release the module reference.
49 Description: server LBUG when using old ost_num_threads parameter
50 Details : Accept the old ost_num_threads parameter but warn that it
51 is deprecated, and fix an off-by-one error that caused an LBUG.
56 Description: Transient SCSI error results in persistent IO issue
57 Details : iobuf->dr_error is not reinitialized to 0 between two
61 Frequency : sometimes when underlying device returns I/O errors
63 Description: OSTs not going read-only during write failures
64 Details : OSTs are not remounted read-only when the journal commit threads
65 get I/O errors because fsfilt_ext3 calls journal_start/stop()
66 instead of the ext3 wrappers.
70 Description: poor connect scaling with increasing client count
71 Details : Don't run filter_grant_sanity_check for more than 100 exports
72 to improve scaling for large numbers of clients.
75 Frequency : SLES10 only
77 Description: sanity-quota.sh quotacheck failed: rc = -22
78 Details : Quotas cannot be enabled on SLES10.
81 Frequency : liblustre clients only
83 Description: getdirentries does not give error when run on compute nodes
84 Details : getdirentries does not fail when the size specified as an argument
85 is too small to contain at least one entry
87 Severity : enhancement
89 Description: Add LNET router traceability for debug purposes
90 Details : If a checksum failure occurs with a router as part of the
91 IO path, the NID of the last router that forwarded the bulk data
92 is printed so it can be identified.
97 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
98 Details : Due to a race condition, liblustre clients were occasionally
101 Severity : enhancement
103 Description: lfs setstripe use optional parameters instead of postional
106 Severity : enhancement
108 Description: Nanosecond timestamp support for ldiskfs
109 Details : The on-disk ldiskfs filesystem has added support for nanosecond
110 resolution timestamps. There is not yet support for this at
111 the Lustre filesystem level.
114 Frequency : during server recovery
116 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
117 Details : request with rq_no_resend flag not awake l_wait_event if they get a
121 Frequency : nfs export on patchless client
123 Description: connectathon hang when test nfs export over patchless client
124 Details : Disconnected dentry cannot be found with lookup, so we do not need
125 to unhash it or make it invalid
128 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
129 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
130 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
135 Description: open req refcounting wrong on reconnect
136 Details : If reconnect happened between getting open reply from server and
137 call to mdc_set_replay_data in ll_file_open, we will schedule
138 replay for unreferenced request that we are about to free.
139 Subsequent close will crash in variety of ways.
140 Check that request is still eligible for replay in
141 mdc_set_replay_data().
146 Description: disable writes to filesystem when reading health_check file
147 Details : the default for reading the health_check proc file has changed
148 to NOT do a journal transaction and write to disk, because this
149 can cause reads of the /proc file to hang and block HA state
150 checking on a healthy but otherwise heavily loaded system. It
151 is possible to return to the previous behaviour during configure
152 with --enable-health-write.
154 Severity : enhancement
156 Description: 64-bit inode version
157 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
158 to the inode. This will be required for version-based recovery.
163 Description: MDS fails to start if a duplicate client export is detected
164 Details : in some rare cases it was possible for a client to connect to
165 an MDS multiple times. Upon recovery the MDS would detect this
166 and fail during startup. Handle this more gracefully.
168 Severity : enhancement
170 Description: Add -o localflock option to simulate old noflock
172 Details : This will achieve local-only flock/fcntl locks
178 Description: log_commit_thread vs filter_destroy race leads to crash
179 Details : Take import reference before releasing llog record semaphore
184 Description: Wrong request locking in request set processing
185 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
186 imp_delayed_list, in this place should be used imp_lock.
189 Frequency : when reconnection
191 Description: Grant Leak when osc reconnect to OST
192 Details : When osc reconnect ost, OST(filter) should check whether it
193 should grant more space to client by comparing fed_grant and
194 cl_avail_grant, and return the granted space to client instead
195 of "new granted" space, because client will call osc_init_grant
196 to update the client grant space info.
199 Frequency : when client reconnect to OST
201 Description: Grant Leak when osc do resend and replay bulk write
202 Details : When osc reconnect to OST, OST(filter)should clear grant info of
203 bulk write request, because the grant info while be sync between
204 OSC and OST when reconnect, and we should ignore the grant info
205 these of resend/replay write req.
210 Description: Grant space more than avaiable left space sometimes.
211 Details : When then OST is about to be full, if two bulk writing from
212 different clients came to OST. Accord the avaliable space of the
213 OST, the first req should be permitted, and the second one
214 should be denied by ENOSPC. But if the seconde arrived before
215 the first one is commited. The OST might wrongly permit second
216 writing, which will cause grant space > avaiable space.
219 Frequency : when client is evicted
221 Description: Grant might be wrongly erased when osc is evicted by OST
222 Details : when the import is evicted by server, it will fork another
223 thread ptlrpc_invalidate_import_thread to invalidate the
224 import, where the grant will be set to 0. While the original
225 thread will update the grant it got when connecting. So if
226 the former happened latter, the grant will be wrongly errased
227 because of this race.
232 Description: Checking Stale with correct fid
233 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
234 to check whether it is stale, because sometimes, we want the
235 enqueue happened anyway, and op_data.fid2 will not be initialized.
237 Severity : enhancement
239 Description: update patchless client
240 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
243 Frequency : only with 2.4 kernel
245 Description: random memory corruption
246 Details : size of struct ll_inode_info is to big for union inode.u and this
247 can be cause of random memory corruption.
252 Description: Memory leak in recovery
253 Details : Lov_mds_md was not free in an error handler in mds_create_object.
254 It should also check obd_fail before fsfilt_start, otherwise if
255 fsfilt_start return -EROFS,(failover mds during mds recovery).
256 then the req will return with repmsg->transno = 0 and rc = EROFS.
257 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
258 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
259 be freed no matter whether fsfilt_commit success or not.
262 Frequency : only with huge count clients
264 Description: Prevents from taking the superblock lock in llap_from_page for
266 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
267 for llap_from_page call in ll_removepage prevents from taking the
268 superblock lock for a soon died page.
273 Description: Not check open intent error before release open handle
274 Details : in some rare cases, the open intent error is not checked before
275 release open handle, which may cause
276 ASSERTION(open_req->rq_transno != 0), because it tries to release
277 the failed open handle.
282 Description: Set cat log bitmap only after create log success.
283 Details : in some rare cases, the cat log bitmap is set too early. and it
284 should be set only after create log success.
288 Description: Accessing a block bevice can re-enable I/O when Lustre is
289 tearing down a device.
290 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
295 Description: service threads may hog cpus when there are a lot of requests
297 Details : Insert cond_resched to give other threads a chance to use some of
303 Description: the cat log was not initialized in recovery
304 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
305 unlink log on mds will not be initialized until mds post
306 recovery. And also in mds post recovery, the unlink log will
307 initialization will be done asynchronausly, so there will be race
308 between add unlink log and unlink log initialization.
312 Description: brw_stats were being printed incorrectly
313 Details : brw_stats were being printed as log2 but all of them were not
314 recorded as log2. Also remove some code duplication arising from
315 filter_tally_{read,write}.
319 Frequency : rare, only in recovery.
320 Description: ASSERTION(req->rq_type != LI_POISON) failed
321 Details : imp_lock should be held while iterating over imp_sending_list for
322 prevent destroy request after get timeout in ptlrpc_queue_wait.
326 Description: replay-single.sh test 52 fails
327 Details : A lock's skiplist need to be cleanup when it being unlinked
328 from its resource list.
332 Description: Short directio read returns full requested size rather than
334 Details : Direct I/O operations should return actual amount of bytes
335 transferred rather than requested size.
337 Severity : enhancement
339 Description: metadata RPC reduction (e.g. for rm performance)
340 Details : decrease the amount of synchronous RPC between clients and servers
341 by canceling conflicing lock before the operation on the client side
342 and packing thier handles into the main operation RPC to server.
344 --------------------------------------------------------------------------------
346 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
351 Frequency : on some architectures
353 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
354 Details : Uninitialized flags sometimes cause configuration commands to
358 Frequency : patchless clients only
360 Description: missing __iget() symbol export
361 Details : The __iget() symbol export is missing. To avoid the need for
362 this on patchless clients the deathrow inode reaper is turned
363 off, and we depend on the VM to clean up old inodes. This
364 dependency was during via the fix for bug 12181.
366 --------------------------------------------------------------------------------
368 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
370 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
371 INTEROPERATE with older versions automatically. In many cases a
372 special upgrade step is needed. Please read the
373 user documentation before upgrading any part of a live system.
374 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
375 version will not interoperate with 1.6 betas before beta5 (1.5.95).
376 * WARNING: Lustre configuration and startup changes are required with
377 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
382 Severity : enhancement
384 Description: MountConf
385 Details : Lustre configuration is now managed via mkfs and mount
386 commands instead of lmc and lconf. New obd types (MGS, MGC)
387 are added for dynamic configuration management. See
388 https://mail.clusterfs.com/wikis/lustre/MountConf for
391 Severity : enhancement
393 Description: dynamic OST addition
394 Details : OSTs can now be added to a live filesystem
396 Severity : enhancement
398 Description: startup order invariance
399 Details : MDTs and OSTs can be started in any order. Clients only
400 require the MDT to complete startup.
402 Severity : enhancement
404 Description: parallel, asynchronous orphan cleanup
405 Details : orphan cleanup is now performed in separate threads for each
406 OST, allowing parallel non-blocking operation.
408 Severity : enhancement
410 Description: optimized stripe assignment
411 Details : stripe assignments are now made based on ost space available,
412 ost previous usage, and OSS previous usage, in order to try
413 to optimize storage space and networking resources.
415 Severity : enhancement
417 Description: Permanently set tunables
418 Details : All writable /proc/fs/lustre tunables can now be permanently
419 set on a per-server basis, at mkfs time or on a live
422 Severity : enhancement
424 Description: Lustre message v2
425 Details : Add lustre message format v2.
427 Severity : enhancement
429 Description: client OST exclusion list
430 Details : Clients can be started with a list of OSTs that should be
431 declared "inactive" for known non-responsive OSTs.
435 Description: SPEC SFS validation failure on NFS v2 over lustre.
436 Details : Changes the blocksize for regular files to be 2x RPC size,
437 and not depend on stripe size.
439 Severity : enhancement
441 Description: Multiple MD RPCs in flight.
442 Details : Further unserialise some read-only MDS RPCs - learn about intents.
443 To avoid overly-overloading MDS, introduce a limit on number of
444 MDS RPCs in flight for a single client and add /proc controls
445 to adjust this limit.
447 Severity : enhancement
449 Description: client read/write statistics
450 Details : Add client read/write call usage stats for performance
451 analysis of user processes.
452 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
453 file access. extents_stats shows chunk size distribution.
454 extents_stats_per_process show chunk size distribution per
457 Severity : enhancement
459 Description: mds statistics
460 Details : Add detailed mds operations statistics in
461 /proc/fs/lustre/mds/*/stats.
465 Description: Failure of copying files with lustre special EAs.
466 Details : Client side always return success for setxattr call for lustre
467 special xattr (currently only "trusted.lov").
472 Description: Refcount LNET uuids
473 Details : The global LNET uuid list grew linearly with every startup;
474 refcount repeated list entries instead of always adding to
477 Severity : enhancement
479 Description: Dynamic service threads
480 Details : Within a small range, start extra service threads
481 automatically when the request queue builds up.
484 Frequency : mixed-endian client/server environments
486 Description: mixed-endian crashes
487 Details : The new msg_v2 system had some failures in mixed-endian
490 Severity : enhancement
492 Description: Easy OST removal
493 Details : OSTs can be permanently deactivated with e.g. 'lctl
494 conf_param lustre-OST0001.osc.active=0'
496 Severity : enhancement
498 Description: MGS proc entries
499 Details : Added basic proc entries for the MGS showing what filesystems
502 Severity : enhancement
504 Description: provide MGS failover
505 Details : Added config lock reacquisition after MGS server failover.
507 Severity : enhancement
509 Description: add Linux 2.4 support
510 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
514 Description: a large application tries to do I/O to the same resource and dies
516 Details : Check the req->rq_arrival time after the call to
517 ost_brw_lock_get(), but before we do anything about
518 processing it & sending the BULK transfer request. This
519 should help move old stale pending locks off the queue as
520 quickly as obd_timeout.
523 Frequency : when an incorrect nid is specified during startup
525 Description: ptlrpc connect to non-existant node causes kernel crash
526 Details : LNET can't be re-entered from an event callback, which
527 happened when we expire a message after the export has been
528 cleaned up. Instead, hand the zombie cleanup off to another
531 Severity : enhancement
533 Description: plain/inodebits lock performance improvement
534 Details : Grouping plain/inodebits in granted list by their request modes
535 and bits policy, thus improving the performance of search through
539 Frequency : only if OST filesystem is corrupted
541 Description: client incorrectly hits assertion in ptlrpc_replay_req()
542 Details : for a short time RPCs with bulk IO are in the replay list,
543 but replay of bulk IOs is unimplemented. If the OST filesystem
544 is corrupted due to disk cache incoherency and then replay is
545 started it is possible to trip an assertion. Avoid putting
546 committed RPCs into the replay list at all to avoid this issue.
549 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
551 Description: System hang on startup
552 Details : This bug allowed the liblustre (e.g. catamount) client to
553 return to the app before handling all startup RPCs. This
554 could leave the node unresponsive to lustre network traffic
555 and manifested as a server ptllnd timeout.
557 Severity : enhancement
559 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
560 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
561 boolean to control whether to print peer debug info when a
562 client's RPC times out.
565 Frequency : only for kernels with patches from Lustre below 1.4.3
567 Description: Remove old rdonly API
568 Details : Remove old rdonly API which unused from at least lustre 1.4.3
571 Frequency : only for devices with external journals
573 Description: Set external device read-only also
574 Details : During a commanded failover stop, we set the disk device
575 read-only while the server shuts down. We now also set any
576 external journal device read-only at the same time.
579 Frequency : when upgrading from 1.4 while trying to change parameters
581 Description: The wrong (new) MDC name was used when setting parameters for
582 upgraded MDT's. Also allows changing of OSC (and MDC)
583 parameters if --writeconf is specified at tunefs upgrade time.
586 Frequency : when setting specific ost indicies
588 Description: QOS code breaks on skipped indicies
589 Details : Add checks for missing OST indicies in the QOS code, so OSTs
590 created with --index need not be sequential.
592 Severity : enhancement
594 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
595 Details : The uninit_groups feature works in conjunction with the kernel
596 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
597 up the pass1 processing of e2fsck. This is a read-only feature
598 in ldiskfs2 only, so older kernels and current ldiskfs cannot
599 mount filesystems that have had this feature enabled.
601 Severity : enhancement
603 Description: Improve multi-block allocation algorithm to avoid fragmentation
604 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
605 allocation locality and avoid filesystem fragmentation.
607 ------------------------------------------------------------------------------
609 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
611 * Support for kernels:
612 2.4.21-47.0.1.EL (RHEL 3)
614 2.6.9-42.0.10.EL (RHEL 4)
615 2.6.12.6 vanilla (kernel.org)
616 2.6.16.27-0.9 (SLES 10)
617 * Recommended e2fsprogs version: 1.39.cfs5
619 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
623 Frequency : occasional, depends on client load and configuration
624 Bugzilla : 12181, 12203
625 Description: data loss for recently-modified files
627 Details : In some cases it is possible that recently written or created
628 files may not be written to disk in a timely manner (this should
629 normally be within 30s unless client IO load is very high).
630 The problem appears as zero-length files or files that are a
631 multiple of 1MB in size after a client crash or client eviction
632 that are missing data at the end of the file.
634 This problem is more likely to be hit on clients where files are
635 repeatedly created and unlinked in the same directory, clients
636 have a large amount of RAM, have many CPUs, the filesystem has
637 many OSTs, the clients are rebooted frequently, and/or the files
638 are not accessed by other nodes after being written.
640 The presence of the problem can be detected by looking at
641 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
642 smaller than the second (nr_unused) then dirty files will not
643 be flushed automatically to disk. "sync; sleep 10" should be
644 run several times on the node before unmounting it to update
645 Lustre (this is also safe to run on nodes without this problem).
647 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
648 that can cause this same problem, so customers using that kernel
649 also need to update the kernel in addition to Lustre. In order
650 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
652 It is normal that files written just before a client crash (less
653 than 30s) may not yet have been flushed to disk, even for local
657 Frequency : frequent on thin XT3 nodes
659 Description: UUID collision on thin XT3 Linux nodes
660 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
661 randomly, since we relied on an insufficiently-seeded PRNG.
666 Description: OSS hangs after "All ost request buffers busy"
667 Details : A deadlock between quota and journal operations caused OSS
668 hangs after printing "All ost request buffers busy."
671 Frequency : always on liblustre builds
673 Description: Cleanup compiler warnings on liblustre
676 Frequency : always on liblustre builds on XT3
678 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
682 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
683 > 32000 subdirectories
684 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
685 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
686 subdirectory count crosses 32000. This will aid e2fsck to
687 correctly handle more than 32000 subdirectories.
690 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
692 Description: System hang on startup
693 Details : This bug allowed the liblustre (e.g. catamount) client to
694 return to the app before handling all startup RPCs. This
695 could leave the node unresponsive to lustre network traffic
696 and manifested as a server ptllnd timeout.
698 Severity : enhancement
700 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
701 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
702 boolean to control whether to print peer debug info when a
703 client's RPC times out.
708 Description: make O_SYNC working on 2.6 kernels
709 Details : 2.6 kernels use different method for mark pages for write,
710 so need add a code to lustre for O_SYNC work.
715 Description: Failure to close file and release space on NFS
716 Details : Put inode details into lock acquired in ll_intent_file_open.
717 Use mdc_intent_lock in ll_intent_open to properly
718 detect all kind of errors unhandled by mdc_enqueue.
723 Description: proc file read during shutdown sometimes raced obd removal,
725 Details : Add lock to prevent obd access after proc file removal.
728 Frequency : Only for files larger than 4GB on 32-bit clients.
730 Description: improperly doing page alignment of locks
731 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
732 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
737 Description: RPCs being resent when they shouldn't be
738 Details : Some RPCs that should not be resent are being resent. This
739 can cause inconsistencies in the RPC state machine. Do not
740 resend such requests.
743 Frequency : rare, only with NFS export
745 Description: Crash on NFS re-export node
746 Details : under very unusual load conditions an assertion is hit in
747 ll_intent_file_open()
750 Frequency : only if OST filesystem is corrupted
752 Description: client incorrectly hits assertion in ptlrpc_replay_req()
753 Details : for a short time RPCs with bulk IO are in the replay list,
754 but replay of bulk IOs is unimplemented. If the OST filesystem
755 is corrupted due to disk cache incoherency and then replay is
756 started it is possible to trip an assertion. Avoid putting
757 committed RPCs into the replay list at all to avoid this issue.
762 Description: large O_DIRECT requests fail under memory pressure/fragmentation
763 Details : Large single O_DIRECT read and write calls can fail to allocate
764 a sufficiently large buffer to process the request. In case of
765 allocation failure the allocation is retried with a smaller
766 buffer and broken into smaller requests.
768 Severity : enhancement
770 Description: Add -o localflock option to simulate old noflock behaviour.
771 Details : This will achieve local-only flock/fcntl locks coherentness.
776 Description: versioning check is incomplete
777 Details : Checking the version difference of client vs. server, report
778 error if the gap is too big.
783 Description: add support PG_writeback bit
784 Details : add support for PG_writeback bit for Lustre, for more carefull
785 work with page cache in 2.6 kernel. This also fix some deadlocks
786 and remove hack for work O_SYNC with 2.6 kernel.
788 Severity : enhancement
790 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
791 Details : The uninit_groups feature works in conjunction with the kernel
792 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
793 up the pass1 processing of e2fsck. This is a read-only feature
794 in ldiskfs2 only, so older kernels and current ldiskfs cannot
795 mount filesystems that have had this feature enabled.
797 Severity : enhancement
799 Description: Improve multi-block allocation algorithm to avoid fragmentation
800 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
801 allocation locality and avoid filesystem fragmentation.
803 ------------------------------------------------------------------------------
805 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
807 * Support for kernels:
808 2.6.9-42.0.3.EL (RHEL 4)
810 2.4.21-47.0.1.EL (RHEL 3)
811 2.6.12.6 vanilla (kernel.org)
812 2.6.16.21-0.8 (SLES10)
813 * Recommended e2fsprogs version: 1.39.cfs2-0
815 * The backwards-compatible /proc/sys/portals symlink has been removed
816 in this release. Before upgrading, please ensure that you change
817 any configuration scripts or /etc/sysctl.conf files that access
818 /proc/sys/portals/* or sysctl portals.* to use the corresponding
819 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
820 in advance of the upgrade on any system running Lustre 1.4.6 or
821 newer, since /proc/sys/lnet was added in that version.
822 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
826 Frequency : only when quota is used
828 Description: avoid scanning export list for quota master
829 Details : Change the algorithms to avoid scanning export list in order
830 to improve the efficiency.
833 Frequency : MDS failover only, very rarely
835 Description: "went back in time" messages on mds failover
836 Details : The greatest transno may be lost when the current operation
837 finishes with an error (transno==0) and the client's last_rcvd
838 record is over-written. Save the greatest transno in the
839 mds_last_transno for this case.
842 Frequency : always for specific kernels and striping counts
844 Description: client may get "Matching packet too big" without ACL support
845 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
846 when trying to access files in certain configurations. The
847 clients should in fact be denied when mounting because they do
851 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
853 Description: many clients connecting with IO in progress causes connect timeouts
854 Details : Avoid synchronous journal commits to avoid delays caused by many
855 clients connecting/disconnecting when bulk IO is in progress.
856 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
857 OST_IO_PORTAL to avoid delays behind potentially many pending
861 Frequency : occasionally with multiple writers to a single file
863 Description: shared writes to file may result in wrong size reported by stat()
864 Details : Allow growing of kms when extent lock is cancelled
867 Frequency : always with random mmap IO to multi-striped file
869 Description: mmap write might be lost if we are writing to a 'hole' in stripe
870 Details : Only if the hole is at the end of OST object so that kms is too
871 small. Fix is to increase kms accordingly in ll_nopage.
874 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
876 Description: writes to a missing object would leak memory on the OST
877 Details : If there is an inconsistency between the MDS and OST filesystems,
878 such that the MDS references an object that doesn't exist, writes
879 to that object will leak memory due to incorrect cleanup in the
880 error handling path, eventually running out of memory on the OST.
885 Description: Creating too long symlink causes lustre errors
886 Details : Check symlink and name lengths before sending requests to MDS.
889 Frequency : only if flock is enabled (not on by default)
891 Description: posix locks not released on fd closure on 2.6.9+
892 Details : We failed to add posix locks to list of inode locks on 2.6.9+
893 kernels, this caused such locks not to be released on fd close and
894 then assertions on fs unmount about still used locks.
897 Frequency : MDS failover only, very rarely
899 Description: clients may get ASSERTION(granted_lock != NULL)
900 Details : When request was taking a long time, and a client was resending
901 a getattr by name lock request. The were multiple lock requests
902 with the same client lock handle and
903 mds_getattr_name->fixup_handle_for_resent_request found one of the
904 lock handles but later failed with ASSERTION(granted_lock != NULL).
909 Description: handle->h_buffer_credits > 0, assertion failure
910 Details : h_buffer_credits is zero after truncate, causing assertion
911 failure. This patch extends the transaction or creates a new
915 Frequency : NFS re-export or patchless client
916 Bugzilla : 11179, 10796
917 Description: Crash on NFS re-export node (__d_move)
918 Details : We do not want to hash the dentry if we don't have a lock.
919 But if this dentry is later used in d_move, we'd hit uninitialised
920 list head d_hash, so we just do this to init d_hash field but
921 leave dentry unhashed.
924 Frequency : NFS re-export or patchless client
926 Description: NFS exports has problem with symbolic link
927 Details : lustre client didn't properly install dentry when re-exported
928 to NFS or running patchless client.
931 Frequency : NFS re-export or patchless client
933 Description: Various nfs/patchless fixes.
934 Details : fixes reuse disconected alias for lookup process - this fixes
935 warning "find_exported_dentry: npd != pd",
936 fix permission error with open files at nfs.
937 fix apply umask when do revalidate.
940 Frequency : occasional
942 Description: Crash on NFS re-export node
943 Details : calling clear_page() on the wrong pointer triggered oops in
944 generic_mapping_read().
947 Frequency : rarely, using O_DIRECT IO
949 Description: unaligned directio crashes client with LASSERT
950 Details : check for unaligned buffers before trying any requests.
953 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
955 Description: stale data returned from RAID cache
956 Details : If only a small amount of IO is done to the RAID device before
957 reading it again it is possible to get stale data from the RAID
958 cache instead of reading it from disk.
961 Frequency : always for sles10 kernel
963 Description: sles10 support
964 Details : ll_follow_link: compile fixes and using of nd_set_link
968 Frequency : depends on arch, kernel and compiler version, always on sles10
971 Description: recursive or deep enough symlinks cause stack overflow
972 Details : getting rid of large stack-allocated variable in
976 Frequency : depends on hardware
978 Description: lustre write performance loss in the SLES10 kernel
979 Details : the performance loss is caused by using of write barriers in the
980 ext3 code. The SLES10 kernel turns barrier support on by
981 default. The fix is to undo that change for ldiskfs.
983 ------------------------------------------------------------------------------
985 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
987 * Support for kernels:
988 2.6.9-42.0.3EL (RHEL 4)
990 2.4.21-47.0.1.EL (RHEL 3)
991 2.6.12.6 vanilla (kernel.org)
995 Frequency : quota enabled and large files being deleted
997 Description: releasing more than 4GB of quota at once hangs OST
998 Details : If a user deletes more than 4GB of files on a single OST it
999 will cause the OST to spin in an infinite loop. Release
1000 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1005 Description: statfs data retrieved from /proc may be stale or zero
1006 Details : When reading per-device statfs data from /proc, in the
1007 {kbytes,files}_{total,free,avail} files, it may appear
1008 as zero or be out of date.
1011 Frequency : systems with MD RAID1 external journal devices
1013 Description: lconf's call to blkid is confused by RAID1 journal devices
1014 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1015 of returning all block devices that match the journal UUID.
1018 Frequency : always, for aggregate stripe size over 4GB
1020 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1021 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1022 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1025 Frequency : always if "lfs find" used on a local file/directory
1027 Description: "lfs find" segfaults if used on a local file/directory
1028 Details : The case where a directory component was not specified wasn't
1029 handled correctly. Handle this properly.
1032 Frequency : always on ppc64
1034 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1035 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1036 perform endianness conversion.
1039 Frequency : rarely (truncate to non-zero file size after write under load)
1040 Bugzilla : 10730, 10687
1041 Description: Files padded with zeros to next 4K multiple
1042 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1043 it is possible that files that are truncated to a non-zero size
1044 immediately after being written are filled with zero bytes beyond
1045 the truncated size. No file data is lost.
1047 Severity : enhancement
1049 Description: Allow recovery/failover for liblustre clients.
1050 Details : liblustre clients were unaware of failover configurations until
1053 Severity : enhancement
1055 Description: user file locks should fail when not mounting with flock option
1056 Details : Set up an error-returning stub in ll_file_operations.lock field
1057 to prevent incorrect behaviour when client is mounted without
1058 flock option. Also, set up properly f_op->flock field for
1062 Frequency : always on ia64
1064 Description: "lfs df" loops on printing out MDS statfs information
1065 Details : The obd_ioctl_data was not initialized and in some systems
1066 this caused a failure during the ioctl that did not return
1067 an error. Initialize the struct and return an error on failure.
1070 Frequency : SLES 9 only
1072 Description: Error of copying files with lustre special EAs as root
1073 Details : Client side always return success for setxattr call for lustre
1074 special xattr (currently only "trusted.lov").
1077 Frequency : rarely on clusters with both ia64+i386 clients
1079 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1080 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1081 clients are doing concurrent writes to the same file it is
1082 possible that the ia64 clients may LASSERT because the OST
1083 extent locks are not PAGE_SIZE aligned. Ensure that grown
1084 locks are always aligned on the request boundary.
1087 Frequency : specific use, occasional
1089 Description: Overwriting in use executable truncates on-disk binary image
1090 Details : If one node attempts to overwrite an executable in use by
1091 another node, we now correctly return ETXTBSY instead of
1092 truncating the file.
1094 Severity : enhancement
1096 Description: Async OSC create to avoid the blocking unnecessarily.
1097 Details : If a OST has no remain object, system will block on the creating
1098 when need to create a new object on this OST. Now, ways use
1099 pre-created objects when available, instead of blocking on an
1100 empty osc while others are not empty. If we must block, we block
1101 for the shortest possible period of time.
1106 Description: chmod on Lustre root is propagated to other clients
1107 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1108 invalid by the follow_mount time.
1111 Frequency : liblustre clients only
1113 Description: Race in 'instant cancel' lock handling could lead to such locks
1114 never to be granted in case of SMP MDS
1115 Details : Do not destroy not yet granted but cbpending locks in
1119 Frequency : replay/resend of open
1121 Description: non null lock assetion failure in mds_intent_policy
1122 Details : Trying to replay/resend lockless open requests resulted in
1123 mds_open() returning 0 with no lock. Now it sets a flag if
1124 a lock is going to be returned.
1126 Severity : enhancement
1128 Description: Checksum enhancements
1129 Details : New checksum enhancements allow for resending RPCs that failed
1132 Severity : enhancement
1134 Description: Tunables on number of dirty pages in cacche
1135 Details : Allow to set limit on number of dirty pages cached.
1140 Description: client crash on unmount - lock still has references
1141 Details : In some error handling cases it was possible to leak a lock
1142 reference on a client while accessing a file. This was not
1143 harmful to the client during operation, but would cause the
1144 client to crash when the filesystem is unmounted.
1147 Frequency : specific case, rare
1149 Description: ETXTBSY on mds though file not in use by client
1150 Details : ETXTBSY is no longer incorrectly returned when attempting to
1151 chmod or chown a directory that the user previously tried to
1152 execute or a currently-executing binary.
1155 Frequency : extremely rare except on liblustre-based clients
1157 Description: Lustre space not freed when files are deleted
1158 Details : Clean up open-unlinked files after client eviction. Previously
1159 the unlink was skipped and the files remained as orphans.
1164 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1165 Details : In some cases it was possible to send a blocking callback to
1166 a client doing a glimpse, even though that client didn't get
1167 a lock granted. When the glimpse lock is cancelled on the OST
1168 the freed lock is left on the waiting list and corrupted the list.
1171 Frequency : all core dumps
1173 Description: Broke core dumps to lustre
1174 Details : Negative dentry may be unhashed if parent does not have UPDATE
1175 lock, but some callers, e.g. do_coredump, expect dentry to be
1176 hashed after successful create, hash it in ll_create_it.
1178 ------------------------------------------------------------------------------
1180 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1182 * Support for kernels:
1183 2.6.9-42.0.2.EL (RHEL 4)
1184 2.6.5-7.276 (SLES 9)
1185 2.4.21-40.EL (RHEL 3)
1186 2.6.12.6 vanilla (kernel.org)
1190 Frequency : always on RHEL 3
1192 Description: Number of open files grows over time
1193 Details : The number of open files grows over time, whether or not
1194 Lustre is started. This was due to a filp leak introduced
1195 by one of our kernel patches.
1197 ------------------------------------------------------------------------------
1199 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1201 * Support for kernels:
1202 2.6.9-42.EL (RHEL 4)
1203 2.6.5-7.276 (SLES 9)
1204 2.4.21-40.EL (RHEL 3)
1205 2.6.12.6 vanilla (kernel.org)
1210 Bugzilla : 5719, 9635, 9792, 9684
1211 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1212 Details : If a server is under heavy load and cannot reply to new
1213 connection requests before the client resends the (re)connect,
1214 the connection handling code can behave badly if two service
1215 threads are concurrently handing separate (re)connections from
1216 the same client. Add better locking to the connection handling
1217 code, and ensure that only a single connection will be processed
1218 for a given client UUID, even if the lock is dropped.
1220 Severity : enhancement
1222 Description: add TCP zero-copy support to kernel
1223 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1224 sends if the hardware supports scatter-gather and checksumming.
1225 This allows socklnd to do client-write and server-read more
1226 efficiently and reduce CPU utilization from skbuf copying.
1229 Frequency : only if NFS exporting from client
1231 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1232 Details : If setting attributes on a file created under NFS that had
1233 never been opened it would be possible to oops the client
1234 if the file had no objects.
1237 Frequency : always for liblustre
1239 Description: liblustre client does MDS+OSTs setattr RPC for each write
1240 Details : When doing a write from a liblustre client, the client
1241 incorrectly issued an RPC to the MDS and each OST the file was
1242 striped over in order to update the timestamps. When writing
1243 with small chunks and many clients this could overwhelm the MDS
1244 with RPCs. In all cases it would slow down the write because
1245 these RPCs are unnecessary.
1247 Severity : enhancement
1249 Description: allow number of MDS service threads to be changed at module load
1250 Details : It is now possible to change the number of MDS service threads
1251 running. Adding "options mds mds_num_threads={N}" to the MDS's
1252 /etc/modprobe.conf will set the number of threads for the next
1253 time Lustre is restarted (assuming the "mds" module is also
1254 reloaded at that time). The default number of threads will
1255 stay the same, 32 for most systems.
1260 Description: OST crash if filesystem is unformatted or corrupt
1261 Details : If an OST is started on a device that has never been formatted
1262 or if the filesystem is corrupt and cannot even mount then the
1263 error handling cleanup routines would dereference a NULL pointer.
1268 Description: NULL pointer deref in llap_from_page.
1269 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1270 incorrect) mapping if the page was truncated/reclaimed while it was
1271 searched for. Check for this condition and skip such pages when
1272 doing readahead. Introduce extra check to llap_from_page() to
1273 verify page->mapping->host is non-NULL (so page is not anonymous).
1276 Frequency : Sometimes when using sys_sendfile
1278 Description: "page not covered by a lock" warnings from ll_readpage
1279 Details : sendfile called ll_readpage without right page locks present.
1280 Now we introduced ll_file_sendfile that does necessary locking
1281 around call to generic_file_sendfile() much like we do in
1285 Frequency : with certain MDS communication failures at client mount time
1287 Description: NULL pointer deref after failed client mount
1288 Details : a client connection request may delayed by the network layer
1289 and not be sent until after the PTLRPC layer has timed out the
1290 request. If the client fails the mount immediately it will try
1291 to clean up before the network times out the request. Add a
1292 reference from the request import to the obd device and delay
1293 the cleanup until the network drops the request.
1296 Frequency : occasionally during client (re)connect
1298 Description: assertion failure during client (re)connect
1299 Details : processing a client connection request may be delayed by the
1300 client or server longer than the client connect timeout. This
1301 causes the client to resend the connection request. If the
1302 original connection request is replied in this interval, the
1303 client may trip an assertion failure in ptlrpc_connect_interpret()
1304 which thought it would be the only running connect process.
1307 Frequency : only with obd_echo servers and clients that are rebooted
1309 Description: kernel BUG accessing uninitialized data structure
1310 Details : When running an obd_echo server it did not start the ping_evictor
1311 thread, and when a client was evicted an uninitialized data
1312 structure was accessed. Start the ping_evictor in the RPC
1313 service startup instead of the OBD startup.
1315 Severity : enhancement
1316 Bugzilla : 10193 (patchless)
1317 Description: Remove dependency on various unexported kernel interfaces.
1318 Details : No longer need reparent_to_init, exit_mm, exit_files,
1319 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1322 Frequency : rare (only users of deprecated and unsupported LDAP config)
1324 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1325 Details : LDAP apparently contains 'lustreName' attributes instead of
1326 'name'. A simple remapping of the name is sufficient.
1329 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1331 Description: waiting_locks_callback trips kernel BUG if client is evicted
1332 Details : Running with the dump_on_timeout debug flag turned on makes
1333 it possible that the waiting_locks_callback() can try to dump
1334 the Lustre kernel debug logs from an interrupt handler. Defer
1335 this log dumping to the expired_lock_main() thread.
1337 Severity : enhancement
1339 Description: Support NFS exporting on 2.6 kernels.
1340 Details : Implement non-rawops metadata methods for NFS server to use without
1341 changing NFS server code.
1344 Frequency : very rare (synthetic metadata workload only)
1346 Description: two racing renames might cause an MDS thread to deadlock
1347 Details : Running the "racer" program may cause one MDS thread to rename
1348 a file from being the source of a rename to being the target of
1349 a rename at exactly the same time that another thread is doing
1350 so, and the second thread has already enqueued these locks after
1351 doing a lookup of the target and is trying to relock them in
1352 order. Ensure that we don't try to re-lock the same resource.
1355 Frequency : only very large systems with liblustre clients
1357 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1358 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1359 all OSTs in parallel. This allows the eviction of stale liblustre
1360 clients to proceed much faster than if they were done in series,
1361 and also offers similar improvements for other set_info RPCs.
1366 Description: excessive CPU usage during initial read phase on client
1367 Details : During the initial read phase on a client, it would agressively
1368 retry readahead on the file, consuming too much CPU and impacting
1369 performance (since 1.4.5.8). Improve the readahead algorithm
1370 to avoid this, and also improve some other common cases (read
1371 of small files in particular, where "small" is files smaller than
1372 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1377 Description: MDS crash when receiving packet with unknown intent.
1378 Details : Do not LBUG in unknown intent case, just return -EFAULT
1380 Severity : enhancement
1381 Bugzilla : 9293, 9385
1382 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1383 Details : Do not serialize getattr (non-intent version) and statfs.
1386 Frequency : occasional, when OST network is overloaded/intermittent
1388 Description: client evicted by OST after bulk IO timeout
1389 Details : If a client sends a bulk IO request (read or write) the OST
1390 may evict the client if it is unresposive to its data GET/PUT
1391 request. This is incorrect if the network is overloaded (takes
1392 too long to transfer the RPC data) or dropped the OST GET/PUT
1393 request. There is no need to evict the client at all, since
1394 the pinger and/or lock callbacks will handle this, and the
1395 client can restart the bulk request.
1398 Frequency : Always when mmapping file with no objects
1400 Description: client crashes when mmapping file with no objects
1401 Details : Check that we actually have objects in a file before doing any
1402 operations on objects in ll_vm_open, ll_vm_close and
1408 Description: Request leak when working with deleted CWD
1409 Details : Introduce advanced request refcount tracking for requests
1410 referenced from lustre intent.
1412 Severity : Enhancement
1414 Description: Cache open file handles on client.
1415 Details : MDS now will return special lock along with openhandle, if
1416 requested and client is allowed to hold openhandle, even if unused,
1417 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1418 closing files for every read/write openration.
1420 Severity : Enhancement
1422 Description: Cache open negative dentries on client when possible.
1423 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1424 negative dentries on lock revocation.
1429 Description: Remounting a client read-only wasn't possible with a zconf mount
1430 Details : It wasn't possible to remount a client read-only with llmount.
1432 Severity : enhancement
1433 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1434 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1435 MPICH is included to add Lustre-specific ADIO interfaces.
1436 This is based closely on the UFS ADIO layer and only differs
1437 in file creation, in order to allow the OST striping to be set.
1438 This is user-contributed code and not supported by CFS.
1443 Description: extended inode attributes (immutable, append-only) work improperly
1444 when 2.4 and 2.6 kernels are used on client/server or vice versa
1445 Details : Introduce kernel-independent values for these flags.
1447 Severity : enhancement
1450 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1451 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1452 granularity. Now, Fractional MB granularity is be supported,
1453 this is very useful for low memory system.
1455 Severity : enhancement
1457 Description: Getattr by fid
1458 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1459 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1460 have locks and avoids getattr rpc requests for it on every stat.
1463 Frequency : Always, for filesystems larger than 2TB
1465 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1466 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1467 filesystems larger than 2TB to be mounted with mballoc enabled.
1470 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1472 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1473 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1474 filesystems larger than 2TB is mounted then the kernel will
1475 silently corrupt the start of the filesystem. CONFIG_LBD is
1476 enabled for all CFS-supported kernels, but the possibility of
1477 this happening with a modified kernel config exists.
1479 Severity : enhancement
1481 Description: add client O_DIRECT support for 2.6 kernels
1482 Details : It is now possible to do O_DIRECT reads and writes to files
1483 in the Lustre client mountpoint on 2.6 kernel clients.
1485 Severity : enhancement
1487 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1488 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1489 parallel, not waiting for response from every OST before sending
1490 a rpc to the next OST.
1495 Description: setattr vs write race when updating file timestamps
1496 Details : Client processes that update a file timestamp into the past
1497 right after writing to the file (e.g. tar) it is possible that
1498 the updated file modification time can be reset to the current
1499 time due to a race between processing the setattr and write RPC.
1501 Severity : enhancement
1503 Description: Bring 'lfs find' closer in line with regular Linux find.
1504 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1505 -print0 options and obtains all the needed info through the lustre
1508 Severity : enhancement
1510 Description: support up to 1024 configured devices on one node
1511 Details : change obd_dev array from statically allocated to dynamically
1512 allocated structs as they are first used to reduce memory usage
1517 Description: Flush dirty partially truncated pages during truncate
1518 Details : Immediatelly flush partially truncated pages in filter_setattr,
1519 this way we completely avoid having any pages in page cache on OST
1520 and can retire ugly workarounds during writes to flush such pages.
1525 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1526 Details : protect inode from truncation within vfs_unlink() context
1527 just take a reference before calling vfs_unlink() and release it
1528 when parent's i_sem is free.
1533 Description: last_id value checked outside lock on OST caused LASSERT failure
1534 Details : If there were multiple MDS->OST object precreate requests in
1535 flight, it was possible that the OST's last object id was checked
1536 outside a lock and incorrectly tripped an assertion. Move checks
1537 inside locks, and discard old precreate requests.
1540 Frequency : always, if extents are used on OSTs
1542 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1543 Details : OSTs using the extents format would not zero the high 16 bits of
1544 the index physical block number. This is not a problem for any
1545 OST filesystems smaller than 16TB, and no kernels support ext3
1546 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1547 new/modified files) and can be fixed for existing filesystems
1548 with e2fsprogs-1.39-cfs1.
1553 Description: import connection selection may be incorrect if timer wraps
1554 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1555 import connections to be ignored if the 32-bit jiffies counter
1556 wraps. Use a 64-bit jiffies counter.
1559 Frequency : very large clusters immediately after boot
1561 Description: LNET request buffers exhausted under heavy short-term load
1562 Details : If a large number of client requests are generated on a service
1563 that has previously never seen so many requests it is possible
1564 that the request buffer growth cannot keep up with the spike in
1565 demand. Instead of dropping incoming requests, they are held in
1566 the LND until the RPC service can accept more requests.
1569 Frequency : Sometimes during replay
1571 Description: Assertion failure in ll_local_open after replay.
1572 Details : If replay happened on an open request reply before we were able
1573 to set replay handler, reply will become not swabbed tripping the
1574 assertion in ll_local_open. Now we set the handler right after
1575 recognising of open request
1578 Frequency : very rare
1580 Description: kernel reports "badness in vsnprintf"
1581 Details : Reading from the "recovery_status" /proc file in small chunks
1582 may cause a negative length in lprocfs_obd_rd_recovery_status()
1583 call to vsnprintf() (which is otherwise harmless). Exit early
1584 if there is no more space in the output buffer.
1586 Severity : enhancement
1588 Description: clear OBD RPC statistics by writing to them
1589 Details : It is now possible to clear the OBD RPC statistics by writing
1590 to the "stats" file.
1595 Description: Client mtime is not the same on different clients after utimes
1596 Details : In some cases, the client was using the utimes() syscall on
1597 a file cached on another node. The clients now validate the
1598 ctime from the MDS + OSTs to determine which one is right.
1603 Description: Inability to activate failout mode
1604 Details : lconf script incorrectly assumed that in pythong string's numeric
1605 value is used in comparisons.
1608 Frequency : always with multiple stripes per file
1610 Description: Inefficient object allocation for mutli-stripe files
1611 Details : When selecting which OSTs to stripe files over, for files with
1612 a stripe count that divides evenly into the number of OSTs,
1613 the MDS is always picking the same starting OST for each file.
1614 Return the OST selection heuristic to the original design.
1619 Description: mount failures may take full timeout to return an error
1620 Details : Under some heavy load conditions it is possible that a
1621 failed mount can wait for the full obd_timeout interval,
1622 possibly several minutes, before reporting an error.
1623 Instead return an error as soon as the status is known.
1625 Frequency : quota enabled and large files being deleted
1627 Description: releasing more than 4GB of quota at once hangs OST
1628 Details : If a user deletes more than 4GB of files on a single OST it
1629 will cause the OST to spin in an infinite loop. Release
1630 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1635 Description: statfs data retrieved from /proc may be stale or zero
1636 Details : When reading per-device statfs data from /proc, in the
1637 {kbytes,files}_{total,free,avail} files, it may appear
1638 as zero or be out of date.
1641 Frequency : systems with MD RAID1 external journal devices
1643 Description: lconf's call to blkid is confused by RAID1 journal devices
1644 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1645 of returning all block devices that match the journal UUID.
1648 Frequency : always, for aggregate stripe size over 4GB
1650 Description: assertion fails when trying to use 4GB stripe size
1651 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
1652 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
1655 Frequency : always on ppc64
1657 Description: the first write on an ext3 filesystem with mballoc got stuck
1658 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1659 perform endianness conversion.
1661 ------------------------------------------------------------------------------
1663 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1665 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1666 INTEROPERATE with older versions automatically. Please read the
1667 user documentation before upgrading any part of a live system.
1668 * WARNING: Lustre networking configuration changes are required with
1669 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1672 * Support for kernels:
1673 2.6.9-22.0.2.EL (RHEL 4)
1674 2.6.5-7.244 (SLES 9)
1675 2.6.12.6 vanilla (kernel.org)
1678 Severity : enhancement
1679 Bugzilla : 7981/8208
1680 Description: Introduced Lustre Networking (LNET)
1681 Details : LNET is new networking infrastructure for Lustre, it includes
1682 a reorganized network configuration mode (see the user
1683 documentation for full details) as well as support for routing
1684 between different network fabrics. Lustre Networking Devices
1685 (LNDs) for the supported network fabrics have also been
1686 created for this new infrastructure.
1688 Severity : enhancement
1689 Description: Introduced Access control lists
1690 Details : clients can set ACLs on files and directories in order to have
1691 more fine-grained permissions than the standard Unix UGO+RWX.
1692 The MDS must be started with the "-o acl" mount option.
1694 Severity : enhancement
1695 Description: Introduced filesystem quotas
1696 Details : Administrators may now establish per-user quotas on the
1699 Severity : enhancement
1701 Description: Configuration change for the XT3
1702 The PTLLND is now used to run Lustre over Portals on the XT3
1703 The configure option(s) --with-cray-portals are no longer used.
1704 Rather --with-portals=<path-to-portals-includes> is used to
1705 enable building on the XT3. In addition to enable XT3 specific
1706 features the option --enable-cray-xt3 must be used.
1711 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1712 Details : A race between allocating a new llcd and re-getting the llcd_lock
1713 allowed another thread to grab newly-allocated llcd.
1715 Severity : enhancement
1717 Description: 2.6 OST async journal commit and locking fix to improve performance
1718 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1719 2.6 kernels are now async as they already were in 2.4 kernels so
1720 that they can commit concurrently with the network bulk transfer.
1721 For block-allocated files the filter allocation semaphore is held
1722 to avoid filesystem fragmentation during allocation. BKL lock
1723 removed for 2.6 xattr operations where it is no longer needed.
1728 Description: lconf incorrectly determined whether two IP networks could talk
1729 Details : In some more complicated routing and multiple-network
1730 configurations, lconf will avoid trying to make a network
1731 connection to a disjoint part of the IP space. It was doing the
1732 math incorrectly for one set of cases.
1737 Description: Fix for potential infinite loop processing records in an llog.
1738 Details : If an llog record is corrupted/zeroed, it is possible to loop
1739 forever in llog_process(). Validate the llog record length
1740 and skip the remainder of the block on error.
1743 Frequency : occasional (liblustre only)
1745 Description: liblustre could not open files whose last component is a symlink
1746 Details : sysio_path_walk() would incorrectly pass the open intent to
1747 intermediate path components.
1750 Frequency : rare (liblustre only with non-standard tuning)
1751 Bugzilla : 7201 (7350)
1752 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1753 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1754 after it has been released, only do this for referenced locks
1756 Severity : enhancement
1758 Description: specifying an (invalid) directory default stripe_size of -1
1759 would reset the directory default striping
1760 Details : stripe_size -1 was used internally to signal directory stripe
1761 removal, now use "all default" to signal dir stripe removal
1762 as a directory striping of "all default" is not useful
1765 Frequency : common for large clusters running liblustre clients
1767 Description: doing an ls when liblustre clients are running is slow
1768 Details : sending a glimpse AST to a liblustre client waits for every AST
1769 to time out, as liblustre clients will not respond. Since they
1770 cannot cache data we refresh the OST lock LVB from disk instead.
1772 Severity : enhancement
1774 Description: doing an ls at the same time as file IO can be slow
1775 Details : enqueue and other "small" requests can be blocked behind many
1776 large IO requests. Create a new OST IO portal for non-IO
1777 requests so they can be processed faster.
1780 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1782 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1783 created file with mode 000
1784 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1785 allow an ftruncate() as a non-root user to a file with mode 000.
1786 Lustre now respects this flag to disable mode checks when
1787 truncating a file owned by the user
1790 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1793 Description: Revoking locks from clients that went dead or catatonic might take
1795 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1796 cancellation of such locks instant on servers without waiting for
1797 any reply from clients. Clients drops these locks when cancel
1798 notification from server is received without replying.
1801 Frequency : liblustre-only, when liblustre client dies or becomes busy
1803 Description: Doing ls on Linux clients can take a long time with active
1805 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1806 granting such locks to it in the first place if possible. Locks
1807 are taken by proxy on the OST during the read or write and
1808 dropped immediately afterward. Add connect flags handling, do
1809 not grant locks to liblustre clients for glimpse ASTs.
1811 Severity : enhancement
1813 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1814 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1815 sequential reads and behaves badly with applications doing random
1816 reads. Improve it by reading ahead at least read region, and
1817 avoiding excessive large RPC for small reads.
1819 Severity : enhancement
1821 Description: Creating more than 1000 files for a single job may cause a load
1822 imbalance on the OSTs if there are also a large number of OSTs.
1823 Details : qos_prep_create() uses an OST index reseed value that is an
1824 even multiple of the number of available OSTs so that if the
1825 reseed happens in the middle of the object allocation it will
1826 still utilize the OSTs as uniformly as possible.
1831 Description: OST or MDS may oops in ping_evictor_main()
1832 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1833 but doesn't restart at beginning of obd_exports_timed list
1836 Severity : enhancement
1838 Description: improve by-nid export eviction on the MDS and OST
1839 Details : allow multiple exports with the same NID to be evicted at one
1840 time without re-searching the exports list.
1843 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1845 Description: MDS may oops in groups_free()
1846 Details : in rare race conditions a newly allocated group_info struct is
1847 freed again, and this can be NULL. The 2.4 compatibility code
1848 for groups_free() checked for a NULL pointer, but 2.6 did not.
1851 Frequency : common for liblustre clients doing little filesystem IO
1852 Bugzilla : 9352, 7313
1853 Description: server may evict liblustre clients accessing contended locks
1854 Details : if a client is granted a lock or receives a completion AST
1855 with a blocking AST already set it would not reply to the AST
1856 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1859 Frequency : lfs setstripe, only systems with more than 160 OSTs
1861 Description: unable to set striping with a starting offset beyond OST 160
1862 Details : llapi_create_file() incorrectly limited the starting stripe
1863 index to the maximum single-file stripe count.
1866 Frequency : LDAP users only
1868 Description: lconf did not handle in-kernel recovery with LDAP properly
1869 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1871 Severity : enhancement
1873 Description: bind OST threads to NUMA nodes to improve performance
1874 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1875 node and do their allocations there to localize memory access
1877 Severity : enhancement
1879 Description: llmount can determine client NID directly from Myrinet (GM)
1880 Details : the client NID code from gmnalnid was moved directly into
1881 llmount, removing the need to use this or specifying the
1882 client NID explicitly when mounting GM clients with zeroconf
1885 Frequency : if client is started with down MDS
1887 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1888 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1889 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1890 ptlrpc_import_delayed_req() because we didn't check if the
1891 request was interrupted, and we also didn't break out of the
1892 event loop if there was no timeout
1897 Description: data loss during non-page-aligned writes to a single file from
1898 both multiple nodes and multiple threads on one node at same time
1899 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1900 inconsistency led to false short-reads, that were cached and later
1901 used by ->prepare_write() to fill in partially written page,
1902 leading to data loss.
1905 Frequency : always, if lconf --abort_recovery used
1907 Description: lconf --abort_recovery fails with 'Operation not supported'
1908 Details : lconf was attempting to abort recovery on the MDT device and not
1911 Severity : enhancement
1913 Description: remove cleanup logs
1914 Details : replace lconf-generated cleanup logs with lustre internal
1915 cleanup routines. Eliminates the need for client-cleanup and
1918 Severity : enhancement
1920 Description: add support for EAs (user and system) on lustre filesystems
1921 Details : it is now possible to store extended attributes in the Lustre
1922 client filesystem, and with the user_xattr mount option it
1923 is possible to allow users to store EAs on their files also
1925 Severity : enhancement
1927 Description: Add possibility (config option) to show minimal available OST free
1929 Details : When compiled with --enable-mindf configure option, statfs(2)
1930 (and so, df) will return least minimal free space available from
1931 all OSTs as amount of free space on FS, instead of summary of
1932 free spaces of all OSTs.
1934 Severity : enhancement
1936 Description: do not expand extent locks acquired on OST-side
1937 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1938 by server: they are not cached anyway.
1941 Frequency : when mmap is used/binaries executed from Lustre
1943 Description: Unmmap pages before throwing them away from read cache.
1944 Details : llap_shrink cache now attempts to unmap pages before discarding
1945 them (if unmapping failed - do not discard). SLES9 kernel has
1946 extra checks that trigger if this unmapping is not done first.
1951 Description: lconf didn't resolve symlinks before checking to see whether a
1952 given mountpoint was already in use
1955 Frequency : when migrating failover services
1956 Bugzilla : 6395, 9514
1957 Description: When migrating a subset of services from a node (e.g. failback
1958 from a failover service node) the remaining services would
1959 time out and evict clients.
1960 Details : lconf --force (implied by --failover) sets the global obd_timeout
1961 to 5 seconds in order to quickly disconnect, but this caused
1962 other RPCs to time out too quickly. Do not change the global
1963 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1965 Severity : enhancement
1966 Frequency : if MDS is started with down OST
1967 Bugzilla : 9439,5706
1968 Description: Allow startup/shutdown of an MDS without depending on the
1969 availability of the OSTs.
1970 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1971 Add appropriate locking and lov-osc refcounts for safe
1972 cleaning. Add osc abort_inflight calls in case the
1973 synchronize never started.
1976 Frequency : occasional (Cray XT3 only)
1978 Description: root not authorized to access files in CRAY_PORTALS environment
1979 Details : The client process capabilities were not honoured on the MDS in
1980 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1981 been accessed by an authorized user then root was able to access
1982 the file on the local client also. The root user capabilities
1983 are now allowed on the MDS, as this environment has secure UID.
1986 Frequency : occasional
1988 Description: ldiskfs "too long searching" message happens too often
1989 Details : A debugging message (otherwise harmless) prints too often on
1990 the OST console. This has been reduced to only happen when
1991 there are fragmentation problems on the filesystem.
1996 Description: Division by zero in statfs when all OSCs are inactive
1997 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
1998 lov_statfs divides by value returned by lov_get_stripecnt().
2002 Bugzilla : 9489, 3273
2003 Description: First write from each client to each OST was only 4kB in size,
2004 to initialize client writeback cache, which caused sub-optimal
2005 RPCs and poor layout on disk for the first writen file.
2006 Details : Clients now request an initial cache grant at (re)connect time
2007 and so that they can start streaming writes to the cache right
2008 away and always do full-sized RPCs if there is enough data.
2009 If the OST is rebooted the client also re-establishes its grant
2010 so that client cached writes will be honoured under the grant.
2015 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2016 Details : Now I/O RPCs go to different portal number and (presumably) fast
2017 lock requests (and glimses) and other RPCs get their own service
2018 threads pool that should be able to service those RPCs
2021 Severity : enhancement
2023 Description: Ability to exchange lustre version between client and servers and
2024 issue warnings at client side if client is too old. Also for
2025 liblustre clients there is ability to refuse connection of too old
2027 Details : New 'version' field is added to connect data structure that is
2028 filled with version info. That info is later checked by server and
2032 Frequency : rare, liblustre only.
2033 Bugzilla : 9296, 9581
2034 Description: Two simultaneous writes from liblustre at offset within same page
2035 might proceed at the same time overwriting eachother with stale
2037 Details : I/O lock withing llu_file_prwv was released too early, before data
2038 actually was hitting the wire. Extended lock-holding time until
2039 server acknowledges receiving data.
2042 Frequency : extremely rare. Never observed in practice.
2044 Description: avoid generating lustre_handle cookie of 0.
2045 Details : class_handle_hash() generates handle cookies by incrementing
2046 global counter, and can hit 0 occasionaly (this is unlikely, but
2047 not impossible, because initial value of cookie counter is
2048 selected randonly). Value of 0 is used as a sentinel meaning
2049 "unassigned handle" --- avoid it. Also coalesce two critical
2050 sections in this function into one.
2052 Severity : enhancement
2054 Description: allow liblustre clients to delegate truncate locking to OST
2055 Details : To avoid overhead of locking, liblustre client instructs OST to
2056 take extent lock in ost_punch() on client's behalf. New connection
2057 flag is added to handle backward compatibility.
2059 Severity : enhancement
2060 Bugzilla : 4928, 7341, 9758
2061 Description: allow number of OST service threads to be specified
2062 Details : a module parameter allows the number of OST service threads
2063 to be specified via "options ost ost_num_threads={N}" in the
2064 OSS's /etc/modules.conf or /etc/modprobe.conf.
2068 Bugzilla : 6146, 9635, 9895
2069 Description: servers crash with bad pointer in target_handle_connect()
2070 Details : In rare cases when a client is reconnecting it was possible that
2071 the connection request was the last reference for that export.
2072 We would temporarily drop the export reference and get a new
2073 one, but this may have been the last reference and the export
2074 was just destroyed. Get new reference before dropping old one.
2076 Severity : enhancement
2077 Frequency : if client is started with failover MDS
2079 Description: Allow multiple MDS hostnames in the mount command
2080 Details : Try to read the configuration from all specified MDS
2081 hostnames during a client mount in case the "primary"
2084 Severity : enhancement
2086 Description: Stop sending data to evicted clients as soon as possible.
2087 Details : Check if the client we are about to send or are sending data to
2088 was evicted already. (Check is done every second of waiting,
2089 for which l_wait_event interface was extended to allow checking
2090 of exit condition at specified intervals).
2093 Frequency : rare, normally only when NFS exporting is done from client
2095 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2097 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2100 Frequency : always, if srand() is called before liblustre initialization
2102 Description: Liblustre uses system PRNG disturbing its usage by user application
2103 Details : Introduce internal to lustre fast and high-quality PRNG for
2104 lustre usage and make liblustre and some other places in generic
2105 lustre code to use it.
2107 Severity : enhancement
2108 Bugzilla : 9477, 9557, 9870
2109 Description: Verify that the MDS configuration logs are updated when xml is
2110 Details : Check if the .xml configuration logs are newer than the config
2111 logs stored on the MDS and report an error if this is the case.
2112 Request --write-conf, or allow starting with --old_conf.
2114 Severity : enhancement
2116 Description: Handle symlinks in the path when checking if Lustre is mounted.
2117 Details : Resolve intermediate symlinks when checking if a client has
2118 mounted a filesystem to avoid duplicate client mounts.
2123 Description: lconf can hit an error exception but still return success.
2124 Details : The lconf command catches the Command error exception at the top
2125 level script context and will exit with the associated exit
2126 status, but doesn't ensure that this exit status is non-zero.
2131 Description: failure of ptlrpc thread startup can cause oops
2132 Details : Starting a ptlrpc service thread can fail if there are a large
2133 number of threads or the server memory is very fragmented.
2134 Handle this without oopsing.
2137 Frequency : always, only if liblustre and non-default acceptor port was used
2139 Description: liblustre cannot connect to servers with non-default acceptor port
2140 Details : tcpnal_set_default_params() was not called and was therefore
2141 ignoring the environment varaible TCPNAL_PORT, as well as other
2142 TCPNAL_ environment variables
2147 Description: two objects could be created on the same OST for a single file
2148 Details : If an OST is down, in some cases it was possible to create two
2149 objects on a single OST for a single file. No problems other
2150 than potential performance impact and spurious error messages.
2154 Bugzilla : 5681, 9562
2155 Description: Client may oops in ll_unhash_aliases
2156 Details : Client dcache may become inconsistent in race condition.
2157 In some cases "getcwd" can fail if the current directory is
2163 Description: Inode refcounting problems in NFS export code
2164 Details : link_raw functions used to call d_instantiate without obtaining
2165 extra inode reference first.
2169 Bugzilla : 9942, 9903
2170 Description: Referencing freed requests leading to crash, memleaks with NFS.
2171 Details : We used to require that call to ll_revalidate_it was always
2172 followed by ll_lookup_it. Also with revalidate_special() it is
2173 possible to call ll_revalidate_it() twice for the same dentry
2174 even if first occurence returned success. This fix changes semantic
2175 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2176 reference on a request referred from the intent.
2177 ll_intent_release() then releases such a request.
2180 Frequency : rare, normally benchmark loads only
2182 Description: unlinked inodes were kept in memory on the client
2183 Details : If a client is repeatedly creating and unlinking files it
2184 can accumulate a lot of stale inodes in the inode slab cache.
2185 If there is no other client load running this can cause the
2186 client node to run out of memory. Instead flush old inodes
2187 from client cache that have the same inode number as a new inode.
2190 Frequency : SLES9 2.6.5 kernel and long filenames only
2191 Bugzilla : 9969, 10379
2192 Description: utime reports stale NFS file handle
2193 Details : SLES9 uses out-of-dentry names in some cases, which confused
2194 the lustre dentry revalidation. Change it to always use the
2198 Frequency : rare, unless heavy write-truncate concurrency is continuous
2199 Bugzilla : 4180, 6984, 7171, 9963, 9331
2200 Description: OST becomes very slow and/or deadlocked during object unlink
2201 Details : filter_destroy() was holding onto the parent directory lock
2202 while truncating+unlinking objects. For very large objects this
2203 may block other threads for a long time and slow overall OST
2204 responsiveness. It may also be possible to get a lock ordering
2205 deadlock in this case, or run out of journal credits because of
2206 the combined truncate+unlink. Solution is to do object truncate
2207 first in one transaction without parent lock, and then do the
2208 final unlink in a new transaction with the parent lock. This
2209 reduces the lock hold time dramatically.
2212 Frequency : rare, 2.4 kernels only
2214 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2215 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2216 before unmounting their filesystem in order to avoid deadlock.
2217 The kernel_locked() function in 2.4 kernels only checks whether
2218 the kernel lock is held, not whether it is this process that is
2219 holding it as 2.6 kernels do.
2224 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2225 Details : The client ptlrpc code may be trying to reconnect to a down
2226 server before a previous connection attempt has timed out.
2227 Increase the reconnect interval to be longer than the connection
2228 timeout interval to avoid sending duplicate connections to
2232 Frequency : echo_client brw_test command
2234 Description: fix echo_client to work with OST preallocated code
2235 Details : OST preallocation code (5137) didn't take echo_client IO path
2236 into account: echo_client calls filter methods outside of any
2237 OST thread and, hence, there is no per-thread preallocated
2238 pages and buffers to use. Solution: hijack pga pages for IO. As
2239 a byproduct, this avoids unnecessary data copying.
2243 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2244 Description: Client can oops in mdc_commit_close() after open replay
2245 Details : It was possible for the MDS to return an open request with no
2246 transaction number in mds_finish_transno() if the client was
2247 evicted, but without actually returning an error. Clients
2248 would later try to replay that open and may trip an assertion
2249 Simplify the client close codepath, and always return an error
2250 from the MDS in case the open is not successful.
2253 Frequency : rare, 2.6 OSTs only
2255 Description: OST may deadlock under high load on fragmented files
2256 Details : If there was a heavy load and highly-fragmented OST filesystems
2257 it was possible to have all the OST threads deadlock waiting on
2258 allocation of biovecs, because the biovecs were not released
2259 until the entire RPC IO was completed. Instead, release biovecs
2260 as soon as they are complete to ensure forward IO progress.
2262 Severity : enhancement
2264 Description: Support for specifying external journal device at mount
2265 Details : If an OST or MDS device is formatted with an external journal
2266 device, this device major/minor is stored in the ext3 superblock
2267 and may not be valid for failover. Allow detecting and
2268 specifying the external journal at mount time.
2273 Description: Mounting an MDS with pending unlinked files may cause oops
2274 Details : target_finish_recovery() calls mds_postrecov() which returned
2275 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2276 considers this an error and immediately begins cleaning up the
2277 lov, just after starting the mds_lov process
2279 Severity : enhancement
2281 Description: Implement 'lfs df' to report actual free space on per-OST basis
2282 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2283 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2284 usage of objects; '-h' to report in human readable format.
2286 ------------------------------------------------------------------------------
2288 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2295 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2296 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2297 at startup during orphan unlinking. Instead, skip buddy update
2298 before setup, as it will be regenerated after recovery is complete.
2303 Description: If an OST is inactive, its locks might reference stale inodes.
2304 Details : lov_change_cbdata() must iterate over all namespaces, even if
2305 they are inactive to clear inode references from the lock.
2307 Severity : enhancement
2308 Frequency : occasional, if non-standard max_dirty_mb used
2310 Description: Client will block write RPCs if not enough grant
2311 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2312 then the client will block writes while waiting for another RPC
2313 to complete instead of consuming its dirty limit. With change
2314 we get improved performance when max_dirty_mb is small.
2316 Severity : enhancement
2317 Bugzilla : 3389, 6253
2318 Description: Add support for supplementary groups on the MDS.
2319 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2320 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2321 lookups for user supplementary groups into a cache.
2325 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2326 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2327 for liblustre O_EXCL is enforced by MDS.
2331 Description: readdir never returns NULL in liblustre.
2332 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2333 dentry in struct dirent.
2337 Frequency : liblustre only, depends on application IO pattern
2338 Description: liblustre clients evicted if not contacting servers
2339 Details : Don't put liblustre clients into the ping_evictor list, so
2340 they will not be evicted by the pinger ever.
2342 Severity : enhancement
2344 Description: Add ability to evict clients by NID from MDS.
2345 Details : By echoing "nid:$NID" string into
2346 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2347 $NID would be instantly evicted from this MDS and from all active
2348 OSTs connected to it.
2352 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2353 Details : lookup_it_finish() used to query file size from OSTs that was not
2358 Description: service threads change working directory to that of init
2359 Details : Starting lustre service threads may pin the working directory
2360 of the parent thread, making that filesystem busy. Threads
2361 now change to the working directory of init to avoid this.
2365 Frequency : during shutdown only
2366 Description: shutdown with a failed MDS or OST can cause unmount to hang
2367 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2372 Frequency : 2.6 only
2373 Description: chmod/chown may include an extra supplementary group
2374 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2375 supplementary group and if none is specified this is used.
2378 Bugzilla : 5479 (6816)
2380 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2381 Details : If lookup is in progress on a file that is unlinked we might try
2382 to revalidate the inode and fail in revalidate after lookup is
2383 complete and ll_file_open() enqueues the open again but
2384 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2387 Frequency : always, if lconf --abort_recovery used
2389 Description: lconf --abort_recovery fails with 'Operation not supported'
2390 Details : lconf was attempting to abort recovery on the MDT device and not
2393 ------------------------------------------------------------------------------
2395 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2400 Frequency : rare (only unsupported configurations with a node running as an
2402 Bugzilla : 6514, 5137
2403 Description: Mounting a Lustre file system on a node running as an OST could
2405 Details : OSTs now preallocates memory needed to write out data at
2406 startup, instead of when needed, to avoid having to
2407 allocate memory in possibly low memory situations.
2408 Specifically, if the file system is mounted on on OST,
2409 memory pressure could force it to try to write out data,
2410 which it needed to allocate memory to do. Due to the low
2411 memory, it would be unable to do so and the node would
2412 become unresponsive.
2414 Severity : enhancement
2416 Description: Addition of lconf --service command line option
2417 Details : lconf now accepts a '--service <arg>' option, which is
2418 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2420 Severity : enhancement
2422 Description: Failover mode is now the default for OSTs.
2423 Details : By default, OSTs will now run in failover mode. To return to
2424 the old behaviour, add '--failout' to the lmc line for OSTs.
2426 Severity : enhancement
2428 Description: Health checks are now provided for MDS and OSTs
2429 Details : Additional detailed health check information on MSD and OSTs
2430 is now provided through the procfs health_check value.
2433 Frequency : occasional, depends on IO load
2435 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2436 after numerous create/delete cycles
2437 Details : The ext3 inode allocation policy would not allocate new inodes
2438 very well on the OSTs because there are no new directories
2439 being created. Instead we look for groups with free space if
2440 the parent directories are nearly full.
2445 Description: Network or server problems during mount may cause partially
2446 mounted clients instead of returning an error.
2447 Details : The config llog parsing code may overwrite the error return
2448 code during mount error handling, returning success instead
2454 Description: MDS can fail to allocate large reply buffers
2455 Details : After long uptimes the MDS can fail to allocate large reply
2456 buffers (e.g. zconf client mount config records) due to memory
2457 fragmentation or consumption by the buffer cache. Preallocate
2458 some large reply buffers so that these replies can be sent even
2459 under memory pressure.
2463 Frequency : rare (liblustre)
2464 Description: fsx running with liblustre complained that using truncate() to
2465 extend the file doesn't work. This patch corrects that issue.
2466 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2467 ATTR_SIZE and lsm use in llu_setattr_raw.
2471 Frequency : rare, only 2.6 kernels
2472 Description: Unusual file access patterns on the MDS may result in inode
2473 data being lost in very rare circumstances.
2474 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2475 code in the 2.6 kernel caused the inode and/or EA data not to
2476 be read from disk, causing single-file corruption.
2480 Frequency : rare, only 2.6 filesystems using extents
2481 Description: Heavy concurrent write and delete load may cause data corruption.
2482 Details : It was possible under high-load situations to have an extent
2483 metadata block in the block device cache from a just-unlinked
2484 file overwrite a newly-allocated data block. We now unmap any
2485 metadata buffers that alias just-allocated data blocks.
2489 Frequency : filesystems with default stripe_count larger than 77
2490 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2491 Details : lconf specifies an inode size of 4096 bytes when the default
2492 stripe_count is larger than 77. This conflicts with the default
2493 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2494 this case to avoid pinning too much memory for large EAs.
2496 ------------------------------------------------------------------------------
2498 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2503 Frequency : rare (extremely heavy IO load with hundreds of clients)
2505 Description: Client is evicted, gets IO error writing to file
2506 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2507 caused clients to be evicted for AST timeouts. The fixes in
2508 bug 5192 mean we no longer need to have such short AST timeouts
2509 so ldlm_timeout has been increased.
2512 Frequency : occasional during --force or --failover shutdown under load
2513 Bugzilla : 5949, 4834
2514 Description: Server oops/LBUG if stopped with --force or --failover under load
2515 Details : a collection of import/export refcount and cleanup ordering
2516 issues fixed for safer force cleanup
2519 Frequency : only filesystems larger than 120 OSTs
2520 Bugzilla : 5990, 6223
2521 Description: lfs getstripe would oops on a very large filesystem
2522 Details : lov_getconfig used kfree on vmalloc'd memory
2525 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2526 Bugzilla : 6242, 6243
2527 Description: reading from files that had been truncated to a non-zero size
2528 but never opened returned no data
2529 Details : ll_file_read() reads zeros from no-object files to EOF
2534 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2535 Details : There's little harm in aborting MDS/OSS recovery and letting it
2536 try again, so I removed the LASSERT and return an error instead.
2538 Severity : enhancement
2540 Description: New debugging infrastructure for tracking down data corruption
2541 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2542 (b) cover more of the client-side code path, and (c) try to narrow
2543 down where problems occurred
2547 Bugzilla : 3819, 4364, 4397, 6313
2548 Description: Racing close and eviction MDS could cause assertion in mds_close
2549 Details : It was possible to get multiple mfd references during close and
2550 client eviction, leading to one thread referencing a freed mfd.
2552 Severity: : enhancement
2553 Bugzilla : 3262, 6359
2554 Description: Attempts to reconnect to servers are now more aggressive.
2555 Details : This builds on the enhanced upcall-less recovery that was added
2556 in 1.4.2. When trying to reconnect to servers, clients will
2557 now try each server in the failover group every 10 seconds. By
2558 default, clients would previously try one server every 25 seconds.
2563 Description: After recovery, certain operations trigger a failed
2564 assertion on a client.
2565 Details : Failing over an mds, using lconf -d --failover, while a
2566 client was doing a readdir() call would cause the client to
2567 LBUG after recovery completed and the readdir() was resent.
2569 Severity : enhancement
2571 Description: Default groups are now added by lconf
2572 Details : You can now run lconf --group <servicename> without having to
2573 manually add groups with lmc.
2576 Frequency : occasional
2578 Description: Nodes with an elan id of 0 trigger a failed assertion
2581 Frequency : always when accessing e.g. tty/console device nodes
2583 Description: tty and some other devices nodes cannot be used on lustre
2584 Details : file's private_data field is used by device data and lustre
2585 values in there got lost. New field was added to struct file to
2586 store fs-specific private data.
2589 Frequency : when exporting Lustre via NFS
2591 Description: NFSD failed occasionally when looking up a path component
2592 Details : NFSD is looking up ".." which was broken in ext3 directories
2593 that had grown large enough to become hashed.
2596 Frequency : Clusters with multiple interfaces not on the same subnet
2598 Description: Nodes will repeatedly try to reconnect to an interface which it
2599 cannot reach and report an error to the log.
2600 Details : Extra peer list entries will be created by lconf with some peers
2601 unreachable. lconf now validates the peer before adding it.
2604 Frequency : Only if a default stripe is set on the filesystem root.
2606 Description: Setting a default stripe on the filesystem root prevented the
2607 filesystem from being remounted.
2608 Details : The client was sending extra request flags in the root getattr
2609 request and did not allocate a reply buffer for the dir EA.
2612 Frequency : occasional, higher if lots of files are accessed by one client
2613 Bugzilla : 6159, 6097
2614 Description: Client trips assertion regarding lsm mismatch/magic
2615 Details : While revalidating inodes the VFS looks up inodes with ifind()
2616 and in rare cases can find an inode that is being freed.
2617 The ll_test_inode() code will free the lsm during ifind()
2618 when it finds an existing inode and then the VFS later attaches
2619 this free lsm to a new inode.
2623 Bugzilla : 6422, 7030
2624 Description: MDS deadlock between mkdir and client eviction
2625 Details : Creating a new file via mkdir or mknod (starting a transaction
2626 and getting the ns lock) can deadlock with client eviction
2627 (gets ns lock and trying to finish a synchronous transaction).
2630 Frequency : occasional
2631 Description: While starting a server, the fsfilt_ext3 module could not be
2633 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2634 kernels. Previously, lconf would still use the ext3 name
2635 when trying to load modules. Now, it will correctly use
2636 ext3 on 2.4 and ldiskfs on 2.6.
2638 Severity : enhancement
2639 Description: The default stripe count has been changed to 1
2640 Details : The interpretation of the default stripe count (0, to lfs
2641 or lmc) has been changed to mean striping across a single
2642 OST, rather than all available. For general usage we have
2643 found a stripe count of 1 or 2 works best.
2645 Severity : enhancement
2646 Description: Add support for compiling against Cray portals.
2647 Details : Conditional compiling for some areas that are different
2651 Frequency : occasional
2652 Bugzilla : 6409, 6834
2653 Description: Creating files with an explicit stripe count may lead to
2654 a failed assertion on the MDS
2655 Details : If some OSTs are full or unavailable, creating files may
2656 trigger a failed assertion on the MDS. Now, Lustre will
2657 try to use other servers or return an error to the
2661 Frequency : occasional
2663 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2664 caused lock timeout during readahead (since 1.4.2).
2665 Details : Processes doing readahead might match a lock that hasn't been
2666 granted yet if there are overlapping and conflicting lock
2667 requests. The readahead process waits on ungranted lock
2668 (original lock is CBPENDING), while OST waits for that process
2669 to cancel CBPENDING read lock and eventually evicts client.
2671 Severity : enhancement
2673 Description: Initial enabling of flock support for clients
2674 Details : Implements fcntl advisory locking and file status functions.
2675 This feature is provided as an optional mount flag (default
2676 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2677 locking are implemented yet, and those that are are not guaranteed
2678 to be completely correct in production environments.
2679 mount -t lustre -o [flock|noflock] ...
2682 Frequency : occasional
2684 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2685 assertion in the ext3 JBD (journaling) layer.
2686 Details : The b_committed_data struct is protected by the big kernel lock
2687 in 2.4 kernels, serializing journal_commit_transaction() and
2688 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2689 access to this struct. In 2.6 kernels there is finer grained
2690 locking to improve SMP performance of the JBD layer.
2694 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2696 -----------------------------------------------------------------------------
2698 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2700 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2701 versions of Lustre. Please update all servers and clients to
2702 version 1.4.2 or later at the same time. You must also run
2703 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2704 to update the configuration logs.
2706 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2707 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2708 - swab llog records of type '0' so we get proper header size/idx (5861)
2709 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2710 - fix rename of one directory over another leaking an inode (5953)
2711 - avoid SetPageDirty on 2.6 (5981)
2712 - don't re-add just-being-destroyed locks to the waiting list (5653)
2713 - when creating new directories, inherit the parent's custom
2714 striping settings if present parent (3048)
2715 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2716 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2717 - don't hold client locks on temporary worklist from l_lru (5666)
2718 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2719 - automatically evict dead clients (5921)
2720 - Update file size properly in create+truncate+fstat case (6196)
2721 - Do not unhash mountpoint dentries, do not allow removal of
2723 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2724 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2725 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2726 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2727 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2728 - Correctly handle reads of files with no objects (6243)
2729 - lctl recover will also mark a device active if deactivate used (5933)
2731 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2732 - allow --write-conf on an MDS with different nettype than client (5619)
2733 - don't write config llogs to MDS for mounts not from that MDS (5617)
2734 - lconf should create multiple TCP connections from a client (5201)
2735 - init scripts are now turned off by default; run chkconfig --on
2736 lustre and chkconfig --on lustrefs to use them
2737 - upcalls are no longer needed for clients to recover to failover
2739 - add --abort-recovery option to lconf to abort recovery on device
2741 - add support for an arbitrary number of OSTs (3026)
2742 - Quota support protocol changes.
2743 - forward compatibility changes to wire structs (6007)
2744 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2745 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2746 - improved "lustre status" script
2747 - initialize blocksize for non-regular files (6062)
2748 - added --disable-server and --disable-client configure options (5782)
2749 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2750 - Vanilla 2.4.29 support
2751 - increase maximum number of obd devices to 520 (6242)
2752 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2753 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2754 SLES 9 2.6.5 kernels
2755 - limit stripes per file to 160 (the maximum EA size) (6093)
2757 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2760 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2761 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2762 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2763 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2764 - small ext3 extents cleanups and fixes (5733)
2765 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2766 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2767 - avoid needless client->OST connect, fix handle mismatch (5317)
2768 - fix DLM error path that led to out-of-sync client, long delays (5779)
2769 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2770 - fix several locking issues related to i_size (5492,5624,5654,5672)
2771 - don't move pending lock onto export if it is already evicted (5683)
2772 - fix kernel oops when creating .foo in unlinked directory (5548)
2773 - fix deadlock in obdfilter statistics vs. object create (5811)
2774 - use time_{before,after} to avoid timer jiffies wrap (5882)
2775 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2776 - Do not leak request if server was not able to process it (5154)
2777 - If mds_open unable to find parent dir, make that negative lookup(5154)
2778 - don't create new directories with extent-mapping (5909, 5936)
2780 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2781 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2782 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2783 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2784 - support for RHEL4 kernel on i686 (5773)
2785 - provide error messages when incompatible logs are encountered (5898)
2787 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2788 * version 1.4.0.10 (1.4.1 release candidate 1)
2790 - don't keep a lock reference when lock is not granted (4238)
2791 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2792 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2793 - drop import inflight refcount on signal_completed_replay error (5255)
2794 - unlock page after async write error during send (3677)
2795 - handle missing objects in filter_preprw_read properly (5265)
2796 - no transno return for symlink open, don't save no-trasno open (3440)
2797 - don't try to complete elan receive that already failed (4012)
2798 - free RPC server reply state on error (5406)
2799 - clean up thread from ptlrpc_start_thread() on error (5160)
2800 - readahead could read extra page into cache that wasn't ejected (5388)
2801 - prevent races in class_attach/setup/cleanup/detach (5260)
2802 - don't dereference de->d_inode after l_dput of de (5458)
2803 - use "int" for stripe value returned from lock_to_stripe (5544)
2804 - mballoc allocation and error-checking fixes in 2.6 (5504)
2805 - block device patches to fix I/O request sizes in 2.6 (5482)
2806 - look up hostnames for IB nals (5602)
2807 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2808 - don't start multiple acceptors for the same port (5277)
2809 - fix incorrect LASSERT in mds_getattr_name (5635)
2810 - export a proc file for general "ping" checking (5628)
2811 - fix "lfs check" to not block when the MDS is down (5628)
2813 - service request history (4965)
2814 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2815 - create an "evict_client" /proc entry on OSTs, like the MDS has
2816 - fix mount usage message, return errors per mount(8) (5168)
2817 - change grep [] to grep "[]" in tests so they work in more UMLs
2818 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2819 - remove ancient LOV_MAGIC_V0 EA support (5047)
2820 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2821 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2822 - allow readahead allocations to fail when low on memory (5383)
2823 - mmap locking landed again, after considerable improvement (2828)
2824 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2826 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2829 - send OST transaction number in read/write reply to free req (4966)
2830 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2831 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2832 - the watchdog thread now runs as interruptible (5246)
2833 - flock/lockf fixes (but it's still disabled, pending 5135)
2834 - don't use EXT3 constants in llite code (5094)
2835 - memory shortage at startup could cause assertion (5176)
2837 - reorganization of lov code
2838 - single portals codebase
2840 - add extents/mballoc support (5025)
2841 - direct I/O reads in the obdfilter (4048)
2842 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2844 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2847 - send OST transaction number in read/write reply to free req (4966)
2848 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2849 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2850 - the watchdog thread now runs as interruptible (5246)
2851 - handle missing objects in filter_preprw_read properly (5265)
2852 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2853 - drop import inflight refcount on signal_completed_replay error (5255)
2854 - unlock page after async write error during send (3677)
2855 - return original error code on reconstructed replies (3761)
2856 - no transno return for symlink open, don't save no-trasno open (3440)
2858 - add pid to ldlm debugging output (4922)
2859 - bump the watchdog timeouts -- we can't handle 30sec yet
2860 - extra debugging for orphan dentry/inode bug (5259)
2862 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2865 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2866 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2867 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2868 - fix NULL dereference in /proc/sys/portals/routes (4827)
2869 - allow failed mdc_close() operations to be interrupted (4561)
2870 - stop precreate on OST before MDS would time out on it (4778)
2871 - don't send partial-page writes before EOF from client (4410)
2872 - discard client grant for sub-page writes on large-page clients (4520)
2873 - don't free dentries not owned by NFS code, check generation (4806)
2874 - fix lsm leak if mds_create_objects() fails (4801)
2875 - limit debug_daemon file size, always print CERROR messages (4789)
2876 - use transno after validating reply (3892)
2877 - process timed out requests if import state changes (3754)
2878 - update mtime on OST during writes, return in glimpse (4829)
2879 - add mkfsoptions to LDAP (4679)
2880 - use ->max_readahead method instead of zapping global ra (5039)
2881 - don't interrupt __l_wait_event() during strace
2883 - add software watchdogs to catch hung threads quickly (4941)
2884 - make lustrefs init script start after nfs is mounted
2885 - fix CWARN/ERROR duplication (4930)
2886 - return async write errors to application if possible (2248)
2887 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2888 - print NAL number in %x format (4645)
2889 - update barely-supported suse-2.4.21-171 series (4842)
2890 - support for sles 9 %post scripts
2891 - support for building 2.6 kernel-source packages
2892 - support for sles km_* packages
2894 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2897 - ignore -ENOENT errors in osc_destroy (3639)
2898 - notify osc create thread that OSC is being cleaned up (4600)
2899 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2900 - reconstruct ost_handle() like mds_handle() (4657)
2901 - create a new thread to do import eviction to avoid deadlock (3969)
2902 - let lconf resolve symlinked-to devices (4629)
2903 - don't unlink "objects" from directory with default EA (4554)
2904 - hold socknal file ref over connect in case target is down (4394)
2905 - allow more than 32000 subdirectories in a single directory (3244)
2906 - fix blocks count for O_DIRECT writes (3751)
2907 - OST returns ENOSPC from object create when no space left (4539)
2908 - don't send truncate RPC if file size isn't changing (4410)
2909 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2910 - bind to privileged port in socknal and tcpnal (3689)
2912 - rate limit CERROR/CWARN console message to avoid overload (4519)
2913 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2914 - basic mmap support (3918)
2915 - kernel patch series update from b1_4 (4711)
2917 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2920 - avoid crash during MDS cleanup with OST shut down (2775)
2921 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2922 - don't use bad inodes on the MDS (3744)
2923 - dynamic object preallocation to improve recovery speed (4236)
2924 - don't hold spinlock over lock dumping or change debug flags (4401)
2925 - don't zero obd_dev when it is force cleaned (3651)
2926 - print grants to console if they go negative (4431)
2927 - "lctl deactivate" will stop automatic recovery attempts (3406)
2928 - look for existing locks in ldlm_handle_enqueue() (3764)
2929 - don't resolve lock handle twice in recovery avoiding race (4401)
2930 - revalidate should check working dir is a directory (4134)
2932 - don't always mark "slow" obdfilter messages as errors (4418)
2934 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2937 - don't close LustreDB during write_conf until it is done (3860)
2938 - fix typo in lconf for_each_profile (3821)
2939 - allow dumping logs from multiple threads at one time (3820)
2940 - don't allow multiple threads in OSC recovery (3812)
2941 - fix debug_size parameters (3864)
2942 - fix mds_postrecov to initialize import for llog ctxt (3121)
2943 - replace config semaphore with spinlock (3306)
2944 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2945 - don't allow enqueue to complete on a destroyed export (3822)
2946 - down write_lock before checking llog header bitmap (3825)
2947 - recover from lock replay timeout (3764)
2948 - up llog sem before sending rpc (3652)
2949 - reduce ns lock hold times when setting kms (3267)
2950 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2951 - fix NULL deref and obd_dev leak on setup error (3312)
2952 - replace some LBUG about llog ops with error handling (3841)
2953 - don't match INVALID dentries from d_lookup and spin (3784)
2954 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2955 - fix invalid assertion in ptlrpc_set_wait (3880)
2957 - add libwrap support for the TCP acceptor (3996)
2958 - add /proc/sys/portals/routes for non-root route listing (3994)
2959 - allow setting MDS UUID in .xml (2580)
2960 - print the stack of a process that LBUGs (4228)
2962 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2965 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2966 - make sure to unset replay flag from failed open requests (3440)
2967 - if default stripe count is 0, use OST count for inode size (3636)
2968 - update parent mtime/ctime on client for create/unlink (2611)
2969 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2970 - free recovery state on server during a forced cleanup (3571)
2971 - unregister_reply for resent reqs (3063)
2972 - loop back devices mounting and status check on 2.6 (3563)
2973 - fix resource-creation race that can provoke i_size == 0 (3513)
2974 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2975 - more debugging for page-accounting assertion (3746)
2976 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2977 - avoid deadlock after precreation failure (3758)
2978 - fix race and lock order deadlock in orphan handling (3450, 3750)
2979 - add validity checks when grabbing inodes from l_ast_data (3599)
2981 - add /proc/.../recovery_status to obdfilter (3428)
2982 - lightweight CDEBUG infrastructure, debug daemon (3668)
2983 - change default OSC RPC parameters to be better on small clusters
2984 - turn off OST read cache for files smaller than 32MB
2985 - install man pages and include them in rpms (3100)
2986 - add new init script for (un)mounting lustre filesystems (2593)
2987 - run chkconfig in %post for init scripts (3701)
2988 - drop scimac NAL (unmaintained)
2990 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
2993 - clean kiobufs before and after use (3485)
2994 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
2995 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
2996 - change init script to fail more clearly if not run as root (1528)
2997 - allow clients to reconnect during replay (1742)
2998 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
2999 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3000 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3001 - fix cond_resched (3554)
3002 - only evict unfinished clients after recovery (3515)
3003 - allow bulk resend, prevent data loss (3570)
3004 - dynamic ptlrpc request buffer allocation (2102)
3005 - don't allow unlinking open directory if it isn't empty (2904)
3006 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3007 - remove extraneous obd dereference causing LASSERT failure (3334)
3008 - don't use get_cycles() when creating temp. files on the mds (3156)
3009 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3010 - handle EEXIST for set-stripe, set proper directory name (3336)
3012 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3013 - fix ksocknal_fmb_callback() error messages (2918)
3015 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3018 - don't copy lvb into (possibly NULL) reply on error (2983)
3019 - don't deref dentry after dput, don't free lvb on error (2922)
3020 - use the kms to determine writeback rpc length (2947)
3021 - increment oti_logcookies when osc is inactive (2948)
3022 - update client's i_blocks count via lvb messages (2543)
3023 - handle intent open/close of special files properly (1557)
3024 - mount MDS with errors=remount-ro, like obdfilter (2009)
3025 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3026 - don't use cancelling-locks' kms values (2947)
3027 - use highest lock extent for kms, not last one (2925)
3028 - don't dereference ERR_PTR() dentry in error handling path (3107)
3029 - fix thread race in portals_debug_dumplog() (3122)
3030 - create lprocfs device entries at setup instead of at attach (1519)
3031 - common AST error handler, don't evict client on completion race (3145)
3032 - zero nameidata in detach_mnt in 2.6 (3118)
3033 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3034 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3035 - initialize RPC timeout timer earlier for 2.6 (3219)
3036 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3037 - print nal/nid for unknown nid (3258)
3038 - additional checks for oscc recovery before doing precreate (3284)
3039 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3040 - don't crash in mdc_close for bad permissions on open (3285)
3041 - zero i_rdev for non-device files (3147)
3042 - clear page->private before handing to FS, better assertion (3119)
3043 - tune the read pipeline (3236)
3044 - fix incorrect decref of invalidated dentry (2350)
3045 - provide read-ahead stats and refine rpc in flight stats (3328)
3046 - don't hold journal transaction open across create RPC (3313)
3047 - update atime on MDS at close time (3265)
3048 - close LDAP connection when recovering to avoid server load (3315)
3049 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3050 - don't leak open file on MDS after open resend (3325)
3051 - serialize filter_precreate and filter_destroy_precreated (3329)
3052 - loop device shouldn't call sync_dev() for nul device (3092)
3053 - clear page cache after eviction (2766)
3054 - resynchronize MDS->OST in background (2824)
3055 - refuse to mount the same filesystem twice on same mountpoint (3394)
3056 - allow llmount to create routes for mounting behind routers (3320)
3057 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3058 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3059 - fix rare problem with rename on htree directories (3417)
3061 - allow default OST striping configuration per directory (1414)
3062 - fix compilation for qswnal for 2.6 kernels (3125)
3063 - increase maximum number of MDS request buffers for large systems
3064 - change liblustreapi to be useful for external progs like lfsck (3098)
3065 - increase local configuration timeout for slow disks (3353)
3066 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3068 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3071 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3072 - don't overwrite extent policy data in reply if lock was blocked (2901)
3073 - drop filter export grants atomically with removal from device (2663)
3074 - del obd_self_export from work_list in class_disconnect_exports (2908)
3075 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3076 - swab reply message in mdc_close, other PPC fixes (2464)
3077 - fix destroying of named logs (2325)
3078 - overwrite old logs when running lconf --write_conf (2264)
3079 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3080 - fix race in target_handle_connect (2898)
3081 - mds_reint_create() should take same inode create lock (2926)
3082 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3083 - don't close files for self_export to avoid uninitialized obd (2936)
3084 - allow MDS with the same name as client node (2939)
3085 - hold dentry reference for closed log files for unlink (2325)
3086 - reserve space for all logs during transactions (2059)
3087 - don't evict page beyond end of stripe extent (2925)
3088 - don't oops on a deleted current working directory (2399)
3089 - handle hard links to targets without a parent properly (2517)
3090 - don't dereference NULL lock when racing during eviction (2867)
3091 - don't grow lock extents when lots of conflicting locks (2919)
3093 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3096 - account for cache space usage on clients to avoid data loss (974)
3097 - lfsck support in lustre kernel code (2349)
3098 - reduce journal credits needed for BRW writes (2370)
3099 - orphan handling to avoid losing space on client/server crashes
3100 - ptlrpcd can be blocked, stopping ALL progress (2477)
3101 - use lock value blocks to assist in proper KMS, faster stat (1021)
3102 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3103 - recovery for initial connections (2355)
3104 - fixes for mds_cleanup_orphans (1934)
3105 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3106 - block all file creations until orphan recovery completes (1901)
3107 - client remove rq_connection from request struct (2423)
3108 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3109 - recovery timer race (2670)
3110 - mdc_close recovey bug (2532)
3111 - ptlrpc cleanup bug (2710)
3112 - mds timeout on local locks (2588)
3113 - namespace lock held during RPCs (2431)
3114 - handle interrupted sync write properly (2503)
3115 - don't try to handle a message that hasn't been replied to (2699)
3116 - client assert failure during cleanup after abort recovery (2701)
3117 - leak mdc device after failed mount (2712)
3118 - ptlrpc_check_set allows timedout requests to complete (2714)
3119 - wait for inflight reqs when ptlrpcd finishes (2710)
3120 - make sure unregistered services are removed from the srv_list
3121 - reset bulk XID's when resending them (caught by 1138 test)
3122 - unregister_bulk after timeout
3123 - fix lconf error (2694)
3124 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3125 - readahead locks pages, leaves pending causing memory pressure (2673)
3126 - increase OST request buffers to 4096 on large machines (2729)
3127 - fix up permission of existing directories in simple_mkdir (2661)
3128 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3129 - don't assign transno to errored transactions (2742)
3130 - don't delete objects on OST if given a bogus objid from MDS (2751)
3131 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3132 - if rq_no_resend, then timeout request after recovery (2432)
3133 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3134 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3135 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3136 - MDS recovery shouldn't delete objects that it has given out (2730)
3137 - if enqueue arrives after completion, don't clobber LVB (2819)
3138 - don't unlock pages twice when trigger_group_io returns error (2814)
3139 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3140 - don't write pages to disk if there was an error (1450)
3141 - don't ping imports that have recovery disabled (2676)
3142 - take buffered bytes into account when balancing socknal conn (2817)
3143 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3144 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3145 - remove little-endian swabbing of llog records (1987)
3146 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3147 - retry reposting request buffers if they fail (1191)
3148 - grow extent at grant time to avoid granting a revoked lock (2809)
3149 - lock revoke doesn't evict page if covered by a second lock (2765)
3150 - disable VM readahead to avoid reading outside lock extents (2805)
3152 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3153 - updated kernel patches for hp-2.4.20 kernel (2681)
3155 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3158 - fix truncated write corruption (2366)
3159 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3161 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3162 - ensure len(uuid) < 37 in lmc (1171)
3163 - fix ia64 OOPS in llog_test (2255)
3164 - zero end of page at obdfilter for partial page writes (2648)
3165 - don't leave stale dentries around after renames (bug 2428)
3166 - fix timeouts when evicting a client with a single lock held (2642)
3167 - set deadline for the initial HELLO message to drain (2634)
3168 - print out dotted-quad IP addresses in the socknal (2302)
3170 - additional debugging for MDS client eviction problem (2443)
3171 - fix mkfsoptions support for osts (2603, 2604)
3173 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3176 - add series for the vanilla 2.6.0 kernel
3177 - add series for the vanilla 2.4.24 kernel
3178 - add series for a cray x86/64 UL kernel drop
3179 - fix xattr patches for the vanilla 2.4.19 series
3181 - generate true UUIDs in lmc (1171)
3182 - have portals stack dumping break in UML (2466)
3183 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3184 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3185 - wake callback waiting threads on client eviction (2460)
3186 - Add --ptldebug and --subsystem to lmc (1719)
3187 - update assertion to allow safe interrupt allocation
3188 - set rq_no_resend for cancel requests (2432)
3189 - recalculate ptlrpcd timeout after resend (2494)
3190 - call vfs_rmdir when removing pending directories (2368)
3191 - fix renaming a file to itself (2429)
3192 - lmc creates a default one-stripe lov (2454)
3193 - expand procfs space to handle large clusters (2326)
3194 - increase UML stack to avoid overflow
3195 - update lconf's list of debug and subsystem masks
3196 - fix lfs find --obd (2510)
3197 - /proc tunable for disabling filter read caching (2591)
3198 - stop rpm packages from altering slapd.conf (2301)
3199 - disable nagle in the socknal under 0conf (2578)
3200 - choose mds inode size based on stripe count (2572)
3201 - fix kernel-source rpm problems (2516)
3203 - add --disable-doc to avoid pdf generation (2421)
3204 - update documentation, tests, type-os, comments
3205 - avoid format warnings on ia64
3206 - remove the TOE NAL
3207 - tiny code cleanups by removing unused fields
3209 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3212 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3213 - remove the most blatant lies from BUILDING file (2371)
3214 - change default debug level to reasonable production setting
3215 - reduce client side cache size to reduce cache flush time
3216 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3217 - make TCP zerocopy and pinger support enabled by default (2476)
3218 - sync writes completed after process exits caused crashes (2319)
3219 - maintain correct mount count on the MDS (2356)
3220 - backout 1557, because 2316 wasn't really fixed
3221 - better file I/O statistics gathering in /proc
3222 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3223 - another kernel patch to fix zero-copy TCP function export
3224 - don't take duplicate lock when processing re-sent getattr (2420)
3225 - lctl uses obd_self_export instead of creating new conn (2353)
3226 - MDS/OST recovery case which requires object creation asserted (2425)
3227 - move lfs from /usr/sbin to /usr/bin in packages
3228 - fix race between mds_client_add and mds_client_free (2417)
3229 - use kmalloc instead of slabs in portals (2430)
3230 - don't create duplicate records when a failover MDS is present (2442)
3231 - remove unnecessary mount age check (2332)
3232 - don't remove directory inodes from locks prematurely (2451)
3233 - don't break if MDS service name is the same as hostname (2103)
3234 - fix races in client write RPC generation when cache full (2482)
3236 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3239 - remove now-unused request->rq_obd (278)
3240 - if an allocation fails, print out how much memory we've used (1933)
3241 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3242 - add the "configurable stack size" patch to most series files (1256)
3243 - ability to write large log records, for 100+ OST configs (2306)
3244 - fix NULL deref when filter_prep fails (2314)
3245 - fix operator precedence error in filter_sync
3246 - dynamic allocation of socknal TX descriptors (2315)
3247 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3248 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3249 - ensure that CWARN messages in Portals always get to the syslog
3250 - __init/__exit are not for prototype decls (ldlm_init/exit)
3251 - x86-64 compile warning fixes
3252 - fix gateway LMC keyword conflict (2318)
3253 - fix MDS lock inversions in getattr/reint paths (1844)
3254 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3255 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3256 - DLM race condition prevented some lock evictions (2328)
3257 - ENOMEM detection and retry on socknal sends (2230)
3258 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3259 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3260 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3261 - better error messages when a client is rejected during recovery (1505)
3262 - avoid cancelling locks which were never granted, after failure (2330)
3263 - fix i_sem/journal inversion in mds_client_add (2333)
3264 - fix truncate/getattr lock cycle deadlock (2334)
3265 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3266 - fix two rare exit paths which could leak an l_lock() ref (2321)
3267 - fencepost error in MDS/OST orphan recovery (2226)
3268 - make log record alignment 8 bytes (1988)
3269 - lstripe now fails when requested offset > ost_count (2237)
3270 - ensure that all kernel series have a complete list.h (1607)
3271 - fix crashes in special-file operations (2316)
3272 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3274 - allow configurable automake binary, for testing new versions
3275 - small update to the lfs documentation
3277 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3279 * fix negative export reference count in fsfilt_sync (2312)
3281 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3282 * release candidate 0.9.1
3284 - orphans are moved into the PENDING directory for possible recovery
3285 - replayed opens now open by fid for orphan/rename safety (1042)
3286 - last close of an orphan inode generates a transno (683)
3287 - chdir() and mount() now pin the directory entry (1020)
3288 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3289 - discard very old requests without processing them (1502)
3290 - remove some common, well-understood CERRORs (1505)
3291 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3292 - clear "grant" flags in OST replies until OST grant code lands (1644)
3293 - fix read performance by not clobbering i_blksize on client (1598)
3294 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3295 - make LRU size a /proc tunable, clears locks when reduced (707)
3296 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3297 - server locks take references on exports now (1558)
3298 - build fixes for 2.4.20-rh trees (1663)
3299 - return an error from lov_create if all OSCs are inactive (1751)
3300 - fix import levels when a reconnect happens without a timeout (1597)
3301 - exit early from mds_open if we get a lookup error (1749)
3302 - partial page read at EOF wouldn't wait for disk before sending (1642)
3303 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3304 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3305 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3306 - write/truncate lock inversion (1639)
3307 - Don't auto-load obdclass, portals modules during cleanup (1495)
3308 - fix timestamps from jumping to "now" (1763)
3309 - extra journal assertions (1648)
3310 - add an extra multiunlink test (1771)
3311 - fix read_record/write_record API (1776)
3312 - fix leak of offset_extent, possible incorrect i_size later (1772)
3313 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3314 - Debugging for the kqswnal_get_idle_tx problems (1820)
3315 - Allow recovery to be attempted multiple times (1536)
3316 - Write out MDS last_rcvd file after it is first created (1600)
3317 - Fix tx_descriptor leak in failed transmit situations (1827)
3318 - ext3 journaling fixes for assertion failure after IO error (1871)
3319 - class_export_put() on freed export after completion AST error (1896)
3320 - Fix revalidate looping in VFS (1322)
3321 - Don't access a freed export during MDS_REINT timeout (1521)
3322 - Add open-unlink recovery support on the MDS (1673,1764)
3323 - Return an error if no MDS data was read from last_rcvd (1946)
3324 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3325 - Don't setup a disk device that doesn't match exported UUID (317)
3326 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3327 - avoid committing NULL handle in force close
3328 - local.sh is now a one-stripe LOV configuration
3329 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3330 - fix ext3 htree duplicate directory entry corruption (1516)
3331 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3332 - update to new LOV EA format (2097)
3333 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3335 2003-06-15 Phil Schwan <phil@clusterfs.com>
3338 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3339 - per-import/export recovery handling (958, 931, 959)
3340 - multiple last-rcvd slots, for serving multiple FSes (949)
3341 - connections are again shared between multiple imp/exports (963, 964)
3342 - "umount -f" would hang if any requests needed to be sent (393, 978)
3343 - avoid pinning large req buffer by copying for queued messages (989)
3344 - add "uuid" to "lctl device" command to help upcalls (991)
3345 - "open" RPCs with transnos would confuse recovery counters (1037)
3346 - do proper endian conversion of all wire messages (288, 340, 891)
3347 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3348 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3349 - fix ptlrpc_connection leak in target_handle_connect (1174)
3350 - fix import refcounting bug in OST and MDS cleanup (1134)
3351 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3352 - fix very unlikely obd_types race condition (501)
3353 - remove osc_open hack for echo_client (1187)
3354 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3355 - a failure in read_inode2 leads to deadlock (1139)
3356 - cancel ack-locks as soon as transaction is committed (1072)
3357 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3358 - make sure to commitrw after any preprw to avoid deadlock (1162)
3359 - failing to execute a file in a lustre FS would lock inode (1203)
3360 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3361 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3362 - obd_uuid2tgt would walk off the end of the list (1255)
3363 - on IA64 the osc would give portals incorrect bulk size (1258)
3364 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3365 - fix lock inversion caused by new llite matching code (1282)
3366 - limit the number of dirty pages on a client to 10MB (1286)
3367 - timed out locks were not being corrected cancelled (1289)
3368 - fix O_DIRECT above 4GB on IA-32 (1292)
3369 * major user-visible changes
3370 - fail out/fail over policy now controlled by the upcall (993)
3372 - add OBD_PING to check server availability and failure (954)
3373 - lustre messages are now sent in sending host order (288, 340, 891)
3374 - add eadatalen to MDS getattr reply (340)
3375 - OST read replies may contain second buffer, with per-page status (593)
3377 2003-03-11 Phil Schwan <phil@clusterfs.com>
3380 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3381 - failed open()s could cause deadlock; fixed (867, 869)
3382 - stop cancelling OST locks when files are closed (481)
3383 - overlapping XID spaces caused network corruption (851, 853)
3384 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3385 - setattr_raw added, to do single-RPC, server-side setattrs
3386 - lmc/lconf syntax change for OST UUIDs
3387 - fix crashy race condition between ptlrpc_free_req and osc_close
3388 - don't use request in mdc_enqueue if we hit a timeout (889)
3389 - don't set the inode i_size for regular files from the MDS (896)
3390 - handle out of order completion AST (842)
3391 - don't LBUG if a lock request times out after receiving AST (913)
3392 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3393 - if a bad lock AST arrives, send an error instead of dropping entirely
3394 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3395 - fix leak in bulk IO when only partially completed (899, 900, 926)
3396 - fix O_DIRECT for ia64 (55)
3397 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3398 - C-z after timeout could hang a process forever; fixed (977)
3400 - client-side I/O cache (678, 924, 929, 941, 970)
3402 - READPAGE and SETATTRs which don't take server-side locks get
3405 2003-02-11 Phil Schwan <phil@clusterfs.com>
3408 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3409 - Fix fsfilt_extN_readpage() to read a full page of directory
3410 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3411 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3412 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3413 - Most intent operations were removed, in favour of a new RPC mode
3414 that does a single RPC to the server and bypasses most of the VFS
3415 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3416 - Aggressively cancel local locks on DLM servers
3417 - mds_reint_unlink sends EA to the client if it's the last nlink.
3418 client uses that EA to unlink OST objects.
3419 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3420 - recursive symlinks were fixed (439)
3421 - fixed NULL deref in DEBUG_REQ
3422 - filter_update_lastobjid no longer calls sync, which annoyed extN
3423 - fixed multi-client small-writes to a single file problem (445)
3424 - fixed mtime updates during file writes (607)
3425 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3426 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3427 - fixed Linux OST size reporting problem (444, 656)
3428 - OST now updates object mtime with writes or setattr (607, 619)
3429 - client verifies file size before zeroing page past EOF (445)
3430 - OST now writes last allocated objid to disk with allocation (108)
3431 - LOV on echo now works (409)
3433 - mds_reint_unlink sends a new buffer, with the EA included. this
3434 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3435 set if a regular file was being unlinked, and it was the last link
3436 - use PtlGet from the target for bulk writes (315)
3437 - OST now updates object mtime with writes or setattr (607, 619)
3438 - LDLM now has a grant-time callback to revalidate locked items, if
3440 - Many MDS operations were reorganized to combat race conditions
3442 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3443 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3445 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3448 - Fully reactivate OST imports after reconnection (512, others)
3449 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3450 - More graceful error handling for truncating on dead OST (515)
3451 - Don't error out unless we're actually accessing dead stripes (474)
3452 - Fix garbage sizes when stripes are missing (410)
3453 - LRU counters were broken, causing constant lock purge (433, 432)
3454 - garbage on read from stripes with failed OSTs (441)
3455 - mark OSCs as active before reconnecting during recovery (438)
3456 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3457 - lfind did not preserve OST order in output (443)
3458 - symlinks cause hung clients, incorrect data (439)
3459 - stop dereferencing request after dropping refcount (457)
3460 - don't LASSERT(spin_is_locked) on non-SMP (455)
3461 - fixes for many rename() bugs
3462 - fstat didn't correctly synchronize attributes (399)
3463 - server must handle lock cancellation during blocking AST prep (487)
3464 - bulk descriptors were free()d too soon (511)
3465 - fix paths in lconf, which would load incorrect modules (451, 507)
3466 - fix confusing lconf 'host not found' error message (386)
3467 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3468 - fix race condition in mdc_blocking_ast() for inode access (526)
3469 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3470 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3471 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3472 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3473 - fix lstripe in several ways (kernel side) (527)
3474 - fix request leak in ldlm_cli_enqueue (262)
3475 - incorrect OSC was marked inactive after OST failure
3476 - call mds_fs_cleanup before unmounting filesystem (524)
3477 - fix races between taking ns_lock and ldlm_lock_change_resource
3478 - fix races updating LOV export open file list
3479 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3480 - fix recovery NULL deref in ldlm_cli_cancel_unused
3481 - fix some DLM races by using new hash table for lock handles (419)
3482 - permit the client to specify desired inodes, at replay
3483 - duplicate requests when we queue them for replay reintegration
3484 - fix last_rcvd offset calculation
3485 - sync after each recovered transaction, so we always make progress
3486 - never, not always, ERESTART requests without transnos
3487 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3489 - skip replay if the MDS says that the client is already connected
3490 - don't check for a recovery-enabled export to match lctl's UUID
3491 - don't INC_USE_COUNT for phantom exports
3492 - don't crash when cleaning up phantom exports (567)
3493 - don't double-finish or set replay data for errored mdc_open requests
3494 - abort requests when they time out, so we don't get old replies
3495 - send/receive replies for AST messages again
3496 - if the client says that it doesn't have the lock, cancel it on the
3498 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3499 mark it as destroyed, it will all work out when decref is called
3500 - fix module use counts (22, 581)
3502 - ASTs now expect a reply (server cancels lock on error reply)
3504 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3507 - fix many simultaneous client startup (392)
3508 - fix dentry->d_it clobbering
3509 - credentials weren't being shipped for readdir/getattr operations
3510 - remove invalid assertions triggered during some concurrent MD
3512 - proper Lustre versions added (336, 389)
3513 - fix memory leak for create error case (398)
3514 - fix LOV locking bug that would get cli/srv out of sync
3515 - fix echo client over LOV (409)
3516 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3517 - fix double-O_EXCL intent crash (424)
3518 - avoid sending multiple lock CANCELs (352)
3520 - MDS can do multi-client recovery (modulo bugs in new code)
3522 - many updates, edits, cleanups
3524 2002-11-18 Phil Schwan <phil@clusterfs.com>
3527 - fix null d_it dereference (346)
3528 - fix full OST/dbench hang (333)
3529 - fix permission problem with file removal (286)
3530 - fix removal of OSCs from LOV when they fail
3531 - fix NULL deref during bulk timeout (214)
3532 - fix problems related to multiple filesystems on one MDS (241)
3533 - fixed serious subtle metadata locking bugs
3534 - free locks on clients when inodes are removed due to memory
3536 - fix inode pointer in lock data (285)
3537 - partial support for multiple MDS on a single host (241)
3538 - data locks weren't cancelled at clear_inode time (290, 311)
3539 - intent locks could lead to unbounded lock growth (205)
3540 - added a maximum lock count, an LRU list, and a flusher
3541 - fix multiple rename (365)
3542 - properly abstracted the echo client
3543 - OSC locked 1 byte too many; fixed
3544 - rewrote brw callback code:
3545 - fixed recovery bugs related to LOVs (306)
3546 - fixed too-many-pages-in-one-write crash (191)
3547 - fixed (again) crash in sync_io_timeout (214)
3548 - probably fixed callback-related race (385)
3550 - Add capability to MDS protocol
3551 - LDLM cancellations and callbacks on different portals
3553 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3556 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3557 - defer open object destruction to close time (601981, 138)
3558 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3559 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3560 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3561 - don't call obd_setattr in ll_file_release for destroyed objects
3562 * protocol change to lustre_msg: move |version| and add |flags|
3563 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3564 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3565 * added replay of create, unlink, link and rename operations during
3566 MDS failover; recovery should be much more robust now
3567 * remove failed OSCs from LOVs (only lov_create uses this so far)
3568 * the lustre-HOWTO was brought (more) up to date (582544)
3570 2002-10-23 Phil Schwan <phil@clusterfs.com>
3573 - in-use dentries weren't being reused properly (617851)
3574 - prevent multiple LDLM setup (599178)
3575 - fix LOV size calculations for truncate (617853)
3576 - fix client handling of MDS intent errors (POSIX)
3577 - fix permission bug in lovstripe.c test (624321)
3578 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3579 - truncate past end of file could corrupt data
3580 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3581 - a race in recovery could return ETIMEDOUT to apps (623947)
3582 - building outside the source directory was fixed
3583 * the lustre-HOWTO was brought (more) up to date (582544)
3584 * major progress was made on recovery functionality
3586 2002-10-10 Phil Schwan <phil@clusterfs.com>
3589 - recovery deadlock fix
3590 - rm -rf causes LBUG fix (617817)
3591 - file open by multiple tasks fix (618962)
3592 - directory permissions bugs (602707 and 620007)
3593 - journal_stop fixed with locking (611313)
3594 - O_APPEND failures resolved (618273, perhaps 614459)
3595 - lconf PATH fix (619770)
3596 - IA64 build fix (621450)
3597 - RPC buffer sizes scale with amount of memory
3599 2002-10-01 Phil Schwan <phil@clusterfs.com>
3602 - locks would be cancelled without throwing away data pages,
3603 resulting in inconsistent data (605627)
3604 - inode attributes were not always being refreshed (605627, 612449)
3605 - lconf now continues to cleanup after lctl reports an error
3606 - MDS now enforces user permissions (602707)
3607 - lprocfs cleanup fixed, but not yet enabled (614157)
3608 - fixed infinite server hang, should a client not respond to an AST
3609 - avoid going into recovery if user calls readlink() with a buffer
3610 that's too small (613941)
3611 - AST RPCs no longer require replies (614867) -- this may be changed
3612 - don't crash server if client sends an IOV that's too big (611336)
3613 - fixed lock conversion deadlock (611892)
3614 - fixed the following of symlinks (614622)
3615 * recovery: the server can remove locks from a client that dies, other
3616 clients can make progress
3617 * more extN patch fixes
3618 * compile-time configurable ptlrpc buffer allocations
3620 - collaborative read cache document
3621 - Lustre Lite Performance CDR document-in-progress
3623 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3626 - fix typo in patch-2.4.18
3628 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3631 - clear ptlrpc request each time in handle_incoming_request()
3632 - unlink of files now destroys the object on the OST
3634 2002-09-19 Peter Braam <braam@clusterfs.com>
3636 * add hard link support
3637 * change obdfile creation method
3638 * kernel patch changed
3640 2002-09-19 Peter Braam <braam@clusterfs.com>
3643 - stack overflow bug in extN fixed
3645 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3647 * documentation updates
3648 - add man pages for config tools
3649 - update tests/README to describe testing with new config tools
3650 - finish metadata API descriptions
3651 * bug fixes and cleanups
3652 - statfs workaround for 16TB limit
3653 - LOV stripe allocation improved, can stripe on subset of OSTs
3654 - LOV file size/IO offset was wrong for files > 4GB in size
3655 - object EA data was being dropped, caused files to be unreadable
3656 - memory overflow with non-LOV OST caused memory corruption
3657 - fixed regression tests to work with new config tools, obdfilter
3658 - fixed bug when directory size became larger than 1 block
3659 - fixed bug (for single client case) when PWD was deleted
3660 - invalidate local directory pages when doing intent-based ops
3661 - avoid LDLM oops when lock callback contained bad data
3663 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3665 * documentation updates
3666 * bug fixes and cleanups
3667 - configuration tools
3670 - 64-bit compile warnings
3671 - 64-bit internal statfs data
3673 * test_brw on persistent OST devices
3675 * lprocfs (disabled)
3677 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3679 * documentation updates
3680 * bug fixes and cleanups
3681 * configuration tools
3683 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3686 * many small fixes to 0_5_4
3687 * io/network handling
3688 * thinkos in MDS operations
3690 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3693 * crucial basic fixes to 0.5.3
3694 * IOR, Iozone work over Elan
3697 2002-08-07 Phil Schwan <phil@clusterfs.com>
3698 * version 0_5_3, our first alpha
3699 * we use the new Portals iovs
3700 * documentation updates
3701 * bug fixes and cleanups
3702 * small changes in the DLM wire protocol
3704 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3705 * version 0_5_1 with some initial stability,
3706 * locking on MD and file I/O.
3707 * documentation updates
3708 * several bug fixes since 0.5.0
3709 * small changes in wire protocol
3711 2002-07-18 Phil Schwan <phil@clusterfs.com>
3713 * delivered as Lustre Light Alpha
3714 * fixed a crash after handling invalid MDS requests
3715 * fixed directory pages for architectures with non-4k pages sizes
3717 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3718 * release version v0_4_4
3719 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3721 * Fixes a number of interruption problems with OST operations.
3722 * Update documentation for portals header changes
3723 * Move all wire protocol structs/defines to lustre_idl.h
3724 * Fixes symlink length bug.
3725 * Add tcpdump to repository.
3727 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3728 * release version v0_4_3
3729 * Fixes statfs for inodes on extN.
3730 * Fixes bug in runtests which would delete /etc/hosts.
3731 * Use 64-bit object IDs wherever possible (not into VFS though)
3732 Remove ost_get_info, which is unused by lustre, and out of date.
3734 2002-07-03 Peter Braam <braam@clusterfs.com>
3735 * release version v0_4_2 Fixes a lookup error (type not passed)
3736 * move forward to head of Portals
3737 * move forward to latest Lustre kernel
3739 2002-06-25 Peter Braam <braam@clusterfs.com>
3740 * release version v0_4_1. Hopefully stable on single node use.
git://git.whamcloud.com / fs / lustre-release.git / blob