1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4 2.4.21-47.0.1.EL (RHEL 3)
6 2.6.9-42.0.10.EL (RHEL 4)
7 2.6.12.6 vanilla (kernel.org)
8 2.6.16.27-0.9 (SLES 10)
9 * Client support for unpatched kernels:
10 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
11 2.6.9-42.0.10.EL (RHEL 4)
12 2.6.16 - 2.6.21 vanilla (kernel.org)
13 * Recommended e2fsprogs version: 1.39.cfs7
14 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
18 Frequency : 2.6.18 servers only
20 Description: ll_kern_mount() doesn't release the module reference
21 Details : The ldiskfs module reference count never drops down to 0
22 because ll_kern_mount() doesn't release the module reference.
27 Description: server LBUG when using old ost_num_threads parameter
28 Details : Accept the old ost_num_threads parameter but warn that it
29 is deprecated, and fix an off-by-one error that caused an LBUG.
34 Description: Transient SCSI error results in persistent IO issue
35 Details : iobuf->dr_error is not reinitialized to 0 between two
39 Frequency : sometimes when underlying device returns I/O errors
41 Description: OSTs not going read-only during write failures
42 Details : OSTs are not remounted read-only when the journal commit threads
43 get I/O errors because fsfilt_ext3 calls journal_start/stop()
44 instead of the ext3 wrappers.
48 Description: poor connect scaling with increasing client count
49 Details : Don't run filter_grant_sanity_check for more than 100 exports
50 to improve scaling for large numbers of clients.
53 Frequency : SLES10 only
55 Description: sanity-quota.sh quotacheck failed: rc = -22
56 Details : Quotas cannot be enabled on SLES10.
59 Frequency : liblustre clients only
61 Description: getdirentries does not give error when run on compute nodes
62 Details : getdirentries does not fail when the size specified as an argument
63 is too small to contain at least one entry
65 Severity : enhancement
67 Description: Add LNET router traceability for debug purposes
68 Details : If a checksum failure occurs with a router as part of the
69 IO path, the NID of the last router that forwarded the bulk data
70 is printed so it can be identified.
75 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
76 Details : Due to a race condition, liblustre clients were occasionally
79 Severity : enhancement
81 Description: lfs setstripe use optional parameters instead of postional
84 Severity : enhancement
86 Description: Nanosecond timestamp support for ldiskfs
87 Details : The on-disk ldiskfs filesystem has added support for nanosecond
88 resolution timestamps. There is not yet support for this at
89 the Lustre filesystem level.
92 Frequency : during server recovery
94 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
95 Details : request with rq_no_resend flag not awake l_wait_event if they get a
99 Frequency : nfs export on patchless client
101 Description: connectathon hang when test nfs export over patchless client
102 Details : Disconnected dentry cannot be found with lookup, so we do not need
103 to unhash it or make it invalid
106 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
107 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
108 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
113 Description: open req refcounting wrong on reconnect
114 Details : If reconnect happened between getting open reply from server and
115 call to mdc_set_replay_data in ll_file_open, we will schedule
116 replay for unreferenced request that we are about to free.
117 Subsequent close will crash in variety of ways.
118 Check that request is still eligible for replay in
119 mdc_set_replay_data().
124 Description: disable writes to filesystem when reading health_check file
125 Details : the default for reading the health_check proc file has changed
126 to NOT do a journal transaction and write to disk, because this
127 can cause reads of the /proc file to hang and block HA state
128 checking on a healthy but otherwise heavily loaded system. It
129 is possible to return to the previous behaviour during configure
130 with --enable-health-write.
132 Severity : enhancement
134 Description: 64-bit inode version
135 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
136 to the inode. This will be required for version-based recovery.
141 Description: MDS fails to start if a duplicate client export is detected
142 Details : in some rare cases it was possible for a client to connect to
143 an MDS multiple times. Upon recovery the MDS would detect this
144 and fail during startup. Handle this more gracefully.
146 Severity : enhancement
148 Description: Add -o localflock option to simulate old noflock
150 Details : This will achieve local-only flock/fcntl locks
156 Description: log_commit_thread vs filter_destroy race leads to crash
157 Details : Take import reference before releasing llog record semaphore
162 Description: Wrong request locking in request set processing
163 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
164 imp_delayed_list, in this place should be used imp_lock.
167 Frequency : when reconnection
169 Description: Grant Leak when osc reconnect to OST
170 Details : When osc reconnect ost, OST(filter) should check whether it
171 should grant more space to client by comparing fed_grant and
172 cl_avail_grant, and return the granted space to client instead
173 of "new granted" space, because client will call osc_init_grant
174 to update the client grant space info.
177 Frequency : when client reconnect to OST
179 Description: Grant Leak when osc do resend and replay bulk write
180 Details : When osc reconnect to OST, OST(filter)should clear grant info of
181 bulk write request, because the grant info while be sync between
182 OSC and OST when reconnect, and we should ignore the grant info
183 these of resend/replay write req.
188 Description: Grant space more than avaiable left space sometimes.
189 Details : When then OST is about to be full, if two bulk writing from
190 different clients came to OST. Accord the avaliable space of the
191 OST, the first req should be permitted, and the second one
192 should be denied by ENOSPC. But if the seconde arrived before
193 the first one is commited. The OST might wrongly permit second
194 writing, which will cause grant space > avaiable space.
197 Frequency : when client is evicted
199 Description: Grant might be wrongly erased when osc is evicted by OST
200 Details : when the import is evicted by server, it will fork another
201 thread ptlrpc_invalidate_import_thread to invalidate the
202 import, where the grant will be set to 0. While the original
203 thread will update the grant it got when connecting. So if
204 the former happened latter, the grant will be wrongly errased
205 because of this race.
210 Description: Checking Stale with correct fid
211 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
212 to check whether it is stale, because sometimes, we want the
213 enqueue happened anyway, and op_data.fid2 will not be initialized.
215 Severity : enhancement
217 Description: update patchless client
218 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
221 Frequency : only with 2.4 kernel
223 Description: random memory corruption
224 Details : size of struct ll_inode_info is to big for union inode.u and this
225 can be cause of random memory corruption.
230 Description: Memory leak in recovery
231 Details : Lov_mds_md was not free in an error handler in mds_create_object.
232 It should also check obd_fail before fsfilt_start, otherwise if
233 fsfilt_start return -EROFS,(failover mds during mds recovery).
234 then the req will return with repmsg->transno = 0 and rc = EROFS.
235 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
236 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
237 be freed no matter whether fsfilt_commit success or not.
240 Frequency : only with huge count clients
242 Description: Prevents from taking the superblock lock in llap_from_page for
244 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
245 for llap_from_page call in ll_removepage prevents from taking the
246 superblock lock for a soon died page.
251 Description: Not check open intent error before release open handle
252 Details : in some rare cases, the open intent error is not checked before
253 release open handle, which may cause
254 ASSERTION(open_req->rq_transno != 0), because it tries to release
255 the failed open handle.
260 Description: Set cat log bitmap only after create log success.
261 Details : in some rare cases, the cat log bitmap is set too early. and it
262 should be set only after create log success.
266 Description: Accessing a block bevice can re-enable I/O when Lustre is
267 tearing down a device.
268 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
273 Description: service threads may hog cpus when there are a lot of requests
275 Details : Insert cond_resched to give other threads a chance to use some of
281 Description: the cat log was not initialized in recovery
282 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
283 unlink log on mds will not be initialized until mds post
284 recovery. And also in mds post recovery, the unlink log will
285 initialization will be done asynchronausly, so there will be race
286 between add unlink log and unlink log initialization.
290 Description: brw_stats were being printed incorrectly
291 Details : brw_stats were being printed as log2 but all of them were not
292 recorded as log2. Also remove some code duplication arising from
293 filter_tally_{read,write}.
296 --------------------------------------------------------------------------------
298 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
303 Frequency : on some architectures
305 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
306 Details : Uninitialized flags sometimes cause configuration commands to
310 Frequency : patchless clients only
312 Description: missing __iget() symbol export
313 Details : The __iget() symbol export is missing. To avoid the need for
314 this on patchless clients the deathrow inode reaper is turned
315 off, and we depend on the VM to clean up old inodes. This
316 dependency was during via the fix for bug 12181.
318 --------------------------------------------------------------------------------
320 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
322 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
323 INTEROPERATE with older versions automatically. In many cases a
324 special upgrade step is needed. Please read the
325 user documentation before upgrading any part of a 1.4.x system.
326 * WARNING: Lustre configuration and startup changes are required with
327 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
329 * Support for kernels:
330 2.4.21-47.0.1.EL (RHEL 3)
332 2.6.9-42.0.10.EL (RHEL 4)
333 2.6.12.6 vanilla (kernel.org)
334 2.6.16.27-0.9 (SLES10)
335 * Client support for unpatched kernels:
336 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
337 2.6.16 - 2.6.19 vanilla (kernel.org)
338 2.6.9-42.0.8EL (RHEL 4)
339 * Recommended e2fsprogs version: 1.39.cfs6
340 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
343 Severity : enhancement
345 Description: MountConf
346 Details : Lustre configuration is now managed via mkfs and mount
347 commands instead of lmc and lconf. New obd types (MGS, MGC)
348 are added for dynamic configuration management. See
349 https://mail.clusterfs.com/wikis/lustre/MountConf for
352 Severity : enhancement
354 Description: dynamic OST addition
355 Details : OSTs can now be added to a live filesystem
357 Severity : enhancement
359 Description: startup order invariance
360 Details : MDTs and OSTs can be started in any order. Clients only
361 require the MDT to complete startup.
363 Severity : enhancement
365 Description: parallel, asynchronous orphan cleanup
366 Details : orphan cleanup is now performed in separate threads for each
367 OST, allowing parallel non-blocking operation.
369 Severity : enhancement
371 Description: optimized stripe assignment
372 Details : stripe assignments are now made based on ost space available,
373 ost previous usage, and OSS previous usage, in order to try
374 to optimize storage space and networking resources.
376 Severity : enhancement
378 Description: Permanently set tunables
379 Details : All writable /proc/fs/lustre tunables can now be permanently
380 set on a per-server basis, at mkfs time or on a live
383 Severity : enhancement
385 Description: Lustre message v2
386 Details : Add lustre message format v2.
388 Severity : enhancement
390 Description: client OST exclusion list
391 Details : Clients can be started with a list of OSTs that should be
392 declared "inactive" for known non-responsive OSTs.
396 Description: ENOENT returned for valid filehandle during dbench.
397 Details : Check if a directory has children when invalidating dentries
398 associated with an inode during lock cancellation. This fixes
399 an incorrect ENOENT sometimes seen for valid filehandles during
403 Frequency : SFS test only (otherwise harmless)
405 Description: SPEC SFS validation failure on NFS v2 over lustre.
406 Details : Changes the blocksize for regular files to be 2x RPC size,
407 and not depend on stripe size.
409 Severity : enhancement
411 Description: fine-grained SMP locking inside DLM
412 Details : Improve DLM performance on SMP systems by removing the single
413 per-namespace lock and replace it with per-resource locks.
415 Severity : enhancement
417 Description: don't hold multiple extent locks at one time
418 Details : To avoid client eviction during large writes, locks are not
419 held on multiple stripes at one time or for very large writes.
420 Otherwise, clients can block waiting for a lock on a failed OST
421 while holding locks on other OSTs and be evicted.
423 Severity : enhancement
425 Description: Multiple MD RPCs in flight.
426 Details : Further unserialise some read-only MDT RPCs - learn about intents.
427 To avoid overly-overloading MDT, introduce a limit on number of
428 MDT RPCs in flight for a single client and add /proc controls
429 to adjust this limit.
431 Severity : enhancement
433 Description: client read/write statistics
434 Details : Add client read/write call usage stats for performance
435 analysis of user processes.
436 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
437 file access. extents_stats shows chunk size distribution.
438 extents_stats_per_process show chunk size distribution per
441 Severity : enhancement
443 Description: per-client statistics on server
444 Details : Add ldlm and operations statistics for each client in
445 /proc/fs/lustre/mds|obdfilter/*/exports/
447 Severity : enhancement
449 Description: improved MDT statistics
450 Details : Add detailed MDT operations statistics in
451 /proc/fs/lustre/mds/*/stats
453 Severity : enhancement
455 Description: VFS operations stats
456 Details : Add client VFS call stats, trackable by pid, ppid, or gid
457 /proc/fs/lustre/llite/*/vfs_ops_stats
458 /proc/fs/lustre/llite/*/vfs_track_[pid|ppid|gid]
463 Description: Fix client-side osc byte counters
464 Details : The osc read/write byte counters in
465 /proc/fs/lustre/osc/*/stats are now working
468 Frequency : always as root on SLES
470 Description: Failure of copying files with lustre special EAs.
471 Details : Client side always return success for setxattr call for lustre
472 special xattr (currently only "trusted.lov").
477 Description: Refcount LNET uuids
478 Details : The global LNET uuid list grew linearly with every startup;
479 refcount repeated list entries instead of always adding to
482 Severity : enhancement
484 Description: Dynamic service threads
485 Details : Within a small range, start extra service threads
486 automatically when the request queue builds up.
489 Frequency : mixed-endian client/server environments
491 Description: mixed-endian crashes
492 Details : The new msg_v2 system had some failures in mixed-endian
495 Severity : enhancement
497 Description: Easy OST removal
498 Details : OSTs can be permanently deactivated with e.g. 'lctl
499 conf_param lustre-OST0001.osc.active=0'
501 Severity : enhancement
503 Description: MGS proc entries
504 Details : Added basic proc entries for the MGS showing what filesystems
507 Severity : enhancement
509 Description: provide MGS failover
510 Details : Added config lock reacquisition after MGS server failover.
512 Severity : enhancement
514 Description: add Linux 2.4 support
515 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
519 Description: a large application tries to do I/O to the same resource and dies
521 Details : Check the req->rq_arrival time after the call to
522 ost_brw_lock_get(), but before we do anything about
523 processing it & sending the BULK transfer request. This
524 should help move old stale pending locks off the queue as
525 quickly as obd_timeout.
528 Frequency : when an incorrect nid is specified during startup
530 Description: ptlrpc connect to non-existant node causes kernel crash
531 Details : LNET can't be re-entered from an event callback, which
532 happened when we expire a message after the export has been
533 cleaned up. Instead, hand the zombie cleanup off to another
536 Severity : enhancement
538 Description: plain/inodebits lock performance improvement
539 Details : Grouping plain/inodebits in granted list by their request modes
540 and bits policy, thus improving the performance of search through
544 Frequency : only if OST filesystem is corrupted
546 Description: client incorrectly hits assertion in ptlrpc_replay_req()
547 Details : for a short time RPCs with bulk IO are in the replay list,
548 but replay of bulk IOs is unimplemented. If the OST filesystem
549 is corrupted due to disk cache incoherency and then replay is
550 started it is possible to trip an assertion. Avoid putting
551 committed RPCs into the replay list at all to avoid this issue.
554 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
556 Description: System hang on startup
557 Details : This bug allowed the liblustre (e.g. catamount) client to
558 return to the app before handling all startup RPCs. This
559 could leave the node unresponsive to lustre network traffic
560 and manifested as a server ptllnd timeout.
562 Severity : enhancement
564 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
565 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
566 boolean to control whether to print peer debug info when a
567 client's RPC times out.
570 Frequency : only for kernels with patches from Lustre below 1.4.3
572 Description: Remove old rdonly API
573 Details : Remove old rdonly API which unused from at least lustre 1.4.3
576 Frequency : only for devices with external journals
578 Description: Set external device read-only also
579 Details : During a commanded failover stop, we set the disk device
580 read-only while the server shuts down. We now also set any
581 external journal device read-only at the same time.
584 Frequency : when upgrading from 1.4 while trying to change parameters
586 Description: The wrong (new) MDC name was used when setting parameters for
587 upgraded MDT's. Also allows changing of OSC (and MDC)
588 parameters if --writeconf is specified at tunefs upgrade time.
591 Frequency : when setting specific ost indicies
593 Description: QOS code breaks on skipped indicies
594 Details : Add checks for missing OST indicies in the QOS code, so OSTs
595 created with --index need not be sequential.
597 Severity : enhancement
599 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
600 Details : The uninit_groups feature works in conjunction with the kernel
601 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
602 up the pass1 processing of e2fsck. This is a read-only feature
603 in ldiskfs2 only, so older kernels and current ldiskfs cannot
604 mount filesystems that have had this feature enabled.
606 Severity : enhancement
608 Description: Improve multi-block allocation algorithm to avoid fragmentation
609 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
610 allocation locality and avoid filesystem fragmentation.
612 ------------------------------------------------------------------------------
614 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
616 * Support for kernels:
617 2.4.21-47.0.1.EL (RHEL 3)
619 2.6.9-42.0.10.EL (RHEL 4)
620 2.6.12.6 vanilla (kernel.org)
621 2.6.16.27-0.9 (SLES 10)
622 * Recommended e2fsprogs version: 1.39.cfs5
624 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
628 Frequency : occasional, depends on client load and configuration
629 Bugzilla : 12181, 12203
630 Description: data loss for recently-modified files
632 Details : In some cases it is possible that recently written or created
633 files may not be written to disk in a timely manner (this should
634 normally be within 30s unless client IO load is very high).
635 The problem appears as zero-length files or files that are a
636 multiple of 1MB in size after a client crash or client eviction
637 that are missing data at the end of the file.
639 This problem is more likely to be hit on clients where files are
640 repeatedly created and unlinked in the same directory, clients
641 have a large amount of RAM, have many CPUs, the filesystem has
642 many OSTs, the clients are rebooted frequently, and/or the files
643 are not accessed by other nodes after being written.
645 The presence of the problem can be detected by looking at
646 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
647 smaller than the second (nr_unused) then dirty files will not
648 be flushed automatically to disk. "sync; sleep 10" should be
649 run several times on the node before unmounting it to update
650 Lustre (this is also safe to run on nodes without this problem).
652 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
653 that can cause this same problem, so customers using that kernel
654 also need to update the kernel in addition to Lustre. In order
655 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
657 It is normal that files written just before a client crash (less
658 than 30s) may not yet have been flushed to disk, even for local
662 Frequency : frequent on thin XT3 nodes
664 Description: UUID collision on thin XT3 Linux nodes
665 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
666 randomly, since we relied on an insufficiently-seeded PRNG.
671 Description: OSS hangs after "All ost request buffers busy"
672 Details : A deadlock between quota and journal operations caused OSS
673 hangs after printing "All ost request buffers busy."
676 Frequency : always on liblustre builds
678 Description: Cleanup compiler warnings on liblustre
681 Frequency : always on liblustre builds on XT3
683 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
687 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
688 > 32000 subdirectories
689 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
690 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
691 subdirectory count crosses 32000. This will aid e2fsck to
692 correctly handle more than 32000 subdirectories.
695 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
697 Description: System hang on startup
698 Details : This bug allowed the liblustre (e.g. catamount) client to
699 return to the app before handling all startup RPCs. This
700 could leave the node unresponsive to lustre network traffic
701 and manifested as a server ptllnd timeout.
703 Severity : enhancement
705 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
706 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
707 boolean to control whether to print peer debug info when a
708 client's RPC times out.
713 Description: make O_SYNC working on 2.6 kernels
714 Details : 2.6 kernels use different method for mark pages for write,
715 so need add a code to lustre for O_SYNC work.
720 Description: Failure to close file and release space on NFS
721 Details : Put inode details into lock acquired in ll_intent_file_open.
722 Use mdc_intent_lock in ll_intent_open to properly
723 detect all kind of errors unhandled by mdc_enqueue.
728 Description: proc file read during shutdown sometimes raced obd removal,
730 Details : Add lock to prevent obd access after proc file removal.
733 Frequency : Only for files larger than 4GB on 32-bit clients.
735 Description: improperly doing page alignment of locks
736 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
737 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
742 Description: RPCs being resent when they shouldn't be
743 Details : Some RPCs that should not be resent are being resent. This
744 can cause inconsistencies in the RPC state machine. Do not
745 resend such requests.
748 Frequency : rare, only with NFS export
750 Description: Crash on NFS re-export node
751 Details : under very unusual load conditions an assertion is hit in
752 ll_intent_file_open()
755 Frequency : only if OST filesystem is corrupted
757 Description: client incorrectly hits assertion in ptlrpc_replay_req()
758 Details : for a short time RPCs with bulk IO are in the replay list,
759 but replay of bulk IOs is unimplemented. If the OST filesystem
760 is corrupted due to disk cache incoherency and then replay is
761 started it is possible to trip an assertion. Avoid putting
762 committed RPCs into the replay list at all to avoid this issue.
767 Description: large O_DIRECT requests fail under memory pressure/fragmentation
768 Details : Large single O_DIRECT read and write calls can fail to allocate
769 a sufficiently large buffer to process the request. In case of
770 allocation failure the allocation is retried with a smaller
771 buffer and broken into smaller requests.
773 Severity : enhancement
775 Description: Add -o localflock option to simulate old noflock behaviour.
776 Details : This will achieve local-only flock/fcntl locks coherentness.
781 Description: versioning check is incomplete
782 Details : Checking the version difference of client vs. server, report
783 error if the gap is too big.
788 Description: add support PG_writeback bit
789 Details : add support for PG_writeback bit for Lustre, for more carefull
790 work with page cache in 2.6 kernel. This also fix some deadlocks
791 and remove hack for work O_SYNC with 2.6 kernel.
793 Severity : enhancement
795 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
796 Details : The uninit_groups feature works in conjunction with the kernel
797 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
798 up the pass1 processing of e2fsck. This is a read-only feature
799 in ldiskfs2 only, so older kernels and current ldiskfs cannot
800 mount filesystems that have had this feature enabled.
802 Severity : enhancement
804 Description: Improve multi-block allocation algorithm to avoid fragmentation
805 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
806 allocation locality and avoid filesystem fragmentation.
808 ------------------------------------------------------------------------------
810 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
812 * Support for kernels:
813 2.6.9-42.0.3.EL (RHEL 4)
815 2.4.21-47.0.1.EL (RHEL 3)
816 2.6.12.6 vanilla (kernel.org)
817 2.6.16.21-0.8 (SLES10)
818 * Recommended e2fsprogs version: 1.39.cfs2-0
820 * The backwards-compatible /proc/sys/portals symlink has been removed
821 in this release. Before upgrading, please ensure that you change
822 any configuration scripts or /etc/sysctl.conf files that access
823 /proc/sys/portals/* or sysctl portals.* to use the corresponding
824 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
825 in advance of the upgrade on any system running Lustre 1.4.6 or
826 newer, since /proc/sys/lnet was added in that version.
827 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
831 Frequency : only when quota is used
833 Description: avoid scanning export list for quota master
834 Details : Change the algorithms to avoid scanning export list in order
835 to improve the efficiency.
838 Frequency : MDS failover only, very rarely
840 Description: "went back in time" messages on mds failover
841 Details : The greatest transno may be lost when the current operation
842 finishes with an error (transno==0) and the client's last_rcvd
843 record is over-written. Save the greatest transno in the
844 mds_last_transno for this case.
847 Frequency : always for specific kernels and striping counts
849 Description: client may get "Matching packet too big" without ACL support
850 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
851 when trying to access files in certain configurations. The
852 clients should in fact be denied when mounting because they do
856 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
858 Description: many clients connecting with IO in progress causes connect timeouts
859 Details : Avoid synchronous journal commits to avoid delays caused by many
860 clients connecting/disconnecting when bulk IO is in progress.
861 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
862 OST_IO_PORTAL to avoid delays behind potentially many pending
866 Frequency : occasionally with multiple writers to a single file
868 Description: shared writes to file may result in wrong size reported by stat()
869 Details : Allow growing of kms when extent lock is cancelled
872 Frequency : always with random mmap IO to multi-striped file
874 Description: mmap write might be lost if we are writing to a 'hole' in stripe
875 Details : Only if the hole is at the end of OST object so that kms is too
876 small. Fix is to increase kms accordingly in ll_nopage.
879 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
881 Description: writes to a missing object would leak memory on the OST
882 Details : If there is an inconsistency between the MDS and OST filesystems,
883 such that the MDS references an object that doesn't exist, writes
884 to that object will leak memory due to incorrect cleanup in the
885 error handling path, eventually running out of memory on the OST.
890 Description: Creating too long symlink causes lustre errors
891 Details : Check symlink and name lengths before sending requests to MDS.
894 Frequency : only if flock is enabled (not on by default)
896 Description: posix locks not released on fd closure on 2.6.9+
897 Details : We failed to add posix locks to list of inode locks on 2.6.9+
898 kernels, this caused such locks not to be released on fd close and
899 then assertions on fs unmount about still used locks.
902 Frequency : MDS failover only, very rarely
904 Description: clients may get ASSERTION(granted_lock != NULL)
905 Details : When request was taking a long time, and a client was resending
906 a getattr by name lock request. The were multiple lock requests
907 with the same client lock handle and
908 mds_getattr_name->fixup_handle_for_resent_request found one of the
909 lock handles but later failed with ASSERTION(granted_lock != NULL).
914 Description: handle->h_buffer_credits > 0, assertion failure
915 Details : h_buffer_credits is zero after truncate, causing assertion
916 failure. This patch extends the transaction or creates a new
920 Frequency : NFS re-export or patchless client
921 Bugzilla : 11179, 10796
922 Description: Crash on NFS re-export node (__d_move)
923 Details : We do not want to hash the dentry if we don't have a lock.
924 But if this dentry is later used in d_move, we'd hit uninitialised
925 list head d_hash, so we just do this to init d_hash field but
926 leave dentry unhashed.
929 Frequency : NFS re-export or patchless client
931 Description: NFS exports has problem with symbolic link
932 Details : lustre client didn't properly install dentry when re-exported
933 to NFS or running patchless client.
936 Frequency : NFS re-export or patchless client
938 Description: Various nfs/patchless fixes.
939 Details : fixes reuse disconected alias for lookup process - this fixes
940 warning "find_exported_dentry: npd != pd",
941 fix permission error with open files at nfs.
942 fix apply umask when do revalidate.
945 Frequency : occasional
947 Description: Crash on NFS re-export node
948 Details : calling clear_page() on the wrong pointer triggered oops in
949 generic_mapping_read().
952 Frequency : rarely, using O_DIRECT IO
954 Description: unaligned directio crashes client with LASSERT
955 Details : check for unaligned buffers before trying any requests.
958 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
960 Description: stale data returned from RAID cache
961 Details : If only a small amount of IO is done to the RAID device before
962 reading it again it is possible to get stale data from the RAID
963 cache instead of reading it from disk.
966 Frequency : always for sles10 kernel
968 Description: sles10 support
969 Details : ll_follow_link: compile fixes and using of nd_set_link
973 Frequency : depends on arch, kernel and compiler version, always on sles10
976 Description: recursive or deep enough symlinks cause stack overflow
977 Details : getting rid of large stack-allocated variable in
981 Frequency : depends on hardware
983 Description: lustre write performance loss in the SLES10 kernel
984 Details : the performance loss is caused by using of write barriers in the
985 ext3 code. The SLES10 kernel turns barrier support on by
986 default. The fix is to undo that change for ldiskfs.
988 ------------------------------------------------------------------------------
990 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
992 * Support for kernels:
993 2.6.9-42.0.3EL (RHEL 4)
995 2.4.21-47.0.1.EL (RHEL 3)
996 2.6.12.6 vanilla (kernel.org)
1000 Frequency : quota enabled and large files being deleted
1002 Description: releasing more than 4GB of quota at once hangs OST
1003 Details : If a user deletes more than 4GB of files on a single OST it
1004 will cause the OST to spin in an infinite loop. Release
1005 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1010 Description: statfs data retrieved from /proc may be stale or zero
1011 Details : When reading per-device statfs data from /proc, in the
1012 {kbytes,files}_{total,free,avail} files, it may appear
1013 as zero or be out of date.
1016 Frequency : systems with MD RAID1 external journal devices
1018 Description: lconf's call to blkid is confused by RAID1 journal devices
1019 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1020 of returning all block devices that match the journal UUID.
1023 Frequency : always, for aggregate stripe size over 4GB
1025 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1026 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1027 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1030 Frequency : always if "lfs find" used on a local file/directory
1032 Description: "lfs find" segfaults if used on a local file/directory
1033 Details : The case where a directory component was not specified wasn't
1034 handled correctly. Handle this properly.
1037 Frequency : always on ppc64
1039 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1040 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1041 perform endianness conversion.
1044 Frequency : rarely (truncate to non-zero file size after write under load)
1045 Bugzilla : 10730, 10687
1046 Description: Files padded with zeros to next 4K multiple
1047 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1048 it is possible that files that are truncated to a non-zero size
1049 immediately after being written are filled with zero bytes beyond
1050 the truncated size. No file data is lost.
1052 Severity : enhancement
1054 Description: Allow recovery/failover for liblustre clients.
1055 Details : liblustre clients were unaware of failover configurations until
1058 Severity : enhancement
1060 Description: user file locks should fail when not mounting with flock option
1061 Details : Set up an error-returning stub in ll_file_operations.lock field
1062 to prevent incorrect behaviour when client is mounted without
1063 flock option. Also, set up properly f_op->flock field for
1067 Frequency : always on ia64
1069 Description: "lfs df" loops on printing out MDS statfs information
1070 Details : The obd_ioctl_data was not initialized and in some systems
1071 this caused a failure during the ioctl that did not return
1072 an error. Initialize the struct and return an error on failure.
1075 Frequency : SLES 9 only
1077 Description: Error of copying files with lustre special EAs as root
1078 Details : Client side always return success for setxattr call for lustre
1079 special xattr (currently only "trusted.lov").
1082 Frequency : rarely on clusters with both ia64+i386 clients
1084 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1085 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1086 clients are doing concurrent writes to the same file it is
1087 possible that the ia64 clients may LASSERT because the OST
1088 extent locks are not PAGE_SIZE aligned. Ensure that grown
1089 locks are always aligned on the request boundary.
1092 Frequency : specific use, occasional
1094 Description: Overwriting in use executable truncates on-disk binary image
1095 Details : If one node attempts to overwrite an executable in use by
1096 another node, we now correctly return ETXTBSY instead of
1097 truncating the file.
1099 Severity : enhancement
1101 Description: Async OSC create to avoid the blocking unnecessarily.
1102 Details : If a OST has no remain object, system will block on the creating
1103 when need to create a new object on this OST. Now, ways use
1104 pre-created objects when available, instead of blocking on an
1105 empty osc while others are not empty. If we must block, we block
1106 for the shortest possible period of time.
1111 Description: chmod on Lustre root is propagated to other clients
1112 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1113 invalid by the follow_mount time.
1116 Frequency : liblustre clients only
1118 Description: Race in 'instant cancel' lock handling could lead to such locks
1119 never to be granted in case of SMP MDS
1120 Details : Do not destroy not yet granted but cbpending locks in
1124 Frequency : replay/resend of open
1126 Description: non null lock assetion failure in mds_intent_policy
1127 Details : Trying to replay/resend lockless open requests resulted in
1128 mds_open() returning 0 with no lock. Now it sets a flag if
1129 a lock is going to be returned.
1131 Severity : enhancement
1133 Description: Checksum enhancements
1134 Details : New checksum enhancements allow for resending RPCs that failed
1137 Severity : enhancement
1139 Description: Tunables on number of dirty pages in cacche
1140 Details : Allow to set limit on number of dirty pages cached.
1145 Description: client crash on unmount - lock still has references
1146 Details : In some error handling cases it was possible to leak a lock
1147 reference on a client while accessing a file. This was not
1148 harmful to the client during operation, but would cause the
1149 client to crash when the filesystem is unmounted.
1152 Frequency : specific case, rare
1154 Description: ETXTBSY on mds though file not in use by client
1155 Details : ETXTBSY is no longer incorrectly returned when attempting to
1156 chmod or chown a directory that the user previously tried to
1157 execute or a currently-executing binary.
1160 Frequency : extremely rare except on liblustre-based clients
1162 Description: Lustre space not freed when files are deleted
1163 Details : Clean up open-unlinked files after client eviction. Previously
1164 the unlink was skipped and the files remained as orphans.
1169 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1170 Details : In some cases it was possible to send a blocking callback to
1171 a client doing a glimpse, even though that client didn't get
1172 a lock granted. When the glimpse lock is cancelled on the OST
1173 the freed lock is left on the waiting list and corrupted the list.
1176 Frequency : all core dumps
1178 Description: Broke core dumps to lustre
1179 Details : Negative dentry may be unhashed if parent does not have UPDATE
1180 lock, but some callers, e.g. do_coredump, expect dentry to be
1181 hashed after successful create, hash it in ll_create_it.
1183 ------------------------------------------------------------------------------
1185 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1187 * Support for kernels:
1188 2.6.9-42.0.2.EL (RHEL 4)
1189 2.6.5-7.276 (SLES 9)
1190 2.4.21-40.EL (RHEL 3)
1191 2.6.12.6 vanilla (kernel.org)
1195 Frequency : always on RHEL 3
1197 Description: Number of open files grows over time
1198 Details : The number of open files grows over time, whether or not
1199 Lustre is started. This was due to a filp leak introduced
1200 by one of our kernel patches.
1202 ------------------------------------------------------------------------------
1204 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
1206 * Support for kernels:
1207 2.6.9-42.EL (RHEL 4)
1208 2.6.5-7.267 (SLES 9)
1209 2.4.21-40.EL (RHEL 3)
1210 2.6.12.6 vanilla (kernel.org)
1215 Bugzilla : 5719, 9635, 9792, 9684
1216 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1217 Details : If a server is under heavy load and cannot reply to new
1218 connection requests before the client resends the (re)connect,
1219 the connection handling code can behave badly if two service
1220 threads are concurrently handing separate (re)connections from
1221 the same client. Add better locking to the connection handling
1222 code, and ensure that only a single connection will be processed
1223 for a given client UUID, even if the lock is dropped.
1225 Severity : enhancement
1227 Description: add TCP zero-copy support to kernel
1228 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1229 sends if the hardware supports scatter-gather and checksumming.
1230 This allows socklnd to do client-write and server-read more
1231 efficiently and reduce CPU utilization from skbuf copying.
1234 Frequency : only if NFS exporting from client
1236 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1237 Details : If setting attributes on a file created under NFS that had
1238 never been opened it would be possible to oops the client
1239 if the file had no objects.
1243 Bugzilla : 9326, 10402, 10897
1244 Description: client crash in ptlrpcd_wake() thread when sending async RPC
1245 Details : It is possible that ptlrpcd_wake() dereferences a freed async
1246 RPC. In rare cases the ptlrpcd thread alread processed the RPC
1247 before ptlrpcd_wake() was called and the request was freed.
1250 Frequency : always for liblustre
1252 Description: liblustre client does MDS+OSTs setattr RPC for each write
1253 Details : When doing a write from a liblustre client, the client
1254 incorrectly issued an RPC to the MDS and each OST the file was
1255 striped over in order to update the timestamps. When writing
1256 with small chunks and many clients this could overwhelm the MDS
1257 with RPCs. In all cases it would slow down the write because
1258 these RPCs are unnecessary.
1260 Severity : enhancement
1262 Description: allow number of MDS service threads to be changed at module load
1263 Details : It is now possible to change the number of MDS service threads
1264 running. Adding "options mds mds_num_threads={N}" to the MDS's
1265 /etc/modprobe.conf will set the number of threads for the next
1266 time Lustre is restarted (assuming the "mds" module is also
1267 reloaded at that time). The default number of threads will
1268 stay the same, 32 for most systems.
1273 Description: OST crash if filesystem is unformatted or corrupt
1274 Details : If an OST is started on a device that has never been formatted
1275 or if the filesystem is corrupt and cannot even mount then the
1276 error handling cleanup routines would dereference a NULL pointer.
1281 Description: NULL pointer deref in llap_from_page.
1282 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1283 incorrect) mapping if the page was truncated/reclaimed while it was
1284 searched for. Check for this condition and skip such pages when
1285 doing readahead. Introduce extra check to llap_from_page() to
1286 verify page->mapping->host is non-NULL (so page is not anonymous).
1289 Frequency : Sometimes when using sys_sendfile
1291 Description: "page not covered by a lock" warnings from ll_readpage
1292 Details : sendfile called ll_readpage without right page locks present.
1293 Now we introduced ll_file_sendfile that does necessary locking
1294 around call to generic_file_sendfile() much like we do in
1298 Frequency : with certain MDS communication failures at client mount time
1300 Description: NULL pointer deref after failed client mount
1301 Details : a client connection request may delayed by the network layer
1302 and not be sent until after the PTLRPC layer has timed out the
1303 request. If the client fails the mount immediately it will try
1304 to clean up before the network times out the request. Add a
1305 reference from the request import to the obd device and delay
1306 the cleanup until the network drops the request.
1309 Frequency : occasionally during client (re)connect
1311 Description: assertion failure during client (re)connect
1312 Details : processing a client connection request may be delayed by the
1313 client or server longer than the client connect timeout. This
1314 causes the client to resend the connection request. If the
1315 original connection request is replied in this interval, the
1316 client may trip an assertion failure in ptlrpc_connect_interpret()
1317 which thought it would be the only running connect process.
1320 Frequency : only with obd_echo servers and clients that are rebooted
1322 Description: kernel BUG accessing uninitialized data structure
1323 Details : When running an obd_echo server it did not start the ping_evictor
1324 thread, and when a client was evicted an uninitialized data
1325 structure was accessed. Start the ping_evictor in the RPC
1326 service startup instead of the OBD startup.
1328 Severity : enhancement
1329 Bugzilla : 10193 (patchless)
1330 Description: Remove dependency on various unexported kernel interfaces.
1331 Details : No longer need reparent_to_init, exit_mm, exit_files,
1332 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1335 Frequency : rare (only users of deprecated and unsupported LDAP config)
1337 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1338 Details : LDAP apparently contains 'lustreName' attributes instead of
1339 'name'. A simple remapping of the name is sufficient.
1342 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1344 Description: waiting_locks_callback trips kernel BUG if client is evicted
1345 Details : Running with the dump_on_timeout debug flag turned on makes
1346 it possible that the waiting_locks_callback() can try to dump
1347 the Lustre kernel debug logs from an interrupt handler. Defer
1348 this log dumping to the expired_lock_main() thread.
1350 Severity : enhancement
1352 Description: Support NFS exporting on 2.6 kernels.
1353 Details : Implement non-rawops metadata methods for NFS server to use without
1354 changing NFS server code.
1357 Frequency : very rare (synthetic metadata workload only)
1359 Description: two racing renames might cause an MDS thread to deadlock
1360 Details : Running the "racer" program may cause one MDS thread to rename
1361 a file from being the source of a rename to being the target of
1362 a rename at exactly the same time that another thread is doing
1363 so, and the second thread has already enqueued these locks after
1364 doing a lookup of the target and is trying to relock them in
1365 order. Ensure that we don't try to re-lock the same resource.
1368 Frequency : only very large systems with liblustre clients
1370 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1371 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1372 all OSTs in parallel. This allows the eviction of stale liblustre
1373 clients to proceed much faster than if they were done in series,
1374 and also offers similar improvements for other set_info RPCs.
1379 Description: excessive CPU usage during initial read phase on client
1380 Details : During the initial read phase on a client, it would agressively
1381 retry readahead on the file, consuming too much CPU and impacting
1382 performance (since 1.4.5.8). Improve the readahead algorithm
1383 to avoid this, and also improve some other common cases (read
1384 of small files in particular, where "small" is files smaller than
1385 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1390 Description: MDS crash when receiving packet with unknown intent.
1391 Details : Do not LBUG in unknown intent case, just return -EFAULT
1393 Severity : enhancement
1394 Bugzilla : 9293, 9385
1395 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1396 Details : Do not serialize getattr (non-intent version) and statfs.
1399 Frequency : occasional, when OST network is overloaded/intermittent
1401 Description: client evicted by OST after bulk IO timeout
1402 Details : If a client sends a bulk IO request (read or write) the OST
1403 may evict the client if it is unresposive to its data GET/PUT
1404 request. This is incorrect if the network is overloaded (takes
1405 too long to transfer the RPC data) or dropped the OST GET/PUT
1406 request. There is no need to evict the client at all, since
1407 the pinger and/or lock callbacks will handle this, and the
1408 client can restart the bulk request.
1411 Frequency : Always when mmapping file with no objects
1413 Description: client crashes when mmapping file with no objects
1414 Details : Check that we actually have objects in a file before doing any
1415 operations on objects in ll_vm_open, ll_vm_close and
1421 Description: Request leak when working with deleted CWD
1422 Details : Introduce advanced request refcount tracking for requests
1423 referenced from lustre intent.
1425 Severity : Enhancement
1427 Description: Cache open file handles on client.
1428 Details : MDS now will return special lock along with openhandle, if
1429 requested and client is allowed to hold openhandle, even if unused,
1430 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1431 closing files for every read/write openration.
1433 Severity : Enhancement
1435 Description: Cache open negative dentries on client when possible.
1436 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1437 negative dentries on lock revocation.
1442 Description: Remounting a client read-only wasn't possible with a zconf mount
1443 Details : It wasn't possible to remount a client read-only with llmount.
1445 Severity : enhancement
1446 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1447 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1448 MPICH is included to add Lustre-specific ADIO interfaces.
1449 This is based closely on the UFS ADIO layer and only differs
1450 in file creation, in order to allow the OST striping to be set.
1451 This is user-contributed code and not supported by CFS.
1456 Description: extended inode attributes (immutable, append-only) work improperly
1457 when 2.4 and 2.6 kernels are used on client/server or vice versa
1458 Details : Introduce kernel-independent values for these flags.
1460 Severity : enhancement
1463 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1464 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1465 granularity. Now, Fractional MB granularity is be supported,
1466 this is very useful for low memory system.
1468 Severity : enhancement
1470 Description: Getattr by fid
1471 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1472 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1473 have locks and avoids getattr rpc requests for it on every stat.
1476 Frequency : Always, for filesystems larger than 2TB
1478 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1479 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1480 filesystems larger than 2TB to be mounted with mballoc enabled.
1483 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1485 Description: filesystem corruption for non-standard kernels and very large OSTs
1486 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1487 filesystems larger than 2TB is mounted then the kernel will
1488 silently corrupt the start of the filesystem. CONFIG_LBD is
1489 enabled for all CFS-supported kernels, but the possibility of
1490 this happening with a modified kernel config exists.
1492 Severity : enhancement
1494 Description: add client O_DIRECT support for 2.6 kernels
1495 Details : It is now possible to do O_DIRECT reads and writes to files
1496 in the Lustre client mountpoint on 2.6 kernel clients.
1498 Severity : enhancement
1500 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1501 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1502 parallel, not waiting for response from every OST before sending
1503 a rpc to the next OST.
1508 Description: setattr vs write race when updating file timestamps
1509 Details : Client processes that update a file timestamp into the past
1510 right after writing to the file (e.g. tar) it is possible that
1511 the updated file modification time can be reset to the current
1512 time due to a race between processing the setattr and write RPC.
1514 Severity : enhancement
1516 Description: Bring 'lfs find' closer in line with regular Linux find.
1517 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1518 -print0 options and obtains all the needed info through the lustre
1521 Severity : enhancement
1523 Description: support up to 1024 configured devices on one node
1524 Details : change obd_dev array from statically allocated to dynamically
1525 allocated structs as they are first used to reduce memory usage
1530 Description: Flush dirty partially truncated pages during truncate
1531 Details : Immediatelly flush partially truncated pages in filter_setattr,
1532 this way we completely avoid having any pages in page cache on OST
1533 and can retire ugly workarounds during writes to flush such pages.
1538 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1539 Details : protect inode from truncation within vfs_unlink() context
1540 just take a reference before calling vfs_unlink() and release it
1541 when parent's i_sem is free.
1544 Frequency : always, if extents are used on OSTs
1546 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1547 Details : OSTs using the extents format would not zero the high 16 bits of
1548 the index physical block number. This is not a problem for any
1549 OST filesystems smaller than 16TB, and no kernels support ext3
1550 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1551 new/modified files) and can be fixed for existing filesystems
1552 with e2fsprogs-1.39-cfs1.
1557 Description: import connection selection may be incorrect if timer wraps
1558 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1559 import connections to be ignored if the 32-bit jiffies counter
1560 wraps. Use a 64-bit jiffies counter.
1563 Frequency : during server recovery
1565 Description: crash after server is denying duplicate export
1566 Details : If clients are resending connect requests to the server, the
1567 server refuses to allow a client to connect multiple times.
1568 Fixed a bug in the handling of this case.
1571 Frequency : very large clusters immediately after boot
1573 Description: LNET request buffers exhausted under heavy short-term load
1574 Details : If a large number of client requests are generated on a service
1575 that has previously never seen so many requests it is possible
1576 that the request buffer growth cannot keep up with the spike in
1577 demand. Instead of dropping incoming requests, they are held in
1578 the LND until the RPC service can accept more requests.
1581 Frequency : Sometimes during replay
1583 Description: Assertion failure in ll_local_open after replay.
1584 Details : If replay happened on an open request reply before we were able
1585 to set replay handler, reply will become not swabbed tripping the
1586 assertion in ll_local_open. Now we set the handler right after
1587 recognising of open request
1590 Frequency : very rare
1592 Description: kernel reports "badness in vsnprintf"
1593 Details : Reading from the "recovery_status" /proc file in small chunks
1594 may cause a negative length in lprocfs_obd_rd_recovery_status()
1595 call to vsnprintf() (which is otherwise harmless). Exit early
1596 if there is no more space in the output buffer.
1598 Severity : enhancement
1600 Description: clear OBD RPC statistics by writing to them
1601 Details : It is now possible to clear the OBD RPC statistics by writing
1602 to the "stats" file.
1607 Description: Client mtime is not the same on different clients after utimes
1608 Details : In some cases, the client was using the utimes() syscall on
1609 a file cached on another node. The clients now validate the
1610 ctime from the MDS + OSTs to determine which one is right.
1615 Description: Inability to activate failout mode
1616 Details : lconf script incorrectly assumed that in python string's numeric
1617 value is used in comparisons.
1620 Frequency : always with multiple stripes per file
1622 Description: Inefficient object allocation for mutli-stripe files
1623 Details : When selecting which OSTs to stripe files over, for files with
1624 a stripe count that divides evenly into the number of OSTs,
1625 the MDS is always picking the same starting OST for each file.
1626 Return the OST selection heuristic to the original design.
1631 Description: mount failures may take full timeout to return an error
1632 Details : Under some heavy load conditions it is possible that a
1633 failed mount can wait for the full obd_timeout interval,
1634 possibly several minutes, before reporting an error.
1635 Instead return an error as soon as the status is known.
1637 ------------------------------------------------------------------------------
1639 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
1641 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1642 INTEROPERATE with older versions automatically. Please read the
1643 user documentation before upgrading any part of a live system.
1644 * WARNING: Lustre networking configuration changes are required with
1645 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1648 * Support for kernels:
1649 2.6.9-22.0.2.EL (RHEL 4)
1650 2.6.5-7.244 (SLES 9)
1651 2.6.12.6 vanilla (kernel.org)
1654 Severity : enhancement
1655 Bugzilla : 7981/8208
1656 Description: Introduced Lustre Networking (LNET)
1657 Details : LNET is new networking infrastructure for Lustre, it includes
1658 a reorganized network configuration mode (see the user
1659 documentation for full details) as well as support for routing
1660 between different network fabrics. Lustre Networking Devices
1661 (LNDs) for the supported network fabrics have also been
1662 created for this new infrastructure.
1664 Severity : enhancement
1665 Description: Introduced Access control lists
1666 Details : clients can set ACLs on files and directories in order to have
1667 more fine-grained permissions than the standard Unix UGO+RWX.
1668 The MDS must be started with the "-o acl" mount option.
1670 Severity : enhancement
1671 Description: Introduced filesystem quotas
1672 Details : Administrators may now establish per-user quotas on the
1675 Severity : enhancement
1677 Description: Configuration change for the XT3
1678 The PTLLND is now used to run Lustre over Portals on the XT3
1679 The configure option(s) --with-cray-portals are no longer used.
1680 Rather --with-portals=<path-to-portals-includes> is used to
1681 enable building on the XT3. In addition to enable XT3 specific
1682 features the option --enable-cray-xt3 must be used.
1687 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1688 Details : A race between allocating a new llcd and re-getting the llcd_lock
1689 allowed another thread to grab newly-allocated llcd.
1691 Severity : enhancement
1693 Description: 2.6 OST async journal commit and locking fix to improve performance
1694 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1695 2.6 kernels are now async as they already were in 2.4 kernels so
1696 that they can commit concurrently with the network bulk transfer.
1697 For block-allocated files the filter allocation semaphore is held
1698 to avoid filesystem fragmentation during allocation. BKL lock
1699 removed for 2.6 xattr operations where it is no longer needed.
1704 Description: lconf incorrectly determined whether two IP networks could talk
1705 Details : In some more complicated routing and multiple-network
1706 configurations, lconf will avoid trying to make a network
1707 connection to a disjoint part of the IP space. It was doing the
1708 math incorrectly for one set of cases.
1713 Description: Fix for potential infinite loop processing records in an llog.
1714 Details : If an llog record is corrupted/zeroed, it is possible to loop
1715 forever in llog_process(). Validate the llog record length
1716 and skip the remainder of the block on error.
1719 Frequency : occasional (liblustre only)
1721 Description: liblustre could not open files whose last component is a symlink
1722 Details : sysio_path_walk() would incorrectly pass the open intent to
1723 intermediate path components.
1726 Frequency : rare (liblustre only with non-standard tuning)
1727 Bugzilla : 7201 (7350)
1728 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1729 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1730 after it has been released, only do this for referenced locks
1732 Severity : enhancement
1734 Description: specifying an (invalid) directory default stripe_size of -1
1735 would reset the directory default striping
1736 Details : stripe_size -1 was used internally to signal directory stripe
1737 removal, now use "all default" to signal dir stripe removal
1738 as a directory striping of "all default" is not useful
1741 Frequency : common for large clusters running liblustre clients
1743 Description: doing an ls when liblustre clients are running is slow
1744 Details : sending a glimpse AST to a liblustre client waits for every AST
1745 to time out, as liblustre clients will not respond. Since they
1746 cannot cache data we refresh the OST lock LVB from disk instead.
1748 Severity : enhancement
1750 Description: doing an ls at the same time as file IO can be slow
1751 Details : enqueue and other "small" requests can be blocked behind many
1752 large IO requests. Create a new OST IO portal for non-IO
1753 requests so they can be processed faster.
1756 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1758 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1759 created file with mode 000
1760 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1761 allow an ftruncate() as a non-root user to a file with mode 000.
1762 Lustre now respects this flag to disable mode checks when
1763 truncating a file owned by the user
1766 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1769 Description: Revoking locks from clients that went dead or catatonic might take
1771 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1772 cancellation of such locks instant on servers without waiting for
1773 any reply from clients. Clients drops these locks when cancel
1774 notification from server is received without replying.
1777 Frequency : liblustre-only, when liblustre client dies or becomes busy
1779 Description: Doing ls on Linux clients can take a long time with active
1781 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1782 granting such locks to it in the first place if possible. Locks
1783 are taken by proxy on the OST during the read or write and
1784 dropped immediately afterward. Add connect flags handling, do
1785 not grant locks to liblustre clients for glimpse ASTs.
1787 Severity : enhancement
1789 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1790 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1791 sequential reads and behaves badly with applications doing random
1792 reads. Improve it by reading ahead at least read region, and
1793 avoiding excessive large RPC for small reads.
1795 Severity : enhancement
1797 Description: Creating more than 1000 files for a single job may cause a load
1798 imbalance on the OSTs if there are also a large number of OSTs.
1799 Details : qos_prep_create() uses an OST index reseed value that is an
1800 even multiple of the number of available OSTs so that if the
1801 reseed happens in the middle of the object allocation it will
1802 still utilize the OSTs as uniformly as possible.
1807 Description: OST or MDS may oops in ping_evictor_main()
1808 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1809 but doesn't restart at beginning of obd_exports_timed list
1812 Severity : enhancement
1814 Description: improve by-nid export eviction on the MDS and OST
1815 Details : allow multiple exports with the same NID to be evicted at one
1816 time without re-searching the exports list.
1819 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1821 Description: MDS may oops in groups_free()
1822 Details : in rare race conditions a newly allocated group_info struct is
1823 freed again, and this can be NULL. The 2.4 compatibility code
1824 for groups_free() checked for a NULL pointer, but 2.6 did not.
1827 Frequency : common for liblustre clients doing little filesystem IO
1828 Bugzilla : 9352, 7313
1829 Description: server may evict liblustre clients accessing contended locks
1830 Details : if a client is granted a lock or receives a completion AST
1831 with a blocking AST already set it would not reply to the AST
1832 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1835 Frequency : lfs setstripe, only systems with more than 160 OSTs
1837 Description: unable to set striping with a starting offset beyond OST 160
1838 Details : llapi_create_file() incorrectly limited the starting stripe
1839 index to the maximum single-file stripe count.
1842 Frequency : LDAP users only
1844 Description: lconf did not handle in-kernel recovery with LDAP properly
1845 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1847 Severity : enhancement
1849 Description: bind OST threads to NUMA nodes to improve performance
1850 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1851 node and do their allocations there to localize memory access
1853 Severity : enhancement
1855 Description: llmount can determine client NID directly from Myrinet (GM)
1856 Details : the client NID code from gmnalnid was moved directly into
1857 llmount, removing the need to use this or specifying the
1858 client NID explicitly when mounting GM clients with zeroconf
1861 Frequency : if client is started with down MDS
1863 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1864 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1865 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1866 ptlrpc_import_delayed_req() because we didn't check if the
1867 request was interrupted, and we also didn't break out of the
1868 event loop if there was no timeout
1873 Description: data loss during non-page-aligned writes to a single file from
1874 both multiple nodes and multiple threads on one node at same time
1875 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1876 inconsistency led to false short-reads, that were cached and later
1877 used by ->prepare_write() to fill in partially written page,
1878 leading to data loss.
1881 Frequency : always, if lconf --abort_recovery used
1883 Description: lconf --abort_recovery fails with 'Operation not supported'
1884 Details : lconf was attempting to abort recovery on the MDT device and not
1887 Severity : enhancement
1889 Description: remove cleanup logs
1890 Details : replace lconf-generated cleanup logs with lustre internal
1891 cleanup routines. Eliminates the need for client-cleanup and
1894 Severity : enhancement
1896 Description: add support for EAs (user and system) on lustre filesystems
1897 Details : it is now possible to store extended attributes in the Lustre
1898 client filesystem, and with the user_xattr mount option it
1899 is possible to allow users to store EAs on their files also
1901 Severity : enhancement
1903 Description: Add possibility (config option) to show minimal available OST free
1905 Details : When compiled with --enable-mindf configure option, statfs(2)
1906 (and so, df) will return least minimal free space available from
1907 all OSTs as amount of free space on FS, instead of summary of
1908 free spaces of all OSTs.
1910 Severity : enhancement
1912 Description: do not expand extent locks acquired on OST-side
1913 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1914 by server: they are not cached anyway.
1917 Frequency : when mmap is used/binaries executed from Lustre
1919 Description: Unmmap pages before throwing them away from read cache.
1920 Details : llap_shrink cache now attempts to unmap pages before discarding
1921 them (if unmapping failed - do not discard). SLES9 kernel has
1922 extra checks that trigger if this unmapping is not done first.
1927 Description: lconf didn't resolve symlinks before checking to see whether a
1928 given mountpoint was already in use
1931 Frequency : when migrating failover services
1932 Bugzilla : 6395, 9514
1933 Description: When migrating a subset of services from a node (e.g. failback
1934 from a failover service node) the remaining services would
1935 time out and evict clients.
1936 Details : lconf --force (implied by --failover) sets the global obd_timeout
1937 to 5 seconds in order to quickly disconnect, but this caused
1938 other RPCs to time out too quickly. Do not change the global
1939 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1941 Severity : enhancement
1942 Frequency : if MDS is started with down OST
1943 Bugzilla : 9439,5706
1944 Description: Allow startup/shutdown of an MDS without depending on the
1945 availability of the OSTs.
1946 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1947 Add appropriate locking and lov-osc refcounts for safe
1948 cleaning. Add osc abort_inflight calls in case the
1949 synchronize never started.
1952 Frequency : occasional (Cray XT3 only)
1954 Description: root not authorized to access files in CRAY_PORTALS environment
1955 Details : The client process capabilities were not honoured on the MDS in
1956 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1957 been accessed by an authorized user then root was able to access
1958 the file on the local client also. The root user capabilities
1959 are now allowed on the MDS, as this environment has secure UID.
1962 Frequency : occasional
1964 Description: ldiskfs "too long searching" message happens too often
1965 Details : A debugging message (otherwise harmless) prints too often on
1966 the OST console. This has been reduced to only happen when
1967 there are fragmentation problems on the filesystem.
1972 Description: Division by zero in statfs when all OSCs are inactive
1973 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
1974 lov_statfs divides by value returned by lov_get_stripecnt().
1978 Bugzilla : 9489, 3273
1979 Description: First write from each client to each OST was only 4kB in size,
1980 to initialize client writeback cache, which caused sub-optimal
1981 RPCs and poor layout on disk for the first writen file.
1982 Details : Clients now request an initial cache grant at (re)connect time
1983 and so that they can start streaming writes to the cache right
1984 away and always do full-sized RPCs if there is enough data.
1985 If the OST is rebooted the client also re-establishes its grant
1986 so that client cached writes will be honoured under the grant.
1991 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
1992 Details : Now I/O RPCs go to different portal number and (presumably) fast
1993 lock requests (and glimses) and other RPCs get their own service
1994 threads pool that should be able to service those RPCs
1997 Severity : enhancement
1999 Description: Ability to exchange lustre version between client and servers and
2000 issue warnings at client side if client is too old. Also for
2001 liblustre clients there is ability to refuse connection of too old
2003 Details : New 'version' field is added to connect data structure that is
2004 filled with version info. That info is later checked by server and
2008 Frequency : rare, liblustre only.
2009 Bugzilla : 9296, 9581
2010 Description: Two simultaneous writes from liblustre at offset within same page
2011 might proceed at the same time overwriting eachother with stale
2013 Details : I/O lock withing llu_file_prwv was released too early, before data
2014 actually was hitting the wire. Extended lock-holding time until
2015 server acknowledges receiving data.
2018 Frequency : extremely rare. Never observed in practice.
2020 Description: avoid generating lustre_handle cookie of 0.
2021 Details : class_handle_hash() generates handle cookies by incrementing
2022 global counter, and can hit 0 occasionaly (this is unlikely, but
2023 not impossible, because initial value of cookie counter is
2024 selected randonly). Value of 0 is used as a sentinel meaning
2025 "unassigned handle" --- avoid it. Also coalesce two critical
2026 sections in this function into one.
2028 Severity : enhancement
2030 Description: allow liblustre clients to delegate truncate locking to OST
2031 Details : To avoid overhead of locking, liblustre client instructs OST to
2032 take extent lock in ost_punch() on client's behalf. New connection
2033 flag is added to handle backward compatibility.
2035 Severity : enhancement
2036 Bugzilla : 4928, 7341, 9758
2037 Description: allow number of OST service threads to be specified
2038 Details : a module parameter allows the number of OST service threads
2039 to be specified via "options ost ost_num_threads={N}" in the
2040 OSS's /etc/modules.conf or /etc/modprobe.conf.
2044 Bugzilla : 6146, 9635, 9895
2045 Description: servers crash with bad pointer in target_handle_connect()
2046 Details : In rare cases when a client is reconnecting it was possible that
2047 the connection request was the last reference for that export.
2048 We would temporarily drop the export reference and get a new
2049 one, but this may have been the last reference and the export
2050 was just destroyed. Get new reference before dropping old one.
2052 Severity : enhancement
2053 Frequency : if client is started with failover MDS
2055 Description: Allow multiple MDS hostnames in the mount command
2056 Details : Try to read the configuration from all specified MDS
2057 hostnames during a client mount in case the "primary"
2060 Severity : enhancement
2062 Description: Stop sending data to evicted clients as soon as possible.
2063 Details : Check if the client we are about to send or are sending data to
2064 was evicted already. (Check is done every second of waiting,
2065 for which l_wait_event interface was extended to allow checking
2066 of exit condition at specified intervals).
2069 Frequency : rare, normally only when NFS exporting is done from client
2071 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2073 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2076 Frequency : always, if srand() is called before liblustre initialization
2078 Description: Liblustre uses system PRNG disturbing its usage by user application
2079 Details : Introduce internal to lustre fast and high-quality PRNG for
2080 lustre usage and make liblustre and some other places in generic
2081 lustre code to use it.
2083 Severity : enhancement
2084 Bugzilla : 9477, 9557, 9870
2085 Description: Verify that the MDS configuration logs are updated when xml is
2086 Details : Check if the .xml configuration logs are newer than the config
2087 logs stored on the MDS and report an error if this is the case.
2088 Request --write-conf, or allow starting with --old_conf.
2090 Severity : enhancement
2092 Description: Handle symlinks in the path when checking if Lustre is mounted.
2093 Details : Resolve intermediate symlinks when checking if a client has
2094 mounted a filesystem to avoid duplicate client mounts.
2099 Description: lconf can hit an error exception but still return success.
2100 Details : The lconf command catches the Command error exception at the top
2101 level script context and will exit with the associated exit
2102 status, but doesn't ensure that this exit status is non-zero.
2107 Description: failure of ptlrpc thread startup can cause oops
2108 Details : Starting a ptlrpc service thread can fail if there are a large
2109 number of threads or the server memory is very fragmented.
2110 Handle this without oopsing.
2113 Frequency : always, only if liblustre and non-default acceptor port was used
2115 Description: liblustre cannot connect to servers with non-default acceptor port
2116 Details : tcpnal_set_default_params() was not called and was therefore
2117 ignoring the environment varaible TCPNAL_PORT, as well as other
2118 TCPNAL_ environment variables
2123 Description: two objects could be created on the same OST for a single file
2124 Details : If an OST is down, in some cases it was possible to create two
2125 objects on a single OST for a single file. No problems other
2126 than potential performance impact and spurious error messages.
2130 Bugzilla : 5681, 9562
2131 Description: Client may oops in ll_unhash_aliases
2132 Details : Client dcache may become inconsistent in race condition.
2133 In some cases "getcwd" can fail if the current directory is
2139 Description: Inode refcounting problems in NFS export code
2140 Details : link_raw functions used to call d_instantiate without obtaining
2141 extra inode reference first.
2145 Bugzilla : 9942, 9903
2146 Description: Referencing freed requests leading to crash, memleaks with NFS.
2147 Details : We used to require that call to ll_revalidate_it was always
2148 followed by ll_lookup_it. Also with revalidate_special() it is
2149 possible to call ll_revalidate_it() twice for the same dentry
2150 even if first occurence returned success. This fix changes semantic
2151 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2152 reference on a request referred from the intent.
2153 ll_intent_release() then releases such a request.
2156 Frequency : rare, normally benchmark loads only
2158 Description: unlinked inodes were kept in memory on the client
2159 Details : If a client is repeatedly creating and unlinking files it
2160 can accumulate a lot of stale inodes in the inode slab cache.
2161 If there is no other client load running this can cause the
2162 client node to run out of memory. Instead flush old inodes
2163 from client cache that have the same inode number as a new inode.
2166 Frequency : SLES9 2.6.5 kernel and long filenames only
2167 Bugzilla : 9969, 10379
2168 Description: utime reports stale NFS file handle
2169 Details : SLES9 uses out-of-dentry names in some cases, which confused
2170 the lustre dentry revalidation. Change it to always use the
2174 Frequency : rare, unless heavy write-truncate concurrency is continuous
2175 Bugzilla : 4180, 6984, 7171, 9963, 9331
2176 Description: OST becomes very slow and/or deadlocked during object unlink
2177 Details : filter_destroy() was holding onto the parent directory lock
2178 while truncating+unlinking objects. For very large objects this
2179 may block other threads for a long time and slow overall OST
2180 responsiveness. It may also be possible to get a lock ordering
2181 deadlock in this case, or run out of journal credits because of
2182 the combined truncate+unlink. Solution is to do object truncate
2183 first in one transaction without parent lock, and then do the
2184 final unlink in a new transaction with the parent lock. This
2185 reduces the lock hold time dramatically.
2188 Frequency : rare, 2.4 kernels only
2190 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2191 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2192 before unmounting their filesystem in order to avoid deadlock.
2193 The kernel_locked() function in 2.4 kernels only checks whether
2194 the kernel lock is held, not whether it is this process that is
2195 holding it as 2.6 kernels do.
2200 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2201 Details : The client ptlrpc code may be trying to reconnect to a down
2202 server before a previous connection attempt has timed out.
2203 Increase the reconnect interval to be longer than the connection
2204 timeout interval to avoid sending duplicate connections to
2208 Frequency : echo_client brw_test command
2210 Description: fix echo_client to work with OST preallocated code
2211 Details : OST preallocation code (5137) didn't take echo_client IO path
2212 into account: echo_client calls filter methods outside of any
2213 OST thread and, hence, there is no per-thread preallocated
2214 pages and buffers to use. Solution: hijack pga pages for IO. As
2215 a byproduct, this avoids unnecessary data copying.
2219 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2220 Description: Client can oops in mdc_commit_close() after open replay
2221 Details : It was possible for the MDS to return an open request with no
2222 transaction number in mds_finish_transno() if the client was
2223 evicted, but without actually returning an error. Clients
2224 would later try to replay that open and may trip an assertion
2225 Simplify the client close codepath, and always return an error
2226 from the MDS in case the open is not successful.
2229 Frequency : rare, 2.6 OSTs only
2231 Description: OST may deadlock under high load on fragmented files
2232 Details : If there was a heavy load and highly-fragmented OST filesystems
2233 it was possible to have all the OST threads deadlock waiting on
2234 allocation of biovecs, because the biovecs were not released
2235 until the entire RPC IO was completed. Instead, release biovecs
2236 as soon as they are complete to ensure forward IO progress.
2238 Severity : enhancement
2240 Description: Support for specifying external journal device at mount
2241 Details : If an OST or MDS device is formatted with an external journal
2242 device, this device major/minor is stored in the ext3 superblock
2243 and may not be valid for failover. Allow detecting and
2244 specifying the external journal at mount time.
2249 Description: Mounting an MDS with pending unlinked files may cause oops
2250 Details : target_finish_recovery() calls mds_postrecov() which returned
2251 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2252 considers this an error and immediately begins cleaning up the
2253 lov, just after starting the mds_lov process
2255 Severity : enhancement
2257 Description: Implement 'lfs df' to report actual free space on per-OST basis
2258 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2259 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2260 usage of objects; '-h' to report in human readable format.
2262 ------------------------------------------------------------------------------
2264 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
2271 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2272 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2273 at startup during orphan unlinking. Instead, skip buddy update
2274 before setup, as it will be regenerated after recovery is complete.
2279 Description: If an OST is inactive, its locks might reference stale inodes.
2280 Details : lov_change_cbdata() must iterate over all namespaces, even if
2281 they are inactive to clear inode references from the lock.
2283 Severity : enhancement
2284 Frequency : occasional, if non-standard max_dirty_mb used
2286 Description: Client will block write RPCs if not enough grant
2287 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2288 then the client will block writes while waiting for another RPC
2289 to complete instead of consuming its dirty limit. With change
2290 we get improved performance when max_dirty_mb is small.
2292 Severity : enhancement
2293 Bugzilla : 3389, 6253
2294 Description: Add support for supplementary groups on the MDS.
2295 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2296 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2297 lookups for user supplementary groups into a cache.
2301 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2302 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2303 for liblustre O_EXCL is enforced by MDS.
2307 Description: readdir never returns NULL in liblustre.
2308 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2309 dentry in struct dirent.
2313 Frequency : liblustre only, depends on application IO pattern
2314 Description: liblustre clients evicted if not contacting servers
2315 Details : Don't put liblustre clients into the ping_evictor list, so
2316 they will not be evicted by the pinger ever.
2318 Severity : enhancement
2320 Description: Add ability to evict clients by NID from MDS.
2321 Details : By echoing "nid:$NID" string into
2322 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2323 $NID would be instantly evicted from this MDS and from all active
2324 OSTs connected to it.
2328 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2329 Details : lookup_it_finish() used to query file size from OSTs that was not
2334 Description: service threads change working directory to that of init
2335 Details : Starting lustre service threads may pin the working directory
2336 of the parent thread, making that filesystem busy. Threads
2337 now change to the working directory of init to avoid this.
2341 Frequency : during shutdown only
2342 Description: shutdown with a failed MDS or OST can cause unmount to hang
2343 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2348 Frequency : 2.6 only
2349 Description: chmod/chown may include an extra supplementary group
2350 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2351 supplementary group and if none is specified this is used.
2354 Bugzilla : 5479 (6816)
2356 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2357 Details : If lookup is in progress on a file that is unlinked we might try
2358 to revalidate the inode and fail in revalidate after lookup is
2359 complete and ll_file_open() enqueues the open again but
2360 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2363 Frequency : always, if lconf --abort_recovery used
2365 Description: lconf --abort_recovery fails with 'Operation not supported'
2366 Details : lconf was attempting to abort recovery on the MDT device and not
2369 ------------------------------------------------------------------------------
2371 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2376 Frequency : rare (only unsupported configurations with a node running as an
2378 Bugzilla : 6514, 5137
2379 Description: Mounting a Lustre file system on a node running as an OST could
2381 Details : OSTs now preallocates memory needed to write out data at
2382 startup, instead of when needed, to avoid having to
2383 allocate memory in possibly low memory situations.
2384 Specifically, if the file system is mounted on on OST,
2385 memory pressure could force it to try to write out data,
2386 which it needed to allocate memory to do. Due to the low
2387 memory, it would be unable to do so and the node would
2388 become unresponsive.
2390 Severity : enhancement
2392 Description: Addition of lconf --service command line option
2393 Details : lconf now accepts a '--service <arg>' option, which is
2394 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2396 Severity : enhancement
2398 Description: Failover mode is now the default for OSTs.
2399 Details : By default, OSTs will now run in failover mode. To return to
2400 the old behaviour, add '--failout' to the lmc line for OSTs.
2402 Severity : enhancement
2404 Description: Health checks are now provided for MDS and OSTs
2405 Details : Additional detailed health check information on MSD and OSTs
2406 is now provided through the procfs health_check value.
2409 Frequency : occasional, depends on IO load
2411 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2412 after numerous create/delete cycles
2413 Details : The ext3 inode allocation policy would not allocate new inodes
2414 very well on the OSTs because there are no new directories
2415 being created. Instead we look for groups with free space if
2416 the parent directories are nearly full.
2421 Description: Network or server problems during mount may cause partially
2422 mounted clients instead of returning an error.
2423 Details : The config llog parsing code may overwrite the error return
2424 code during mount error handling, returning success instead
2430 Description: MDS can fail to allocate large reply buffers
2431 Details : After long uptimes the MDS can fail to allocate large reply
2432 buffers (e.g. zconf client mount config records) due to memory
2433 fragmentation or consumption by the buffer cache. Preallocate
2434 some large reply buffers so that these replies can be sent even
2435 under memory pressure.
2439 Frequency : rare (liblustre)
2440 Description: fsx running with liblustre complained that using truncate() to
2441 extend the file doesn't work. This patch corrects that issue.
2442 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2443 ATTR_SIZE and lsm use in llu_setattr_raw.
2447 Frequency : rare, only 2.6 kernels
2448 Description: Unusual file access patterns on the MDS may result in inode
2449 data being lost in very rare circumstances.
2450 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2451 code in the 2.6 kernel caused the inode and/or EA data not to
2452 be read from disk, causing single-file corruption.
2456 Frequency : rare, only 2.6 filesystems using extents
2457 Description: Heavy concurrent write and delete load may cause data corruption.
2458 Details : It was possible under high-load situations to have an extent
2459 metadata block in the block device cache from a just-unlinked
2460 file overwrite a newly-allocated data block. We now unmap any
2461 metadata buffers that alias just-allocated data blocks.
2465 Frequency : filesystems with default stripe_count larger than 77
2466 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2467 Details : lconf specifies an inode size of 4096 bytes when the default
2468 stripe_count is larger than 77. This conflicts with the default
2469 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2470 this case to avoid pinning too much memory for large EAs.
2472 ------------------------------------------------------------------------------
2474 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2479 Frequency : rare (extremely heavy IO load with hundreds of clients)
2481 Description: Client is evicted, gets IO error writing to file
2482 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2483 caused clients to be evicted for AST timeouts. The fixes in
2484 bug 5192 mean we no longer need to have such short AST timeouts
2485 so ldlm_timeout has been increased.
2488 Frequency : occasional during --force or --failover shutdown under load
2489 Bugzilla : 5949, 4834
2490 Description: Server oops/LBUG if stopped with --force or --failover under load
2491 Details : a collection of import/export refcount and cleanup ordering
2492 issues fixed for safer force cleanup
2495 Frequency : only filesystems larger than 120 OSTs
2496 Bugzilla : 5990, 6223
2497 Description: lfs getstripe would oops on a very large filesystem
2498 Details : lov_getconfig used kfree on vmalloc'd memory
2501 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2502 Bugzilla : 6242, 6243
2503 Description: reading from files that had been truncated to a non-zero size
2504 but never opened returned no data
2505 Details : ll_file_read() reads zeros from no-object files to EOF
2510 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2511 Details : There's little harm in aborting MDS/OSS recovery and letting it
2512 try again, so I removed the LASSERT and return an error instead.
2514 Severity : enhancement
2516 Description: New debugging infrastructure for tracking down data corruption
2517 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2518 (b) cover more of the client-side code path, and (c) try to narrow
2519 down where problems occurred
2523 Bugzilla : 3819, 4364, 4397, 6313
2524 Description: Racing close and eviction MDS could cause assertion in mds_close
2525 Details : It was possible to get multiple mfd references during close and
2526 client eviction, leading to one thread referencing a freed mfd.
2528 Severity: : enhancement
2529 Bugzilla : 3262, 6359
2530 Description: Attempts to reconnect to servers are now more aggressive.
2531 Details : This builds on the enhanced upcall-less recovery that was added
2532 in 1.4.2. When trying to reconnect to servers, clients will
2533 now try each server in the failover group every 10 seconds. By
2534 default, clients would previously try one server every 25 seconds.
2539 Description: After recovery, certain operations trigger a failed
2540 assertion on a client.
2541 Details : Failing over an mds, using lconf -d --failover, while a
2542 client was doing a readdir() call would cause the client to
2543 LBUG after recovery completed and the readdir() was resent.
2545 Severity : enhancement
2547 Description: Default groups are now added by lconf
2548 Details : You can now run lconf --group <servicename> without having to
2549 manually add groups with lmc.
2552 Frequency : occasional
2554 Description: Nodes with an elan id of 0 trigger a failed assertion
2557 Frequency : always when accessing e.g. tty/console device nodes
2559 Description: tty and some other devices nodes cannot be used on lustre
2560 Details : file's private_data field is used by device data and lustre
2561 values in there got lost. New field was added to struct file to
2562 store fs-specific private data.
2565 Frequency : when exporting Lustre via NFS
2567 Description: NFSD failed occasionally when looking up a path component
2568 Details : NFSD is looking up ".." which was broken in ext3 directories
2569 that had grown large enough to become hashed.
2572 Frequency : Clusters with multiple interfaces not on the same subnet
2574 Description: Nodes will repeatedly try to reconnect to an interface which it
2575 cannot reach and report an error to the log.
2576 Details : Extra peer list entries will be created by lconf with some peers
2577 unreachable. lconf now validates the peer before adding it.
2580 Frequency : Only if a default stripe is set on the filesystem root.
2582 Description: Setting a default stripe on the filesystem root prevented the
2583 filesystem from being remounted.
2584 Details : The client was sending extra request flags in the root getattr
2585 request and did not allocate a reply buffer for the dir EA.
2588 Frequency : occasional, higher if lots of files are accessed by one client
2589 Bugzilla : 6159, 6097
2590 Description: Client trips assertion regarding lsm mismatch/magic
2591 Details : While revalidating inodes the VFS looks up inodes with ifind()
2592 and in rare cases can find an inode that is being freed.
2593 The ll_test_inode() code will free the lsm during ifind()
2594 when it finds an existing inode and then the VFS later attaches
2595 this free lsm to a new inode.
2599 Bugzilla : 6422, 7030
2600 Description: MDS deadlock between mkdir and client eviction
2601 Details : Creating a new file via mkdir or mknod (starting a transaction
2602 and getting the ns lock) can deadlock with client eviction
2603 (gets ns lock and trying to finish a synchronous transaction).
2606 Frequency : occasional
2607 Description: While starting a server, the fsfilt_ext3 module could not be
2609 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2610 kernels. Previously, lconf would still use the ext3 name
2611 when trying to load modules. Now, it will correctly use
2612 ext3 on 2.4 and ldiskfs on 2.6.
2614 Severity : enhancement
2615 Description: The default stripe count has been changed to 1
2616 Details : The interpretation of the default stripe count (0, to lfs
2617 or lmc) has been changed to mean striping across a single
2618 OST, rather than all available. For general usage we have
2619 found a stripe count of 1 or 2 works best.
2621 Severity : enhancement
2622 Description: Add support for compiling against Cray portals.
2623 Details : Conditional compiling for some areas that are different
2627 Frequency : occasional
2628 Bugzilla : 6409, 6834
2629 Description: Creating files with an explicit stripe count may lead to
2630 a failed assertion on the MDS
2631 Details : If some OSTs are full or unavailable, creating files may
2632 trigger a failed assertion on the MDS. Now, Lustre will
2633 try to use other servers or return an error to the
2637 Frequency : occasional
2639 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2640 caused lock timeout during readahead (since 1.4.2).
2641 Details : Processes doing readahead might match a lock that hasn't been
2642 granted yet if there are overlapping and conflicting lock
2643 requests. The readahead process waits on ungranted lock
2644 (original lock is CBPENDING), while OST waits for that process
2645 to cancel CBPENDING read lock and eventually evicts client.
2647 Severity : enhancement
2649 Description: Initial enabling of flock support for clients
2650 Details : Implements fcntl advisory locking and file status functions.
2651 This feature is provided as an optional mount flag (default
2652 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2653 locking are implemented yet, and those that are are not guaranteed
2654 to be completely correct in production environments.
2655 mount -t lustre -o [flock|noflock] ...
2658 Frequency : occasional
2660 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2661 assertion in the ext3 JBD (journaling) layer.
2662 Details : The b_committed_data struct is protected by the big kernel lock
2663 in 2.4 kernels, serializing journal_commit_transaction() and
2664 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2665 access to this struct. In 2.6 kernels there is finer grained
2666 locking to improve SMP performance of the JBD layer.
2670 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2672 -----------------------------------------------------------------------------
2674 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2676 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2677 versions of Lustre. Please update all servers and clients to
2678 version 1.4.2 or later at the same time. You must also run
2679 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2680 to update the configuration logs.
2682 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2683 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2684 - swab llog records of type '0' so we get proper header size/idx (5861)
2685 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2686 - fix rename of one directory over another leaking an inode (5953)
2687 - avoid SetPageDirty on 2.6 (5981)
2688 - don't re-add just-being-destroyed locks to the waiting list (5653)
2689 - when creating new directories, inherit the parent's custom
2690 striping settings if present parent (3048)
2691 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2692 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2693 - don't hold client locks on temporary worklist from l_lru (5666)
2694 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2695 - automatically evict dead clients (5921)
2696 - Update file size properly in create+truncate+fstat case (6196)
2697 - Do not unhash mountpoint dentries, do not allow removal of
2699 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2700 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2701 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2702 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2703 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2704 - Correctly handle reads of files with no objects (6243)
2705 - lctl recover will also mark a device active if deactivate used (5933)
2707 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2708 - allow --write-conf on an MDS with different nettype than client (5619)
2709 - don't write config llogs to MDS for mounts not from that MDS (5617)
2710 - lconf should create multiple TCP connections from a client (5201)
2711 - init scripts are now turned off by default; run chkconfig --on
2712 lustre and chkconfig --on lustrefs to use them
2713 - upcalls are no longer needed for clients to recover to failover
2715 - add --abort-recovery option to lconf to abort recovery on device
2717 - add support for an arbitrary number of OSTs (3026)
2718 - Quota support protocol changes.
2719 - forward compatibility changes to wire structs (6007)
2720 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2721 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2722 - improved "lustre status" script
2723 - initialize blocksize for non-regular files (6062)
2724 - added --disable-server and --disable-client configure options (5782)
2725 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2726 - Vanilla 2.4.29 support
2727 - increase maximum number of obd devices to 520 (6242)
2728 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2729 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2730 SLES 9 2.6.5 kernels
2731 - limit stripes per file to 160 (the maximum EA size) (6093)
2733 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2736 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2737 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2738 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2739 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2740 - small ext3 extents cleanups and fixes (5733)
2741 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2742 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2743 - avoid needless client->OST connect, fix handle mismatch (5317)
2744 - fix DLM error path that led to out-of-sync client, long delays (5779)
2745 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2746 - fix several locking issues related to i_size (5492,5624,5654,5672)
2747 - don't move pending lock onto export if it is already evicted (5683)
2748 - fix kernel oops when creating .foo in unlinked directory (5548)
2749 - fix deadlock in obdfilter statistics vs. object create (5811)
2750 - use time_{before,after} to avoid timer jiffies wrap (5882)
2751 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2752 - Do not leak request if server was not able to process it (5154)
2753 - If mds_open unable to find parent dir, make that negative lookup(5154)
2754 - don't create new directories with extent-mapping (5909, 5936)
2756 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2757 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2758 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2759 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2760 - support for RHEL4 kernel on i686 (5773)
2761 - provide error messages when incompatible logs are encountered (5898)
2763 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2764 * version 1.4.0.10 (1.4.1 release candidate 1)
2766 - don't keep a lock reference when lock is not granted (4238)
2767 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2768 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2769 - drop import inflight refcount on signal_completed_replay error (5255)
2770 - unlock page after async write error during send (3677)
2771 - handle missing objects in filter_preprw_read properly (5265)
2772 - no transno return for symlink open, don't save no-trasno open (3440)
2773 - don't try to complete elan receive that already failed (4012)
2774 - free RPC server reply state on error (5406)
2775 - clean up thread from ptlrpc_start_thread() on error (5160)
2776 - readahead could read extra page into cache that wasn't ejected (5388)
2777 - prevent races in class_attach/setup/cleanup/detach (5260)
2778 - don't dereference de->d_inode after l_dput of de (5458)
2779 - use "int" for stripe value returned from lock_to_stripe (5544)
2780 - mballoc allocation and error-checking fixes in 2.6 (5504)
2781 - block device patches to fix I/O request sizes in 2.6 (5482)
2782 - look up hostnames for IB nals (5602)
2783 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2784 - don't start multiple acceptors for the same port (5277)
2785 - fix incorrect LASSERT in mds_getattr_name (5635)
2786 - export a proc file for general "ping" checking (5628)
2787 - fix "lfs check" to not block when the MDS is down (5628)
2789 - service request history (4965)
2790 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2791 - create an "evict_client" /proc entry on OSTs, like the MDS has
2792 - fix mount usage message, return errors per mount(8) (5168)
2793 - change grep [] to grep "[]" in tests so they work in more UMLs
2794 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2795 - remove ancient LOV_MAGIC_V0 EA support (5047)
2796 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2797 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2798 - allow readahead allocations to fail when low on memory (5383)
2799 - mmap locking landed again, after considerable improvement (2828)
2800 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2802 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2805 - send OST transaction number in read/write reply to free req (4966)
2806 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2807 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2808 - the watchdog thread now runs as interruptible (5246)
2809 - flock/lockf fixes (but it's still disabled, pending 5135)
2810 - don't use EXT3 constants in llite code (5094)
2811 - memory shortage at startup could cause assertion (5176)
2813 - reorganization of lov code
2814 - single portals codebase
2816 - add extents/mballoc support (5025)
2817 - direct I/O reads in the obdfilter (4048)
2818 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2820 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2823 - send OST transaction number in read/write reply to free req (4966)
2824 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2825 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2826 - the watchdog thread now runs as interruptible (5246)
2827 - handle missing objects in filter_preprw_read properly (5265)
2828 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2829 - drop import inflight refcount on signal_completed_replay error (5255)
2830 - unlock page after async write error during send (3677)
2831 - return original error code on reconstructed replies (3761)
2832 - no transno return for symlink open, don't save no-trasno open (3440)
2834 - add pid to ldlm debugging output (4922)
2835 - bump the watchdog timeouts -- we can't handle 30sec yet
2836 - extra debugging for orphan dentry/inode bug (5259)
2838 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2841 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2842 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2843 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2844 - fix NULL dereference in /proc/sys/portals/routes (4827)
2845 - allow failed mdc_close() operations to be interrupted (4561)
2846 - stop precreate on OST before MDS would time out on it (4778)
2847 - don't send partial-page writes before EOF from client (4410)
2848 - discard client grant for sub-page writes on large-page clients (4520)
2849 - don't free dentries not owned by NFS code, check generation (4806)
2850 - fix lsm leak if mds_create_objects() fails (4801)
2851 - limit debug_daemon file size, always print CERROR messages (4789)
2852 - use transno after validating reply (3892)
2853 - process timed out requests if import state changes (3754)
2854 - update mtime on OST during writes, return in glimpse (4829)
2855 - add mkfsoptions to LDAP (4679)
2856 - use ->max_readahead method instead of zapping global ra (5039)
2857 - don't interrupt __l_wait_event() during strace
2859 - add software watchdogs to catch hung threads quickly (4941)
2860 - make lustrefs init script start after nfs is mounted
2861 - fix CWARN/ERROR duplication (4930)
2862 - return async write errors to application if possible (2248)
2863 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2864 - print NAL number in %x format (4645)
2865 - update barely-supported suse-2.4.21-171 series (4842)
2866 - support for sles 9 %post scripts
2867 - support for building 2.6 kernel-source packages
2868 - support for sles km_* packages
2870 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2873 - ignore -ENOENT errors in osc_destroy (3639)
2874 - notify osc create thread that OSC is being cleaned up (4600)
2875 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2876 - reconstruct ost_handle() like mds_handle() (4657)
2877 - create a new thread to do import eviction to avoid deadlock (3969)
2878 - let lconf resolve symlinked-to devices (4629)
2879 - don't unlink "objects" from directory with default EA (4554)
2880 - hold socknal file ref over connect in case target is down (4394)
2881 - allow more than 32000 subdirectories in a single directory (3244)
2882 - fix blocks count for O_DIRECT writes (3751)
2883 - OST returns ENOSPC from object create when no space left (4539)
2884 - don't send truncate RPC if file size isn't changing (4410)
2885 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2886 - bind to privileged port in socknal and tcpnal (3689)
2888 - rate limit CERROR/CWARN console message to avoid overload (4519)
2889 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2890 - basic mmap support (3918)
2891 - kernel patch series update from b1_4 (4711)
2893 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2896 - avoid crash during MDS cleanup with OST shut down (2775)
2897 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2898 - don't use bad inodes on the MDS (3744)
2899 - dynamic object preallocation to improve recovery speed (4236)
2900 - don't hold spinlock over lock dumping or change debug flags (4401)
2901 - don't zero obd_dev when it is force cleaned (3651)
2902 - print grants to console if they go negative (4431)
2903 - "lctl deactivate" will stop automatic recovery attempts (3406)
2904 - look for existing locks in ldlm_handle_enqueue() (3764)
2905 - don't resolve lock handle twice in recovery avoiding race (4401)
2906 - revalidate should check working dir is a directory (4134)
2908 - don't always mark "slow" obdfilter messages as errors (4418)
2910 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2913 - don't close LustreDB during write_conf until it is done (3860)
2914 - fix typo in lconf for_each_profile (3821)
2915 - allow dumping logs from multiple threads at one time (3820)
2916 - don't allow multiple threads in OSC recovery (3812)
2917 - fix debug_size parameters (3864)
2918 - fix mds_postrecov to initialize import for llog ctxt (3121)
2919 - replace config semaphore with spinlock (3306)
2920 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2921 - don't allow enqueue to complete on a destroyed export (3822)
2922 - down write_lock before checking llog header bitmap (3825)
2923 - recover from lock replay timeout (3764)
2924 - up llog sem before sending rpc (3652)
2925 - reduce ns lock hold times when setting kms (3267)
2926 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2927 - fix NULL deref and obd_dev leak on setup error (3312)
2928 - replace some LBUG about llog ops with error handling (3841)
2929 - don't match INVALID dentries from d_lookup and spin (3784)
2930 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2931 - fix invalid assertion in ptlrpc_set_wait (3880)
2933 - add libwrap support for the TCP acceptor (3996)
2934 - add /proc/sys/portals/routes for non-root route listing (3994)
2935 - allow setting MDS UUID in .xml (2580)
2936 - print the stack of a process that LBUGs (4228)
2938 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2941 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2942 - make sure to unset replay flag from failed open requests (3440)
2943 - if default stripe count is 0, use OST count for inode size (3636)
2944 - update parent mtime/ctime on client for create/unlink (2611)
2945 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2946 - free recovery state on server during a forced cleanup (3571)
2947 - unregister_reply for resent reqs (3063)
2948 - loop back devices mounting and status check on 2.6 (3563)
2949 - fix resource-creation race that can provoke i_size == 0 (3513)
2950 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2951 - more debugging for page-accounting assertion (3746)
2952 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2953 - avoid deadlock after precreation failure (3758)
2954 - fix race and lock order deadlock in orphan handling (3450, 3750)
2955 - add validity checks when grabbing inodes from l_ast_data (3599)
2957 - add /proc/.../recovery_status to obdfilter (3428)
2958 - lightweight CDEBUG infrastructure, debug daemon (3668)
2959 - change default OSC RPC parameters to be better on small clusters
2960 - turn off OST read cache for files smaller than 32MB
2961 - install man pages and include them in rpms (3100)
2962 - add new init script for (un)mounting lustre filesystems (2593)
2963 - run chkconfig in %post for init scripts (3701)
2964 - drop scimac NAL (unmaintained)
2966 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
2969 - clean kiobufs before and after use (3485)
2970 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
2971 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
2972 - change init script to fail more clearly if not run as root (1528)
2973 - allow clients to reconnect during replay (1742)
2974 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
2975 - don't do DNS lookups on NIDs too small for IP addresses (3442)
2976 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
2977 - fix cond_resched (3554)
2978 - only evict unfinished clients after recovery (3515)
2979 - allow bulk resend, prevent data loss (3570)
2980 - dynamic ptlrpc request buffer allocation (2102)
2981 - don't allow unlinking open directory if it isn't empty (2904)
2982 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
2983 - remove extraneous obd dereference causing LASSERT failure (3334)
2984 - don't use get_cycles() when creating temp. files on the mds (3156)
2985 - hold i_sem when setting i_size in ll_extent_lock() (3564)
2986 - handle EEXIST for set-stripe, set proper directory name (3336)
2988 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
2989 - fix ksocknal_fmb_callback() error messages (2918)
2991 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
2994 - don't copy lvb into (possibly NULL) reply on error (2983)
2995 - don't deref dentry after dput, don't free lvb on error (2922)
2996 - use the kms to determine writeback rpc length (2947)
2997 - increment oti_logcookies when osc is inactive (2948)
2998 - update client's i_blocks count via lvb messages (2543)
2999 - handle intent open/close of special files properly (1557)
3000 - mount MDS with errors=remount-ro, like obdfilter (2009)
3001 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3002 - don't use cancelling-locks' kms values (2947)
3003 - use highest lock extent for kms, not last one (2925)
3004 - don't dereference ERR_PTR() dentry in error handling path (3107)
3005 - fix thread race in portals_debug_dumplog() (3122)
3006 - create lprocfs device entries at setup instead of at attach (1519)
3007 - common AST error handler, don't evict client on completion race (3145)
3008 - zero nameidata in detach_mnt in 2.6 (3118)
3009 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3010 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3011 - initialize RPC timeout timer earlier for 2.6 (3219)
3012 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3013 - print nal/nid for unknown nid (3258)
3014 - additional checks for oscc recovery before doing precreate (3284)
3015 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3016 - don't crash in mdc_close for bad permissions on open (3285)
3017 - zero i_rdev for non-device files (3147)
3018 - clear page->private before handing to FS, better assertion (3119)
3019 - tune the read pipeline (3236)
3020 - fix incorrect decref of invalidated dentry (2350)
3021 - provide read-ahead stats and refine rpc in flight stats (3328)
3022 - don't hold journal transaction open across create RPC (3313)
3023 - update atime on MDS at close time (3265)
3024 - close LDAP connection when recovering to avoid server load (3315)
3025 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3026 - don't leak open file on MDS after open resend (3325)
3027 - serialize filter_precreate and filter_destroy_precreated (3329)
3028 - loop device shouldn't call sync_dev() for nul device (3092)
3029 - clear page cache after eviction (2766)
3030 - resynchronize MDS->OST in background (2824)
3031 - refuse to mount the same filesystem twice on same mountpoint (3394)
3032 - allow llmount to create routes for mounting behind routers (3320)
3033 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3034 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3035 - fix rare problem with rename on htree directories (3417)
3037 - allow default OST striping configuration per directory (1414)
3038 - fix compilation for qswnal for 2.6 kernels (3125)
3039 - increase maximum number of MDS request buffers for large systems
3040 - change liblustreapi to be useful for external progs like lfsck (3098)
3041 - increase local configuration timeout for slow disks (3353)
3042 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3044 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3047 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3048 - don't overwrite extent policy data in reply if lock was blocked (2901)
3049 - drop filter export grants atomically with removal from device (2663)
3050 - del obd_self_export from work_list in class_disconnect_exports (2908)
3051 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3052 - swab reply message in mdc_close, other PPC fixes (2464)
3053 - fix destroying of named logs (2325)
3054 - overwrite old logs when running lconf --write_conf (2264)
3055 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3056 - fix race in target_handle_connect (2898)
3057 - mds_reint_create() should take same inode create lock (2926)
3058 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3059 - don't close files for self_export to avoid uninitialized obd (2936)
3060 - allow MDS with the same name as client node (2939)
3061 - hold dentry reference for closed log files for unlink (2325)
3062 - reserve space for all logs during transactions (2059)
3063 - don't evict page beyond end of stripe extent (2925)
3064 - don't oops on a deleted current working directory (2399)
3065 - handle hard links to targets without a parent properly (2517)
3066 - don't dereference NULL lock when racing during eviction (2867)
3067 - don't grow lock extents when lots of conflicting locks (2919)
3069 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3072 - account for cache space usage on clients to avoid data loss (974)
3073 - lfsck support in lustre kernel code (2349)
3074 - reduce journal credits needed for BRW writes (2370)
3075 - orphan handling to avoid losing space on client/server crashes
3076 - ptlrpcd can be blocked, stopping ALL progress (2477)
3077 - use lock value blocks to assist in proper KMS, faster stat (1021)
3078 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3079 - recovery for initial connections (2355)
3080 - fixes for mds_cleanup_orphans (1934)
3081 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3082 - block all file creations until orphan recovery completes (1901)
3083 - client remove rq_connection from request struct (2423)
3084 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3085 - recovery timer race (2670)
3086 - mdc_close recovey bug (2532)
3087 - ptlrpc cleanup bug (2710)
3088 - mds timeout on local locks (2588)
3089 - namespace lock held during RPCs (2431)
3090 - handle interrupted sync write properly (2503)
3091 - don't try to handle a message that hasn't been replied to (2699)
3092 - client assert failure during cleanup after abort recovery (2701)
3093 - leak mdc device after failed mount (2712)
3094 - ptlrpc_check_set allows timedout requests to complete (2714)
3095 - wait for inflight reqs when ptlrpcd finishes (2710)
3096 - make sure unregistered services are removed from the srv_list
3097 - reset bulk XID's when resending them (caught by 1138 test)
3098 - unregister_bulk after timeout
3099 - fix lconf error (2694)
3100 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3101 - readahead locks pages, leaves pending causing memory pressure (2673)
3102 - increase OST request buffers to 4096 on large machines (2729)
3103 - fix up permission of existing directories in simple_mkdir (2661)
3104 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3105 - don't assign transno to errored transactions (2742)
3106 - don't delete objects on OST if given a bogus objid from MDS (2751)
3107 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3108 - if rq_no_resend, then timeout request after recovery (2432)
3109 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3110 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3111 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3112 - MDS recovery shouldn't delete objects that it has given out (2730)
3113 - if enqueue arrives after completion, don't clobber LVB (2819)
3114 - don't unlock pages twice when trigger_group_io returns error (2814)
3115 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3116 - don't write pages to disk if there was an error (1450)
3117 - don't ping imports that have recovery disabled (2676)
3118 - take buffered bytes into account when balancing socknal conn (2817)
3119 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3120 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3121 - remove little-endian swabbing of llog records (1987)
3122 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3123 - retry reposting request buffers if they fail (1191)
3124 - grow extent at grant time to avoid granting a revoked lock (2809)
3125 - lock revoke doesn't evict page if covered by a second lock (2765)
3126 - disable VM readahead to avoid reading outside lock extents (2805)
3128 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3129 - updated kernel patches for hp-2.4.20 kernel (2681)
3131 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3134 - fix truncated write corruption (2366)
3135 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3137 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3138 - ensure len(uuid) < 37 in lmc (1171)
3139 - fix ia64 OOPS in llog_test (2255)
3140 - zero end of page at obdfilter for partial page writes (2648)
3141 - don't leave stale dentries around after renames (bug 2428)
3142 - fix timeouts when evicting a client with a single lock held (2642)
3143 - set deadline for the initial HELLO message to drain (2634)
3144 - print out dotted-quad IP addresses in the socknal (2302)
3146 - additional debugging for MDS client eviction problem (2443)
3147 - fix mkfsoptions support for osts (2603, 2604)
3149 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3152 - add series for the vanilla 2.6.0 kernel
3153 - add series for the vanilla 2.4.24 kernel
3154 - add series for a cray x86/64 UL kernel drop
3155 - fix xattr patches for the vanilla 2.4.19 series
3157 - generate true UUIDs in lmc (1171)
3158 - have portals stack dumping break in UML (2466)
3159 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3160 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3161 - wake callback waiting threads on client eviction (2460)
3162 - Add --ptldebug and --subsystem to lmc (1719)
3163 - update assertion to allow safe interrupt allocation
3164 - set rq_no_resend for cancel requests (2432)
3165 - recalculate ptlrpcd timeout after resend (2494)
3166 - call vfs_rmdir when removing pending directories (2368)
3167 - fix renaming a file to itself (2429)
3168 - lmc creates a default one-stripe lov (2454)
3169 - expand procfs space to handle large clusters (2326)
3170 - increase UML stack to avoid overflow
3171 - update lconf's list of debug and subsystem masks
3172 - fix lfs find --obd (2510)
3173 - /proc tunable for disabling filter read caching (2591)
3174 - stop rpm packages from altering slapd.conf (2301)
3175 - disable nagle in the socknal under 0conf (2578)
3176 - choose mds inode size based on stripe count (2572)
3177 - fix kernel-source rpm problems (2516)
3179 - add --disable-doc to avoid pdf generation (2421)
3180 - update documentation, tests, type-os, comments
3181 - avoid format warnings on ia64
3182 - remove the TOE NAL
3183 - tiny code cleanups by removing unused fields
3185 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3188 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3189 - remove the most blatant lies from BUILDING file (2371)
3190 - change default debug level to reasonable production setting
3191 - reduce client side cache size to reduce cache flush time
3192 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3193 - make TCP zerocopy and pinger support enabled by default (2476)
3194 - sync writes completed after process exits caused crashes (2319)
3195 - maintain correct mount count on the MDS (2356)
3196 - backout 1557, because 2316 wasn't really fixed
3197 - better file I/O statistics gathering in /proc
3198 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3199 - another kernel patch to fix zero-copy TCP function export
3200 - don't take duplicate lock when processing re-sent getattr (2420)
3201 - lctl uses obd_self_export instead of creating new conn (2353)
3202 - MDS/OST recovery case which requires object creation asserted (2425)
3203 - move lfs from /usr/sbin to /usr/bin in packages
3204 - fix race between mds_client_add and mds_client_free (2417)
3205 - use kmalloc instead of slabs in portals (2430)
3206 - don't create duplicate records when a failover MDS is present (2442)
3207 - remove unnecessary mount age check (2332)
3208 - don't remove directory inodes from locks prematurely (2451)
3209 - don't break if MDS service name is the same as hostname (2103)
3210 - fix races in client write RPC generation when cache full (2482)
3212 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3215 - remove now-unused request->rq_obd (278)
3216 - if an allocation fails, print out how much memory we've used (1933)
3217 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3218 - add the "configurable stack size" patch to most series files (1256)
3219 - ability to write large log records, for 100+ OST configs (2306)
3220 - fix NULL deref when filter_prep fails (2314)
3221 - fix operator precedence error in filter_sync
3222 - dynamic allocation of socknal TX descriptors (2315)
3223 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3224 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3225 - ensure that CWARN messages in Portals always get to the syslog
3226 - __init/__exit are not for prototype decls (ldlm_init/exit)
3227 - x86-64 compile warning fixes
3228 - fix gateway LMC keyword conflict (2318)
3229 - fix MDS lock inversions in getattr/reint paths (1844)
3230 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3231 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3232 - DLM race condition prevented some lock evictions (2328)
3233 - ENOMEM detection and retry on socknal sends (2230)
3234 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3235 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3236 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3237 - better error messages when a client is rejected during recovery (1505)
3238 - avoid cancelling locks which were never granted, after failure (2330)
3239 - fix i_sem/journal inversion in mds_client_add (2333)
3240 - fix truncate/getattr lock cycle deadlock (2334)
3241 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3242 - fix two rare exit paths which could leak an l_lock() ref (2321)
3243 - fencepost error in MDS/OST orphan recovery (2226)
3244 - make log record alignment 8 bytes (1988)
3245 - lstripe now fails when requested offset > ost_count (2237)
3246 - ensure that all kernel series have a complete list.h (1607)
3247 - fix crashes in special-file operations (2316)
3248 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3250 - allow configurable automake binary, for testing new versions
3251 - small update to the lfs documentation
3253 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3255 * fix negative export reference count in fsfilt_sync (2312)
3257 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3258 * release candidate 0.9.1
3260 - orphans are moved into the PENDING directory for possible recovery
3261 - replayed opens now open by fid for orphan/rename safety (1042)
3262 - last close of an orphan inode generates a transno (683)
3263 - chdir() and mount() now pin the directory entry (1020)
3264 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3265 - discard very old requests without processing them (1502)
3266 - remove some common, well-understood CERRORs (1505)
3267 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3268 - clear "grant" flags in OST replies until OST grant code lands (1644)
3269 - fix read performance by not clobbering i_blksize on client (1598)
3270 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3271 - make LRU size a /proc tunable, clears locks when reduced (707)
3272 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3273 - server locks take references on exports now (1558)
3274 - build fixes for 2.4.20-rh trees (1663)
3275 - return an error from lov_create if all OSCs are inactive (1751)
3276 - fix import levels when a reconnect happens without a timeout (1597)
3277 - exit early from mds_open if we get a lookup error (1749)
3278 - partial page read at EOF wouldn't wait for disk before sending (1642)
3279 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3280 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3281 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3282 - write/truncate lock inversion (1639)
3283 - Don't auto-load obdclass, portals modules during cleanup (1495)
3284 - fix timestamps from jumping to "now" (1763)
3285 - extra journal assertions (1648)
3286 - add an extra multiunlink test (1771)
3287 - fix read_record/write_record API (1776)
3288 - fix leak of offset_extent, possible incorrect i_size later (1772)
3289 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3290 - Debugging for the kqswnal_get_idle_tx problems (1820)
3291 - Allow recovery to be attempted multiple times (1536)
3292 - Write out MDS last_rcvd file after it is first created (1600)
3293 - Fix tx_descriptor leak in failed transmit situations (1827)
3294 - ext3 journaling fixes for assertion failure after IO error (1871)
3295 - class_export_put() on freed export after completion AST error (1896)
3296 - Fix revalidate looping in VFS (1322)
3297 - Don't access a freed export during MDS_REINT timeout (1521)
3298 - Add open-unlink recovery support on the MDS (1673,1764)
3299 - Return an error if no MDS data was read from last_rcvd (1946)
3300 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3301 - Don't setup a disk device that doesn't match exported UUID (317)
3302 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3303 - avoid committing NULL handle in force close
3304 - local.sh is now a one-stripe LOV configuration
3305 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3306 - fix ext3 htree duplicate directory entry corruption (1516)
3307 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3308 - update to new LOV EA format (2097)
3309 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3311 2003-06-15 Phil Schwan <phil@clusterfs.com>
3314 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3315 - per-import/export recovery handling (958, 931, 959)
3316 - multiple last-rcvd slots, for serving multiple FSes (949)
3317 - connections are again shared between multiple imp/exports (963, 964)
3318 - "umount -f" would hang if any requests needed to be sent (393, 978)
3319 - avoid pinning large req buffer by copying for queued messages (989)
3320 - add "uuid" to "lctl device" command to help upcalls (991)
3321 - "open" RPCs with transnos would confuse recovery counters (1037)
3322 - do proper endian conversion of all wire messages (288, 340, 891)
3323 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3324 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3325 - fix ptlrpc_connection leak in target_handle_connect (1174)
3326 - fix import refcounting bug in OST and MDS cleanup (1134)
3327 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3328 - fix very unlikely obd_types race condition (501)
3329 - remove osc_open hack for echo_client (1187)
3330 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3331 - a failure in read_inode2 leads to deadlock (1139)
3332 - cancel ack-locks as soon as transaction is committed (1072)
3333 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3334 - make sure to commitrw after any preprw to avoid deadlock (1162)
3335 - failing to execute a file in a lustre FS would lock inode (1203)
3336 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3337 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3338 - obd_uuid2tgt would walk off the end of the list (1255)
3339 - on IA64 the osc would give portals incorrect bulk size (1258)
3340 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3341 - fix lock inversion caused by new llite matching code (1282)
3342 - limit the number of dirty pages on a client to 10MB (1286)
3343 - timed out locks were not being corrected cancelled (1289)
3344 - fix O_DIRECT above 4GB on IA-32 (1292)
3345 * major user-visible changes
3346 - fail out/fail over policy now controlled by the upcall (993)
3348 - add OBD_PING to check server availability and failure (954)
3349 - lustre messages are now sent in sending host order (288, 340, 891)
3350 - add eadatalen to MDS getattr reply (340)
3351 - OST read replies may contain second buffer, with per-page status (593)
3353 2003-03-11 Phil Schwan <phil@clusterfs.com>
3356 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3357 - failed open()s could cause deadlock; fixed (867, 869)
3358 - stop cancelling OST locks when files are closed (481)
3359 - overlapping XID spaces caused network corruption (851, 853)
3360 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3361 - setattr_raw added, to do single-RPC, server-side setattrs
3362 - lmc/lconf syntax change for OST UUIDs
3363 - fix crashy race condition between ptlrpc_free_req and osc_close
3364 - don't use request in mdc_enqueue if we hit a timeout (889)
3365 - don't set the inode i_size for regular files from the MDS (896)
3366 - handle out of order completion AST (842)
3367 - don't LBUG if a lock request times out after receiving AST (913)
3368 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3369 - if a bad lock AST arrives, send an error instead of dropping entirely
3370 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3371 - fix leak in bulk IO when only partially completed (899, 900, 926)
3372 - fix O_DIRECT for ia64 (55)
3373 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3374 - C-z after timeout could hang a process forever; fixed (977)
3376 - client-side I/O cache (678, 924, 929, 941, 970)
3378 - READPAGE and SETATTRs which don't take server-side locks get
3381 2003-02-11 Phil Schwan <phil@clusterfs.com>
3384 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3385 - Fix fsfilt_extN_readpage() to read a full page of directory
3386 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3387 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3388 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3389 - Most intent operations were removed, in favour of a new RPC mode
3390 that does a single RPC to the server and bypasses most of the VFS
3391 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3392 - Aggressively cancel local locks on DLM servers
3393 - mds_reint_unlink sends EA to the client if it's the last nlink.
3394 client uses that EA to unlink OST objects.
3395 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3396 - recursive symlinks were fixed (439)
3397 - fixed NULL deref in DEBUG_REQ
3398 - filter_update_lastobjid no longer calls sync, which annoyed extN
3399 - fixed multi-client small-writes to a single file problem (445)
3400 - fixed mtime updates during file writes (607)
3401 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3402 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3403 - fixed Linux OST size reporting problem (444, 656)
3404 - OST now updates object mtime with writes or setattr (607, 619)
3405 - client verifies file size before zeroing page past EOF (445)
3406 - OST now writes last allocated objid to disk with allocation (108)
3407 - LOV on echo now works (409)
3409 - mds_reint_unlink sends a new buffer, with the EA included. this
3410 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3411 set if a regular file was being unlinked, and it was the last link
3412 - use PtlGet from the target for bulk writes (315)
3413 - OST now updates object mtime with writes or setattr (607, 619)
3414 - LDLM now has a grant-time callback to revalidate locked items, if
3416 - Many MDS operations were reorganized to combat race conditions
3418 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3419 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3421 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3424 - Fully reactivate OST imports after reconnection (512, others)
3425 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3426 - More graceful error handling for truncating on dead OST (515)
3427 - Don't error out unless we're actually accessing dead stripes (474)
3428 - Fix garbage sizes when stripes are missing (410)
3429 - LRU counters were broken, causing constant lock purge (433, 432)
3430 - garbage on read from stripes with failed OSTs (441)
3431 - mark OSCs as active before reconnecting during recovery (438)
3432 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3433 - lfind did not preserve OST order in output (443)
3434 - symlinks cause hung clients, incorrect data (439)
3435 - stop dereferencing request after dropping refcount (457)
3436 - don't LASSERT(spin_is_locked) on non-SMP (455)
3437 - fixes for many rename() bugs
3438 - fstat didn't correctly synchronize attributes (399)
3439 - server must handle lock cancellation during blocking AST prep (487)
3440 - bulk descriptors were free()d too soon (511)
3441 - fix paths in lconf, which would load incorrect modules (451, 507)
3442 - fix confusing lconf 'host not found' error message (386)
3443 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3444 - fix race condition in mdc_blocking_ast() for inode access (526)
3445 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3446 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3447 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3448 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3449 - fix lstripe in several ways (kernel side) (527)
3450 - fix request leak in ldlm_cli_enqueue (262)
3451 - incorrect OSC was marked inactive after OST failure
3452 - call mds_fs_cleanup before unmounting filesystem (524)
3453 - fix races between taking ns_lock and ldlm_lock_change_resource
3454 - fix races updating LOV export open file list
3455 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3456 - fix recovery NULL deref in ldlm_cli_cancel_unused
3457 - fix some DLM races by using new hash table for lock handles (419)
3458 - permit the client to specify desired inodes, at replay
3459 - duplicate requests when we queue them for replay reintegration
3460 - fix last_rcvd offset calculation
3461 - sync after each recovered transaction, so we always make progress
3462 - never, not always, ERESTART requests without transnos
3463 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3465 - skip replay if the MDS says that the client is already connected
3466 - don't check for a recovery-enabled export to match lctl's UUID
3467 - don't INC_USE_COUNT for phantom exports
3468 - don't crash when cleaning up phantom exports (567)
3469 - don't double-finish or set replay data for errored mdc_open requests
3470 - abort requests when they time out, so we don't get old replies
3471 - send/receive replies for AST messages again
3472 - if the client says that it doesn't have the lock, cancel it on the
3474 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3475 mark it as destroyed, it will all work out when decref is called
3476 - fix module use counts (22, 581)
3478 - ASTs now expect a reply (server cancels lock on error reply)
3480 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3483 - fix many simultaneous client startup (392)
3484 - fix dentry->d_it clobbering
3485 - credentials weren't being shipped for readdir/getattr operations
3486 - remove invalid assertions triggered during some concurrent MD
3488 - proper Lustre versions added (336, 389)
3489 - fix memory leak for create error case (398)
3490 - fix LOV locking bug that would get cli/srv out of sync
3491 - fix echo client over LOV (409)
3492 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3493 - fix double-O_EXCL intent crash (424)
3494 - avoid sending multiple lock CANCELs (352)
3496 - MDS can do multi-client recovery (modulo bugs in new code)
3498 - many updates, edits, cleanups
3500 2002-11-18 Phil Schwan <phil@clusterfs.com>
3503 - fix null d_it dereference (346)
3504 - fix full OST/dbench hang (333)
3505 - fix permission problem with file removal (286)
3506 - fix removal of OSCs from LOV when they fail
3507 - fix NULL deref during bulk timeout (214)
3508 - fix problems related to multiple filesystems on one MDS (241)
3509 - fixed serious subtle metadata locking bugs
3510 - free locks on clients when inodes are removed due to memory
3512 - fix inode pointer in lock data (285)
3513 - partial support for multiple MDS on a single host (241)
3514 - data locks weren't cancelled at clear_inode time (290, 311)
3515 - intent locks could lead to unbounded lock growth (205)
3516 - added a maximum lock count, an LRU list, and a flusher
3517 - fix multiple rename (365)
3518 - properly abstracted the echo client
3519 - OSC locked 1 byte too many; fixed
3520 - rewrote brw callback code:
3521 - fixed recovery bugs related to LOVs (306)
3522 - fixed too-many-pages-in-one-write crash (191)
3523 - fixed (again) crash in sync_io_timeout (214)
3524 - probably fixed callback-related race (385)
3526 - Add capability to MDS protocol
3527 - LDLM cancellations and callbacks on different portals
3529 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3532 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3533 - defer open object destruction to close time (601981, 138)
3534 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3535 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3536 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3537 - don't call obd_setattr in ll_file_release for destroyed objects
3538 * protocol change to lustre_msg: move |version| and add |flags|
3539 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3540 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3541 * added replay of create, unlink, link and rename operations during
3542 MDS failover; recovery should be much more robust now
3543 * remove failed OSCs from LOVs (only lov_create uses this so far)
3544 * the lustre-HOWTO was brought (more) up to date (582544)
3546 2002-10-23 Phil Schwan <phil@clusterfs.com>
3549 - in-use dentries weren't being reused properly (617851)
3550 - prevent multiple LDLM setup (599178)
3551 - fix LOV size calculations for truncate (617853)
3552 - fix client handling of MDS intent errors (POSIX)
3553 - fix permission bug in lovstripe.c test (624321)
3554 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3555 - truncate past end of file could corrupt data
3556 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3557 - a race in recovery could return ETIMEDOUT to apps (623947)
3558 - building outside the source directory was fixed
3559 * the lustre-HOWTO was brought (more) up to date (582544)
3560 * major progress was made on recovery functionality
3562 2002-10-10 Phil Schwan <phil@clusterfs.com>
3565 - recovery deadlock fix
3566 - rm -rf causes LBUG fix (617817)
3567 - file open by multiple tasks fix (618962)
3568 - directory permissions bugs (602707 and 620007)
3569 - journal_stop fixed with locking (611313)
3570 - O_APPEND failures resolved (618273, perhaps 614459)
3571 - lconf PATH fix (619770)
3572 - IA64 build fix (621450)
3573 - RPC buffer sizes scale with amount of memory
3575 2002-10-01 Phil Schwan <phil@clusterfs.com>
3578 - locks would be cancelled without throwing away data pages,
3579 resulting in inconsistent data (605627)
3580 - inode attributes were not always being refreshed (605627, 612449)
3581 - lconf now continues to cleanup after lctl reports an error
3582 - MDS now enforces user permissions (602707)
3583 - lprocfs cleanup fixed, but not yet enabled (614157)
3584 - fixed infinite server hang, should a client not respond to an AST
3585 - avoid going into recovery if user calls readlink() with a buffer
3586 that's too small (613941)
3587 - AST RPCs no longer require replies (614867) -- this may be changed
3588 - don't crash server if client sends an IOV that's too big (611336)
3589 - fixed lock conversion deadlock (611892)
3590 - fixed the following of symlinks (614622)
3591 * recovery: the server can remove locks from a client that dies, other
3592 clients can make progress
3593 * more extN patch fixes
3594 * compile-time configurable ptlrpc buffer allocations
3596 - collaborative read cache document
3597 - Lustre Lite Performance CDR document-in-progress
3599 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3602 - fix typo in patch-2.4.18
3604 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3607 - clear ptlrpc request each time in handle_incoming_request()
3608 - unlink of files now destroys the object on the OST
3610 2002-09-19 Peter Braam <braam@clusterfs.com>
3612 * add hard link support
3613 * change obdfile creation method
3614 * kernel patch changed
3616 2002-09-19 Peter Braam <braam@clusterfs.com>
3619 - stack overflow bug in extN fixed
3621 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3623 * documentation updates
3624 - add man pages for config tools
3625 - update tests/README to describe testing with new config tools
3626 - finish metadata API descriptions
3627 * bug fixes and cleanups
3628 - statfs workaround for 16TB limit
3629 - LOV stripe allocation improved, can stripe on subset of OSTs
3630 - LOV file size/IO offset was wrong for files > 4GB in size
3631 - object EA data was being dropped, caused files to be unreadable
3632 - memory overflow with non-LOV OST caused memory corruption
3633 - fixed regression tests to work with new config tools, obdfilter
3634 - fixed bug when directory size became larger than 1 block
3635 - fixed bug (for single client case) when PWD was deleted
3636 - invalidate local directory pages when doing intent-based ops
3637 - avoid LDLM oops when lock callback contained bad data
3639 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3641 * documentation updates
3642 * bug fixes and cleanups
3643 - configuration tools
3646 - 64-bit compile warnings
3647 - 64-bit internal statfs data
3649 * test_brw on persistent OST devices
3651 * lprocfs (disabled)
3653 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3655 * documentation updates
3656 * bug fixes and cleanups
3657 * configuration tools
3659 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3662 * many small fixes to 0_5_4
3663 * io/network handling
3664 * thinkos in MDS operations
3666 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3669 * crucial basic fixes to 0.5.3
3670 * IOR, Iozone work over Elan
3673 2002-08-07 Phil Schwan <phil@clusterfs.com>
3674 * version 0_5_3, our first alpha
3675 * we use the new Portals iovs
3676 * documentation updates
3677 * bug fixes and cleanups
3678 * small changes in the DLM wire protocol
3680 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3681 * version 0_5_1 with some initial stability,
3682 * locking on MD and file I/O.
3683 * documentation updates
3684 * several bug fixes since 0.5.0
3685 * small changes in wire protocol
3687 2002-07-18 Phil Schwan <phil@clusterfs.com>
3689 * delivered as Lustre Light Alpha
3690 * fixed a crash after handling invalid MDS requests
3691 * fixed directory pages for architectures with non-4k pages sizes
3693 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3694 * release version v0_4_4
3695 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3697 * Fixes a number of interruption problems with OST operations.
3698 * Update documentation for portals header changes
3699 * Move all wire protocol structs/defines to lustre_idl.h
3700 * Fixes symlink length bug.
3701 * Add tcpdump to repository.
3703 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3704 * release version v0_4_3
3705 * Fixes statfs for inodes on extN.
3706 * Fixes bug in runtests which would delete /etc/hosts.
3707 * Use 64-bit object IDs wherever possible (not into VFS though)
3708 Remove ost_get_info, which is unused by lustre, and out of date.
3710 2002-07-03 Peter Braam <braam@clusterfs.com>
3711 * release version v0_4_2 Fixes a lookup error (type not passed)
3712 * move forward to head of Portals
3713 * move forward to latest Lustre kernel
3715 2002-06-25 Peter Braam <braam@clusterfs.com>
3716 * release version v0_4_1. Hopefully stable on single node use.