1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5 2.6.16.46-0.14 (SLES 10)
6 2.6.18.8 vanilla (kernel.org)
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.9-42.0.10.EL (RHEL 4)
10 2.6.16 - 2.6.21 vanilla (kernel.org)
11 * Recommended e2fsprogs version: 1.39.cfs8
12 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
16 Description: don't use GFP_* in generic Lustre code.
17 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
19 --------------------------------------------------------------------------------
21 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
23 * Support for kernels:
26 2.6.16.46-0.14 (SLES 10)
27 2.6.18.8 vanilla (kernel.org)
28 * Client support for unpatched kernels:
29 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
30 2.6.16 - 2.6.22 vanilla (kernel.org)
31 * Due to recently discovered recovery problems, we do not recommend
32 using patchless RHEL 4 clients with this or any earlier release.
33 * Recommended e2fsprogs version: 1.39.cfs8
34 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
35 * Starting with this release, the ldiskfs backing filesystem required
36 by Lustre is now in its own package, lustre-ldiskfs. This package
37 should be installed. It is versioned separately from Lustre and
38 may be released separately in future.
40 Severity : enhancement
42 Description: add optional extra BUILD_VERSION info
43 Details : add a new environment variable (namely LUSTRE_VERS) which allows
44 to override the lustre version.
47 Frequency : 2.6.18 servers only
49 Description: ll_kern_mount() doesn't release the module reference
50 Details : The ldiskfs module reference count never drops down to 0
51 because ll_kern_mount() doesn't release the module reference.
56 Description: server LBUG when using old ost_num_threads parameter
57 Details : Accept the old ost_num_threads parameter but warn that it
58 is deprecated, and fix an off-by-one error that caused an LBUG.
63 Description: Transient SCSI error results in persistent IO issue
64 Details : iobuf->dr_error is not reinitialized to 0 between two
68 Frequency : sometimes when underlying device returns I/O errors
70 Description: OSTs not going read-only during write failures
71 Details : OSTs are not remounted read-only when the journal commit threads
72 get I/O errors because fsfilt_ext3 calls journal_start/stop()
73 instead of the ext3 wrappers.
77 Description: poor connect scaling with increasing client count
78 Details : Don't run filter_grant_sanity_check for more than 100 exports
79 to improve scaling for large numbers of clients.
82 Frequency : SLES10 only
84 Description: sanity-quota.sh quotacheck failed: rc = -22
85 Details : Quotas cannot be enabled on SLES10.
88 Frequency : liblustre clients only
90 Description: getdirentries does not give error when run on compute nodes
91 Details : getdirentries does not fail when the size specified as an argument
92 is too small to contain at least one entry
94 Severity : enhancement
96 Description: Add LNET router traceability for debug purposes
97 Details : If a checksum failure occurs with a router as part of the
98 IO path, the NID of the last router that forwarded the bulk data
99 is printed so it can be identified.
104 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
105 Details : Due to a race condition, liblustre clients were occasionally
108 Severity : enhancement
110 Description: lfs setstripe use optional parameters instead of postional
113 Severity : enhancement
115 Description: Nanosecond timestamp support for ldiskfs
116 Details : The on-disk ldiskfs filesystem has added support for nanosecond
117 resolution timestamps. There is not yet support for this at
118 the Lustre filesystem level.
121 Frequency : during server recovery
123 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
124 Details : request with rq_no_resend flag not awake l_wait_event if they get a
128 Frequency : nfs export on patchless client
130 Description: connectathon hang when test nfs export over patchless client
131 Details : Disconnected dentry cannot be found with lookup, so we do not need
132 to unhash it or make it invalid
135 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
136 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
137 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
142 Description: open req refcounting wrong on reconnect
143 Details : If reconnect happened between getting open reply from server and
144 call to mdc_set_replay_data in ll_file_open, we will schedule
145 replay for unreferenced request that we are about to free.
146 Subsequent close will crash in variety of ways.
147 Check that request is still eligible for replay in
148 mdc_set_replay_data().
153 Description: disable writes to filesystem when reading health_check file
154 Details : the default for reading the health_check proc file has changed
155 to NOT do a journal transaction and write to disk, because this
156 can cause reads of the /proc file to hang and block HA state
157 checking on a healthy but otherwise heavily loaded system. It
158 is possible to return to the previous behaviour during configure
159 with --enable-health-write.
161 Severity : enhancement
163 Description: 64-bit inode version
164 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
165 to the inode. This will be required for version-based recovery.
170 Description: MDS fails to start if a duplicate client export is detected
171 Details : in some rare cases it was possible for a client to connect to
172 an MDS multiple times. Upon recovery the MDS would detect this
173 and fail during startup. Handle this more gracefully.
175 Severity : enhancement
177 Description: Add -o localflock option to simulate old noflock
179 Details : This will achieve local-only flock/fcntl locks
185 Description: log_commit_thread vs filter_destroy race leads to crash
186 Details : Take import reference before releasing llog record semaphore
191 Description: Wrong request locking in request set processing
192 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
193 imp_delayed_list, in this place should be used imp_lock.
196 Frequency : when reconnection
198 Description: Grant Leak when osc reconnect to OST
199 Details : When osc reconnect ost, OST(filter) should check whether it
200 should grant more space to client by comparing fed_grant and
201 cl_avail_grant, and return the granted space to client instead
202 of "new granted" space, because client will call osc_init_grant
203 to update the client grant space info.
206 Frequency : when client reconnect to OST
208 Description: Grant Leak when osc do resend and replay bulk write
209 Details : When osc reconnect to OST, OST(filter)should clear grant info of
210 bulk write request, because the grant info while be sync between
211 OSC and OST when reconnect, and we should ignore the grant info
212 these of resend/replay write req.
217 Description: Grant space more than avaiable left space sometimes.
218 Details : When then OST is about to be full, if two bulk writing from
219 different clients came to OST. Accord the avaliable space of the
220 OST, the first req should be permitted, and the second one
221 should be denied by ENOSPC. But if the seconde arrived before
222 the first one is commited. The OST might wrongly permit second
223 writing, which will cause grant space > avaiable space.
226 Frequency : when client is evicted
228 Description: Grant might be wrongly erased when osc is evicted by OST
229 Details : when the import is evicted by server, it will fork another
230 thread ptlrpc_invalidate_import_thread to invalidate the
231 import, where the grant will be set to 0. While the original
232 thread will update the grant it got when connecting. So if
233 the former happened latter, the grant will be wrongly errased
234 because of this race.
239 Description: Checking Stale with correct fid
240 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
241 to check whether it is stale, because sometimes, we want the
242 enqueue happened anyway, and op_data.fid2 will not be initialized.
244 Severity : enhancement
246 Description: update patchless client
247 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
250 Frequency : only with 2.4 kernel
252 Description: random memory corruption
253 Details : size of struct ll_inode_info is to big for union inode.u and this
254 can be cause of random memory corruption.
259 Description: Memory leak in recovery
260 Details : Lov_mds_md was not free in an error handler in mds_create_object.
261 It should also check obd_fail before fsfilt_start, otherwise if
262 fsfilt_start return -EROFS,(failover mds during mds recovery).
263 then the req will return with repmsg->transno = 0 and rc = EROFS.
264 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
265 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
266 be freed no matter whether fsfilt_commit success or not.
269 Frequency : only with huge count clients
271 Description: Prevents from taking the superblock lock in llap_from_page for
273 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
274 for llap_from_page call in ll_removepage prevents from taking the
275 superblock lock for a soon died page.
280 Description: Not check open intent error before release open handle
281 Details : in some rare cases, the open intent error is not checked before
282 release open handle, which may cause
283 ASSERTION(open_req->rq_transno != 0), because it tries to release
284 the failed open handle.
289 Description: Set cat log bitmap only after create log success.
290 Details : in some rare cases, the cat log bitmap is set too early. and it
291 should be set only after create log success.
295 Description: Accessing a block bevice can re-enable I/O when Lustre is
296 tearing down a device.
297 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
302 Description: service threads may hog cpus when there are a lot of requests
304 Details : Insert cond_resched to give other threads a chance to use some of
310 Description: the cat log was not initialized in recovery
311 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
312 unlink log on mds will not be initialized until mds post
313 recovery. And also in mds post recovery, the unlink log will
314 initialization will be done asynchronausly, so there will be race
315 between add unlink log and unlink log initialization.
319 Description: brw_stats were being printed incorrectly
320 Details : brw_stats were being printed as log2 but all of them were not
321 recorded as log2. Also remove some code duplication arising from
322 filter_tally_{read,write}.
326 Frequency : rare, only in recovery.
327 Description: ASSERTION(req->rq_type != LI_POISON) failed
328 Details : imp_lock should be held while iterating over imp_sending_list for
329 prevent destroy request after get timeout in ptlrpc_queue_wait.
333 Description: replay-single.sh test 52 fails
334 Details : A lock's skiplist need to be cleanup when it being unlinked
335 from its resource list.
339 Description: Short directio read returns full requested size rather than
341 Details : Direct I/O operations should return actual amount of bytes
342 transferred rather than requested size.
344 Severity : enhancement
346 Description: metadata RPC reduction (e.g. for rm performance)
347 Details : decrease the amount of synchronous RPC between clients and servers
348 by canceling conflicing lock before the operation on the client side
349 and packing thier handles into the main operation RPC to server.
351 --------------------------------------------------------------------------------
353 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
358 Frequency : on some architectures
360 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
361 Details : Uninitialized flags sometimes cause configuration commands to
365 Frequency : patchless clients only
367 Description: missing __iget() symbol export
368 Details : The __iget() symbol export is missing. To avoid the need for
369 this on patchless clients the deathrow inode reaper is turned
370 off, and we depend on the VM to clean up old inodes. This
371 dependency was during via the fix for bug 12181.
373 --------------------------------------------------------------------------------
375 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
377 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
378 INTEROPERATE with older versions automatically. In many cases a
379 special upgrade step is needed. Please read the
380 user documentation before upgrading any part of a live system.
381 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
382 version will not interoperate with 1.6 betas before beta5 (1.5.95).
383 * WARNING: Lustre configuration and startup changes are required with
384 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
389 Severity : enhancement
391 Description: MountConf
392 Details : Lustre configuration is now managed via mkfs and mount
393 commands instead of lmc and lconf. New obd types (MGS, MGC)
394 are added for dynamic configuration management. See
395 https://mail.clusterfs.com/wikis/lustre/MountConf for
398 Severity : enhancement
400 Description: dynamic OST addition
401 Details : OSTs can now be added to a live filesystem
403 Severity : enhancement
405 Description: startup order invariance
406 Details : MDTs and OSTs can be started in any order. Clients only
407 require the MDT to complete startup.
409 Severity : enhancement
411 Description: parallel, asynchronous orphan cleanup
412 Details : orphan cleanup is now performed in separate threads for each
413 OST, allowing parallel non-blocking operation.
415 Severity : enhancement
417 Description: optimized stripe assignment
418 Details : stripe assignments are now made based on ost space available,
419 ost previous usage, and OSS previous usage, in order to try
420 to optimize storage space and networking resources.
422 Severity : enhancement
424 Description: Permanently set tunables
425 Details : All writable /proc/fs/lustre tunables can now be permanently
426 set on a per-server basis, at mkfs time or on a live
429 Severity : enhancement
431 Description: Lustre message v2
432 Details : Add lustre message format v2.
434 Severity : enhancement
436 Description: client OST exclusion list
437 Details : Clients can be started with a list of OSTs that should be
438 declared "inactive" for known non-responsive OSTs.
442 Description: SPEC SFS validation failure on NFS v2 over lustre.
443 Details : Changes the blocksize for regular files to be 2x RPC size,
444 and not depend on stripe size.
446 Severity : enhancement
448 Description: Multiple MD RPCs in flight.
449 Details : Further unserialise some read-only MDS RPCs - learn about intents.
450 To avoid overly-overloading MDS, introduce a limit on number of
451 MDS RPCs in flight for a single client and add /proc controls
452 to adjust this limit.
454 Severity : enhancement
456 Description: client read/write statistics
457 Details : Add client read/write call usage stats for performance
458 analysis of user processes.
459 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
460 file access. extents_stats shows chunk size distribution.
461 extents_stats_per_process show chunk size distribution per
464 Severity : enhancement
466 Description: mds statistics
467 Details : Add detailed mds operations statistics in
468 /proc/fs/lustre/mds/*/stats.
472 Description: Failure of copying files with lustre special EAs.
473 Details : Client side always return success for setxattr call for lustre
474 special xattr (currently only "trusted.lov").
479 Description: Refcount LNET uuids
480 Details : The global LNET uuid list grew linearly with every startup;
481 refcount repeated list entries instead of always adding to
484 Severity : enhancement
486 Description: Dynamic service threads
487 Details : Within a small range, start extra service threads
488 automatically when the request queue builds up.
491 Frequency : mixed-endian client/server environments
493 Description: mixed-endian crashes
494 Details : The new msg_v2 system had some failures in mixed-endian
497 Severity : enhancement
499 Description: Easy OST removal
500 Details : OSTs can be permanently deactivated with e.g. 'lctl
501 conf_param lustre-OST0001.osc.active=0'
503 Severity : enhancement
505 Description: MGS proc entries
506 Details : Added basic proc entries for the MGS showing what filesystems
509 Severity : enhancement
511 Description: provide MGS failover
512 Details : Added config lock reacquisition after MGS server failover.
514 Severity : enhancement
516 Description: add Linux 2.4 support
517 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
521 Description: a large application tries to do I/O to the same resource and dies
523 Details : Check the req->rq_arrival time after the call to
524 ost_brw_lock_get(), but before we do anything about
525 processing it & sending the BULK transfer request. This
526 should help move old stale pending locks off the queue as
527 quickly as obd_timeout.
530 Frequency : when an incorrect nid is specified during startup
532 Description: ptlrpc connect to non-existant node causes kernel crash
533 Details : LNET can't be re-entered from an event callback, which
534 happened when we expire a message after the export has been
535 cleaned up. Instead, hand the zombie cleanup off to another
538 Severity : enhancement
540 Description: plain/inodebits lock performance improvement
541 Details : Grouping plain/inodebits in granted list by their request modes
542 and bits policy, thus improving the performance of search through
546 Frequency : only if OST filesystem is corrupted
548 Description: client incorrectly hits assertion in ptlrpc_replay_req()
549 Details : for a short time RPCs with bulk IO are in the replay list,
550 but replay of bulk IOs is unimplemented. If the OST filesystem
551 is corrupted due to disk cache incoherency and then replay is
552 started it is possible to trip an assertion. Avoid putting
553 committed RPCs into the replay list at all to avoid this issue.
556 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
558 Description: System hang on startup
559 Details : This bug allowed the liblustre (e.g. catamount) client to
560 return to the app before handling all startup RPCs. This
561 could leave the node unresponsive to lustre network traffic
562 and manifested as a server ptllnd timeout.
564 Severity : enhancement
566 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
567 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
568 boolean to control whether to print peer debug info when a
569 client's RPC times out.
572 Frequency : only for kernels with patches from Lustre below 1.4.3
574 Description: Remove old rdonly API
575 Details : Remove old rdonly API which unused from at least lustre 1.4.3
578 Frequency : only for devices with external journals
580 Description: Set external device read-only also
581 Details : During a commanded failover stop, we set the disk device
582 read-only while the server shuts down. We now also set any
583 external journal device read-only at the same time.
586 Frequency : when upgrading from 1.4 while trying to change parameters
588 Description: The wrong (new) MDC name was used when setting parameters for
589 upgraded MDT's. Also allows changing of OSC (and MDC)
590 parameters if --writeconf is specified at tunefs upgrade time.
593 Frequency : when setting specific ost indicies
595 Description: QOS code breaks on skipped indicies
596 Details : Add checks for missing OST indicies in the QOS code, so OSTs
597 created with --index need not be sequential.
599 Severity : enhancement
601 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
602 Details : The uninit_groups feature works in conjunction with the kernel
603 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
604 up the pass1 processing of e2fsck. This is a read-only feature
605 in ldiskfs2 only, so older kernels and current ldiskfs cannot
606 mount filesystems that have had this feature enabled.
608 Severity : enhancement
610 Description: Improve multi-block allocation algorithm to avoid fragmentation
611 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
612 allocation locality and avoid filesystem fragmentation.
614 ------------------------------------------------------------------------------
616 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
618 * Support for kernels:
619 2.4.21-47.0.1.EL (RHEL 3)
621 2.6.9-42.0.10.EL (RHEL 4)
622 2.6.12.6 vanilla (kernel.org)
623 2.6.16.27-0.9 (SLES 10)
624 * Recommended e2fsprogs version: 1.39.cfs5
626 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
630 Frequency : occasional, depends on client load and configuration
631 Bugzilla : 12181, 12203
632 Description: data loss for recently-modified files
634 Details : In some cases it is possible that recently written or created
635 files may not be written to disk in a timely manner (this should
636 normally be within 30s unless client IO load is very high).
637 The problem appears as zero-length files or files that are a
638 multiple of 1MB in size after a client crash or client eviction
639 that are missing data at the end of the file.
641 This problem is more likely to be hit on clients where files are
642 repeatedly created and unlinked in the same directory, clients
643 have a large amount of RAM, have many CPUs, the filesystem has
644 many OSTs, the clients are rebooted frequently, and/or the files
645 are not accessed by other nodes after being written.
647 The presence of the problem can be detected by looking at
648 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
649 smaller than the second (nr_unused) then dirty files will not
650 be flushed automatically to disk. "sync; sleep 10" should be
651 run several times on the node before unmounting it to update
652 Lustre (this is also safe to run on nodes without this problem).
654 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
655 that can cause this same problem, so customers using that kernel
656 also need to update the kernel in addition to Lustre. In order
657 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
659 It is normal that files written just before a client crash (less
660 than 30s) may not yet have been flushed to disk, even for local
664 Frequency : frequent on thin XT3 nodes
666 Description: UUID collision on thin XT3 Linux nodes
667 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
668 randomly, since we relied on an insufficiently-seeded PRNG.
673 Description: OSS hangs after "All ost request buffers busy"
674 Details : A deadlock between quota and journal operations caused OSS
675 hangs after printing "All ost request buffers busy."
678 Frequency : always on liblustre builds
680 Description: Cleanup compiler warnings on liblustre
683 Frequency : always on liblustre builds on XT3
685 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
689 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
690 > 32000 subdirectories
691 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
692 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
693 subdirectory count crosses 32000. This will aid e2fsck to
694 correctly handle more than 32000 subdirectories.
697 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
699 Description: System hang on startup
700 Details : This bug allowed the liblustre (e.g. catamount) client to
701 return to the app before handling all startup RPCs. This
702 could leave the node unresponsive to lustre network traffic
703 and manifested as a server ptllnd timeout.
705 Severity : enhancement
707 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
708 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
709 boolean to control whether to print peer debug info when a
710 client's RPC times out.
715 Description: make O_SYNC working on 2.6 kernels
716 Details : 2.6 kernels use different method for mark pages for write,
717 so need add a code to lustre for O_SYNC work.
722 Description: Failure to close file and release space on NFS
723 Details : Put inode details into lock acquired in ll_intent_file_open.
724 Use mdc_intent_lock in ll_intent_open to properly
725 detect all kind of errors unhandled by mdc_enqueue.
730 Description: proc file read during shutdown sometimes raced obd removal,
732 Details : Add lock to prevent obd access after proc file removal.
735 Frequency : Only for files larger than 4GB on 32-bit clients.
737 Description: improperly doing page alignment of locks
738 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
739 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
744 Description: RPCs being resent when they shouldn't be
745 Details : Some RPCs that should not be resent are being resent. This
746 can cause inconsistencies in the RPC state machine. Do not
747 resend such requests.
750 Frequency : rare, only with NFS export
752 Description: Crash on NFS re-export node
753 Details : under very unusual load conditions an assertion is hit in
754 ll_intent_file_open()
757 Frequency : only if OST filesystem is corrupted
759 Description: client incorrectly hits assertion in ptlrpc_replay_req()
760 Details : for a short time RPCs with bulk IO are in the replay list,
761 but replay of bulk IOs is unimplemented. If the OST filesystem
762 is corrupted due to disk cache incoherency and then replay is
763 started it is possible to trip an assertion. Avoid putting
764 committed RPCs into the replay list at all to avoid this issue.
769 Description: large O_DIRECT requests fail under memory pressure/fragmentation
770 Details : Large single O_DIRECT read and write calls can fail to allocate
771 a sufficiently large buffer to process the request. In case of
772 allocation failure the allocation is retried with a smaller
773 buffer and broken into smaller requests.
775 Severity : enhancement
777 Description: Add -o localflock option to simulate old noflock behaviour.
778 Details : This will achieve local-only flock/fcntl locks coherentness.
783 Description: versioning check is incomplete
784 Details : Checking the version difference of client vs. server, report
785 error if the gap is too big.
790 Description: add support PG_writeback bit
791 Details : add support for PG_writeback bit for Lustre, for more carefull
792 work with page cache in 2.6 kernel. This also fix some deadlocks
793 and remove hack for work O_SYNC with 2.6 kernel.
795 Severity : enhancement
797 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
798 Details : The uninit_groups feature works in conjunction with the kernel
799 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
800 up the pass1 processing of e2fsck. This is a read-only feature
801 in ldiskfs2 only, so older kernels and current ldiskfs cannot
802 mount filesystems that have had this feature enabled.
804 Severity : enhancement
806 Description: Improve multi-block allocation algorithm to avoid fragmentation
807 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
808 allocation locality and avoid filesystem fragmentation.
810 ------------------------------------------------------------------------------
812 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
814 * Support for kernels:
815 2.6.9-42.0.3.EL (RHEL 4)
817 2.4.21-47.0.1.EL (RHEL 3)
818 2.6.12.6 vanilla (kernel.org)
819 2.6.16.21-0.8 (SLES10)
820 * Recommended e2fsprogs version: 1.39.cfs2-0
822 * The backwards-compatible /proc/sys/portals symlink has been removed
823 in this release. Before upgrading, please ensure that you change
824 any configuration scripts or /etc/sysctl.conf files that access
825 /proc/sys/portals/* or sysctl portals.* to use the corresponding
826 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
827 in advance of the upgrade on any system running Lustre 1.4.6 or
828 newer, since /proc/sys/lnet was added in that version.
829 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
833 Frequency : only when quota is used
835 Description: avoid scanning export list for quota master
836 Details : Change the algorithms to avoid scanning export list in order
837 to improve the efficiency.
840 Frequency : MDS failover only, very rarely
842 Description: "went back in time" messages on mds failover
843 Details : The greatest transno may be lost when the current operation
844 finishes with an error (transno==0) and the client's last_rcvd
845 record is over-written. Save the greatest transno in the
846 mds_last_transno for this case.
849 Frequency : always for specific kernels and striping counts
851 Description: client may get "Matching packet too big" without ACL support
852 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
853 when trying to access files in certain configurations. The
854 clients should in fact be denied when mounting because they do
858 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
860 Description: many clients connecting with IO in progress causes connect timeouts
861 Details : Avoid synchronous journal commits to avoid delays caused by many
862 clients connecting/disconnecting when bulk IO is in progress.
863 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
864 OST_IO_PORTAL to avoid delays behind potentially many pending
868 Frequency : occasionally with multiple writers to a single file
870 Description: shared writes to file may result in wrong size reported by stat()
871 Details : Allow growing of kms when extent lock is cancelled
874 Frequency : always with random mmap IO to multi-striped file
876 Description: mmap write might be lost if we are writing to a 'hole' in stripe
877 Details : Only if the hole is at the end of OST object so that kms is too
878 small. Fix is to increase kms accordingly in ll_nopage.
881 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
883 Description: writes to a missing object would leak memory on the OST
884 Details : If there is an inconsistency between the MDS and OST filesystems,
885 such that the MDS references an object that doesn't exist, writes
886 to that object will leak memory due to incorrect cleanup in the
887 error handling path, eventually running out of memory on the OST.
892 Description: Creating too long symlink causes lustre errors
893 Details : Check symlink and name lengths before sending requests to MDS.
896 Frequency : only if flock is enabled (not on by default)
898 Description: posix locks not released on fd closure on 2.6.9+
899 Details : We failed to add posix locks to list of inode locks on 2.6.9+
900 kernels, this caused such locks not to be released on fd close and
901 then assertions on fs unmount about still used locks.
904 Frequency : MDS failover only, very rarely
906 Description: clients may get ASSERTION(granted_lock != NULL)
907 Details : When request was taking a long time, and a client was resending
908 a getattr by name lock request. The were multiple lock requests
909 with the same client lock handle and
910 mds_getattr_name->fixup_handle_for_resent_request found one of the
911 lock handles but later failed with ASSERTION(granted_lock != NULL).
916 Description: handle->h_buffer_credits > 0, assertion failure
917 Details : h_buffer_credits is zero after truncate, causing assertion
918 failure. This patch extends the transaction or creates a new
922 Frequency : NFS re-export or patchless client
923 Bugzilla : 11179, 10796
924 Description: Crash on NFS re-export node (__d_move)
925 Details : We do not want to hash the dentry if we don't have a lock.
926 But if this dentry is later used in d_move, we'd hit uninitialised
927 list head d_hash, so we just do this to init d_hash field but
928 leave dentry unhashed.
931 Frequency : NFS re-export or patchless client
933 Description: NFS exports has problem with symbolic link
934 Details : lustre client didn't properly install dentry when re-exported
935 to NFS or running patchless client.
938 Frequency : NFS re-export or patchless client
940 Description: Various nfs/patchless fixes.
941 Details : fixes reuse disconected alias for lookup process - this fixes
942 warning "find_exported_dentry: npd != pd",
943 fix permission error with open files at nfs.
944 fix apply umask when do revalidate.
947 Frequency : occasional
949 Description: Crash on NFS re-export node
950 Details : calling clear_page() on the wrong pointer triggered oops in
951 generic_mapping_read().
954 Frequency : rarely, using O_DIRECT IO
956 Description: unaligned directio crashes client with LASSERT
957 Details : check for unaligned buffers before trying any requests.
960 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
962 Description: stale data returned from RAID cache
963 Details : If only a small amount of IO is done to the RAID device before
964 reading it again it is possible to get stale data from the RAID
965 cache instead of reading it from disk.
968 Frequency : always for sles10 kernel
970 Description: sles10 support
971 Details : ll_follow_link: compile fixes and using of nd_set_link
975 Frequency : depends on arch, kernel and compiler version, always on sles10
978 Description: recursive or deep enough symlinks cause stack overflow
979 Details : getting rid of large stack-allocated variable in
983 Frequency : depends on hardware
985 Description: lustre write performance loss in the SLES10 kernel
986 Details : the performance loss is caused by using of write barriers in the
987 ext3 code. The SLES10 kernel turns barrier support on by
988 default. The fix is to undo that change for ldiskfs.
990 ------------------------------------------------------------------------------
992 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
994 * Support for kernels:
995 2.6.9-42.0.3EL (RHEL 4)
997 2.4.21-47.0.1.EL (RHEL 3)
998 2.6.12.6 vanilla (kernel.org)
1002 Frequency : quota enabled and large files being deleted
1004 Description: releasing more than 4GB of quota at once hangs OST
1005 Details : If a user deletes more than 4GB of files on a single OST it
1006 will cause the OST to spin in an infinite loop. Release
1007 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1012 Description: statfs data retrieved from /proc may be stale or zero
1013 Details : When reading per-device statfs data from /proc, in the
1014 {kbytes,files}_{total,free,avail} files, it may appear
1015 as zero or be out of date.
1018 Frequency : systems with MD RAID1 external journal devices
1020 Description: lconf's call to blkid is confused by RAID1 journal devices
1021 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1022 of returning all block devices that match the journal UUID.
1025 Frequency : always, for aggregate stripe size over 4GB
1027 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1028 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1029 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1032 Frequency : always if "lfs find" used on a local file/directory
1034 Description: "lfs find" segfaults if used on a local file/directory
1035 Details : The case where a directory component was not specified wasn't
1036 handled correctly. Handle this properly.
1039 Frequency : always on ppc64
1041 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1042 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1043 perform endianness conversion.
1046 Frequency : rarely (truncate to non-zero file size after write under load)
1047 Bugzilla : 10730, 10687
1048 Description: Files padded with zeros to next 4K multiple
1049 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1050 it is possible that files that are truncated to a non-zero size
1051 immediately after being written are filled with zero bytes beyond
1052 the truncated size. No file data is lost.
1054 Severity : enhancement
1056 Description: Allow recovery/failover for liblustre clients.
1057 Details : liblustre clients were unaware of failover configurations until
1060 Severity : enhancement
1062 Description: user file locks should fail when not mounting with flock option
1063 Details : Set up an error-returning stub in ll_file_operations.lock field
1064 to prevent incorrect behaviour when client is mounted without
1065 flock option. Also, set up properly f_op->flock field for
1069 Frequency : always on ia64
1071 Description: "lfs df" loops on printing out MDS statfs information
1072 Details : The obd_ioctl_data was not initialized and in some systems
1073 this caused a failure during the ioctl that did not return
1074 an error. Initialize the struct and return an error on failure.
1077 Frequency : SLES 9 only
1079 Description: Error of copying files with lustre special EAs as root
1080 Details : Client side always return success for setxattr call for lustre
1081 special xattr (currently only "trusted.lov").
1084 Frequency : rarely on clusters with both ia64+i386 clients
1086 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1087 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1088 clients are doing concurrent writes to the same file it is
1089 possible that the ia64 clients may LASSERT because the OST
1090 extent locks are not PAGE_SIZE aligned. Ensure that grown
1091 locks are always aligned on the request boundary.
1094 Frequency : specific use, occasional
1096 Description: Overwriting in use executable truncates on-disk binary image
1097 Details : If one node attempts to overwrite an executable in use by
1098 another node, we now correctly return ETXTBSY instead of
1099 truncating the file.
1101 Severity : enhancement
1103 Description: Async OSC create to avoid the blocking unnecessarily.
1104 Details : If a OST has no remain object, system will block on the creating
1105 when need to create a new object on this OST. Now, ways use
1106 pre-created objects when available, instead of blocking on an
1107 empty osc while others are not empty. If we must block, we block
1108 for the shortest possible period of time.
1113 Description: chmod on Lustre root is propagated to other clients
1114 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1115 invalid by the follow_mount time.
1118 Frequency : liblustre clients only
1120 Description: Race in 'instant cancel' lock handling could lead to such locks
1121 never to be granted in case of SMP MDS
1122 Details : Do not destroy not yet granted but cbpending locks in
1126 Frequency : replay/resend of open
1128 Description: non null lock assetion failure in mds_intent_policy
1129 Details : Trying to replay/resend lockless open requests resulted in
1130 mds_open() returning 0 with no lock. Now it sets a flag if
1131 a lock is going to be returned.
1133 Severity : enhancement
1135 Description: Checksum enhancements
1136 Details : New checksum enhancements allow for resending RPCs that failed
1139 Severity : enhancement
1141 Description: Tunables on number of dirty pages in cacche
1142 Details : Allow to set limit on number of dirty pages cached.
1147 Description: client crash on unmount - lock still has references
1148 Details : In some error handling cases it was possible to leak a lock
1149 reference on a client while accessing a file. This was not
1150 harmful to the client during operation, but would cause the
1151 client to crash when the filesystem is unmounted.
1154 Frequency : specific case, rare
1156 Description: ETXTBSY on mds though file not in use by client
1157 Details : ETXTBSY is no longer incorrectly returned when attempting to
1158 chmod or chown a directory that the user previously tried to
1159 execute or a currently-executing binary.
1162 Frequency : extremely rare except on liblustre-based clients
1164 Description: Lustre space not freed when files are deleted
1165 Details : Clean up open-unlinked files after client eviction. Previously
1166 the unlink was skipped and the files remained as orphans.
1171 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1172 Details : In some cases it was possible to send a blocking callback to
1173 a client doing a glimpse, even though that client didn't get
1174 a lock granted. When the glimpse lock is cancelled on the OST
1175 the freed lock is left on the waiting list and corrupted the list.
1178 Frequency : all core dumps
1180 Description: Broke core dumps to lustre
1181 Details : Negative dentry may be unhashed if parent does not have UPDATE
1182 lock, but some callers, e.g. do_coredump, expect dentry to be
1183 hashed after successful create, hash it in ll_create_it.
1185 ------------------------------------------------------------------------------
1187 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1189 * Support for kernels:
1190 2.6.9-42.0.2.EL (RHEL 4)
1191 2.6.5-7.276 (SLES 9)
1192 2.4.21-40.EL (RHEL 3)
1193 2.6.12.6 vanilla (kernel.org)
1197 Frequency : always on RHEL 3
1199 Description: Number of open files grows over time
1200 Details : The number of open files grows over time, whether or not
1201 Lustre is started. This was due to a filp leak introduced
1202 by one of our kernel patches.
1204 ------------------------------------------------------------------------------
1206 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1208 * Support for kernels:
1209 2.6.9-42.EL (RHEL 4)
1210 2.6.5-7.276 (SLES 9)
1211 2.4.21-40.EL (RHEL 3)
1212 2.6.12.6 vanilla (kernel.org)
1217 Bugzilla : 5719, 9635, 9792, 9684
1218 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1219 Details : If a server is under heavy load and cannot reply to new
1220 connection requests before the client resends the (re)connect,
1221 the connection handling code can behave badly if two service
1222 threads are concurrently handing separate (re)connections from
1223 the same client. Add better locking to the connection handling
1224 code, and ensure that only a single connection will be processed
1225 for a given client UUID, even if the lock is dropped.
1227 Severity : enhancement
1229 Description: add TCP zero-copy support to kernel
1230 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1231 sends if the hardware supports scatter-gather and checksumming.
1232 This allows socklnd to do client-write and server-read more
1233 efficiently and reduce CPU utilization from skbuf copying.
1236 Frequency : only if NFS exporting from client
1238 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1239 Details : If setting attributes on a file created under NFS that had
1240 never been opened it would be possible to oops the client
1241 if the file had no objects.
1244 Frequency : always for liblustre
1246 Description: liblustre client does MDS+OSTs setattr RPC for each write
1247 Details : When doing a write from a liblustre client, the client
1248 incorrectly issued an RPC to the MDS and each OST the file was
1249 striped over in order to update the timestamps. When writing
1250 with small chunks and many clients this could overwhelm the MDS
1251 with RPCs. In all cases it would slow down the write because
1252 these RPCs are unnecessary.
1254 Severity : enhancement
1256 Description: allow number of MDS service threads to be changed at module load
1257 Details : It is now possible to change the number of MDS service threads
1258 running. Adding "options mds mds_num_threads={N}" to the MDS's
1259 /etc/modprobe.conf will set the number of threads for the next
1260 time Lustre is restarted (assuming the "mds" module is also
1261 reloaded at that time). The default number of threads will
1262 stay the same, 32 for most systems.
1267 Description: OST crash if filesystem is unformatted or corrupt
1268 Details : If an OST is started on a device that has never been formatted
1269 or if the filesystem is corrupt and cannot even mount then the
1270 error handling cleanup routines would dereference a NULL pointer.
1275 Description: NULL pointer deref in llap_from_page.
1276 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1277 incorrect) mapping if the page was truncated/reclaimed while it was
1278 searched for. Check for this condition and skip such pages when
1279 doing readahead. Introduce extra check to llap_from_page() to
1280 verify page->mapping->host is non-NULL (so page is not anonymous).
1283 Frequency : Sometimes when using sys_sendfile
1285 Description: "page not covered by a lock" warnings from ll_readpage
1286 Details : sendfile called ll_readpage without right page locks present.
1287 Now we introduced ll_file_sendfile that does necessary locking
1288 around call to generic_file_sendfile() much like we do in
1292 Frequency : with certain MDS communication failures at client mount time
1294 Description: NULL pointer deref after failed client mount
1295 Details : a client connection request may delayed by the network layer
1296 and not be sent until after the PTLRPC layer has timed out the
1297 request. If the client fails the mount immediately it will try
1298 to clean up before the network times out the request. Add a
1299 reference from the request import to the obd device and delay
1300 the cleanup until the network drops the request.
1303 Frequency : occasionally during client (re)connect
1305 Description: assertion failure during client (re)connect
1306 Details : processing a client connection request may be delayed by the
1307 client or server longer than the client connect timeout. This
1308 causes the client to resend the connection request. If the
1309 original connection request is replied in this interval, the
1310 client may trip an assertion failure in ptlrpc_connect_interpret()
1311 which thought it would be the only running connect process.
1314 Frequency : only with obd_echo servers and clients that are rebooted
1316 Description: kernel BUG accessing uninitialized data structure
1317 Details : When running an obd_echo server it did not start the ping_evictor
1318 thread, and when a client was evicted an uninitialized data
1319 structure was accessed. Start the ping_evictor in the RPC
1320 service startup instead of the OBD startup.
1322 Severity : enhancement
1323 Bugzilla : 10193 (patchless)
1324 Description: Remove dependency on various unexported kernel interfaces.
1325 Details : No longer need reparent_to_init, exit_mm, exit_files,
1326 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1329 Frequency : rare (only users of deprecated and unsupported LDAP config)
1331 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1332 Details : LDAP apparently contains 'lustreName' attributes instead of
1333 'name'. A simple remapping of the name is sufficient.
1336 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1338 Description: waiting_locks_callback trips kernel BUG if client is evicted
1339 Details : Running with the dump_on_timeout debug flag turned on makes
1340 it possible that the waiting_locks_callback() can try to dump
1341 the Lustre kernel debug logs from an interrupt handler. Defer
1342 this log dumping to the expired_lock_main() thread.
1344 Severity : enhancement
1346 Description: Support NFS exporting on 2.6 kernels.
1347 Details : Implement non-rawops metadata methods for NFS server to use without
1348 changing NFS server code.
1351 Frequency : very rare (synthetic metadata workload only)
1353 Description: two racing renames might cause an MDS thread to deadlock
1354 Details : Running the "racer" program may cause one MDS thread to rename
1355 a file from being the source of a rename to being the target of
1356 a rename at exactly the same time that another thread is doing
1357 so, and the second thread has already enqueued these locks after
1358 doing a lookup of the target and is trying to relock them in
1359 order. Ensure that we don't try to re-lock the same resource.
1362 Frequency : only very large systems with liblustre clients
1364 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1365 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1366 all OSTs in parallel. This allows the eviction of stale liblustre
1367 clients to proceed much faster than if they were done in series,
1368 and also offers similar improvements for other set_info RPCs.
1373 Description: excessive CPU usage during initial read phase on client
1374 Details : During the initial read phase on a client, it would agressively
1375 retry readahead on the file, consuming too much CPU and impacting
1376 performance (since 1.4.5.8). Improve the readahead algorithm
1377 to avoid this, and also improve some other common cases (read
1378 of small files in particular, where "small" is files smaller than
1379 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1384 Description: MDS crash when receiving packet with unknown intent.
1385 Details : Do not LBUG in unknown intent case, just return -EFAULT
1387 Severity : enhancement
1388 Bugzilla : 9293, 9385
1389 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1390 Details : Do not serialize getattr (non-intent version) and statfs.
1393 Frequency : occasional, when OST network is overloaded/intermittent
1395 Description: client evicted by OST after bulk IO timeout
1396 Details : If a client sends a bulk IO request (read or write) the OST
1397 may evict the client if it is unresposive to its data GET/PUT
1398 request. This is incorrect if the network is overloaded (takes
1399 too long to transfer the RPC data) or dropped the OST GET/PUT
1400 request. There is no need to evict the client at all, since
1401 the pinger and/or lock callbacks will handle this, and the
1402 client can restart the bulk request.
1405 Frequency : Always when mmapping file with no objects
1407 Description: client crashes when mmapping file with no objects
1408 Details : Check that we actually have objects in a file before doing any
1409 operations on objects in ll_vm_open, ll_vm_close and
1415 Description: Request leak when working with deleted CWD
1416 Details : Introduce advanced request refcount tracking for requests
1417 referenced from lustre intent.
1419 Severity : Enhancement
1421 Description: Cache open file handles on client.
1422 Details : MDS now will return special lock along with openhandle, if
1423 requested and client is allowed to hold openhandle, even if unused,
1424 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1425 closing files for every read/write openration.
1427 Severity : Enhancement
1429 Description: Cache open negative dentries on client when possible.
1430 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1431 negative dentries on lock revocation.
1436 Description: Remounting a client read-only wasn't possible with a zconf mount
1437 Details : It wasn't possible to remount a client read-only with llmount.
1439 Severity : enhancement
1440 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1441 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1442 MPICH is included to add Lustre-specific ADIO interfaces.
1443 This is based closely on the UFS ADIO layer and only differs
1444 in file creation, in order to allow the OST striping to be set.
1445 This is user-contributed code and not supported by CFS.
1450 Description: extended inode attributes (immutable, append-only) work improperly
1451 when 2.4 and 2.6 kernels are used on client/server or vice versa
1452 Details : Introduce kernel-independent values for these flags.
1454 Severity : enhancement
1457 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1458 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1459 granularity. Now, Fractional MB granularity is be supported,
1460 this is very useful for low memory system.
1462 Severity : enhancement
1464 Description: Getattr by fid
1465 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1466 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1467 have locks and avoids getattr rpc requests for it on every stat.
1470 Frequency : Always, for filesystems larger than 2TB
1472 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1473 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1474 filesystems larger than 2TB to be mounted with mballoc enabled.
1477 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1479 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1480 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1481 filesystems larger than 2TB is mounted then the kernel will
1482 silently corrupt the start of the filesystem. CONFIG_LBD is
1483 enabled for all CFS-supported kernels, but the possibility of
1484 this happening with a modified kernel config exists.
1486 Severity : enhancement
1488 Description: add client O_DIRECT support for 2.6 kernels
1489 Details : It is now possible to do O_DIRECT reads and writes to files
1490 in the Lustre client mountpoint on 2.6 kernel clients.
1492 Severity : enhancement
1494 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1495 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1496 parallel, not waiting for response from every OST before sending
1497 a rpc to the next OST.
1502 Description: setattr vs write race when updating file timestamps
1503 Details : Client processes that update a file timestamp into the past
1504 right after writing to the file (e.g. tar) it is possible that
1505 the updated file modification time can be reset to the current
1506 time due to a race between processing the setattr and write RPC.
1508 Severity : enhancement
1510 Description: Bring 'lfs find' closer in line with regular Linux find.
1511 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1512 -print0 options and obtains all the needed info through the lustre
1515 Severity : enhancement
1517 Description: support up to 1024 configured devices on one node
1518 Details : change obd_dev array from statically allocated to dynamically
1519 allocated structs as they are first used to reduce memory usage
1524 Description: Flush dirty partially truncated pages during truncate
1525 Details : Immediatelly flush partially truncated pages in filter_setattr,
1526 this way we completely avoid having any pages in page cache on OST
1527 and can retire ugly workarounds during writes to flush such pages.
1532 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1533 Details : protect inode from truncation within vfs_unlink() context
1534 just take a reference before calling vfs_unlink() and release it
1535 when parent's i_sem is free.
1540 Description: last_id value checked outside lock on OST caused LASSERT failure
1541 Details : If there were multiple MDS->OST object precreate requests in
1542 flight, it was possible that the OST's last object id was checked
1543 outside a lock and incorrectly tripped an assertion. Move checks
1544 inside locks, and discard old precreate requests.
1547 Frequency : always, if extents are used on OSTs
1549 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1550 Details : OSTs using the extents format would not zero the high 16 bits of
1551 the index physical block number. This is not a problem for any
1552 OST filesystems smaller than 16TB, and no kernels support ext3
1553 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1554 new/modified files) and can be fixed for existing filesystems
1555 with e2fsprogs-1.39-cfs1.
1560 Description: import connection selection may be incorrect if timer wraps
1561 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1562 import connections to be ignored if the 32-bit jiffies counter
1563 wraps. Use a 64-bit jiffies counter.
1566 Frequency : very large clusters immediately after boot
1568 Description: LNET request buffers exhausted under heavy short-term load
1569 Details : If a large number of client requests are generated on a service
1570 that has previously never seen so many requests it is possible
1571 that the request buffer growth cannot keep up with the spike in
1572 demand. Instead of dropping incoming requests, they are held in
1573 the LND until the RPC service can accept more requests.
1576 Frequency : Sometimes during replay
1578 Description: Assertion failure in ll_local_open after replay.
1579 Details : If replay happened on an open request reply before we were able
1580 to set replay handler, reply will become not swabbed tripping the
1581 assertion in ll_local_open. Now we set the handler right after
1582 recognising of open request
1585 Frequency : very rare
1587 Description: kernel reports "badness in vsnprintf"
1588 Details : Reading from the "recovery_status" /proc file in small chunks
1589 may cause a negative length in lprocfs_obd_rd_recovery_status()
1590 call to vsnprintf() (which is otherwise harmless). Exit early
1591 if there is no more space in the output buffer.
1593 Severity : enhancement
1595 Description: clear OBD RPC statistics by writing to them
1596 Details : It is now possible to clear the OBD RPC statistics by writing
1597 to the "stats" file.
1602 Description: Client mtime is not the same on different clients after utimes
1603 Details : In some cases, the client was using the utimes() syscall on
1604 a file cached on another node. The clients now validate the
1605 ctime from the MDS + OSTs to determine which one is right.
1610 Description: Inability to activate failout mode
1611 Details : lconf script incorrectly assumed that in pythong string's numeric
1612 value is used in comparisons.
1615 Frequency : always with multiple stripes per file
1617 Description: Inefficient object allocation for mutli-stripe files
1618 Details : When selecting which OSTs to stripe files over, for files with
1619 a stripe count that divides evenly into the number of OSTs,
1620 the MDS is always picking the same starting OST for each file.
1621 Return the OST selection heuristic to the original design.
1626 Description: mount failures may take full timeout to return an error
1627 Details : Under some heavy load conditions it is possible that a
1628 failed mount can wait for the full obd_timeout interval,
1629 possibly several minutes, before reporting an error.
1630 Instead return an error as soon as the status is known.
1632 Frequency : quota enabled and large files being deleted
1634 Description: releasing more than 4GB of quota at once hangs OST
1635 Details : If a user deletes more than 4GB of files on a single OST it
1636 will cause the OST to spin in an infinite loop. Release
1637 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1642 Description: statfs data retrieved from /proc may be stale or zero
1643 Details : When reading per-device statfs data from /proc, in the
1644 {kbytes,files}_{total,free,avail} files, it may appear
1645 as zero or be out of date.
1648 Frequency : systems with MD RAID1 external journal devices
1650 Description: lconf's call to blkid is confused by RAID1 journal devices
1651 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1652 of returning all block devices that match the journal UUID.
1655 Frequency : always, for aggregate stripe size over 4GB
1657 Description: assertion fails when trying to use 4GB stripe size
1658 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
1659 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
1662 Frequency : always on ppc64
1664 Description: the first write on an ext3 filesystem with mballoc got stuck
1665 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1666 perform endianness conversion.
1668 ------------------------------------------------------------------------------
1670 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1672 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1673 INTEROPERATE with older versions automatically. Please read the
1674 user documentation before upgrading any part of a live system.
1675 * WARNING: Lustre networking configuration changes are required with
1676 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1679 * Support for kernels:
1680 2.6.9-22.0.2.EL (RHEL 4)
1681 2.6.5-7.244 (SLES 9)
1682 2.6.12.6 vanilla (kernel.org)
1685 Severity : enhancement
1686 Bugzilla : 7981/8208
1687 Description: Introduced Lustre Networking (LNET)
1688 Details : LNET is new networking infrastructure for Lustre, it includes
1689 a reorganized network configuration mode (see the user
1690 documentation for full details) as well as support for routing
1691 between different network fabrics. Lustre Networking Devices
1692 (LNDs) for the supported network fabrics have also been
1693 created for this new infrastructure.
1695 Severity : enhancement
1696 Description: Introduced Access control lists
1697 Details : clients can set ACLs on files and directories in order to have
1698 more fine-grained permissions than the standard Unix UGO+RWX.
1699 The MDS must be started with the "-o acl" mount option.
1701 Severity : enhancement
1702 Description: Introduced filesystem quotas
1703 Details : Administrators may now establish per-user quotas on the
1706 Severity : enhancement
1708 Description: Configuration change for the XT3
1709 The PTLLND is now used to run Lustre over Portals on the XT3
1710 The configure option(s) --with-cray-portals are no longer used.
1711 Rather --with-portals=<path-to-portals-includes> is used to
1712 enable building on the XT3. In addition to enable XT3 specific
1713 features the option --enable-cray-xt3 must be used.
1718 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1719 Details : A race between allocating a new llcd and re-getting the llcd_lock
1720 allowed another thread to grab newly-allocated llcd.
1722 Severity : enhancement
1724 Description: 2.6 OST async journal commit and locking fix to improve performance
1725 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1726 2.6 kernels are now async as they already were in 2.4 kernels so
1727 that they can commit concurrently with the network bulk transfer.
1728 For block-allocated files the filter allocation semaphore is held
1729 to avoid filesystem fragmentation during allocation. BKL lock
1730 removed for 2.6 xattr operations where it is no longer needed.
1735 Description: lconf incorrectly determined whether two IP networks could talk
1736 Details : In some more complicated routing and multiple-network
1737 configurations, lconf will avoid trying to make a network
1738 connection to a disjoint part of the IP space. It was doing the
1739 math incorrectly for one set of cases.
1744 Description: Fix for potential infinite loop processing records in an llog.
1745 Details : If an llog record is corrupted/zeroed, it is possible to loop
1746 forever in llog_process(). Validate the llog record length
1747 and skip the remainder of the block on error.
1750 Frequency : occasional (liblustre only)
1752 Description: liblustre could not open files whose last component is a symlink
1753 Details : sysio_path_walk() would incorrectly pass the open intent to
1754 intermediate path components.
1757 Frequency : rare (liblustre only with non-standard tuning)
1758 Bugzilla : 7201 (7350)
1759 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1760 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1761 after it has been released, only do this for referenced locks
1763 Severity : enhancement
1765 Description: specifying an (invalid) directory default stripe_size of -1
1766 would reset the directory default striping
1767 Details : stripe_size -1 was used internally to signal directory stripe
1768 removal, now use "all default" to signal dir stripe removal
1769 as a directory striping of "all default" is not useful
1772 Frequency : common for large clusters running liblustre clients
1774 Description: doing an ls when liblustre clients are running is slow
1775 Details : sending a glimpse AST to a liblustre client waits for every AST
1776 to time out, as liblustre clients will not respond. Since they
1777 cannot cache data we refresh the OST lock LVB from disk instead.
1779 Severity : enhancement
1781 Description: doing an ls at the same time as file IO can be slow
1782 Details : enqueue and other "small" requests can be blocked behind many
1783 large IO requests. Create a new OST IO portal for non-IO
1784 requests so they can be processed faster.
1787 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1789 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1790 created file with mode 000
1791 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1792 allow an ftruncate() as a non-root user to a file with mode 000.
1793 Lustre now respects this flag to disable mode checks when
1794 truncating a file owned by the user
1797 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1800 Description: Revoking locks from clients that went dead or catatonic might take
1802 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1803 cancellation of such locks instant on servers without waiting for
1804 any reply from clients. Clients drops these locks when cancel
1805 notification from server is received without replying.
1808 Frequency : liblustre-only, when liblustre client dies or becomes busy
1810 Description: Doing ls on Linux clients can take a long time with active
1812 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1813 granting such locks to it in the first place if possible. Locks
1814 are taken by proxy on the OST during the read or write and
1815 dropped immediately afterward. Add connect flags handling, do
1816 not grant locks to liblustre clients for glimpse ASTs.
1818 Severity : enhancement
1820 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1821 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1822 sequential reads and behaves badly with applications doing random
1823 reads. Improve it by reading ahead at least read region, and
1824 avoiding excessive large RPC for small reads.
1826 Severity : enhancement
1828 Description: Creating more than 1000 files for a single job may cause a load
1829 imbalance on the OSTs if there are also a large number of OSTs.
1830 Details : qos_prep_create() uses an OST index reseed value that is an
1831 even multiple of the number of available OSTs so that if the
1832 reseed happens in the middle of the object allocation it will
1833 still utilize the OSTs as uniformly as possible.
1838 Description: OST or MDS may oops in ping_evictor_main()
1839 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1840 but doesn't restart at beginning of obd_exports_timed list
1843 Severity : enhancement
1845 Description: improve by-nid export eviction on the MDS and OST
1846 Details : allow multiple exports with the same NID to be evicted at one
1847 time without re-searching the exports list.
1850 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1852 Description: MDS may oops in groups_free()
1853 Details : in rare race conditions a newly allocated group_info struct is
1854 freed again, and this can be NULL. The 2.4 compatibility code
1855 for groups_free() checked for a NULL pointer, but 2.6 did not.
1858 Frequency : common for liblustre clients doing little filesystem IO
1859 Bugzilla : 9352, 7313
1860 Description: server may evict liblustre clients accessing contended locks
1861 Details : if a client is granted a lock or receives a completion AST
1862 with a blocking AST already set it would not reply to the AST
1863 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1866 Frequency : lfs setstripe, only systems with more than 160 OSTs
1868 Description: unable to set striping with a starting offset beyond OST 160
1869 Details : llapi_create_file() incorrectly limited the starting stripe
1870 index to the maximum single-file stripe count.
1873 Frequency : LDAP users only
1875 Description: lconf did not handle in-kernel recovery with LDAP properly
1876 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1878 Severity : enhancement
1880 Description: bind OST threads to NUMA nodes to improve performance
1881 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1882 node and do their allocations there to localize memory access
1884 Severity : enhancement
1886 Description: llmount can determine client NID directly from Myrinet (GM)
1887 Details : the client NID code from gmnalnid was moved directly into
1888 llmount, removing the need to use this or specifying the
1889 client NID explicitly when mounting GM clients with zeroconf
1892 Frequency : if client is started with down MDS
1894 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1895 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1896 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1897 ptlrpc_import_delayed_req() because we didn't check if the
1898 request was interrupted, and we also didn't break out of the
1899 event loop if there was no timeout
1904 Description: data loss during non-page-aligned writes to a single file from
1905 both multiple nodes and multiple threads on one node at same time
1906 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1907 inconsistency led to false short-reads, that were cached and later
1908 used by ->prepare_write() to fill in partially written page,
1909 leading to data loss.
1912 Frequency : always, if lconf --abort_recovery used
1914 Description: lconf --abort_recovery fails with 'Operation not supported'
1915 Details : lconf was attempting to abort recovery on the MDT device and not
1918 Severity : enhancement
1920 Description: remove cleanup logs
1921 Details : replace lconf-generated cleanup logs with lustre internal
1922 cleanup routines. Eliminates the need for client-cleanup and
1925 Severity : enhancement
1927 Description: add support for EAs (user and system) on lustre filesystems
1928 Details : it is now possible to store extended attributes in the Lustre
1929 client filesystem, and with the user_xattr mount option it
1930 is possible to allow users to store EAs on their files also
1932 Severity : enhancement
1934 Description: Add possibility (config option) to show minimal available OST free
1936 Details : When compiled with --enable-mindf configure option, statfs(2)
1937 (and so, df) will return least minimal free space available from
1938 all OSTs as amount of free space on FS, instead of summary of
1939 free spaces of all OSTs.
1941 Severity : enhancement
1943 Description: do not expand extent locks acquired on OST-side
1944 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1945 by server: they are not cached anyway.
1948 Frequency : when mmap is used/binaries executed from Lustre
1950 Description: Unmmap pages before throwing them away from read cache.
1951 Details : llap_shrink cache now attempts to unmap pages before discarding
1952 them (if unmapping failed - do not discard). SLES9 kernel has
1953 extra checks that trigger if this unmapping is not done first.
1958 Description: lconf didn't resolve symlinks before checking to see whether a
1959 given mountpoint was already in use
1962 Frequency : when migrating failover services
1963 Bugzilla : 6395, 9514
1964 Description: When migrating a subset of services from a node (e.g. failback
1965 from a failover service node) the remaining services would
1966 time out and evict clients.
1967 Details : lconf --force (implied by --failover) sets the global obd_timeout
1968 to 5 seconds in order to quickly disconnect, but this caused
1969 other RPCs to time out too quickly. Do not change the global
1970 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1972 Severity : enhancement
1973 Frequency : if MDS is started with down OST
1974 Bugzilla : 9439,5706
1975 Description: Allow startup/shutdown of an MDS without depending on the
1976 availability of the OSTs.
1977 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1978 Add appropriate locking and lov-osc refcounts for safe
1979 cleaning. Add osc abort_inflight calls in case the
1980 synchronize never started.
1983 Frequency : occasional (Cray XT3 only)
1985 Description: root not authorized to access files in CRAY_PORTALS environment
1986 Details : The client process capabilities were not honoured on the MDS in
1987 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1988 been accessed by an authorized user then root was able to access
1989 the file on the local client also. The root user capabilities
1990 are now allowed on the MDS, as this environment has secure UID.
1993 Frequency : occasional
1995 Description: ldiskfs "too long searching" message happens too often
1996 Details : A debugging message (otherwise harmless) prints too often on
1997 the OST console. This has been reduced to only happen when
1998 there are fragmentation problems on the filesystem.
2003 Description: Division by zero in statfs when all OSCs are inactive
2004 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2005 lov_statfs divides by value returned by lov_get_stripecnt().
2009 Bugzilla : 9489, 3273
2010 Description: First write from each client to each OST was only 4kB in size,
2011 to initialize client writeback cache, which caused sub-optimal
2012 RPCs and poor layout on disk for the first writen file.
2013 Details : Clients now request an initial cache grant at (re)connect time
2014 and so that they can start streaming writes to the cache right
2015 away and always do full-sized RPCs if there is enough data.
2016 If the OST is rebooted the client also re-establishes its grant
2017 so that client cached writes will be honoured under the grant.
2022 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2023 Details : Now I/O RPCs go to different portal number and (presumably) fast
2024 lock requests (and glimses) and other RPCs get their own service
2025 threads pool that should be able to service those RPCs
2028 Severity : enhancement
2030 Description: Ability to exchange lustre version between client and servers and
2031 issue warnings at client side if client is too old. Also for
2032 liblustre clients there is ability to refuse connection of too old
2034 Details : New 'version' field is added to connect data structure that is
2035 filled with version info. That info is later checked by server and
2039 Frequency : rare, liblustre only.
2040 Bugzilla : 9296, 9581
2041 Description: Two simultaneous writes from liblustre at offset within same page
2042 might proceed at the same time overwriting eachother with stale
2044 Details : I/O lock withing llu_file_prwv was released too early, before data
2045 actually was hitting the wire. Extended lock-holding time until
2046 server acknowledges receiving data.
2049 Frequency : extremely rare. Never observed in practice.
2051 Description: avoid generating lustre_handle cookie of 0.
2052 Details : class_handle_hash() generates handle cookies by incrementing
2053 global counter, and can hit 0 occasionaly (this is unlikely, but
2054 not impossible, because initial value of cookie counter is
2055 selected randonly). Value of 0 is used as a sentinel meaning
2056 "unassigned handle" --- avoid it. Also coalesce two critical
2057 sections in this function into one.
2059 Severity : enhancement
2061 Description: allow liblustre clients to delegate truncate locking to OST
2062 Details : To avoid overhead of locking, liblustre client instructs OST to
2063 take extent lock in ost_punch() on client's behalf. New connection
2064 flag is added to handle backward compatibility.
2066 Severity : enhancement
2067 Bugzilla : 4928, 7341, 9758
2068 Description: allow number of OST service threads to be specified
2069 Details : a module parameter allows the number of OST service threads
2070 to be specified via "options ost ost_num_threads={N}" in the
2071 OSS's /etc/modules.conf or /etc/modprobe.conf.
2075 Bugzilla : 6146, 9635, 9895
2076 Description: servers crash with bad pointer in target_handle_connect()
2077 Details : In rare cases when a client is reconnecting it was possible that
2078 the connection request was the last reference for that export.
2079 We would temporarily drop the export reference and get a new
2080 one, but this may have been the last reference and the export
2081 was just destroyed. Get new reference before dropping old one.
2083 Severity : enhancement
2084 Frequency : if client is started with failover MDS
2086 Description: Allow multiple MDS hostnames in the mount command
2087 Details : Try to read the configuration from all specified MDS
2088 hostnames during a client mount in case the "primary"
2091 Severity : enhancement
2093 Description: Stop sending data to evicted clients as soon as possible.
2094 Details : Check if the client we are about to send or are sending data to
2095 was evicted already. (Check is done every second of waiting,
2096 for which l_wait_event interface was extended to allow checking
2097 of exit condition at specified intervals).
2100 Frequency : rare, normally only when NFS exporting is done from client
2102 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2104 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2107 Frequency : always, if srand() is called before liblustre initialization
2109 Description: Liblustre uses system PRNG disturbing its usage by user application
2110 Details : Introduce internal to lustre fast and high-quality PRNG for
2111 lustre usage and make liblustre and some other places in generic
2112 lustre code to use it.
2114 Severity : enhancement
2115 Bugzilla : 9477, 9557, 9870
2116 Description: Verify that the MDS configuration logs are updated when xml is
2117 Details : Check if the .xml configuration logs are newer than the config
2118 logs stored on the MDS and report an error if this is the case.
2119 Request --write-conf, or allow starting with --old_conf.
2121 Severity : enhancement
2123 Description: Handle symlinks in the path when checking if Lustre is mounted.
2124 Details : Resolve intermediate symlinks when checking if a client has
2125 mounted a filesystem to avoid duplicate client mounts.
2130 Description: lconf can hit an error exception but still return success.
2131 Details : The lconf command catches the Command error exception at the top
2132 level script context and will exit with the associated exit
2133 status, but doesn't ensure that this exit status is non-zero.
2138 Description: failure of ptlrpc thread startup can cause oops
2139 Details : Starting a ptlrpc service thread can fail if there are a large
2140 number of threads or the server memory is very fragmented.
2141 Handle this without oopsing.
2144 Frequency : always, only if liblustre and non-default acceptor port was used
2146 Description: liblustre cannot connect to servers with non-default acceptor port
2147 Details : tcpnal_set_default_params() was not called and was therefore
2148 ignoring the environment varaible TCPNAL_PORT, as well as other
2149 TCPNAL_ environment variables
2154 Description: two objects could be created on the same OST for a single file
2155 Details : If an OST is down, in some cases it was possible to create two
2156 objects on a single OST for a single file. No problems other
2157 than potential performance impact and spurious error messages.
2161 Bugzilla : 5681, 9562
2162 Description: Client may oops in ll_unhash_aliases
2163 Details : Client dcache may become inconsistent in race condition.
2164 In some cases "getcwd" can fail if the current directory is
2170 Description: Inode refcounting problems in NFS export code
2171 Details : link_raw functions used to call d_instantiate without obtaining
2172 extra inode reference first.
2176 Bugzilla : 9942, 9903
2177 Description: Referencing freed requests leading to crash, memleaks with NFS.
2178 Details : We used to require that call to ll_revalidate_it was always
2179 followed by ll_lookup_it. Also with revalidate_special() it is
2180 possible to call ll_revalidate_it() twice for the same dentry
2181 even if first occurence returned success. This fix changes semantic
2182 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2183 reference on a request referred from the intent.
2184 ll_intent_release() then releases such a request.
2187 Frequency : rare, normally benchmark loads only
2189 Description: unlinked inodes were kept in memory on the client
2190 Details : If a client is repeatedly creating and unlinking files it
2191 can accumulate a lot of stale inodes in the inode slab cache.
2192 If there is no other client load running this can cause the
2193 client node to run out of memory. Instead flush old inodes
2194 from client cache that have the same inode number as a new inode.
2197 Frequency : SLES9 2.6.5 kernel and long filenames only
2198 Bugzilla : 9969, 10379
2199 Description: utime reports stale NFS file handle
2200 Details : SLES9 uses out-of-dentry names in some cases, which confused
2201 the lustre dentry revalidation. Change it to always use the
2205 Frequency : rare, unless heavy write-truncate concurrency is continuous
2206 Bugzilla : 4180, 6984, 7171, 9963, 9331
2207 Description: OST becomes very slow and/or deadlocked during object unlink
2208 Details : filter_destroy() was holding onto the parent directory lock
2209 while truncating+unlinking objects. For very large objects this
2210 may block other threads for a long time and slow overall OST
2211 responsiveness. It may also be possible to get a lock ordering
2212 deadlock in this case, or run out of journal credits because of
2213 the combined truncate+unlink. Solution is to do object truncate
2214 first in one transaction without parent lock, and then do the
2215 final unlink in a new transaction with the parent lock. This
2216 reduces the lock hold time dramatically.
2219 Frequency : rare, 2.4 kernels only
2221 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2222 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2223 before unmounting their filesystem in order to avoid deadlock.
2224 The kernel_locked() function in 2.4 kernels only checks whether
2225 the kernel lock is held, not whether it is this process that is
2226 holding it as 2.6 kernels do.
2231 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2232 Details : The client ptlrpc code may be trying to reconnect to a down
2233 server before a previous connection attempt has timed out.
2234 Increase the reconnect interval to be longer than the connection
2235 timeout interval to avoid sending duplicate connections to
2239 Frequency : echo_client brw_test command
2241 Description: fix echo_client to work with OST preallocated code
2242 Details : OST preallocation code (5137) didn't take echo_client IO path
2243 into account: echo_client calls filter methods outside of any
2244 OST thread and, hence, there is no per-thread preallocated
2245 pages and buffers to use. Solution: hijack pga pages for IO. As
2246 a byproduct, this avoids unnecessary data copying.
2250 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2251 Description: Client can oops in mdc_commit_close() after open replay
2252 Details : It was possible for the MDS to return an open request with no
2253 transaction number in mds_finish_transno() if the client was
2254 evicted, but without actually returning an error. Clients
2255 would later try to replay that open and may trip an assertion
2256 Simplify the client close codepath, and always return an error
2257 from the MDS in case the open is not successful.
2260 Frequency : rare, 2.6 OSTs only
2262 Description: OST may deadlock under high load on fragmented files
2263 Details : If there was a heavy load and highly-fragmented OST filesystems
2264 it was possible to have all the OST threads deadlock waiting on
2265 allocation of biovecs, because the biovecs were not released
2266 until the entire RPC IO was completed. Instead, release biovecs
2267 as soon as they are complete to ensure forward IO progress.
2269 Severity : enhancement
2271 Description: Support for specifying external journal device at mount
2272 Details : If an OST or MDS device is formatted with an external journal
2273 device, this device major/minor is stored in the ext3 superblock
2274 and may not be valid for failover. Allow detecting and
2275 specifying the external journal at mount time.
2280 Description: Mounting an MDS with pending unlinked files may cause oops
2281 Details : target_finish_recovery() calls mds_postrecov() which returned
2282 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2283 considers this an error and immediately begins cleaning up the
2284 lov, just after starting the mds_lov process
2286 Severity : enhancement
2288 Description: Implement 'lfs df' to report actual free space on per-OST basis
2289 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2290 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2291 usage of objects; '-h' to report in human readable format.
2293 ------------------------------------------------------------------------------
2295 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2302 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2303 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2304 at startup during orphan unlinking. Instead, skip buddy update
2305 before setup, as it will be regenerated after recovery is complete.
2310 Description: If an OST is inactive, its locks might reference stale inodes.
2311 Details : lov_change_cbdata() must iterate over all namespaces, even if
2312 they are inactive to clear inode references from the lock.
2314 Severity : enhancement
2315 Frequency : occasional, if non-standard max_dirty_mb used
2317 Description: Client will block write RPCs if not enough grant
2318 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2319 then the client will block writes while waiting for another RPC
2320 to complete instead of consuming its dirty limit. With change
2321 we get improved performance when max_dirty_mb is small.
2323 Severity : enhancement
2324 Bugzilla : 3389, 6253
2325 Description: Add support for supplementary groups on the MDS.
2326 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2327 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2328 lookups for user supplementary groups into a cache.
2332 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2333 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2334 for liblustre O_EXCL is enforced by MDS.
2338 Description: readdir never returns NULL in liblustre.
2339 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2340 dentry in struct dirent.
2344 Frequency : liblustre only, depends on application IO pattern
2345 Description: liblustre clients evicted if not contacting servers
2346 Details : Don't put liblustre clients into the ping_evictor list, so
2347 they will not be evicted by the pinger ever.
2349 Severity : enhancement
2351 Description: Add ability to evict clients by NID from MDS.
2352 Details : By echoing "nid:$NID" string into
2353 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2354 $NID would be instantly evicted from this MDS and from all active
2355 OSTs connected to it.
2359 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2360 Details : lookup_it_finish() used to query file size from OSTs that was not
2365 Description: service threads change working directory to that of init
2366 Details : Starting lustre service threads may pin the working directory
2367 of the parent thread, making that filesystem busy. Threads
2368 now change to the working directory of init to avoid this.
2372 Frequency : during shutdown only
2373 Description: shutdown with a failed MDS or OST can cause unmount to hang
2374 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2379 Frequency : 2.6 only
2380 Description: chmod/chown may include an extra supplementary group
2381 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2382 supplementary group and if none is specified this is used.
2385 Bugzilla : 5479 (6816)
2387 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2388 Details : If lookup is in progress on a file that is unlinked we might try
2389 to revalidate the inode and fail in revalidate after lookup is
2390 complete and ll_file_open() enqueues the open again but
2391 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2394 Frequency : always, if lconf --abort_recovery used
2396 Description: lconf --abort_recovery fails with 'Operation not supported'
2397 Details : lconf was attempting to abort recovery on the MDT device and not
2400 ------------------------------------------------------------------------------
2402 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2407 Frequency : rare (only unsupported configurations with a node running as an
2409 Bugzilla : 6514, 5137
2410 Description: Mounting a Lustre file system on a node running as an OST could
2412 Details : OSTs now preallocates memory needed to write out data at
2413 startup, instead of when needed, to avoid having to
2414 allocate memory in possibly low memory situations.
2415 Specifically, if the file system is mounted on on OST,
2416 memory pressure could force it to try to write out data,
2417 which it needed to allocate memory to do. Due to the low
2418 memory, it would be unable to do so and the node would
2419 become unresponsive.
2421 Severity : enhancement
2423 Description: Addition of lconf --service command line option
2424 Details : lconf now accepts a '--service <arg>' option, which is
2425 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2427 Severity : enhancement
2429 Description: Failover mode is now the default for OSTs.
2430 Details : By default, OSTs will now run in failover mode. To return to
2431 the old behaviour, add '--failout' to the lmc line for OSTs.
2433 Severity : enhancement
2435 Description: Health checks are now provided for MDS and OSTs
2436 Details : Additional detailed health check information on MSD and OSTs
2437 is now provided through the procfs health_check value.
2440 Frequency : occasional, depends on IO load
2442 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2443 after numerous create/delete cycles
2444 Details : The ext3 inode allocation policy would not allocate new inodes
2445 very well on the OSTs because there are no new directories
2446 being created. Instead we look for groups with free space if
2447 the parent directories are nearly full.
2452 Description: Network or server problems during mount may cause partially
2453 mounted clients instead of returning an error.
2454 Details : The config llog parsing code may overwrite the error return
2455 code during mount error handling, returning success instead
2461 Description: MDS can fail to allocate large reply buffers
2462 Details : After long uptimes the MDS can fail to allocate large reply
2463 buffers (e.g. zconf client mount config records) due to memory
2464 fragmentation or consumption by the buffer cache. Preallocate
2465 some large reply buffers so that these replies can be sent even
2466 under memory pressure.
2470 Frequency : rare (liblustre)
2471 Description: fsx running with liblustre complained that using truncate() to
2472 extend the file doesn't work. This patch corrects that issue.
2473 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2474 ATTR_SIZE and lsm use in llu_setattr_raw.
2478 Frequency : rare, only 2.6 kernels
2479 Description: Unusual file access patterns on the MDS may result in inode
2480 data being lost in very rare circumstances.
2481 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2482 code in the 2.6 kernel caused the inode and/or EA data not to
2483 be read from disk, causing single-file corruption.
2487 Frequency : rare, only 2.6 filesystems using extents
2488 Description: Heavy concurrent write and delete load may cause data corruption.
2489 Details : It was possible under high-load situations to have an extent
2490 metadata block in the block device cache from a just-unlinked
2491 file overwrite a newly-allocated data block. We now unmap any
2492 metadata buffers that alias just-allocated data blocks.
2496 Frequency : filesystems with default stripe_count larger than 77
2497 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2498 Details : lconf specifies an inode size of 4096 bytes when the default
2499 stripe_count is larger than 77. This conflicts with the default
2500 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2501 this case to avoid pinning too much memory for large EAs.
2503 ------------------------------------------------------------------------------
2505 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2510 Frequency : rare (extremely heavy IO load with hundreds of clients)
2512 Description: Client is evicted, gets IO error writing to file
2513 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2514 caused clients to be evicted for AST timeouts. The fixes in
2515 bug 5192 mean we no longer need to have such short AST timeouts
2516 so ldlm_timeout has been increased.
2519 Frequency : occasional during --force or --failover shutdown under load
2520 Bugzilla : 5949, 4834
2521 Description: Server oops/LBUG if stopped with --force or --failover under load
2522 Details : a collection of import/export refcount and cleanup ordering
2523 issues fixed for safer force cleanup
2526 Frequency : only filesystems larger than 120 OSTs
2527 Bugzilla : 5990, 6223
2528 Description: lfs getstripe would oops on a very large filesystem
2529 Details : lov_getconfig used kfree on vmalloc'd memory
2532 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2533 Bugzilla : 6242, 6243
2534 Description: reading from files that had been truncated to a non-zero size
2535 but never opened returned no data
2536 Details : ll_file_read() reads zeros from no-object files to EOF
2541 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2542 Details : There's little harm in aborting MDS/OSS recovery and letting it
2543 try again, so I removed the LASSERT and return an error instead.
2545 Severity : enhancement
2547 Description: New debugging infrastructure for tracking down data corruption
2548 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2549 (b) cover more of the client-side code path, and (c) try to narrow
2550 down where problems occurred
2554 Bugzilla : 3819, 4364, 4397, 6313
2555 Description: Racing close and eviction MDS could cause assertion in mds_close
2556 Details : It was possible to get multiple mfd references during close and
2557 client eviction, leading to one thread referencing a freed mfd.
2559 Severity: : enhancement
2560 Bugzilla : 3262, 6359
2561 Description: Attempts to reconnect to servers are now more aggressive.
2562 Details : This builds on the enhanced upcall-less recovery that was added
2563 in 1.4.2. When trying to reconnect to servers, clients will
2564 now try each server in the failover group every 10 seconds. By
2565 default, clients would previously try one server every 25 seconds.
2570 Description: After recovery, certain operations trigger a failed
2571 assertion on a client.
2572 Details : Failing over an mds, using lconf -d --failover, while a
2573 client was doing a readdir() call would cause the client to
2574 LBUG after recovery completed and the readdir() was resent.
2576 Severity : enhancement
2578 Description: Default groups are now added by lconf
2579 Details : You can now run lconf --group <servicename> without having to
2580 manually add groups with lmc.
2583 Frequency : occasional
2585 Description: Nodes with an elan id of 0 trigger a failed assertion
2588 Frequency : always when accessing e.g. tty/console device nodes
2590 Description: tty and some other devices nodes cannot be used on lustre
2591 Details : file's private_data field is used by device data and lustre
2592 values in there got lost. New field was added to struct file to
2593 store fs-specific private data.
2596 Frequency : when exporting Lustre via NFS
2598 Description: NFSD failed occasionally when looking up a path component
2599 Details : NFSD is looking up ".." which was broken in ext3 directories
2600 that had grown large enough to become hashed.
2603 Frequency : Clusters with multiple interfaces not on the same subnet
2605 Description: Nodes will repeatedly try to reconnect to an interface which it
2606 cannot reach and report an error to the log.
2607 Details : Extra peer list entries will be created by lconf with some peers
2608 unreachable. lconf now validates the peer before adding it.
2611 Frequency : Only if a default stripe is set on the filesystem root.
2613 Description: Setting a default stripe on the filesystem root prevented the
2614 filesystem from being remounted.
2615 Details : The client was sending extra request flags in the root getattr
2616 request and did not allocate a reply buffer for the dir EA.
2619 Frequency : occasional, higher if lots of files are accessed by one client
2620 Bugzilla : 6159, 6097
2621 Description: Client trips assertion regarding lsm mismatch/magic
2622 Details : While revalidating inodes the VFS looks up inodes with ifind()
2623 and in rare cases can find an inode that is being freed.
2624 The ll_test_inode() code will free the lsm during ifind()
2625 when it finds an existing inode and then the VFS later attaches
2626 this free lsm to a new inode.
2630 Bugzilla : 6422, 7030
2631 Description: MDS deadlock between mkdir and client eviction
2632 Details : Creating a new file via mkdir or mknod (starting a transaction
2633 and getting the ns lock) can deadlock with client eviction
2634 (gets ns lock and trying to finish a synchronous transaction).
2637 Frequency : occasional
2638 Description: While starting a server, the fsfilt_ext3 module could not be
2640 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2641 kernels. Previously, lconf would still use the ext3 name
2642 when trying to load modules. Now, it will correctly use
2643 ext3 on 2.4 and ldiskfs on 2.6.
2645 Severity : enhancement
2646 Description: The default stripe count has been changed to 1
2647 Details : The interpretation of the default stripe count (0, to lfs
2648 or lmc) has been changed to mean striping across a single
2649 OST, rather than all available. For general usage we have
2650 found a stripe count of 1 or 2 works best.
2652 Severity : enhancement
2653 Description: Add support for compiling against Cray portals.
2654 Details : Conditional compiling for some areas that are different
2658 Frequency : occasional
2659 Bugzilla : 6409, 6834
2660 Description: Creating files with an explicit stripe count may lead to
2661 a failed assertion on the MDS
2662 Details : If some OSTs are full or unavailable, creating files may
2663 trigger a failed assertion on the MDS. Now, Lustre will
2664 try to use other servers or return an error to the
2668 Frequency : occasional
2670 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2671 caused lock timeout during readahead (since 1.4.2).
2672 Details : Processes doing readahead might match a lock that hasn't been
2673 granted yet if there are overlapping and conflicting lock
2674 requests. The readahead process waits on ungranted lock
2675 (original lock is CBPENDING), while OST waits for that process
2676 to cancel CBPENDING read lock and eventually evicts client.
2678 Severity : enhancement
2680 Description: Initial enabling of flock support for clients
2681 Details : Implements fcntl advisory locking and file status functions.
2682 This feature is provided as an optional mount flag (default
2683 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2684 locking are implemented yet, and those that are are not guaranteed
2685 to be completely correct in production environments.
2686 mount -t lustre -o [flock|noflock] ...
2689 Frequency : occasional
2691 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2692 assertion in the ext3 JBD (journaling) layer.
2693 Details : The b_committed_data struct is protected by the big kernel lock
2694 in 2.4 kernels, serializing journal_commit_transaction() and
2695 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2696 access to this struct. In 2.6 kernels there is finer grained
2697 locking to improve SMP performance of the JBD layer.
2701 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2703 -----------------------------------------------------------------------------
2705 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2707 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2708 versions of Lustre. Please update all servers and clients to
2709 version 1.4.2 or later at the same time. You must also run
2710 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2711 to update the configuration logs.
2713 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2714 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2715 - swab llog records of type '0' so we get proper header size/idx (5861)
2716 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2717 - fix rename of one directory over another leaking an inode (5953)
2718 - avoid SetPageDirty on 2.6 (5981)
2719 - don't re-add just-being-destroyed locks to the waiting list (5653)
2720 - when creating new directories, inherit the parent's custom
2721 striping settings if present parent (3048)
2722 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2723 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2724 - don't hold client locks on temporary worklist from l_lru (5666)
2725 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2726 - automatically evict dead clients (5921)
2727 - Update file size properly in create+truncate+fstat case (6196)
2728 - Do not unhash mountpoint dentries, do not allow removal of
2730 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2731 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2732 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2733 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2734 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2735 - Correctly handle reads of files with no objects (6243)
2736 - lctl recover will also mark a device active if deactivate used (5933)
2738 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2739 - allow --write-conf on an MDS with different nettype than client (5619)
2740 - don't write config llogs to MDS for mounts not from that MDS (5617)
2741 - lconf should create multiple TCP connections from a client (5201)
2742 - init scripts are now turned off by default; run chkconfig --on
2743 lustre and chkconfig --on lustrefs to use them
2744 - upcalls are no longer needed for clients to recover to failover
2746 - add --abort-recovery option to lconf to abort recovery on device
2748 - add support for an arbitrary number of OSTs (3026)
2749 - Quota support protocol changes.
2750 - forward compatibility changes to wire structs (6007)
2751 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2752 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2753 - improved "lustre status" script
2754 - initialize blocksize for non-regular files (6062)
2755 - added --disable-server and --disable-client configure options (5782)
2756 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2757 - Vanilla 2.4.29 support
2758 - increase maximum number of obd devices to 520 (6242)
2759 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2760 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2761 SLES 9 2.6.5 kernels
2762 - limit stripes per file to 160 (the maximum EA size) (6093)
2764 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2767 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2768 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2769 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2770 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2771 - small ext3 extents cleanups and fixes (5733)
2772 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2773 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2774 - avoid needless client->OST connect, fix handle mismatch (5317)
2775 - fix DLM error path that led to out-of-sync client, long delays (5779)
2776 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2777 - fix several locking issues related to i_size (5492,5624,5654,5672)
2778 - don't move pending lock onto export if it is already evicted (5683)
2779 - fix kernel oops when creating .foo in unlinked directory (5548)
2780 - fix deadlock in obdfilter statistics vs. object create (5811)
2781 - use time_{before,after} to avoid timer jiffies wrap (5882)
2782 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2783 - Do not leak request if server was not able to process it (5154)
2784 - If mds_open unable to find parent dir, make that negative lookup(5154)
2785 - don't create new directories with extent-mapping (5909, 5936)
2787 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2788 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2789 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2790 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2791 - support for RHEL4 kernel on i686 (5773)
2792 - provide error messages when incompatible logs are encountered (5898)
2794 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2795 * version 1.4.0.10 (1.4.1 release candidate 1)
2797 - don't keep a lock reference when lock is not granted (4238)
2798 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2799 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2800 - drop import inflight refcount on signal_completed_replay error (5255)
2801 - unlock page after async write error during send (3677)
2802 - handle missing objects in filter_preprw_read properly (5265)
2803 - no transno return for symlink open, don't save no-trasno open (3440)
2804 - don't try to complete elan receive that already failed (4012)
2805 - free RPC server reply state on error (5406)
2806 - clean up thread from ptlrpc_start_thread() on error (5160)
2807 - readahead could read extra page into cache that wasn't ejected (5388)
2808 - prevent races in class_attach/setup/cleanup/detach (5260)
2809 - don't dereference de->d_inode after l_dput of de (5458)
2810 - use "int" for stripe value returned from lock_to_stripe (5544)
2811 - mballoc allocation and error-checking fixes in 2.6 (5504)
2812 - block device patches to fix I/O request sizes in 2.6 (5482)
2813 - look up hostnames for IB nals (5602)
2814 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2815 - don't start multiple acceptors for the same port (5277)
2816 - fix incorrect LASSERT in mds_getattr_name (5635)
2817 - export a proc file for general "ping" checking (5628)
2818 - fix "lfs check" to not block when the MDS is down (5628)
2820 - service request history (4965)
2821 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2822 - create an "evict_client" /proc entry on OSTs, like the MDS has
2823 - fix mount usage message, return errors per mount(8) (5168)
2824 - change grep [] to grep "[]" in tests so they work in more UMLs
2825 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2826 - remove ancient LOV_MAGIC_V0 EA support (5047)
2827 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2828 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2829 - allow readahead allocations to fail when low on memory (5383)
2830 - mmap locking landed again, after considerable improvement (2828)
2831 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2833 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2836 - send OST transaction number in read/write reply to free req (4966)
2837 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2838 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2839 - the watchdog thread now runs as interruptible (5246)
2840 - flock/lockf fixes (but it's still disabled, pending 5135)
2841 - don't use EXT3 constants in llite code (5094)
2842 - memory shortage at startup could cause assertion (5176)
2844 - reorganization of lov code
2845 - single portals codebase
2847 - add extents/mballoc support (5025)
2848 - direct I/O reads in the obdfilter (4048)
2849 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2851 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2854 - send OST transaction number in read/write reply to free req (4966)
2855 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2856 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2857 - the watchdog thread now runs as interruptible (5246)
2858 - handle missing objects in filter_preprw_read properly (5265)
2859 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2860 - drop import inflight refcount on signal_completed_replay error (5255)
2861 - unlock page after async write error during send (3677)
2862 - return original error code on reconstructed replies (3761)
2863 - no transno return for symlink open, don't save no-trasno open (3440)
2865 - add pid to ldlm debugging output (4922)
2866 - bump the watchdog timeouts -- we can't handle 30sec yet
2867 - extra debugging for orphan dentry/inode bug (5259)
2869 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2872 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2873 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2874 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2875 - fix NULL dereference in /proc/sys/portals/routes (4827)
2876 - allow failed mdc_close() operations to be interrupted (4561)
2877 - stop precreate on OST before MDS would time out on it (4778)
2878 - don't send partial-page writes before EOF from client (4410)
2879 - discard client grant for sub-page writes on large-page clients (4520)
2880 - don't free dentries not owned by NFS code, check generation (4806)
2881 - fix lsm leak if mds_create_objects() fails (4801)
2882 - limit debug_daemon file size, always print CERROR messages (4789)
2883 - use transno after validating reply (3892)
2884 - process timed out requests if import state changes (3754)
2885 - update mtime on OST during writes, return in glimpse (4829)
2886 - add mkfsoptions to LDAP (4679)
2887 - use ->max_readahead method instead of zapping global ra (5039)
2888 - don't interrupt __l_wait_event() during strace
2890 - add software watchdogs to catch hung threads quickly (4941)
2891 - make lustrefs init script start after nfs is mounted
2892 - fix CWARN/ERROR duplication (4930)
2893 - return async write errors to application if possible (2248)
2894 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2895 - print NAL number in %x format (4645)
2896 - update barely-supported suse-2.4.21-171 series (4842)
2897 - support for sles 9 %post scripts
2898 - support for building 2.6 kernel-source packages
2899 - support for sles km_* packages
2901 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2904 - ignore -ENOENT errors in osc_destroy (3639)
2905 - notify osc create thread that OSC is being cleaned up (4600)
2906 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2907 - reconstruct ost_handle() like mds_handle() (4657)
2908 - create a new thread to do import eviction to avoid deadlock (3969)
2909 - let lconf resolve symlinked-to devices (4629)
2910 - don't unlink "objects" from directory with default EA (4554)
2911 - hold socknal file ref over connect in case target is down (4394)
2912 - allow more than 32000 subdirectories in a single directory (3244)
2913 - fix blocks count for O_DIRECT writes (3751)
2914 - OST returns ENOSPC from object create when no space left (4539)
2915 - don't send truncate RPC if file size isn't changing (4410)
2916 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2917 - bind to privileged port in socknal and tcpnal (3689)
2919 - rate limit CERROR/CWARN console message to avoid overload (4519)
2920 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2921 - basic mmap support (3918)
2922 - kernel patch series update from b1_4 (4711)
2924 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2927 - avoid crash during MDS cleanup with OST shut down (2775)
2928 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2929 - don't use bad inodes on the MDS (3744)
2930 - dynamic object preallocation to improve recovery speed (4236)
2931 - don't hold spinlock over lock dumping or change debug flags (4401)
2932 - don't zero obd_dev when it is force cleaned (3651)
2933 - print grants to console if they go negative (4431)
2934 - "lctl deactivate" will stop automatic recovery attempts (3406)
2935 - look for existing locks in ldlm_handle_enqueue() (3764)
2936 - don't resolve lock handle twice in recovery avoiding race (4401)
2937 - revalidate should check working dir is a directory (4134)
2939 - don't always mark "slow" obdfilter messages as errors (4418)
2941 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2944 - don't close LustreDB during write_conf until it is done (3860)
2945 - fix typo in lconf for_each_profile (3821)
2946 - allow dumping logs from multiple threads at one time (3820)
2947 - don't allow multiple threads in OSC recovery (3812)
2948 - fix debug_size parameters (3864)
2949 - fix mds_postrecov to initialize import for llog ctxt (3121)
2950 - replace config semaphore with spinlock (3306)
2951 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2952 - don't allow enqueue to complete on a destroyed export (3822)
2953 - down write_lock before checking llog header bitmap (3825)
2954 - recover from lock replay timeout (3764)
2955 - up llog sem before sending rpc (3652)
2956 - reduce ns lock hold times when setting kms (3267)
2957 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2958 - fix NULL deref and obd_dev leak on setup error (3312)
2959 - replace some LBUG about llog ops with error handling (3841)
2960 - don't match INVALID dentries from d_lookup and spin (3784)
2961 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2962 - fix invalid assertion in ptlrpc_set_wait (3880)
2964 - add libwrap support for the TCP acceptor (3996)
2965 - add /proc/sys/portals/routes for non-root route listing (3994)
2966 - allow setting MDS UUID in .xml (2580)
2967 - print the stack of a process that LBUGs (4228)
2969 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2972 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2973 - make sure to unset replay flag from failed open requests (3440)
2974 - if default stripe count is 0, use OST count for inode size (3636)
2975 - update parent mtime/ctime on client for create/unlink (2611)
2976 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2977 - free recovery state on server during a forced cleanup (3571)
2978 - unregister_reply for resent reqs (3063)
2979 - loop back devices mounting and status check on 2.6 (3563)
2980 - fix resource-creation race that can provoke i_size == 0 (3513)
2981 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2982 - more debugging for page-accounting assertion (3746)
2983 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2984 - avoid deadlock after precreation failure (3758)
2985 - fix race and lock order deadlock in orphan handling (3450, 3750)
2986 - add validity checks when grabbing inodes from l_ast_data (3599)
2988 - add /proc/.../recovery_status to obdfilter (3428)
2989 - lightweight CDEBUG infrastructure, debug daemon (3668)
2990 - change default OSC RPC parameters to be better on small clusters
2991 - turn off OST read cache for files smaller than 32MB
2992 - install man pages and include them in rpms (3100)
2993 - add new init script for (un)mounting lustre filesystems (2593)
2994 - run chkconfig in %post for init scripts (3701)
2995 - drop scimac NAL (unmaintained)
2997 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3000 - clean kiobufs before and after use (3485)
3001 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3002 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3003 - change init script to fail more clearly if not run as root (1528)
3004 - allow clients to reconnect during replay (1742)
3005 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3006 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3007 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3008 - fix cond_resched (3554)
3009 - only evict unfinished clients after recovery (3515)
3010 - allow bulk resend, prevent data loss (3570)
3011 - dynamic ptlrpc request buffer allocation (2102)
3012 - don't allow unlinking open directory if it isn't empty (2904)
3013 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3014 - remove extraneous obd dereference causing LASSERT failure (3334)
3015 - don't use get_cycles() when creating temp. files on the mds (3156)
3016 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3017 - handle EEXIST for set-stripe, set proper directory name (3336)
3019 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3020 - fix ksocknal_fmb_callback() error messages (2918)
3022 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3025 - don't copy lvb into (possibly NULL) reply on error (2983)
3026 - don't deref dentry after dput, don't free lvb on error (2922)
3027 - use the kms to determine writeback rpc length (2947)
3028 - increment oti_logcookies when osc is inactive (2948)
3029 - update client's i_blocks count via lvb messages (2543)
3030 - handle intent open/close of special files properly (1557)
3031 - mount MDS with errors=remount-ro, like obdfilter (2009)
3032 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3033 - don't use cancelling-locks' kms values (2947)
3034 - use highest lock extent for kms, not last one (2925)
3035 - don't dereference ERR_PTR() dentry in error handling path (3107)
3036 - fix thread race in portals_debug_dumplog() (3122)
3037 - create lprocfs device entries at setup instead of at attach (1519)
3038 - common AST error handler, don't evict client on completion race (3145)
3039 - zero nameidata in detach_mnt in 2.6 (3118)
3040 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3041 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3042 - initialize RPC timeout timer earlier for 2.6 (3219)
3043 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3044 - print nal/nid for unknown nid (3258)
3045 - additional checks for oscc recovery before doing precreate (3284)
3046 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3047 - don't crash in mdc_close for bad permissions on open (3285)
3048 - zero i_rdev for non-device files (3147)
3049 - clear page->private before handing to FS, better assertion (3119)
3050 - tune the read pipeline (3236)
3051 - fix incorrect decref of invalidated dentry (2350)
3052 - provide read-ahead stats and refine rpc in flight stats (3328)
3053 - don't hold journal transaction open across create RPC (3313)
3054 - update atime on MDS at close time (3265)
3055 - close LDAP connection when recovering to avoid server load (3315)
3056 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3057 - don't leak open file on MDS after open resend (3325)
3058 - serialize filter_precreate and filter_destroy_precreated (3329)
3059 - loop device shouldn't call sync_dev() for nul device (3092)
3060 - clear page cache after eviction (2766)
3061 - resynchronize MDS->OST in background (2824)
3062 - refuse to mount the same filesystem twice on same mountpoint (3394)
3063 - allow llmount to create routes for mounting behind routers (3320)
3064 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3065 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3066 - fix rare problem with rename on htree directories (3417)
3068 - allow default OST striping configuration per directory (1414)
3069 - fix compilation for qswnal for 2.6 kernels (3125)
3070 - increase maximum number of MDS request buffers for large systems
3071 - change liblustreapi to be useful for external progs like lfsck (3098)
3072 - increase local configuration timeout for slow disks (3353)
3073 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3075 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3078 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3079 - don't overwrite extent policy data in reply if lock was blocked (2901)
3080 - drop filter export grants atomically with removal from device (2663)
3081 - del obd_self_export from work_list in class_disconnect_exports (2908)
3082 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3083 - swab reply message in mdc_close, other PPC fixes (2464)
3084 - fix destroying of named logs (2325)
3085 - overwrite old logs when running lconf --write_conf (2264)
3086 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3087 - fix race in target_handle_connect (2898)
3088 - mds_reint_create() should take same inode create lock (2926)
3089 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3090 - don't close files for self_export to avoid uninitialized obd (2936)
3091 - allow MDS with the same name as client node (2939)
3092 - hold dentry reference for closed log files for unlink (2325)
3093 - reserve space for all logs during transactions (2059)
3094 - don't evict page beyond end of stripe extent (2925)
3095 - don't oops on a deleted current working directory (2399)
3096 - handle hard links to targets without a parent properly (2517)
3097 - don't dereference NULL lock when racing during eviction (2867)
3098 - don't grow lock extents when lots of conflicting locks (2919)
3100 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3103 - account for cache space usage on clients to avoid data loss (974)
3104 - lfsck support in lustre kernel code (2349)
3105 - reduce journal credits needed for BRW writes (2370)
3106 - orphan handling to avoid losing space on client/server crashes
3107 - ptlrpcd can be blocked, stopping ALL progress (2477)
3108 - use lock value blocks to assist in proper KMS, faster stat (1021)
3109 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3110 - recovery for initial connections (2355)
3111 - fixes for mds_cleanup_orphans (1934)
3112 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3113 - block all file creations until orphan recovery completes (1901)
3114 - client remove rq_connection from request struct (2423)
3115 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3116 - recovery timer race (2670)
3117 - mdc_close recovey bug (2532)
3118 - ptlrpc cleanup bug (2710)
3119 - mds timeout on local locks (2588)
3120 - namespace lock held during RPCs (2431)
3121 - handle interrupted sync write properly (2503)
3122 - don't try to handle a message that hasn't been replied to (2699)
3123 - client assert failure during cleanup after abort recovery (2701)
3124 - leak mdc device after failed mount (2712)
3125 - ptlrpc_check_set allows timedout requests to complete (2714)
3126 - wait for inflight reqs when ptlrpcd finishes (2710)
3127 - make sure unregistered services are removed from the srv_list
3128 - reset bulk XID's when resending them (caught by 1138 test)
3129 - unregister_bulk after timeout
3130 - fix lconf error (2694)
3131 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3132 - readahead locks pages, leaves pending causing memory pressure (2673)
3133 - increase OST request buffers to 4096 on large machines (2729)
3134 - fix up permission of existing directories in simple_mkdir (2661)
3135 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3136 - don't assign transno to errored transactions (2742)
3137 - don't delete objects on OST if given a bogus objid from MDS (2751)
3138 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3139 - if rq_no_resend, then timeout request after recovery (2432)
3140 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3141 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3142 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3143 - MDS recovery shouldn't delete objects that it has given out (2730)
3144 - if enqueue arrives after completion, don't clobber LVB (2819)
3145 - don't unlock pages twice when trigger_group_io returns error (2814)
3146 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3147 - don't write pages to disk if there was an error (1450)
3148 - don't ping imports that have recovery disabled (2676)
3149 - take buffered bytes into account when balancing socknal conn (2817)
3150 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3151 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3152 - remove little-endian swabbing of llog records (1987)
3153 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3154 - retry reposting request buffers if they fail (1191)
3155 - grow extent at grant time to avoid granting a revoked lock (2809)
3156 - lock revoke doesn't evict page if covered by a second lock (2765)
3157 - disable VM readahead to avoid reading outside lock extents (2805)
3159 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3160 - updated kernel patches for hp-2.4.20 kernel (2681)
3162 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3165 - fix truncated write corruption (2366)
3166 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3168 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3169 - ensure len(uuid) < 37 in lmc (1171)
3170 - fix ia64 OOPS in llog_test (2255)
3171 - zero end of page at obdfilter for partial page writes (2648)
3172 - don't leave stale dentries around after renames (bug 2428)
3173 - fix timeouts when evicting a client with a single lock held (2642)
3174 - set deadline for the initial HELLO message to drain (2634)
3175 - print out dotted-quad IP addresses in the socknal (2302)
3177 - additional debugging for MDS client eviction problem (2443)
3178 - fix mkfsoptions support for osts (2603, 2604)
3180 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3183 - add series for the vanilla 2.6.0 kernel
3184 - add series for the vanilla 2.4.24 kernel
3185 - add series for a cray x86/64 UL kernel drop
3186 - fix xattr patches for the vanilla 2.4.19 series
3188 - generate true UUIDs in lmc (1171)
3189 - have portals stack dumping break in UML (2466)
3190 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3191 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3192 - wake callback waiting threads on client eviction (2460)
3193 - Add --ptldebug and --subsystem to lmc (1719)
3194 - update assertion to allow safe interrupt allocation
3195 - set rq_no_resend for cancel requests (2432)
3196 - recalculate ptlrpcd timeout after resend (2494)
3197 - call vfs_rmdir when removing pending directories (2368)
3198 - fix renaming a file to itself (2429)
3199 - lmc creates a default one-stripe lov (2454)
3200 - expand procfs space to handle large clusters (2326)
3201 - increase UML stack to avoid overflow
3202 - update lconf's list of debug and subsystem masks
3203 - fix lfs find --obd (2510)
3204 - /proc tunable for disabling filter read caching (2591)
3205 - stop rpm packages from altering slapd.conf (2301)
3206 - disable nagle in the socknal under 0conf (2578)
3207 - choose mds inode size based on stripe count (2572)
3208 - fix kernel-source rpm problems (2516)
3210 - add --disable-doc to avoid pdf generation (2421)
3211 - update documentation, tests, type-os, comments
3212 - avoid format warnings on ia64
3213 - remove the TOE NAL
3214 - tiny code cleanups by removing unused fields
3216 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3219 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3220 - remove the most blatant lies from BUILDING file (2371)
3221 - change default debug level to reasonable production setting
3222 - reduce client side cache size to reduce cache flush time
3223 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3224 - make TCP zerocopy and pinger support enabled by default (2476)
3225 - sync writes completed after process exits caused crashes (2319)
3226 - maintain correct mount count on the MDS (2356)
3227 - backout 1557, because 2316 wasn't really fixed
3228 - better file I/O statistics gathering in /proc
3229 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3230 - another kernel patch to fix zero-copy TCP function export
3231 - don't take duplicate lock when processing re-sent getattr (2420)
3232 - lctl uses obd_self_export instead of creating new conn (2353)
3233 - MDS/OST recovery case which requires object creation asserted (2425)
3234 - move lfs from /usr/sbin to /usr/bin in packages
3235 - fix race between mds_client_add and mds_client_free (2417)
3236 - use kmalloc instead of slabs in portals (2430)
3237 - don't create duplicate records when a failover MDS is present (2442)
3238 - remove unnecessary mount age check (2332)
3239 - don't remove directory inodes from locks prematurely (2451)
3240 - don't break if MDS service name is the same as hostname (2103)
3241 - fix races in client write RPC generation when cache full (2482)
3243 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3246 - remove now-unused request->rq_obd (278)
3247 - if an allocation fails, print out how much memory we've used (1933)
3248 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3249 - add the "configurable stack size" patch to most series files (1256)
3250 - ability to write large log records, for 100+ OST configs (2306)
3251 - fix NULL deref when filter_prep fails (2314)
3252 - fix operator precedence error in filter_sync
3253 - dynamic allocation of socknal TX descriptors (2315)
3254 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3255 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3256 - ensure that CWARN messages in Portals always get to the syslog
3257 - __init/__exit are not for prototype decls (ldlm_init/exit)
3258 - x86-64 compile warning fixes
3259 - fix gateway LMC keyword conflict (2318)
3260 - fix MDS lock inversions in getattr/reint paths (1844)
3261 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3262 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3263 - DLM race condition prevented some lock evictions (2328)
3264 - ENOMEM detection and retry on socknal sends (2230)
3265 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3266 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3267 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3268 - better error messages when a client is rejected during recovery (1505)
3269 - avoid cancelling locks which were never granted, after failure (2330)
3270 - fix i_sem/journal inversion in mds_client_add (2333)
3271 - fix truncate/getattr lock cycle deadlock (2334)
3272 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3273 - fix two rare exit paths which could leak an l_lock() ref (2321)
3274 - fencepost error in MDS/OST orphan recovery (2226)
3275 - make log record alignment 8 bytes (1988)
3276 - lstripe now fails when requested offset > ost_count (2237)
3277 - ensure that all kernel series have a complete list.h (1607)
3278 - fix crashes in special-file operations (2316)
3279 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3281 - allow configurable automake binary, for testing new versions
3282 - small update to the lfs documentation
3284 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3286 * fix negative export reference count in fsfilt_sync (2312)
3288 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3289 * release candidate 0.9.1
3291 - orphans are moved into the PENDING directory for possible recovery
3292 - replayed opens now open by fid for orphan/rename safety (1042)
3293 - last close of an orphan inode generates a transno (683)
3294 - chdir() and mount() now pin the directory entry (1020)
3295 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3296 - discard very old requests without processing them (1502)
3297 - remove some common, well-understood CERRORs (1505)
3298 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3299 - clear "grant" flags in OST replies until OST grant code lands (1644)
3300 - fix read performance by not clobbering i_blksize on client (1598)
3301 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3302 - make LRU size a /proc tunable, clears locks when reduced (707)
3303 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3304 - server locks take references on exports now (1558)
3305 - build fixes for 2.4.20-rh trees (1663)
3306 - return an error from lov_create if all OSCs are inactive (1751)
3307 - fix import levels when a reconnect happens without a timeout (1597)
3308 - exit early from mds_open if we get a lookup error (1749)
3309 - partial page read at EOF wouldn't wait for disk before sending (1642)
3310 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3311 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3312 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3313 - write/truncate lock inversion (1639)
3314 - Don't auto-load obdclass, portals modules during cleanup (1495)
3315 - fix timestamps from jumping to "now" (1763)
3316 - extra journal assertions (1648)
3317 - add an extra multiunlink test (1771)
3318 - fix read_record/write_record API (1776)
3319 - fix leak of offset_extent, possible incorrect i_size later (1772)
3320 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3321 - Debugging for the kqswnal_get_idle_tx problems (1820)
3322 - Allow recovery to be attempted multiple times (1536)
3323 - Write out MDS last_rcvd file after it is first created (1600)
3324 - Fix tx_descriptor leak in failed transmit situations (1827)
3325 - ext3 journaling fixes for assertion failure after IO error (1871)
3326 - class_export_put() on freed export after completion AST error (1896)
3327 - Fix revalidate looping in VFS (1322)
3328 - Don't access a freed export during MDS_REINT timeout (1521)
3329 - Add open-unlink recovery support on the MDS (1673,1764)
3330 - Return an error if no MDS data was read from last_rcvd (1946)
3331 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3332 - Don't setup a disk device that doesn't match exported UUID (317)
3333 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3334 - avoid committing NULL handle in force close
3335 - local.sh is now a one-stripe LOV configuration
3336 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3337 - fix ext3 htree duplicate directory entry corruption (1516)
3338 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3339 - update to new LOV EA format (2097)
3340 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3342 2003-06-15 Phil Schwan <phil@clusterfs.com>
3345 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3346 - per-import/export recovery handling (958, 931, 959)
3347 - multiple last-rcvd slots, for serving multiple FSes (949)
3348 - connections are again shared between multiple imp/exports (963, 964)
3349 - "umount -f" would hang if any requests needed to be sent (393, 978)
3350 - avoid pinning large req buffer by copying for queued messages (989)
3351 - add "uuid" to "lctl device" command to help upcalls (991)
3352 - "open" RPCs with transnos would confuse recovery counters (1037)
3353 - do proper endian conversion of all wire messages (288, 340, 891)
3354 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3355 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3356 - fix ptlrpc_connection leak in target_handle_connect (1174)
3357 - fix import refcounting bug in OST and MDS cleanup (1134)
3358 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3359 - fix very unlikely obd_types race condition (501)
3360 - remove osc_open hack for echo_client (1187)
3361 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3362 - a failure in read_inode2 leads to deadlock (1139)
3363 - cancel ack-locks as soon as transaction is committed (1072)
3364 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3365 - make sure to commitrw after any preprw to avoid deadlock (1162)
3366 - failing to execute a file in a lustre FS would lock inode (1203)
3367 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3368 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3369 - obd_uuid2tgt would walk off the end of the list (1255)
3370 - on IA64 the osc would give portals incorrect bulk size (1258)
3371 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3372 - fix lock inversion caused by new llite matching code (1282)
3373 - limit the number of dirty pages on a client to 10MB (1286)
3374 - timed out locks were not being corrected cancelled (1289)
3375 - fix O_DIRECT above 4GB on IA-32 (1292)
3376 * major user-visible changes
3377 - fail out/fail over policy now controlled by the upcall (993)
3379 - add OBD_PING to check server availability and failure (954)
3380 - lustre messages are now sent in sending host order (288, 340, 891)
3381 - add eadatalen to MDS getattr reply (340)
3382 - OST read replies may contain second buffer, with per-page status (593)
3384 2003-03-11 Phil Schwan <phil@clusterfs.com>
3387 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3388 - failed open()s could cause deadlock; fixed (867, 869)
3389 - stop cancelling OST locks when files are closed (481)
3390 - overlapping XID spaces caused network corruption (851, 853)
3391 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3392 - setattr_raw added, to do single-RPC, server-side setattrs
3393 - lmc/lconf syntax change for OST UUIDs
3394 - fix crashy race condition between ptlrpc_free_req and osc_close
3395 - don't use request in mdc_enqueue if we hit a timeout (889)
3396 - don't set the inode i_size for regular files from the MDS (896)
3397 - handle out of order completion AST (842)
3398 - don't LBUG if a lock request times out after receiving AST (913)
3399 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3400 - if a bad lock AST arrives, send an error instead of dropping entirely
3401 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3402 - fix leak in bulk IO when only partially completed (899, 900, 926)
3403 - fix O_DIRECT for ia64 (55)
3404 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3405 - C-z after timeout could hang a process forever; fixed (977)
3407 - client-side I/O cache (678, 924, 929, 941, 970)
3409 - READPAGE and SETATTRs which don't take server-side locks get
3412 2003-02-11 Phil Schwan <phil@clusterfs.com>
3415 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3416 - Fix fsfilt_extN_readpage() to read a full page of directory
3417 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3418 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3419 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3420 - Most intent operations were removed, in favour of a new RPC mode
3421 that does a single RPC to the server and bypasses most of the VFS
3422 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3423 - Aggressively cancel local locks on DLM servers
3424 - mds_reint_unlink sends EA to the client if it's the last nlink.
3425 client uses that EA to unlink OST objects.
3426 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3427 - recursive symlinks were fixed (439)
3428 - fixed NULL deref in DEBUG_REQ
3429 - filter_update_lastobjid no longer calls sync, which annoyed extN
3430 - fixed multi-client small-writes to a single file problem (445)
3431 - fixed mtime updates during file writes (607)
3432 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3433 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3434 - fixed Linux OST size reporting problem (444, 656)
3435 - OST now updates object mtime with writes or setattr (607, 619)
3436 - client verifies file size before zeroing page past EOF (445)
3437 - OST now writes last allocated objid to disk with allocation (108)
3438 - LOV on echo now works (409)
3440 - mds_reint_unlink sends a new buffer, with the EA included. this
3441 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3442 set if a regular file was being unlinked, and it was the last link
3443 - use PtlGet from the target for bulk writes (315)
3444 - OST now updates object mtime with writes or setattr (607, 619)
3445 - LDLM now has a grant-time callback to revalidate locked items, if
3447 - Many MDS operations were reorganized to combat race conditions
3449 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3450 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3452 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3455 - Fully reactivate OST imports after reconnection (512, others)
3456 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3457 - More graceful error handling for truncating on dead OST (515)
3458 - Don't error out unless we're actually accessing dead stripes (474)
3459 - Fix garbage sizes when stripes are missing (410)
3460 - LRU counters were broken, causing constant lock purge (433, 432)
3461 - garbage on read from stripes with failed OSTs (441)
3462 - mark OSCs as active before reconnecting during recovery (438)
3463 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3464 - lfind did not preserve OST order in output (443)
3465 - symlinks cause hung clients, incorrect data (439)
3466 - stop dereferencing request after dropping refcount (457)
3467 - don't LASSERT(spin_is_locked) on non-SMP (455)
3468 - fixes for many rename() bugs
3469 - fstat didn't correctly synchronize attributes (399)
3470 - server must handle lock cancellation during blocking AST prep (487)
3471 - bulk descriptors were free()d too soon (511)
3472 - fix paths in lconf, which would load incorrect modules (451, 507)
3473 - fix confusing lconf 'host not found' error message (386)
3474 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3475 - fix race condition in mdc_blocking_ast() for inode access (526)
3476 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3477 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3478 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3479 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3480 - fix lstripe in several ways (kernel side) (527)
3481 - fix request leak in ldlm_cli_enqueue (262)
3482 - incorrect OSC was marked inactive after OST failure
3483 - call mds_fs_cleanup before unmounting filesystem (524)
3484 - fix races between taking ns_lock and ldlm_lock_change_resource
3485 - fix races updating LOV export open file list
3486 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3487 - fix recovery NULL deref in ldlm_cli_cancel_unused
3488 - fix some DLM races by using new hash table for lock handles (419)
3489 - permit the client to specify desired inodes, at replay
3490 - duplicate requests when we queue them for replay reintegration
3491 - fix last_rcvd offset calculation
3492 - sync after each recovered transaction, so we always make progress
3493 - never, not always, ERESTART requests without transnos
3494 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3496 - skip replay if the MDS says that the client is already connected
3497 - don't check for a recovery-enabled export to match lctl's UUID
3498 - don't INC_USE_COUNT for phantom exports
3499 - don't crash when cleaning up phantom exports (567)
3500 - don't double-finish or set replay data for errored mdc_open requests
3501 - abort requests when they time out, so we don't get old replies
3502 - send/receive replies for AST messages again
3503 - if the client says that it doesn't have the lock, cancel it on the
3505 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3506 mark it as destroyed, it will all work out when decref is called
3507 - fix module use counts (22, 581)
3509 - ASTs now expect a reply (server cancels lock on error reply)
3511 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3514 - fix many simultaneous client startup (392)
3515 - fix dentry->d_it clobbering
3516 - credentials weren't being shipped for readdir/getattr operations
3517 - remove invalid assertions triggered during some concurrent MD
3519 - proper Lustre versions added (336, 389)
3520 - fix memory leak for create error case (398)
3521 - fix LOV locking bug that would get cli/srv out of sync
3522 - fix echo client over LOV (409)
3523 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3524 - fix double-O_EXCL intent crash (424)
3525 - avoid sending multiple lock CANCELs (352)
3527 - MDS can do multi-client recovery (modulo bugs in new code)
3529 - many updates, edits, cleanups
3531 2002-11-18 Phil Schwan <phil@clusterfs.com>
3534 - fix null d_it dereference (346)
3535 - fix full OST/dbench hang (333)
3536 - fix permission problem with file removal (286)
3537 - fix removal of OSCs from LOV when they fail
3538 - fix NULL deref during bulk timeout (214)
3539 - fix problems related to multiple filesystems on one MDS (241)
3540 - fixed serious subtle metadata locking bugs
3541 - free locks on clients when inodes are removed due to memory
3543 - fix inode pointer in lock data (285)
3544 - partial support for multiple MDS on a single host (241)
3545 - data locks weren't cancelled at clear_inode time (290, 311)
3546 - intent locks could lead to unbounded lock growth (205)
3547 - added a maximum lock count, an LRU list, and a flusher
3548 - fix multiple rename (365)
3549 - properly abstracted the echo client
3550 - OSC locked 1 byte too many; fixed
3551 - rewrote brw callback code:
3552 - fixed recovery bugs related to LOVs (306)
3553 - fixed too-many-pages-in-one-write crash (191)
3554 - fixed (again) crash in sync_io_timeout (214)
3555 - probably fixed callback-related race (385)
3557 - Add capability to MDS protocol
3558 - LDLM cancellations and callbacks on different portals
3560 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3563 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3564 - defer open object destruction to close time (601981, 138)
3565 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3566 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3567 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3568 - don't call obd_setattr in ll_file_release for destroyed objects
3569 * protocol change to lustre_msg: move |version| and add |flags|
3570 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3571 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3572 * added replay of create, unlink, link and rename operations during
3573 MDS failover; recovery should be much more robust now
3574 * remove failed OSCs from LOVs (only lov_create uses this so far)
3575 * the lustre-HOWTO was brought (more) up to date (582544)
3577 2002-10-23 Phil Schwan <phil@clusterfs.com>
3580 - in-use dentries weren't being reused properly (617851)
3581 - prevent multiple LDLM setup (599178)
3582 - fix LOV size calculations for truncate (617853)
3583 - fix client handling of MDS intent errors (POSIX)
3584 - fix permission bug in lovstripe.c test (624321)
3585 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3586 - truncate past end of file could corrupt data
3587 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3588 - a race in recovery could return ETIMEDOUT to apps (623947)
3589 - building outside the source directory was fixed
3590 * the lustre-HOWTO was brought (more) up to date (582544)
3591 * major progress was made on recovery functionality
3593 2002-10-10 Phil Schwan <phil@clusterfs.com>
3596 - recovery deadlock fix
3597 - rm -rf causes LBUG fix (617817)
3598 - file open by multiple tasks fix (618962)
3599 - directory permissions bugs (602707 and 620007)
3600 - journal_stop fixed with locking (611313)
3601 - O_APPEND failures resolved (618273, perhaps 614459)
3602 - lconf PATH fix (619770)
3603 - IA64 build fix (621450)
3604 - RPC buffer sizes scale with amount of memory
3606 2002-10-01 Phil Schwan <phil@clusterfs.com>
3609 - locks would be cancelled without throwing away data pages,
3610 resulting in inconsistent data (605627)
3611 - inode attributes were not always being refreshed (605627, 612449)
3612 - lconf now continues to cleanup after lctl reports an error
3613 - MDS now enforces user permissions (602707)
3614 - lprocfs cleanup fixed, but not yet enabled (614157)
3615 - fixed infinite server hang, should a client not respond to an AST
3616 - avoid going into recovery if user calls readlink() with a buffer
3617 that's too small (613941)
3618 - AST RPCs no longer require replies (614867) -- this may be changed
3619 - don't crash server if client sends an IOV that's too big (611336)
3620 - fixed lock conversion deadlock (611892)
3621 - fixed the following of symlinks (614622)
3622 * recovery: the server can remove locks from a client that dies, other
3623 clients can make progress
3624 * more extN patch fixes
3625 * compile-time configurable ptlrpc buffer allocations
3627 - collaborative read cache document
3628 - Lustre Lite Performance CDR document-in-progress
3630 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3633 - fix typo in patch-2.4.18
3635 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3638 - clear ptlrpc request each time in handle_incoming_request()
3639 - unlink of files now destroys the object on the OST
3641 2002-09-19 Peter Braam <braam@clusterfs.com>
3643 * add hard link support
3644 * change obdfile creation method
3645 * kernel patch changed
3647 2002-09-19 Peter Braam <braam@clusterfs.com>
3650 - stack overflow bug in extN fixed
3652 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3654 * documentation updates
3655 - add man pages for config tools
3656 - update tests/README to describe testing with new config tools
3657 - finish metadata API descriptions
3658 * bug fixes and cleanups
3659 - statfs workaround for 16TB limit
3660 - LOV stripe allocation improved, can stripe on subset of OSTs
3661 - LOV file size/IO offset was wrong for files > 4GB in size
3662 - object EA data was being dropped, caused files to be unreadable
3663 - memory overflow with non-LOV OST caused memory corruption
3664 - fixed regression tests to work with new config tools, obdfilter
3665 - fixed bug when directory size became larger than 1 block
3666 - fixed bug (for single client case) when PWD was deleted
3667 - invalidate local directory pages when doing intent-based ops
3668 - avoid LDLM oops when lock callback contained bad data
3670 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3672 * documentation updates
3673 * bug fixes and cleanups
3674 - configuration tools
3677 - 64-bit compile warnings
3678 - 64-bit internal statfs data
3680 * test_brw on persistent OST devices
3682 * lprocfs (disabled)
3684 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3686 * documentation updates
3687 * bug fixes and cleanups
3688 * configuration tools
3690 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3693 * many small fixes to 0_5_4
3694 * io/network handling
3695 * thinkos in MDS operations
3697 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3700 * crucial basic fixes to 0.5.3
3701 * IOR, Iozone work over Elan
3704 2002-08-07 Phil Schwan <phil@clusterfs.com>
3705 * version 0_5_3, our first alpha
3706 * we use the new Portals iovs
3707 * documentation updates
3708 * bug fixes and cleanups
3709 * small changes in the DLM wire protocol
3711 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3712 * version 0_5_1 with some initial stability,
3713 * locking on MD and file I/O.
3714 * documentation updates
3715 * several bug fixes since 0.5.0
3716 * small changes in wire protocol
3718 2002-07-18 Phil Schwan <phil@clusterfs.com>
3720 * delivered as Lustre Light Alpha
3721 * fixed a crash after handling invalid MDS requests
3722 * fixed directory pages for architectures with non-4k pages sizes
3724 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3725 * release version v0_4_4
3726 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3728 * Fixes a number of interruption problems with OST operations.
3729 * Update documentation for portals header changes
3730 * Move all wire protocol structs/defines to lustre_idl.h
3731 * Fixes symlink length bug.
3732 * Add tcpdump to repository.
3734 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3735 * release version v0_4_3
3736 * Fixes statfs for inodes on extN.
3737 * Fixes bug in runtests which would delete /etc/hosts.
3738 * Use 64-bit object IDs wherever possible (not into VFS though)
3739 Remove ost_get_info, which is unused by lustre, and out of date.
3741 2002-07-03 Peter Braam <braam@clusterfs.com>
3742 * release version v0_4_2 Fixes a lookup error (type not passed)
3743 * move forward to head of Portals
3744 * move forward to latest Lustre kernel
3746 2002-06-25 Peter Braam <braam@clusterfs.com>
3747 * release version v0_4_1. Hopefully stable on single node use.
git://git.whamcloud.com / fs / lustre-release.git / blob