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.
15 Frequency : when using more than 256 SCSI disks on a single server
17 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
18 Details : a kernel BUG is hit when using more than 256 SCSI disks.
20 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
22 * Support for kernels:
25 2.6.16.46-0.14 (SLES 10)
26 2.6.18.8 vanilla (kernel.org)
27 * Client support for unpatched kernels:
28 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
29 2.6.16 - 2.6.22 vanilla (kernel.org)
30 * Due to recently discovered recovery problems, we do not recommend
31 using patchless RHEL 4 clients with this or any earlier release.
32 * Recommended e2fsprogs version: 1.39.cfs8
33 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
34 * Starting with this release, the ldiskfs backing filesystem required
35 by Lustre is now in its own package, lustre-ldiskfs. This package
36 should be installed. It is versioned separately from Lustre and
37 may be released separately in future.
39 Severity : enhancement
41 Description: add optional extra BUILD_VERSION info
42 Details : add a new environment variable (namely LUSTRE_VERS) which allows
43 to override the lustre version.
46 Frequency : 2.6.18 servers only
48 Description: ll_kern_mount() doesn't release the module reference
49 Details : The ldiskfs module reference count never drops down to 0
50 because ll_kern_mount() doesn't release the module reference.
55 Description: server LBUG when using old ost_num_threads parameter
56 Details : Accept the old ost_num_threads parameter but warn that it
57 is deprecated, and fix an off-by-one error that caused an LBUG.
62 Description: Transient SCSI error results in persistent IO issue
63 Details : iobuf->dr_error is not reinitialized to 0 between two
67 Frequency : sometimes when underlying device returns I/O errors
69 Description: OSTs not going read-only during write failures
70 Details : OSTs are not remounted read-only when the journal commit threads
71 get I/O errors because fsfilt_ext3 calls journal_start/stop()
72 instead of the ext3 wrappers.
76 Description: poor connect scaling with increasing client count
77 Details : Don't run filter_grant_sanity_check for more than 100 exports
78 to improve scaling for large numbers of clients.
81 Frequency : SLES10 only
83 Description: sanity-quota.sh quotacheck failed: rc = -22
84 Details : Quotas cannot be enabled on SLES10.
87 Frequency : liblustre clients only
89 Description: getdirentries does not give error when run on compute nodes
90 Details : getdirentries does not fail when the size specified as an argument
91 is too small to contain at least one entry
93 Severity : enhancement
95 Description: Add LNET router traceability for debug purposes
96 Details : If a checksum failure occurs with a router as part of the
97 IO path, the NID of the last router that forwarded the bulk data
98 is printed so it can be identified.
103 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
104 Details : Due to a race condition, liblustre clients were occasionally
107 Severity : enhancement
109 Description: lfs setstripe use optional parameters instead of postional
112 Severity : enhancement
114 Description: Nanosecond timestamp support for ldiskfs
115 Details : The on-disk ldiskfs filesystem has added support for nanosecond
116 resolution timestamps. There is not yet support for this at
117 the Lustre filesystem level.
120 Frequency : during server recovery
122 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
123 Details : request with rq_no_resend flag not awake l_wait_event if they get a
127 Frequency : nfs export on patchless client
129 Description: connectathon hang when test nfs export over patchless client
130 Details : Disconnected dentry cannot be found with lookup, so we do not need
131 to unhash it or make it invalid
134 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
135 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
136 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
141 Description: open req refcounting wrong on reconnect
142 Details : If reconnect happened between getting open reply from server and
143 call to mdc_set_replay_data in ll_file_open, we will schedule
144 replay for unreferenced request that we are about to free.
145 Subsequent close will crash in variety of ways.
146 Check that request is still eligible for replay in
147 mdc_set_replay_data().
152 Description: disable writes to filesystem when reading health_check file
153 Details : the default for reading the health_check proc file has changed
154 to NOT do a journal transaction and write to disk, because this
155 can cause reads of the /proc file to hang and block HA state
156 checking on a healthy but otherwise heavily loaded system. It
157 is possible to return to the previous behaviour during configure
158 with --enable-health-write.
160 Severity : enhancement
162 Description: 64-bit inode version
163 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
164 to the inode. This will be required for version-based recovery.
169 Description: MDS fails to start if a duplicate client export is detected
170 Details : in some rare cases it was possible for a client to connect to
171 an MDS multiple times. Upon recovery the MDS would detect this
172 and fail during startup. Handle this more gracefully.
174 Severity : enhancement
176 Description: Add -o localflock option to simulate old noflock
178 Details : This will achieve local-only flock/fcntl locks
184 Description: log_commit_thread vs filter_destroy race leads to crash
185 Details : Take import reference before releasing llog record semaphore
190 Description: Wrong request locking in request set processing
191 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
192 imp_delayed_list, in this place should be used imp_lock.
195 Frequency : when reconnection
197 Description: Grant Leak when osc reconnect to OST
198 Details : When osc reconnect ost, OST(filter) should check whether it
199 should grant more space to client by comparing fed_grant and
200 cl_avail_grant, and return the granted space to client instead
201 of "new granted" space, because client will call osc_init_grant
202 to update the client grant space info.
205 Frequency : when client reconnect to OST
207 Description: Grant Leak when osc do resend and replay bulk write
208 Details : When osc reconnect to OST, OST(filter)should clear grant info of
209 bulk write request, because the grant info while be sync between
210 OSC and OST when reconnect, and we should ignore the grant info
211 these of resend/replay write req.
216 Description: Grant space more than avaiable left space sometimes.
217 Details : When then OST is about to be full, if two bulk writing from
218 different clients came to OST. Accord the avaliable space of the
219 OST, the first req should be permitted, and the second one
220 should be denied by ENOSPC. But if the seconde arrived before
221 the first one is commited. The OST might wrongly permit second
222 writing, which will cause grant space > avaiable space.
225 Frequency : when client is evicted
227 Description: Grant might be wrongly erased when osc is evicted by OST
228 Details : when the import is evicted by server, it will fork another
229 thread ptlrpc_invalidate_import_thread to invalidate the
230 import, where the grant will be set to 0. While the original
231 thread will update the grant it got when connecting. So if
232 the former happened latter, the grant will be wrongly errased
233 because of this race.
238 Description: Checking Stale with correct fid
239 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
240 to check whether it is stale, because sometimes, we want the
241 enqueue happened anyway, and op_data.fid2 will not be initialized.
243 Severity : enhancement
245 Description: update patchless client
246 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
249 Frequency : only with 2.4 kernel
251 Description: random memory corruption
252 Details : size of struct ll_inode_info is to big for union inode.u and this
253 can be cause of random memory corruption.
258 Description: Memory leak in recovery
259 Details : Lov_mds_md was not free in an error handler in mds_create_object.
260 It should also check obd_fail before fsfilt_start, otherwise if
261 fsfilt_start return -EROFS,(failover mds during mds recovery).
262 then the req will return with repmsg->transno = 0 and rc = EROFS.
263 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
264 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
265 be freed no matter whether fsfilt_commit success or not.
268 Frequency : only with huge count clients
270 Description: Prevents from taking the superblock lock in llap_from_page for
272 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
273 for llap_from_page call in ll_removepage prevents from taking the
274 superblock lock for a soon died page.
279 Description: Not check open intent error before release open handle
280 Details : in some rare cases, the open intent error is not checked before
281 release open handle, which may cause
282 ASSERTION(open_req->rq_transno != 0), because it tries to release
283 the failed open handle.
288 Description: Set cat log bitmap only after create log success.
289 Details : in some rare cases, the cat log bitmap is set too early. and it
290 should be set only after create log success.
294 Description: Accessing a block bevice can re-enable I/O when Lustre is
295 tearing down a device.
296 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
301 Description: service threads may hog cpus when there are a lot of requests
303 Details : Insert cond_resched to give other threads a chance to use some of
309 Description: the cat log was not initialized in recovery
310 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
311 unlink log on mds will not be initialized until mds post
312 recovery. And also in mds post recovery, the unlink log will
313 initialization will be done asynchronausly, so there will be race
314 between add unlink log and unlink log initialization.
318 Description: brw_stats were being printed incorrectly
319 Details : brw_stats were being printed as log2 but all of them were not
320 recorded as log2. Also remove some code duplication arising from
321 filter_tally_{read,write}.
325 Frequency : rare, only in recovery.
326 Description: ASSERTION(req->rq_type != LI_POISON) failed
327 Details : imp_lock should be held while iterating over imp_sending_list for
328 prevent destroy request after get timeout in ptlrpc_queue_wait.
332 Description: replay-single.sh test 52 fails
333 Details : A lock's skiplist need to be cleanup when it being unlinked
334 from its resource list.
338 Description: Short directio read returns full requested size rather than
340 Details : Direct I/O operations should return actual amount of bytes
341 transferred rather than requested size.
343 Severity : enhancement
345 Description: metadata RPC reduction (e.g. for rm performance)
346 Details : decrease the amount of synchronous RPC between clients and servers
347 by canceling conflicing lock before the operation on the client side
348 and packing thier handles into the main operation RPC to server.
350 --------------------------------------------------------------------------------
352 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
357 Frequency : on some architectures
359 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
360 Details : Uninitialized flags sometimes cause configuration commands to
364 Frequency : patchless clients only
366 Description: missing __iget() symbol export
367 Details : The __iget() symbol export is missing. To avoid the need for
368 this on patchless clients the deathrow inode reaper is turned
369 off, and we depend on the VM to clean up old inodes. This
370 dependency was during via the fix for bug 12181.
372 --------------------------------------------------------------------------------
374 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
376 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
377 INTEROPERATE with older versions automatically. In many cases a
378 special upgrade step is needed. Please read the
379 user documentation before upgrading any part of a live system.
380 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
381 version will not interoperate with 1.6 betas before beta5 (1.5.95).
382 * WARNING: Lustre configuration and startup changes are required with
383 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
388 Severity : enhancement
390 Description: MountConf
391 Details : Lustre configuration is now managed via mkfs and mount
392 commands instead of lmc and lconf. New obd types (MGS, MGC)
393 are added for dynamic configuration management. See
394 https://mail.clusterfs.com/wikis/lustre/MountConf for
397 Severity : enhancement
399 Description: dynamic OST addition
400 Details : OSTs can now be added to a live filesystem
402 Severity : enhancement
404 Description: startup order invariance
405 Details : MDTs and OSTs can be started in any order. Clients only
406 require the MDT to complete startup.
408 Severity : enhancement
410 Description: parallel, asynchronous orphan cleanup
411 Details : orphan cleanup is now performed in separate threads for each
412 OST, allowing parallel non-blocking operation.
414 Severity : enhancement
416 Description: optimized stripe assignment
417 Details : stripe assignments are now made based on ost space available,
418 ost previous usage, and OSS previous usage, in order to try
419 to optimize storage space and networking resources.
421 Severity : enhancement
423 Description: Permanently set tunables
424 Details : All writable /proc/fs/lustre tunables can now be permanently
425 set on a per-server basis, at mkfs time or on a live
428 Severity : enhancement
430 Description: Lustre message v2
431 Details : Add lustre message format v2.
433 Severity : enhancement
435 Description: client OST exclusion list
436 Details : Clients can be started with a list of OSTs that should be
437 declared "inactive" for known non-responsive OSTs.
441 Description: SPEC SFS validation failure on NFS v2 over lustre.
442 Details : Changes the blocksize for regular files to be 2x RPC size,
443 and not depend on stripe size.
445 Severity : enhancement
447 Description: Multiple MD RPCs in flight.
448 Details : Further unserialise some read-only MDS RPCs - learn about intents.
449 To avoid overly-overloading MDS, introduce a limit on number of
450 MDS RPCs in flight for a single client and add /proc controls
451 to adjust this limit.
453 Severity : enhancement
455 Description: client read/write statistics
456 Details : Add client read/write call usage stats for performance
457 analysis of user processes.
458 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
459 file access. extents_stats shows chunk size distribution.
460 extents_stats_per_process show chunk size distribution per
463 Severity : enhancement
465 Description: mds statistics
466 Details : Add detailed mds operations statistics in
467 /proc/fs/lustre/mds/*/stats.
471 Description: Failure of copying files with lustre special EAs.
472 Details : Client side always return success for setxattr call for lustre
473 special xattr (currently only "trusted.lov").
478 Description: Refcount LNET uuids
479 Details : The global LNET uuid list grew linearly with every startup;
480 refcount repeated list entries instead of always adding to
483 Severity : enhancement
485 Description: Dynamic service threads
486 Details : Within a small range, start extra service threads
487 automatically when the request queue builds up.
490 Frequency : mixed-endian client/server environments
492 Description: mixed-endian crashes
493 Details : The new msg_v2 system had some failures in mixed-endian
496 Severity : enhancement
498 Description: Easy OST removal
499 Details : OSTs can be permanently deactivated with e.g. 'lctl
500 conf_param lustre-OST0001.osc.active=0'
502 Severity : enhancement
504 Description: MGS proc entries
505 Details : Added basic proc entries for the MGS showing what filesystems
508 Severity : enhancement
510 Description: provide MGS failover
511 Details : Added config lock reacquisition after MGS server failover.
513 Severity : enhancement
515 Description: add Linux 2.4 support
516 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
520 Description: a large application tries to do I/O to the same resource and dies
522 Details : Check the req->rq_arrival time after the call to
523 ost_brw_lock_get(), but before we do anything about
524 processing it & sending the BULK transfer request. This
525 should help move old stale pending locks off the queue as
526 quickly as obd_timeout.
529 Frequency : when an incorrect nid is specified during startup
531 Description: ptlrpc connect to non-existant node causes kernel crash
532 Details : LNET can't be re-entered from an event callback, which
533 happened when we expire a message after the export has been
534 cleaned up. Instead, hand the zombie cleanup off to another
537 Severity : enhancement
539 Description: plain/inodebits lock performance improvement
540 Details : Grouping plain/inodebits in granted list by their request modes
541 and bits policy, thus improving the performance of search through
545 Frequency : only if OST filesystem is corrupted
547 Description: client incorrectly hits assertion in ptlrpc_replay_req()
548 Details : for a short time RPCs with bulk IO are in the replay list,
549 but replay of bulk IOs is unimplemented. If the OST filesystem
550 is corrupted due to disk cache incoherency and then replay is
551 started it is possible to trip an assertion. Avoid putting
552 committed RPCs into the replay list at all to avoid this issue.
555 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
557 Description: System hang on startup
558 Details : This bug allowed the liblustre (e.g. catamount) client to
559 return to the app before handling all startup RPCs. This
560 could leave the node unresponsive to lustre network traffic
561 and manifested as a server ptllnd timeout.
563 Severity : enhancement
565 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
566 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
567 boolean to control whether to print peer debug info when a
568 client's RPC times out.
571 Frequency : only for kernels with patches from Lustre below 1.4.3
573 Description: Remove old rdonly API
574 Details : Remove old rdonly API which unused from at least lustre 1.4.3
577 Frequency : only for devices with external journals
579 Description: Set external device read-only also
580 Details : During a commanded failover stop, we set the disk device
581 read-only while the server shuts down. We now also set any
582 external journal device read-only at the same time.
585 Frequency : when upgrading from 1.4 while trying to change parameters
587 Description: The wrong (new) MDC name was used when setting parameters for
588 upgraded MDT's. Also allows changing of OSC (and MDC)
589 parameters if --writeconf is specified at tunefs upgrade time.
592 Frequency : when setting specific ost indicies
594 Description: QOS code breaks on skipped indicies
595 Details : Add checks for missing OST indicies in the QOS code, so OSTs
596 created with --index need not be sequential.
598 Severity : enhancement
600 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
601 Details : The uninit_groups feature works in conjunction with the kernel
602 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
603 up the pass1 processing of e2fsck. This is a read-only feature
604 in ldiskfs2 only, so older kernels and current ldiskfs cannot
605 mount filesystems that have had this feature enabled.
607 Severity : enhancement
609 Description: Improve multi-block allocation algorithm to avoid fragmentation
610 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
611 allocation locality and avoid filesystem fragmentation.
613 ------------------------------------------------------------------------------
615 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
617 * Support for kernels:
618 2.4.21-47.0.1.EL (RHEL 3)
620 2.6.9-42.0.10.EL (RHEL 4)
621 2.6.12.6 vanilla (kernel.org)
622 2.6.16.27-0.9 (SLES 10)
623 * Recommended e2fsprogs version: 1.39.cfs5
625 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
629 Frequency : occasional, depends on client load and configuration
630 Bugzilla : 12181, 12203
631 Description: data loss for recently-modified files
633 Details : In some cases it is possible that recently written or created
634 files may not be written to disk in a timely manner (this should
635 normally be within 30s unless client IO load is very high).
636 The problem appears as zero-length files or files that are a
637 multiple of 1MB in size after a client crash or client eviction
638 that are missing data at the end of the file.
640 This problem is more likely to be hit on clients where files are
641 repeatedly created and unlinked in the same directory, clients
642 have a large amount of RAM, have many CPUs, the filesystem has
643 many OSTs, the clients are rebooted frequently, and/or the files
644 are not accessed by other nodes after being written.
646 The presence of the problem can be detected by looking at
647 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
648 smaller than the second (nr_unused) then dirty files will not
649 be flushed automatically to disk. "sync; sleep 10" should be
650 run several times on the node before unmounting it to update
651 Lustre (this is also safe to run on nodes without this problem).
653 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
654 that can cause this same problem, so customers using that kernel
655 also need to update the kernel in addition to Lustre. In order
656 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
658 It is normal that files written just before a client crash (less
659 than 30s) may not yet have been flushed to disk, even for local
663 Frequency : frequent on thin XT3 nodes
665 Description: UUID collision on thin XT3 Linux nodes
666 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
667 randomly, since we relied on an insufficiently-seeded PRNG.
672 Description: OSS hangs after "All ost request buffers busy"
673 Details : A deadlock between quota and journal operations caused OSS
674 hangs after printing "All ost request buffers busy."
677 Frequency : always on liblustre builds
679 Description: Cleanup compiler warnings on liblustre
682 Frequency : always on liblustre builds on XT3
684 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
688 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
689 > 32000 subdirectories
690 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
691 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
692 subdirectory count crosses 32000. This will aid e2fsck to
693 correctly handle more than 32000 subdirectories.
696 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
698 Description: System hang on startup
699 Details : This bug allowed the liblustre (e.g. catamount) client to
700 return to the app before handling all startup RPCs. This
701 could leave the node unresponsive to lustre network traffic
702 and manifested as a server ptllnd timeout.
704 Severity : enhancement
706 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
707 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
708 boolean to control whether to print peer debug info when a
709 client's RPC times out.
714 Description: make O_SYNC working on 2.6 kernels
715 Details : 2.6 kernels use different method for mark pages for write,
716 so need add a code to lustre for O_SYNC work.
721 Description: Failure to close file and release space on NFS
722 Details : Put inode details into lock acquired in ll_intent_file_open.
723 Use mdc_intent_lock in ll_intent_open to properly
724 detect all kind of errors unhandled by mdc_enqueue.
729 Description: proc file read during shutdown sometimes raced obd removal,
731 Details : Add lock to prevent obd access after proc file removal.
734 Frequency : Only for files larger than 4GB on 32-bit clients.
736 Description: improperly doing page alignment of locks
737 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
738 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
743 Description: RPCs being resent when they shouldn't be
744 Details : Some RPCs that should not be resent are being resent. This
745 can cause inconsistencies in the RPC state machine. Do not
746 resend such requests.
749 Frequency : rare, only with NFS export
751 Description: Crash on NFS re-export node
752 Details : under very unusual load conditions an assertion is hit in
753 ll_intent_file_open()
756 Frequency : only if OST filesystem is corrupted
758 Description: client incorrectly hits assertion in ptlrpc_replay_req()
759 Details : for a short time RPCs with bulk IO are in the replay list,
760 but replay of bulk IOs is unimplemented. If the OST filesystem
761 is corrupted due to disk cache incoherency and then replay is
762 started it is possible to trip an assertion. Avoid putting
763 committed RPCs into the replay list at all to avoid this issue.
768 Description: large O_DIRECT requests fail under memory pressure/fragmentation
769 Details : Large single O_DIRECT read and write calls can fail to allocate
770 a sufficiently large buffer to process the request. In case of
771 allocation failure the allocation is retried with a smaller
772 buffer and broken into smaller requests.
774 Severity : enhancement
776 Description: Add -o localflock option to simulate old noflock behaviour.
777 Details : This will achieve local-only flock/fcntl locks coherentness.
782 Description: versioning check is incomplete
783 Details : Checking the version difference of client vs. server, report
784 error if the gap is too big.
789 Description: add support PG_writeback bit
790 Details : add support for PG_writeback bit for Lustre, for more carefull
791 work with page cache in 2.6 kernel. This also fix some deadlocks
792 and remove hack for work O_SYNC with 2.6 kernel.
794 Severity : enhancement
796 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
797 Details : The uninit_groups feature works in conjunction with the kernel
798 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
799 up the pass1 processing of e2fsck. This is a read-only feature
800 in ldiskfs2 only, so older kernels and current ldiskfs cannot
801 mount filesystems that have had this feature enabled.
803 Severity : enhancement
805 Description: Improve multi-block allocation algorithm to avoid fragmentation
806 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
807 allocation locality and avoid filesystem fragmentation.
809 ------------------------------------------------------------------------------
811 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
813 * Support for kernels:
814 2.6.9-42.0.3.EL (RHEL 4)
816 2.4.21-47.0.1.EL (RHEL 3)
817 2.6.12.6 vanilla (kernel.org)
818 2.6.16.21-0.8 (SLES10)
819 * Recommended e2fsprogs version: 1.39.cfs2-0
821 * The backwards-compatible /proc/sys/portals symlink has been removed
822 in this release. Before upgrading, please ensure that you change
823 any configuration scripts or /etc/sysctl.conf files that access
824 /proc/sys/portals/* or sysctl portals.* to use the corresponding
825 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
826 in advance of the upgrade on any system running Lustre 1.4.6 or
827 newer, since /proc/sys/lnet was added in that version.
828 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
832 Frequency : only when quota is used
834 Description: avoid scanning export list for quota master
835 Details : Change the algorithms to avoid scanning export list in order
836 to improve the efficiency.
839 Frequency : MDS failover only, very rarely
841 Description: "went back in time" messages on mds failover
842 Details : The greatest transno may be lost when the current operation
843 finishes with an error (transno==0) and the client's last_rcvd
844 record is over-written. Save the greatest transno in the
845 mds_last_transno for this case.
848 Frequency : always for specific kernels and striping counts
850 Description: client may get "Matching packet too big" without ACL support
851 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
852 when trying to access files in certain configurations. The
853 clients should in fact be denied when mounting because they do
857 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
859 Description: many clients connecting with IO in progress causes connect timeouts
860 Details : Avoid synchronous journal commits to avoid delays caused by many
861 clients connecting/disconnecting when bulk IO is in progress.
862 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
863 OST_IO_PORTAL to avoid delays behind potentially many pending
867 Frequency : occasionally with multiple writers to a single file
869 Description: shared writes to file may result in wrong size reported by stat()
870 Details : Allow growing of kms when extent lock is cancelled
873 Frequency : always with random mmap IO to multi-striped file
875 Description: mmap write might be lost if we are writing to a 'hole' in stripe
876 Details : Only if the hole is at the end of OST object so that kms is too
877 small. Fix is to increase kms accordingly in ll_nopage.
880 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
882 Description: writes to a missing object would leak memory on the OST
883 Details : If there is an inconsistency between the MDS and OST filesystems,
884 such that the MDS references an object that doesn't exist, writes
885 to that object will leak memory due to incorrect cleanup in the
886 error handling path, eventually running out of memory on the OST.
891 Description: Creating too long symlink causes lustre errors
892 Details : Check symlink and name lengths before sending requests to MDS.
895 Frequency : only if flock is enabled (not on by default)
897 Description: posix locks not released on fd closure on 2.6.9+
898 Details : We failed to add posix locks to list of inode locks on 2.6.9+
899 kernels, this caused such locks not to be released on fd close and
900 then assertions on fs unmount about still used locks.
903 Frequency : MDS failover only, very rarely
905 Description: clients may get ASSERTION(granted_lock != NULL)
906 Details : When request was taking a long time, and a client was resending
907 a getattr by name lock request. The were multiple lock requests
908 with the same client lock handle and
909 mds_getattr_name->fixup_handle_for_resent_request found one of the
910 lock handles but later failed with ASSERTION(granted_lock != NULL).
915 Description: handle->h_buffer_credits > 0, assertion failure
916 Details : h_buffer_credits is zero after truncate, causing assertion
917 failure. This patch extends the transaction or creates a new
921 Frequency : NFS re-export or patchless client
922 Bugzilla : 11179, 10796
923 Description: Crash on NFS re-export node (__d_move)
924 Details : We do not want to hash the dentry if we don't have a lock.
925 But if this dentry is later used in d_move, we'd hit uninitialised
926 list head d_hash, so we just do this to init d_hash field but
927 leave dentry unhashed.
930 Frequency : NFS re-export or patchless client
932 Description: NFS exports has problem with symbolic link
933 Details : lustre client didn't properly install dentry when re-exported
934 to NFS or running patchless client.
937 Frequency : NFS re-export or patchless client
939 Description: Various nfs/patchless fixes.
940 Details : fixes reuse disconected alias for lookup process - this fixes
941 warning "find_exported_dentry: npd != pd",
942 fix permission error with open files at nfs.
943 fix apply umask when do revalidate.
946 Frequency : occasional
948 Description: Crash on NFS re-export node
949 Details : calling clear_page() on the wrong pointer triggered oops in
950 generic_mapping_read().
953 Frequency : rarely, using O_DIRECT IO
955 Description: unaligned directio crashes client with LASSERT
956 Details : check for unaligned buffers before trying any requests.
959 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
961 Description: stale data returned from RAID cache
962 Details : If only a small amount of IO is done to the RAID device before
963 reading it again it is possible to get stale data from the RAID
964 cache instead of reading it from disk.
967 Frequency : always for sles10 kernel
969 Description: sles10 support
970 Details : ll_follow_link: compile fixes and using of nd_set_link
974 Frequency : depends on arch, kernel and compiler version, always on sles10
977 Description: recursive or deep enough symlinks cause stack overflow
978 Details : getting rid of large stack-allocated variable in
982 Frequency : depends on hardware
984 Description: lustre write performance loss in the SLES10 kernel
985 Details : the performance loss is caused by using of write barriers in the
986 ext3 code. The SLES10 kernel turns barrier support on by
987 default. The fix is to undo that change for ldiskfs.
989 ------------------------------------------------------------------------------
991 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
993 * Support for kernels:
994 2.6.9-42.0.3EL (RHEL 4)
996 2.4.21-47.0.1.EL (RHEL 3)
997 2.6.12.6 vanilla (kernel.org)
1001 Frequency : quota enabled and large files being deleted
1003 Description: releasing more than 4GB of quota at once hangs OST
1004 Details : If a user deletes more than 4GB of files on a single OST it
1005 will cause the OST to spin in an infinite loop. Release
1006 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1011 Description: statfs data retrieved from /proc may be stale or zero
1012 Details : When reading per-device statfs data from /proc, in the
1013 {kbytes,files}_{total,free,avail} files, it may appear
1014 as zero or be out of date.
1017 Frequency : systems with MD RAID1 external journal devices
1019 Description: lconf's call to blkid is confused by RAID1 journal devices
1020 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1021 of returning all block devices that match the journal UUID.
1024 Frequency : always, for aggregate stripe size over 4GB
1026 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1027 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1028 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1031 Frequency : always if "lfs find" used on a local file/directory
1033 Description: "lfs find" segfaults if used on a local file/directory
1034 Details : The case where a directory component was not specified wasn't
1035 handled correctly. Handle this properly.
1038 Frequency : always on ppc64
1040 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1041 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1042 perform endianness conversion.
1045 Frequency : rarely (truncate to non-zero file size after write under load)
1046 Bugzilla : 10730, 10687
1047 Description: Files padded with zeros to next 4K multiple
1048 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1049 it is possible that files that are truncated to a non-zero size
1050 immediately after being written are filled with zero bytes beyond
1051 the truncated size. No file data is lost.
1053 Severity : enhancement
1055 Description: Allow recovery/failover for liblustre clients.
1056 Details : liblustre clients were unaware of failover configurations until
1059 Severity : enhancement
1061 Description: user file locks should fail when not mounting with flock option
1062 Details : Set up an error-returning stub in ll_file_operations.lock field
1063 to prevent incorrect behaviour when client is mounted without
1064 flock option. Also, set up properly f_op->flock field for
1068 Frequency : always on ia64
1070 Description: "lfs df" loops on printing out MDS statfs information
1071 Details : The obd_ioctl_data was not initialized and in some systems
1072 this caused a failure during the ioctl that did not return
1073 an error. Initialize the struct and return an error on failure.
1076 Frequency : SLES 9 only
1078 Description: Error of copying files with lustre special EAs as root
1079 Details : Client side always return success for setxattr call for lustre
1080 special xattr (currently only "trusted.lov").
1083 Frequency : rarely on clusters with both ia64+i386 clients
1085 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1086 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1087 clients are doing concurrent writes to the same file it is
1088 possible that the ia64 clients may LASSERT because the OST
1089 extent locks are not PAGE_SIZE aligned. Ensure that grown
1090 locks are always aligned on the request boundary.
1093 Frequency : specific use, occasional
1095 Description: Overwriting in use executable truncates on-disk binary image
1096 Details : If one node attempts to overwrite an executable in use by
1097 another node, we now correctly return ETXTBSY instead of
1098 truncating the file.
1100 Severity : enhancement
1102 Description: Async OSC create to avoid the blocking unnecessarily.
1103 Details : If a OST has no remain object, system will block on the creating
1104 when need to create a new object on this OST. Now, ways use
1105 pre-created objects when available, instead of blocking on an
1106 empty osc while others are not empty. If we must block, we block
1107 for the shortest possible period of time.
1112 Description: chmod on Lustre root is propagated to other clients
1113 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1114 invalid by the follow_mount time.
1117 Frequency : liblustre clients only
1119 Description: Race in 'instant cancel' lock handling could lead to such locks
1120 never to be granted in case of SMP MDS
1121 Details : Do not destroy not yet granted but cbpending locks in
1125 Frequency : replay/resend of open
1127 Description: non null lock assetion failure in mds_intent_policy
1128 Details : Trying to replay/resend lockless open requests resulted in
1129 mds_open() returning 0 with no lock. Now it sets a flag if
1130 a lock is going to be returned.
1132 Severity : enhancement
1134 Description: Checksum enhancements
1135 Details : New checksum enhancements allow for resending RPCs that failed
1138 Severity : enhancement
1140 Description: Tunables on number of dirty pages in cacche
1141 Details : Allow to set limit on number of dirty pages cached.
1146 Description: client crash on unmount - lock still has references
1147 Details : In some error handling cases it was possible to leak a lock
1148 reference on a client while accessing a file. This was not
1149 harmful to the client during operation, but would cause the
1150 client to crash when the filesystem is unmounted.
1153 Frequency : specific case, rare
1155 Description: ETXTBSY on mds though file not in use by client
1156 Details : ETXTBSY is no longer incorrectly returned when attempting to
1157 chmod or chown a directory that the user previously tried to
1158 execute or a currently-executing binary.
1161 Frequency : extremely rare except on liblustre-based clients
1163 Description: Lustre space not freed when files are deleted
1164 Details : Clean up open-unlinked files after client eviction. Previously
1165 the unlink was skipped and the files remained as orphans.
1170 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1171 Details : In some cases it was possible to send a blocking callback to
1172 a client doing a glimpse, even though that client didn't get
1173 a lock granted. When the glimpse lock is cancelled on the OST
1174 the freed lock is left on the waiting list and corrupted the list.
1177 Frequency : all core dumps
1179 Description: Broke core dumps to lustre
1180 Details : Negative dentry may be unhashed if parent does not have UPDATE
1181 lock, but some callers, e.g. do_coredump, expect dentry to be
1182 hashed after successful create, hash it in ll_create_it.
1184 ------------------------------------------------------------------------------
1186 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1188 * Support for kernels:
1189 2.6.9-42.0.2.EL (RHEL 4)
1190 2.6.5-7.276 (SLES 9)
1191 2.4.21-40.EL (RHEL 3)
1192 2.6.12.6 vanilla (kernel.org)
1196 Frequency : always on RHEL 3
1198 Description: Number of open files grows over time
1199 Details : The number of open files grows over time, whether or not
1200 Lustre is started. This was due to a filp leak introduced
1201 by one of our kernel patches.
1203 ------------------------------------------------------------------------------
1205 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1207 * Support for kernels:
1208 2.6.9-42.EL (RHEL 4)
1209 2.6.5-7.276 (SLES 9)
1210 2.4.21-40.EL (RHEL 3)
1211 2.6.12.6 vanilla (kernel.org)
1216 Bugzilla : 5719, 9635, 9792, 9684
1217 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1218 Details : If a server is under heavy load and cannot reply to new
1219 connection requests before the client resends the (re)connect,
1220 the connection handling code can behave badly if two service
1221 threads are concurrently handing separate (re)connections from
1222 the same client. Add better locking to the connection handling
1223 code, and ensure that only a single connection will be processed
1224 for a given client UUID, even if the lock is dropped.
1226 Severity : enhancement
1228 Description: add TCP zero-copy support to kernel
1229 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1230 sends if the hardware supports scatter-gather and checksumming.
1231 This allows socklnd to do client-write and server-read more
1232 efficiently and reduce CPU utilization from skbuf copying.
1235 Frequency : only if NFS exporting from client
1237 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1238 Details : If setting attributes on a file created under NFS that had
1239 never been opened it would be possible to oops the client
1240 if the file had no objects.
1243 Frequency : always for liblustre
1245 Description: liblustre client does MDS+OSTs setattr RPC for each write
1246 Details : When doing a write from a liblustre client, the client
1247 incorrectly issued an RPC to the MDS and each OST the file was
1248 striped over in order to update the timestamps. When writing
1249 with small chunks and many clients this could overwhelm the MDS
1250 with RPCs. In all cases it would slow down the write because
1251 these RPCs are unnecessary.
1253 Severity : enhancement
1255 Description: allow number of MDS service threads to be changed at module load
1256 Details : It is now possible to change the number of MDS service threads
1257 running. Adding "options mds mds_num_threads={N}" to the MDS's
1258 /etc/modprobe.conf will set the number of threads for the next
1259 time Lustre is restarted (assuming the "mds" module is also
1260 reloaded at that time). The default number of threads will
1261 stay the same, 32 for most systems.
1266 Description: OST crash if filesystem is unformatted or corrupt
1267 Details : If an OST is started on a device that has never been formatted
1268 or if the filesystem is corrupt and cannot even mount then the
1269 error handling cleanup routines would dereference a NULL pointer.
1274 Description: NULL pointer deref in llap_from_page.
1275 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1276 incorrect) mapping if the page was truncated/reclaimed while it was
1277 searched for. Check for this condition and skip such pages when
1278 doing readahead. Introduce extra check to llap_from_page() to
1279 verify page->mapping->host is non-NULL (so page is not anonymous).
1282 Frequency : Sometimes when using sys_sendfile
1284 Description: "page not covered by a lock" warnings from ll_readpage
1285 Details : sendfile called ll_readpage without right page locks present.
1286 Now we introduced ll_file_sendfile that does necessary locking
1287 around call to generic_file_sendfile() much like we do in
1291 Frequency : with certain MDS communication failures at client mount time
1293 Description: NULL pointer deref after failed client mount
1294 Details : a client connection request may delayed by the network layer
1295 and not be sent until after the PTLRPC layer has timed out the
1296 request. If the client fails the mount immediately it will try
1297 to clean up before the network times out the request. Add a
1298 reference from the request import to the obd device and delay
1299 the cleanup until the network drops the request.
1302 Frequency : occasionally during client (re)connect
1304 Description: assertion failure during client (re)connect
1305 Details : processing a client connection request may be delayed by the
1306 client or server longer than the client connect timeout. This
1307 causes the client to resend the connection request. If the
1308 original connection request is replied in this interval, the
1309 client may trip an assertion failure in ptlrpc_connect_interpret()
1310 which thought it would be the only running connect process.
1313 Frequency : only with obd_echo servers and clients that are rebooted
1315 Description: kernel BUG accessing uninitialized data structure
1316 Details : When running an obd_echo server it did not start the ping_evictor
1317 thread, and when a client was evicted an uninitialized data
1318 structure was accessed. Start the ping_evictor in the RPC
1319 service startup instead of the OBD startup.
1321 Severity : enhancement
1322 Bugzilla : 10193 (patchless)
1323 Description: Remove dependency on various unexported kernel interfaces.
1324 Details : No longer need reparent_to_init, exit_mm, exit_files,
1325 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1328 Frequency : rare (only users of deprecated and unsupported LDAP config)
1330 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1331 Details : LDAP apparently contains 'lustreName' attributes instead of
1332 'name'. A simple remapping of the name is sufficient.
1335 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1337 Description: waiting_locks_callback trips kernel BUG if client is evicted
1338 Details : Running with the dump_on_timeout debug flag turned on makes
1339 it possible that the waiting_locks_callback() can try to dump
1340 the Lustre kernel debug logs from an interrupt handler. Defer
1341 this log dumping to the expired_lock_main() thread.
1343 Severity : enhancement
1345 Description: Support NFS exporting on 2.6 kernels.
1346 Details : Implement non-rawops metadata methods for NFS server to use without
1347 changing NFS server code.
1350 Frequency : very rare (synthetic metadata workload only)
1352 Description: two racing renames might cause an MDS thread to deadlock
1353 Details : Running the "racer" program may cause one MDS thread to rename
1354 a file from being the source of a rename to being the target of
1355 a rename at exactly the same time that another thread is doing
1356 so, and the second thread has already enqueued these locks after
1357 doing a lookup of the target and is trying to relock them in
1358 order. Ensure that we don't try to re-lock the same resource.
1361 Frequency : only very large systems with liblustre clients
1363 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1364 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1365 all OSTs in parallel. This allows the eviction of stale liblustre
1366 clients to proceed much faster than if they were done in series,
1367 and also offers similar improvements for other set_info RPCs.
1372 Description: excessive CPU usage during initial read phase on client
1373 Details : During the initial read phase on a client, it would agressively
1374 retry readahead on the file, consuming too much CPU and impacting
1375 performance (since 1.4.5.8). Improve the readahead algorithm
1376 to avoid this, and also improve some other common cases (read
1377 of small files in particular, where "small" is files smaller than
1378 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1383 Description: MDS crash when receiving packet with unknown intent.
1384 Details : Do not LBUG in unknown intent case, just return -EFAULT
1386 Severity : enhancement
1387 Bugzilla : 9293, 9385
1388 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1389 Details : Do not serialize getattr (non-intent version) and statfs.
1392 Frequency : occasional, when OST network is overloaded/intermittent
1394 Description: client evicted by OST after bulk IO timeout
1395 Details : If a client sends a bulk IO request (read or write) the OST
1396 may evict the client if it is unresposive to its data GET/PUT
1397 request. This is incorrect if the network is overloaded (takes
1398 too long to transfer the RPC data) or dropped the OST GET/PUT
1399 request. There is no need to evict the client at all, since
1400 the pinger and/or lock callbacks will handle this, and the
1401 client can restart the bulk request.
1404 Frequency : Always when mmapping file with no objects
1406 Description: client crashes when mmapping file with no objects
1407 Details : Check that we actually have objects in a file before doing any
1408 operations on objects in ll_vm_open, ll_vm_close and
1414 Description: Request leak when working with deleted CWD
1415 Details : Introduce advanced request refcount tracking for requests
1416 referenced from lustre intent.
1418 Severity : Enhancement
1420 Description: Cache open file handles on client.
1421 Details : MDS now will return special lock along with openhandle, if
1422 requested and client is allowed to hold openhandle, even if unused,
1423 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1424 closing files for every read/write openration.
1426 Severity : Enhancement
1428 Description: Cache open negative dentries on client when possible.
1429 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1430 negative dentries on lock revocation.
1435 Description: Remounting a client read-only wasn't possible with a zconf mount
1436 Details : It wasn't possible to remount a client read-only with llmount.
1438 Severity : enhancement
1439 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1440 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1441 MPICH is included to add Lustre-specific ADIO interfaces.
1442 This is based closely on the UFS ADIO layer and only differs
1443 in file creation, in order to allow the OST striping to be set.
1444 This is user-contributed code and not supported by CFS.
1449 Description: extended inode attributes (immutable, append-only) work improperly
1450 when 2.4 and 2.6 kernels are used on client/server or vice versa
1451 Details : Introduce kernel-independent values for these flags.
1453 Severity : enhancement
1456 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1457 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1458 granularity. Now, Fractional MB granularity is be supported,
1459 this is very useful for low memory system.
1461 Severity : enhancement
1463 Description: Getattr by fid
1464 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1465 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1466 have locks and avoids getattr rpc requests for it on every stat.
1469 Frequency : Always, for filesystems larger than 2TB
1471 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1472 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1473 filesystems larger than 2TB to be mounted with mballoc enabled.
1476 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1478 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1479 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1480 filesystems larger than 2TB is mounted then the kernel will
1481 silently corrupt the start of the filesystem. CONFIG_LBD is
1482 enabled for all CFS-supported kernels, but the possibility of
1483 this happening with a modified kernel config exists.
1485 Severity : enhancement
1487 Description: add client O_DIRECT support for 2.6 kernels
1488 Details : It is now possible to do O_DIRECT reads and writes to files
1489 in the Lustre client mountpoint on 2.6 kernel clients.
1491 Severity : enhancement
1493 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1494 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1495 parallel, not waiting for response from every OST before sending
1496 a rpc to the next OST.
1501 Description: setattr vs write race when updating file timestamps
1502 Details : Client processes that update a file timestamp into the past
1503 right after writing to the file (e.g. tar) it is possible that
1504 the updated file modification time can be reset to the current
1505 time due to a race between processing the setattr and write RPC.
1507 Severity : enhancement
1509 Description: Bring 'lfs find' closer in line with regular Linux find.
1510 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1511 -print0 options and obtains all the needed info through the lustre
1514 Severity : enhancement
1516 Description: support up to 1024 configured devices on one node
1517 Details : change obd_dev array from statically allocated to dynamically
1518 allocated structs as they are first used to reduce memory usage
1523 Description: Flush dirty partially truncated pages during truncate
1524 Details : Immediatelly flush partially truncated pages in filter_setattr,
1525 this way we completely avoid having any pages in page cache on OST
1526 and can retire ugly workarounds during writes to flush such pages.
1531 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1532 Details : protect inode from truncation within vfs_unlink() context
1533 just take a reference before calling vfs_unlink() and release it
1534 when parent's i_sem is free.
1539 Description: last_id value checked outside lock on OST caused LASSERT failure
1540 Details : If there were multiple MDS->OST object precreate requests in
1541 flight, it was possible that the OST's last object id was checked
1542 outside a lock and incorrectly tripped an assertion. Move checks
1543 inside locks, and discard old precreate requests.
1546 Frequency : always, if extents are used on OSTs
1548 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1549 Details : OSTs using the extents format would not zero the high 16 bits of
1550 the index physical block number. This is not a problem for any
1551 OST filesystems smaller than 16TB, and no kernels support ext3
1552 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1553 new/modified files) and can be fixed for existing filesystems
1554 with e2fsprogs-1.39-cfs1.
1559 Description: import connection selection may be incorrect if timer wraps
1560 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1561 import connections to be ignored if the 32-bit jiffies counter
1562 wraps. Use a 64-bit jiffies counter.
1565 Frequency : very large clusters immediately after boot
1567 Description: LNET request buffers exhausted under heavy short-term load
1568 Details : If a large number of client requests are generated on a service
1569 that has previously never seen so many requests it is possible
1570 that the request buffer growth cannot keep up with the spike in
1571 demand. Instead of dropping incoming requests, they are held in
1572 the LND until the RPC service can accept more requests.
1575 Frequency : Sometimes during replay
1577 Description: Assertion failure in ll_local_open after replay.
1578 Details : If replay happened on an open request reply before we were able
1579 to set replay handler, reply will become not swabbed tripping the
1580 assertion in ll_local_open. Now we set the handler right after
1581 recognising of open request
1584 Frequency : very rare
1586 Description: kernel reports "badness in vsnprintf"
1587 Details : Reading from the "recovery_status" /proc file in small chunks
1588 may cause a negative length in lprocfs_obd_rd_recovery_status()
1589 call to vsnprintf() (which is otherwise harmless). Exit early
1590 if there is no more space in the output buffer.
1592 Severity : enhancement
1594 Description: clear OBD RPC statistics by writing to them
1595 Details : It is now possible to clear the OBD RPC statistics by writing
1596 to the "stats" file.
1601 Description: Client mtime is not the same on different clients after utimes
1602 Details : In some cases, the client was using the utimes() syscall on
1603 a file cached on another node. The clients now validate the
1604 ctime from the MDS + OSTs to determine which one is right.
1609 Description: Inability to activate failout mode
1610 Details : lconf script incorrectly assumed that in pythong string's numeric
1611 value is used in comparisons.
1614 Frequency : always with multiple stripes per file
1616 Description: Inefficient object allocation for mutli-stripe files
1617 Details : When selecting which OSTs to stripe files over, for files with
1618 a stripe count that divides evenly into the number of OSTs,
1619 the MDS is always picking the same starting OST for each file.
1620 Return the OST selection heuristic to the original design.
1625 Description: mount failures may take full timeout to return an error
1626 Details : Under some heavy load conditions it is possible that a
1627 failed mount can wait for the full obd_timeout interval,
1628 possibly several minutes, before reporting an error.
1629 Instead return an error as soon as the status is known.
1631 Frequency : quota enabled and large files being deleted
1633 Description: releasing more than 4GB of quota at once hangs OST
1634 Details : If a user deletes more than 4GB of files on a single OST it
1635 will cause the OST to spin in an infinite loop. Release
1636 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1641 Description: statfs data retrieved from /proc may be stale or zero
1642 Details : When reading per-device statfs data from /proc, in the
1643 {kbytes,files}_{total,free,avail} files, it may appear
1644 as zero or be out of date.
1647 Frequency : systems with MD RAID1 external journal devices
1649 Description: lconf's call to blkid is confused by RAID1 journal devices
1650 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1651 of returning all block devices that match the journal UUID.
1654 Frequency : always, for aggregate stripe size over 4GB
1656 Description: assertion fails when trying to use 4GB stripe size
1657 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
1658 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
1661 Frequency : always on ppc64
1663 Description: the first write on an ext3 filesystem with mballoc got stuck
1664 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1665 perform endianness conversion.
1667 ------------------------------------------------------------------------------
1669 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1671 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1672 INTEROPERATE with older versions automatically. Please read the
1673 user documentation before upgrading any part of a live system.
1674 * WARNING: Lustre networking configuration changes are required with
1675 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1678 * Support for kernels:
1679 2.6.9-22.0.2.EL (RHEL 4)
1680 2.6.5-7.244 (SLES 9)
1681 2.6.12.6 vanilla (kernel.org)
1684 Severity : enhancement
1685 Bugzilla : 7981/8208
1686 Description: Introduced Lustre Networking (LNET)
1687 Details : LNET is new networking infrastructure for Lustre, it includes
1688 a reorganized network configuration mode (see the user
1689 documentation for full details) as well as support for routing
1690 between different network fabrics. Lustre Networking Devices
1691 (LNDs) for the supported network fabrics have also been
1692 created for this new infrastructure.
1694 Severity : enhancement
1695 Description: Introduced Access control lists
1696 Details : clients can set ACLs on files and directories in order to have
1697 more fine-grained permissions than the standard Unix UGO+RWX.
1698 The MDS must be started with the "-o acl" mount option.
1700 Severity : enhancement
1701 Description: Introduced filesystem quotas
1702 Details : Administrators may now establish per-user quotas on the
1705 Severity : enhancement
1707 Description: Configuration change for the XT3
1708 The PTLLND is now used to run Lustre over Portals on the XT3
1709 The configure option(s) --with-cray-portals are no longer used.
1710 Rather --with-portals=<path-to-portals-includes> is used to
1711 enable building on the XT3. In addition to enable XT3 specific
1712 features the option --enable-cray-xt3 must be used.
1717 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1718 Details : A race between allocating a new llcd and re-getting the llcd_lock
1719 allowed another thread to grab newly-allocated llcd.
1721 Severity : enhancement
1723 Description: 2.6 OST async journal commit and locking fix to improve performance
1724 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1725 2.6 kernels are now async as they already were in 2.4 kernels so
1726 that they can commit concurrently with the network bulk transfer.
1727 For block-allocated files the filter allocation semaphore is held
1728 to avoid filesystem fragmentation during allocation. BKL lock
1729 removed for 2.6 xattr operations where it is no longer needed.
1734 Description: lconf incorrectly determined whether two IP networks could talk
1735 Details : In some more complicated routing and multiple-network
1736 configurations, lconf will avoid trying to make a network
1737 connection to a disjoint part of the IP space. It was doing the
1738 math incorrectly for one set of cases.
1743 Description: Fix for potential infinite loop processing records in an llog.
1744 Details : If an llog record is corrupted/zeroed, it is possible to loop
1745 forever in llog_process(). Validate the llog record length
1746 and skip the remainder of the block on error.
1749 Frequency : occasional (liblustre only)
1751 Description: liblustre could not open files whose last component is a symlink
1752 Details : sysio_path_walk() would incorrectly pass the open intent to
1753 intermediate path components.
1756 Frequency : rare (liblustre only with non-standard tuning)
1757 Bugzilla : 7201 (7350)
1758 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1759 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1760 after it has been released, only do this for referenced locks
1762 Severity : enhancement
1764 Description: specifying an (invalid) directory default stripe_size of -1
1765 would reset the directory default striping
1766 Details : stripe_size -1 was used internally to signal directory stripe
1767 removal, now use "all default" to signal dir stripe removal
1768 as a directory striping of "all default" is not useful
1771 Frequency : common for large clusters running liblustre clients
1773 Description: doing an ls when liblustre clients are running is slow
1774 Details : sending a glimpse AST to a liblustre client waits for every AST
1775 to time out, as liblustre clients will not respond. Since they
1776 cannot cache data we refresh the OST lock LVB from disk instead.
1778 Severity : enhancement
1780 Description: doing an ls at the same time as file IO can be slow
1781 Details : enqueue and other "small" requests can be blocked behind many
1782 large IO requests. Create a new OST IO portal for non-IO
1783 requests so they can be processed faster.
1786 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1788 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1789 created file with mode 000
1790 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1791 allow an ftruncate() as a non-root user to a file with mode 000.
1792 Lustre now respects this flag to disable mode checks when
1793 truncating a file owned by the user
1796 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1799 Description: Revoking locks from clients that went dead or catatonic might take
1801 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1802 cancellation of such locks instant on servers without waiting for
1803 any reply from clients. Clients drops these locks when cancel
1804 notification from server is received without replying.
1807 Frequency : liblustre-only, when liblustre client dies or becomes busy
1809 Description: Doing ls on Linux clients can take a long time with active
1811 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1812 granting such locks to it in the first place if possible. Locks
1813 are taken by proxy on the OST during the read or write and
1814 dropped immediately afterward. Add connect flags handling, do
1815 not grant locks to liblustre clients for glimpse ASTs.
1817 Severity : enhancement
1819 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1820 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1821 sequential reads and behaves badly with applications doing random
1822 reads. Improve it by reading ahead at least read region, and
1823 avoiding excessive large RPC for small reads.
1825 Severity : enhancement
1827 Description: Creating more than 1000 files for a single job may cause a load
1828 imbalance on the OSTs if there are also a large number of OSTs.
1829 Details : qos_prep_create() uses an OST index reseed value that is an
1830 even multiple of the number of available OSTs so that if the
1831 reseed happens in the middle of the object allocation it will
1832 still utilize the OSTs as uniformly as possible.
1837 Description: OST or MDS may oops in ping_evictor_main()
1838 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1839 but doesn't restart at beginning of obd_exports_timed list
1842 Severity : enhancement
1844 Description: improve by-nid export eviction on the MDS and OST
1845 Details : allow multiple exports with the same NID to be evicted at one
1846 time without re-searching the exports list.
1849 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1851 Description: MDS may oops in groups_free()
1852 Details : in rare race conditions a newly allocated group_info struct is
1853 freed again, and this can be NULL. The 2.4 compatibility code
1854 for groups_free() checked for a NULL pointer, but 2.6 did not.
1857 Frequency : common for liblustre clients doing little filesystem IO
1858 Bugzilla : 9352, 7313
1859 Description: server may evict liblustre clients accessing contended locks
1860 Details : if a client is granted a lock or receives a completion AST
1861 with a blocking AST already set it would not reply to the AST
1862 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1865 Frequency : lfs setstripe, only systems with more than 160 OSTs
1867 Description: unable to set striping with a starting offset beyond OST 160
1868 Details : llapi_create_file() incorrectly limited the starting stripe
1869 index to the maximum single-file stripe count.
1872 Frequency : LDAP users only
1874 Description: lconf did not handle in-kernel recovery with LDAP properly
1875 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1877 Severity : enhancement
1879 Description: bind OST threads to NUMA nodes to improve performance
1880 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1881 node and do their allocations there to localize memory access
1883 Severity : enhancement
1885 Description: llmount can determine client NID directly from Myrinet (GM)
1886 Details : the client NID code from gmnalnid was moved directly into
1887 llmount, removing the need to use this or specifying the
1888 client NID explicitly when mounting GM clients with zeroconf
1891 Frequency : if client is started with down MDS
1893 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1894 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1895 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1896 ptlrpc_import_delayed_req() because we didn't check if the
1897 request was interrupted, and we also didn't break out of the
1898 event loop if there was no timeout
1903 Description: data loss during non-page-aligned writes to a single file from
1904 both multiple nodes and multiple threads on one node at same time
1905 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1906 inconsistency led to false short-reads, that were cached and later
1907 used by ->prepare_write() to fill in partially written page,
1908 leading to data loss.
1911 Frequency : always, if lconf --abort_recovery used
1913 Description: lconf --abort_recovery fails with 'Operation not supported'
1914 Details : lconf was attempting to abort recovery on the MDT device and not
1917 Severity : enhancement
1919 Description: remove cleanup logs
1920 Details : replace lconf-generated cleanup logs with lustre internal
1921 cleanup routines. Eliminates the need for client-cleanup and
1924 Severity : enhancement
1926 Description: add support for EAs (user and system) on lustre filesystems
1927 Details : it is now possible to store extended attributes in the Lustre
1928 client filesystem, and with the user_xattr mount option it
1929 is possible to allow users to store EAs on their files also
1931 Severity : enhancement
1933 Description: Add possibility (config option) to show minimal available OST free
1935 Details : When compiled with --enable-mindf configure option, statfs(2)
1936 (and so, df) will return least minimal free space available from
1937 all OSTs as amount of free space on FS, instead of summary of
1938 free spaces of all OSTs.
1940 Severity : enhancement
1942 Description: do not expand extent locks acquired on OST-side
1943 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1944 by server: they are not cached anyway.
1947 Frequency : when mmap is used/binaries executed from Lustre
1949 Description: Unmmap pages before throwing them away from read cache.
1950 Details : llap_shrink cache now attempts to unmap pages before discarding
1951 them (if unmapping failed - do not discard). SLES9 kernel has
1952 extra checks that trigger if this unmapping is not done first.
1957 Description: lconf didn't resolve symlinks before checking to see whether a
1958 given mountpoint was already in use
1961 Frequency : when migrating failover services
1962 Bugzilla : 6395, 9514
1963 Description: When migrating a subset of services from a node (e.g. failback
1964 from a failover service node) the remaining services would
1965 time out and evict clients.
1966 Details : lconf --force (implied by --failover) sets the global obd_timeout
1967 to 5 seconds in order to quickly disconnect, but this caused
1968 other RPCs to time out too quickly. Do not change the global
1969 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1971 Severity : enhancement
1972 Frequency : if MDS is started with down OST
1973 Bugzilla : 9439,5706
1974 Description: Allow startup/shutdown of an MDS without depending on the
1975 availability of the OSTs.
1976 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1977 Add appropriate locking and lov-osc refcounts for safe
1978 cleaning. Add osc abort_inflight calls in case the
1979 synchronize never started.
1982 Frequency : occasional (Cray XT3 only)
1984 Description: root not authorized to access files in CRAY_PORTALS environment
1985 Details : The client process capabilities were not honoured on the MDS in
1986 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1987 been accessed by an authorized user then root was able to access
1988 the file on the local client also. The root user capabilities
1989 are now allowed on the MDS, as this environment has secure UID.
1992 Frequency : occasional
1994 Description: ldiskfs "too long searching" message happens too often
1995 Details : A debugging message (otherwise harmless) prints too often on
1996 the OST console. This has been reduced to only happen when
1997 there are fragmentation problems on the filesystem.
2002 Description: Division by zero in statfs when all OSCs are inactive
2003 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2004 lov_statfs divides by value returned by lov_get_stripecnt().
2008 Bugzilla : 9489, 3273
2009 Description: First write from each client to each OST was only 4kB in size,
2010 to initialize client writeback cache, which caused sub-optimal
2011 RPCs and poor layout on disk for the first writen file.
2012 Details : Clients now request an initial cache grant at (re)connect time
2013 and so that they can start streaming writes to the cache right
2014 away and always do full-sized RPCs if there is enough data.
2015 If the OST is rebooted the client also re-establishes its grant
2016 so that client cached writes will be honoured under the grant.
2021 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2022 Details : Now I/O RPCs go to different portal number and (presumably) fast
2023 lock requests (and glimses) and other RPCs get their own service
2024 threads pool that should be able to service those RPCs
2027 Severity : enhancement
2029 Description: Ability to exchange lustre version between client and servers and
2030 issue warnings at client side if client is too old. Also for
2031 liblustre clients there is ability to refuse connection of too old
2033 Details : New 'version' field is added to connect data structure that is
2034 filled with version info. That info is later checked by server and
2038 Frequency : rare, liblustre only.
2039 Bugzilla : 9296, 9581
2040 Description: Two simultaneous writes from liblustre at offset within same page
2041 might proceed at the same time overwriting eachother with stale
2043 Details : I/O lock withing llu_file_prwv was released too early, before data
2044 actually was hitting the wire. Extended lock-holding time until
2045 server acknowledges receiving data.
2048 Frequency : extremely rare. Never observed in practice.
2050 Description: avoid generating lustre_handle cookie of 0.
2051 Details : class_handle_hash() generates handle cookies by incrementing
2052 global counter, and can hit 0 occasionaly (this is unlikely, but
2053 not impossible, because initial value of cookie counter is
2054 selected randonly). Value of 0 is used as a sentinel meaning
2055 "unassigned handle" --- avoid it. Also coalesce two critical
2056 sections in this function into one.
2058 Severity : enhancement
2060 Description: allow liblustre clients to delegate truncate locking to OST
2061 Details : To avoid overhead of locking, liblustre client instructs OST to
2062 take extent lock in ost_punch() on client's behalf. New connection
2063 flag is added to handle backward compatibility.
2065 Severity : enhancement
2066 Bugzilla : 4928, 7341, 9758
2067 Description: allow number of OST service threads to be specified
2068 Details : a module parameter allows the number of OST service threads
2069 to be specified via "options ost ost_num_threads={N}" in the
2070 OSS's /etc/modules.conf or /etc/modprobe.conf.
2074 Bugzilla : 6146, 9635, 9895
2075 Description: servers crash with bad pointer in target_handle_connect()
2076 Details : In rare cases when a client is reconnecting it was possible that
2077 the connection request was the last reference for that export.
2078 We would temporarily drop the export reference and get a new
2079 one, but this may have been the last reference and the export
2080 was just destroyed. Get new reference before dropping old one.
2082 Severity : enhancement
2083 Frequency : if client is started with failover MDS
2085 Description: Allow multiple MDS hostnames in the mount command
2086 Details : Try to read the configuration from all specified MDS
2087 hostnames during a client mount in case the "primary"
2090 Severity : enhancement
2092 Description: Stop sending data to evicted clients as soon as possible.
2093 Details : Check if the client we are about to send or are sending data to
2094 was evicted already. (Check is done every second of waiting,
2095 for which l_wait_event interface was extended to allow checking
2096 of exit condition at specified intervals).
2099 Frequency : rare, normally only when NFS exporting is done from client
2101 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2103 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2106 Frequency : always, if srand() is called before liblustre initialization
2108 Description: Liblustre uses system PRNG disturbing its usage by user application
2109 Details : Introduce internal to lustre fast and high-quality PRNG for
2110 lustre usage and make liblustre and some other places in generic
2111 lustre code to use it.
2113 Severity : enhancement
2114 Bugzilla : 9477, 9557, 9870
2115 Description: Verify that the MDS configuration logs are updated when xml is
2116 Details : Check if the .xml configuration logs are newer than the config
2117 logs stored on the MDS and report an error if this is the case.
2118 Request --write-conf, or allow starting with --old_conf.
2120 Severity : enhancement
2122 Description: Handle symlinks in the path when checking if Lustre is mounted.
2123 Details : Resolve intermediate symlinks when checking if a client has
2124 mounted a filesystem to avoid duplicate client mounts.
2129 Description: lconf can hit an error exception but still return success.
2130 Details : The lconf command catches the Command error exception at the top
2131 level script context and will exit with the associated exit
2132 status, but doesn't ensure that this exit status is non-zero.
2137 Description: failure of ptlrpc thread startup can cause oops
2138 Details : Starting a ptlrpc service thread can fail if there are a large
2139 number of threads or the server memory is very fragmented.
2140 Handle this without oopsing.
2143 Frequency : always, only if liblustre and non-default acceptor port was used
2145 Description: liblustre cannot connect to servers with non-default acceptor port
2146 Details : tcpnal_set_default_params() was not called and was therefore
2147 ignoring the environment varaible TCPNAL_PORT, as well as other
2148 TCPNAL_ environment variables
2153 Description: two objects could be created on the same OST for a single file
2154 Details : If an OST is down, in some cases it was possible to create two
2155 objects on a single OST for a single file. No problems other
2156 than potential performance impact and spurious error messages.
2160 Bugzilla : 5681, 9562
2161 Description: Client may oops in ll_unhash_aliases
2162 Details : Client dcache may become inconsistent in race condition.
2163 In some cases "getcwd" can fail if the current directory is
2169 Description: Inode refcounting problems in NFS export code
2170 Details : link_raw functions used to call d_instantiate without obtaining
2171 extra inode reference first.
2175 Bugzilla : 9942, 9903
2176 Description: Referencing freed requests leading to crash, memleaks with NFS.
2177 Details : We used to require that call to ll_revalidate_it was always
2178 followed by ll_lookup_it. Also with revalidate_special() it is
2179 possible to call ll_revalidate_it() twice for the same dentry
2180 even if first occurence returned success. This fix changes semantic
2181 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2182 reference on a request referred from the intent.
2183 ll_intent_release() then releases such a request.
2186 Frequency : rare, normally benchmark loads only
2188 Description: unlinked inodes were kept in memory on the client
2189 Details : If a client is repeatedly creating and unlinking files it
2190 can accumulate a lot of stale inodes in the inode slab cache.
2191 If there is no other client load running this can cause the
2192 client node to run out of memory. Instead flush old inodes
2193 from client cache that have the same inode number as a new inode.
2196 Frequency : SLES9 2.6.5 kernel and long filenames only
2197 Bugzilla : 9969, 10379
2198 Description: utime reports stale NFS file handle
2199 Details : SLES9 uses out-of-dentry names in some cases, which confused
2200 the lustre dentry revalidation. Change it to always use the
2204 Frequency : rare, unless heavy write-truncate concurrency is continuous
2205 Bugzilla : 4180, 6984, 7171, 9963, 9331
2206 Description: OST becomes very slow and/or deadlocked during object unlink
2207 Details : filter_destroy() was holding onto the parent directory lock
2208 while truncating+unlinking objects. For very large objects this
2209 may block other threads for a long time and slow overall OST
2210 responsiveness. It may also be possible to get a lock ordering
2211 deadlock in this case, or run out of journal credits because of
2212 the combined truncate+unlink. Solution is to do object truncate
2213 first in one transaction without parent lock, and then do the
2214 final unlink in a new transaction with the parent lock. This
2215 reduces the lock hold time dramatically.
2218 Frequency : rare, 2.4 kernels only
2220 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2221 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2222 before unmounting their filesystem in order to avoid deadlock.
2223 The kernel_locked() function in 2.4 kernels only checks whether
2224 the kernel lock is held, not whether it is this process that is
2225 holding it as 2.6 kernels do.
2230 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2231 Details : The client ptlrpc code may be trying to reconnect to a down
2232 server before a previous connection attempt has timed out.
2233 Increase the reconnect interval to be longer than the connection
2234 timeout interval to avoid sending duplicate connections to
2238 Frequency : echo_client brw_test command
2240 Description: fix echo_client to work with OST preallocated code
2241 Details : OST preallocation code (5137) didn't take echo_client IO path
2242 into account: echo_client calls filter methods outside of any
2243 OST thread and, hence, there is no per-thread preallocated
2244 pages and buffers to use. Solution: hijack pga pages for IO. As
2245 a byproduct, this avoids unnecessary data copying.
2249 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2250 Description: Client can oops in mdc_commit_close() after open replay
2251 Details : It was possible for the MDS to return an open request with no
2252 transaction number in mds_finish_transno() if the client was
2253 evicted, but without actually returning an error. Clients
2254 would later try to replay that open and may trip an assertion
2255 Simplify the client close codepath, and always return an error
2256 from the MDS in case the open is not successful.
2259 Frequency : rare, 2.6 OSTs only
2261 Description: OST may deadlock under high load on fragmented files
2262 Details : If there was a heavy load and highly-fragmented OST filesystems
2263 it was possible to have all the OST threads deadlock waiting on
2264 allocation of biovecs, because the biovecs were not released
2265 until the entire RPC IO was completed. Instead, release biovecs
2266 as soon as they are complete to ensure forward IO progress.
2268 Severity : enhancement
2270 Description: Support for specifying external journal device at mount
2271 Details : If an OST or MDS device is formatted with an external journal
2272 device, this device major/minor is stored in the ext3 superblock
2273 and may not be valid for failover. Allow detecting and
2274 specifying the external journal at mount time.
2279 Description: Mounting an MDS with pending unlinked files may cause oops
2280 Details : target_finish_recovery() calls mds_postrecov() which returned
2281 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2282 considers this an error and immediately begins cleaning up the
2283 lov, just after starting the mds_lov process
2285 Severity : enhancement
2287 Description: Implement 'lfs df' to report actual free space on per-OST basis
2288 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2289 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2290 usage of objects; '-h' to report in human readable format.
2292 ------------------------------------------------------------------------------
2294 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2301 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2302 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2303 at startup during orphan unlinking. Instead, skip buddy update
2304 before setup, as it will be regenerated after recovery is complete.
2309 Description: If an OST is inactive, its locks might reference stale inodes.
2310 Details : lov_change_cbdata() must iterate over all namespaces, even if
2311 they are inactive to clear inode references from the lock.
2313 Severity : enhancement
2314 Frequency : occasional, if non-standard max_dirty_mb used
2316 Description: Client will block write RPCs if not enough grant
2317 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2318 then the client will block writes while waiting for another RPC
2319 to complete instead of consuming its dirty limit. With change
2320 we get improved performance when max_dirty_mb is small.
2322 Severity : enhancement
2323 Bugzilla : 3389, 6253
2324 Description: Add support for supplementary groups on the MDS.
2325 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2326 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2327 lookups for user supplementary groups into a cache.
2331 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2332 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2333 for liblustre O_EXCL is enforced by MDS.
2337 Description: readdir never returns NULL in liblustre.
2338 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2339 dentry in struct dirent.
2343 Frequency : liblustre only, depends on application IO pattern
2344 Description: liblustre clients evicted if not contacting servers
2345 Details : Don't put liblustre clients into the ping_evictor list, so
2346 they will not be evicted by the pinger ever.
2348 Severity : enhancement
2350 Description: Add ability to evict clients by NID from MDS.
2351 Details : By echoing "nid:$NID" string into
2352 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2353 $NID would be instantly evicted from this MDS and from all active
2354 OSTs connected to it.
2358 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2359 Details : lookup_it_finish() used to query file size from OSTs that was not
2364 Description: service threads change working directory to that of init
2365 Details : Starting lustre service threads may pin the working directory
2366 of the parent thread, making that filesystem busy. Threads
2367 now change to the working directory of init to avoid this.
2371 Frequency : during shutdown only
2372 Description: shutdown with a failed MDS or OST can cause unmount to hang
2373 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2378 Frequency : 2.6 only
2379 Description: chmod/chown may include an extra supplementary group
2380 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2381 supplementary group and if none is specified this is used.
2384 Bugzilla : 5479 (6816)
2386 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2387 Details : If lookup is in progress on a file that is unlinked we might try
2388 to revalidate the inode and fail in revalidate after lookup is
2389 complete and ll_file_open() enqueues the open again but
2390 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2393 Frequency : always, if lconf --abort_recovery used
2395 Description: lconf --abort_recovery fails with 'Operation not supported'
2396 Details : lconf was attempting to abort recovery on the MDT device and not
2399 ------------------------------------------------------------------------------
2401 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2406 Frequency : rare (only unsupported configurations with a node running as an
2408 Bugzilla : 6514, 5137
2409 Description: Mounting a Lustre file system on a node running as an OST could
2411 Details : OSTs now preallocates memory needed to write out data at
2412 startup, instead of when needed, to avoid having to
2413 allocate memory in possibly low memory situations.
2414 Specifically, if the file system is mounted on on OST,
2415 memory pressure could force it to try to write out data,
2416 which it needed to allocate memory to do. Due to the low
2417 memory, it would be unable to do so and the node would
2418 become unresponsive.
2420 Severity : enhancement
2422 Description: Addition of lconf --service command line option
2423 Details : lconf now accepts a '--service <arg>' option, which is
2424 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2426 Severity : enhancement
2428 Description: Failover mode is now the default for OSTs.
2429 Details : By default, OSTs will now run in failover mode. To return to
2430 the old behaviour, add '--failout' to the lmc line for OSTs.
2432 Severity : enhancement
2434 Description: Health checks are now provided for MDS and OSTs
2435 Details : Additional detailed health check information on MSD and OSTs
2436 is now provided through the procfs health_check value.
2439 Frequency : occasional, depends on IO load
2441 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2442 after numerous create/delete cycles
2443 Details : The ext3 inode allocation policy would not allocate new inodes
2444 very well on the OSTs because there are no new directories
2445 being created. Instead we look for groups with free space if
2446 the parent directories are nearly full.
2451 Description: Network or server problems during mount may cause partially
2452 mounted clients instead of returning an error.
2453 Details : The config llog parsing code may overwrite the error return
2454 code during mount error handling, returning success instead
2460 Description: MDS can fail to allocate large reply buffers
2461 Details : After long uptimes the MDS can fail to allocate large reply
2462 buffers (e.g. zconf client mount config records) due to memory
2463 fragmentation or consumption by the buffer cache. Preallocate
2464 some large reply buffers so that these replies can be sent even
2465 under memory pressure.
2469 Frequency : rare (liblustre)
2470 Description: fsx running with liblustre complained that using truncate() to
2471 extend the file doesn't work. This patch corrects that issue.
2472 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2473 ATTR_SIZE and lsm use in llu_setattr_raw.
2477 Frequency : rare, only 2.6 kernels
2478 Description: Unusual file access patterns on the MDS may result in inode
2479 data being lost in very rare circumstances.
2480 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2481 code in the 2.6 kernel caused the inode and/or EA data not to
2482 be read from disk, causing single-file corruption.
2486 Frequency : rare, only 2.6 filesystems using extents
2487 Description: Heavy concurrent write and delete load may cause data corruption.
2488 Details : It was possible under high-load situations to have an extent
2489 metadata block in the block device cache from a just-unlinked
2490 file overwrite a newly-allocated data block. We now unmap any
2491 metadata buffers that alias just-allocated data blocks.
2495 Frequency : filesystems with default stripe_count larger than 77
2496 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2497 Details : lconf specifies an inode size of 4096 bytes when the default
2498 stripe_count is larger than 77. This conflicts with the default
2499 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2500 this case to avoid pinning too much memory for large EAs.
2502 ------------------------------------------------------------------------------
2504 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2509 Frequency : rare (extremely heavy IO load with hundreds of clients)
2511 Description: Client is evicted, gets IO error writing to file
2512 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2513 caused clients to be evicted for AST timeouts. The fixes in
2514 bug 5192 mean we no longer need to have such short AST timeouts
2515 so ldlm_timeout has been increased.
2518 Frequency : occasional during --force or --failover shutdown under load
2519 Bugzilla : 5949, 4834
2520 Description: Server oops/LBUG if stopped with --force or --failover under load
2521 Details : a collection of import/export refcount and cleanup ordering
2522 issues fixed for safer force cleanup
2525 Frequency : only filesystems larger than 120 OSTs
2526 Bugzilla : 5990, 6223
2527 Description: lfs getstripe would oops on a very large filesystem
2528 Details : lov_getconfig used kfree on vmalloc'd memory
2531 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2532 Bugzilla : 6242, 6243
2533 Description: reading from files that had been truncated to a non-zero size
2534 but never opened returned no data
2535 Details : ll_file_read() reads zeros from no-object files to EOF
2540 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2541 Details : There's little harm in aborting MDS/OSS recovery and letting it
2542 try again, so I removed the LASSERT and return an error instead.
2544 Severity : enhancement
2546 Description: New debugging infrastructure for tracking down data corruption
2547 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2548 (b) cover more of the client-side code path, and (c) try to narrow
2549 down where problems occurred
2553 Bugzilla : 3819, 4364, 4397, 6313
2554 Description: Racing close and eviction MDS could cause assertion in mds_close
2555 Details : It was possible to get multiple mfd references during close and
2556 client eviction, leading to one thread referencing a freed mfd.
2558 Severity: : enhancement
2559 Bugzilla : 3262, 6359
2560 Description: Attempts to reconnect to servers are now more aggressive.
2561 Details : This builds on the enhanced upcall-less recovery that was added
2562 in 1.4.2. When trying to reconnect to servers, clients will
2563 now try each server in the failover group every 10 seconds. By
2564 default, clients would previously try one server every 25 seconds.
2569 Description: After recovery, certain operations trigger a failed
2570 assertion on a client.
2571 Details : Failing over an mds, using lconf -d --failover, while a
2572 client was doing a readdir() call would cause the client to
2573 LBUG after recovery completed and the readdir() was resent.
2575 Severity : enhancement
2577 Description: Default groups are now added by lconf
2578 Details : You can now run lconf --group <servicename> without having to
2579 manually add groups with lmc.
2582 Frequency : occasional
2584 Description: Nodes with an elan id of 0 trigger a failed assertion
2587 Frequency : always when accessing e.g. tty/console device nodes
2589 Description: tty and some other devices nodes cannot be used on lustre
2590 Details : file's private_data field is used by device data and lustre
2591 values in there got lost. New field was added to struct file to
2592 store fs-specific private data.
2595 Frequency : when exporting Lustre via NFS
2597 Description: NFSD failed occasionally when looking up a path component
2598 Details : NFSD is looking up ".." which was broken in ext3 directories
2599 that had grown large enough to become hashed.
2602 Frequency : Clusters with multiple interfaces not on the same subnet
2604 Description: Nodes will repeatedly try to reconnect to an interface which it
2605 cannot reach and report an error to the log.
2606 Details : Extra peer list entries will be created by lconf with some peers
2607 unreachable. lconf now validates the peer before adding it.
2610 Frequency : Only if a default stripe is set on the filesystem root.
2612 Description: Setting a default stripe on the filesystem root prevented the
2613 filesystem from being remounted.
2614 Details : The client was sending extra request flags in the root getattr
2615 request and did not allocate a reply buffer for the dir EA.
2618 Frequency : occasional, higher if lots of files are accessed by one client
2619 Bugzilla : 6159, 6097
2620 Description: Client trips assertion regarding lsm mismatch/magic
2621 Details : While revalidating inodes the VFS looks up inodes with ifind()
2622 and in rare cases can find an inode that is being freed.
2623 The ll_test_inode() code will free the lsm during ifind()
2624 when it finds an existing inode and then the VFS later attaches
2625 this free lsm to a new inode.
2629 Bugzilla : 6422, 7030
2630 Description: MDS deadlock between mkdir and client eviction
2631 Details : Creating a new file via mkdir or mknod (starting a transaction
2632 and getting the ns lock) can deadlock with client eviction
2633 (gets ns lock and trying to finish a synchronous transaction).
2636 Frequency : occasional
2637 Description: While starting a server, the fsfilt_ext3 module could not be
2639 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2640 kernels. Previously, lconf would still use the ext3 name
2641 when trying to load modules. Now, it will correctly use
2642 ext3 on 2.4 and ldiskfs on 2.6.
2644 Severity : enhancement
2645 Description: The default stripe count has been changed to 1
2646 Details : The interpretation of the default stripe count (0, to lfs
2647 or lmc) has been changed to mean striping across a single
2648 OST, rather than all available. For general usage we have
2649 found a stripe count of 1 or 2 works best.
2651 Severity : enhancement
2652 Description: Add support for compiling against Cray portals.
2653 Details : Conditional compiling for some areas that are different
2657 Frequency : occasional
2658 Bugzilla : 6409, 6834
2659 Description: Creating files with an explicit stripe count may lead to
2660 a failed assertion on the MDS
2661 Details : If some OSTs are full or unavailable, creating files may
2662 trigger a failed assertion on the MDS. Now, Lustre will
2663 try to use other servers or return an error to the
2667 Frequency : occasional
2669 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2670 caused lock timeout during readahead (since 1.4.2).
2671 Details : Processes doing readahead might match a lock that hasn't been
2672 granted yet if there are overlapping and conflicting lock
2673 requests. The readahead process waits on ungranted lock
2674 (original lock is CBPENDING), while OST waits for that process
2675 to cancel CBPENDING read lock and eventually evicts client.
2677 Severity : enhancement
2679 Description: Initial enabling of flock support for clients
2680 Details : Implements fcntl advisory locking and file status functions.
2681 This feature is provided as an optional mount flag (default
2682 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2683 locking are implemented yet, and those that are are not guaranteed
2684 to be completely correct in production environments.
2685 mount -t lustre -o [flock|noflock] ...
2688 Frequency : occasional
2690 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2691 assertion in the ext3 JBD (journaling) layer.
2692 Details : The b_committed_data struct is protected by the big kernel lock
2693 in 2.4 kernels, serializing journal_commit_transaction() and
2694 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2695 access to this struct. In 2.6 kernels there is finer grained
2696 locking to improve SMP performance of the JBD layer.
2700 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2702 -----------------------------------------------------------------------------
2704 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2706 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2707 versions of Lustre. Please update all servers and clients to
2708 version 1.4.2 or later at the same time. You must also run
2709 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2710 to update the configuration logs.
2712 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2713 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2714 - swab llog records of type '0' so we get proper header size/idx (5861)
2715 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2716 - fix rename of one directory over another leaking an inode (5953)
2717 - avoid SetPageDirty on 2.6 (5981)
2718 - don't re-add just-being-destroyed locks to the waiting list (5653)
2719 - when creating new directories, inherit the parent's custom
2720 striping settings if present parent (3048)
2721 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2722 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2723 - don't hold client locks on temporary worklist from l_lru (5666)
2724 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2725 - automatically evict dead clients (5921)
2726 - Update file size properly in create+truncate+fstat case (6196)
2727 - Do not unhash mountpoint dentries, do not allow removal of
2729 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2730 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2731 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2732 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2733 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2734 - Correctly handle reads of files with no objects (6243)
2735 - lctl recover will also mark a device active if deactivate used (5933)
2737 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2738 - allow --write-conf on an MDS with different nettype than client (5619)
2739 - don't write config llogs to MDS for mounts not from that MDS (5617)
2740 - lconf should create multiple TCP connections from a client (5201)
2741 - init scripts are now turned off by default; run chkconfig --on
2742 lustre and chkconfig --on lustrefs to use them
2743 - upcalls are no longer needed for clients to recover to failover
2745 - add --abort-recovery option to lconf to abort recovery on device
2747 - add support for an arbitrary number of OSTs (3026)
2748 - Quota support protocol changes.
2749 - forward compatibility changes to wire structs (6007)
2750 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2751 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2752 - improved "lustre status" script
2753 - initialize blocksize for non-regular files (6062)
2754 - added --disable-server and --disable-client configure options (5782)
2755 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2756 - Vanilla 2.4.29 support
2757 - increase maximum number of obd devices to 520 (6242)
2758 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2759 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2760 SLES 9 2.6.5 kernels
2761 - limit stripes per file to 160 (the maximum EA size) (6093)
2763 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2766 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2767 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2768 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2769 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2770 - small ext3 extents cleanups and fixes (5733)
2771 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2772 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2773 - avoid needless client->OST connect, fix handle mismatch (5317)
2774 - fix DLM error path that led to out-of-sync client, long delays (5779)
2775 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2776 - fix several locking issues related to i_size (5492,5624,5654,5672)
2777 - don't move pending lock onto export if it is already evicted (5683)
2778 - fix kernel oops when creating .foo in unlinked directory (5548)
2779 - fix deadlock in obdfilter statistics vs. object create (5811)
2780 - use time_{before,after} to avoid timer jiffies wrap (5882)
2781 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2782 - Do not leak request if server was not able to process it (5154)
2783 - If mds_open unable to find parent dir, make that negative lookup(5154)
2784 - don't create new directories with extent-mapping (5909, 5936)
2786 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2787 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2788 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2789 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2790 - support for RHEL4 kernel on i686 (5773)
2791 - provide error messages when incompatible logs are encountered (5898)
2793 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2794 * version 1.4.0.10 (1.4.1 release candidate 1)
2796 - don't keep a lock reference when lock is not granted (4238)
2797 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2798 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2799 - drop import inflight refcount on signal_completed_replay error (5255)
2800 - unlock page after async write error during send (3677)
2801 - handle missing objects in filter_preprw_read properly (5265)
2802 - no transno return for symlink open, don't save no-trasno open (3440)
2803 - don't try to complete elan receive that already failed (4012)
2804 - free RPC server reply state on error (5406)
2805 - clean up thread from ptlrpc_start_thread() on error (5160)
2806 - readahead could read extra page into cache that wasn't ejected (5388)
2807 - prevent races in class_attach/setup/cleanup/detach (5260)
2808 - don't dereference de->d_inode after l_dput of de (5458)
2809 - use "int" for stripe value returned from lock_to_stripe (5544)
2810 - mballoc allocation and error-checking fixes in 2.6 (5504)
2811 - block device patches to fix I/O request sizes in 2.6 (5482)
2812 - look up hostnames for IB nals (5602)
2813 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2814 - don't start multiple acceptors for the same port (5277)
2815 - fix incorrect LASSERT in mds_getattr_name (5635)
2816 - export a proc file for general "ping" checking (5628)
2817 - fix "lfs check" to not block when the MDS is down (5628)
2819 - service request history (4965)
2820 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2821 - create an "evict_client" /proc entry on OSTs, like the MDS has
2822 - fix mount usage message, return errors per mount(8) (5168)
2823 - change grep [] to grep "[]" in tests so they work in more UMLs
2824 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2825 - remove ancient LOV_MAGIC_V0 EA support (5047)
2826 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2827 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2828 - allow readahead allocations to fail when low on memory (5383)
2829 - mmap locking landed again, after considerable improvement (2828)
2830 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2832 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2835 - send OST transaction number in read/write reply to free req (4966)
2836 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2837 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2838 - the watchdog thread now runs as interruptible (5246)
2839 - flock/lockf fixes (but it's still disabled, pending 5135)
2840 - don't use EXT3 constants in llite code (5094)
2841 - memory shortage at startup could cause assertion (5176)
2843 - reorganization of lov code
2844 - single portals codebase
2846 - add extents/mballoc support (5025)
2847 - direct I/O reads in the obdfilter (4048)
2848 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2850 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2853 - send OST transaction number in read/write reply to free req (4966)
2854 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2855 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2856 - the watchdog thread now runs as interruptible (5246)
2857 - handle missing objects in filter_preprw_read properly (5265)
2858 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2859 - drop import inflight refcount on signal_completed_replay error (5255)
2860 - unlock page after async write error during send (3677)
2861 - return original error code on reconstructed replies (3761)
2862 - no transno return for symlink open, don't save no-trasno open (3440)
2864 - add pid to ldlm debugging output (4922)
2865 - bump the watchdog timeouts -- we can't handle 30sec yet
2866 - extra debugging for orphan dentry/inode bug (5259)
2868 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2871 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2872 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2873 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2874 - fix NULL dereference in /proc/sys/portals/routes (4827)
2875 - allow failed mdc_close() operations to be interrupted (4561)
2876 - stop precreate on OST before MDS would time out on it (4778)
2877 - don't send partial-page writes before EOF from client (4410)
2878 - discard client grant for sub-page writes on large-page clients (4520)
2879 - don't free dentries not owned by NFS code, check generation (4806)
2880 - fix lsm leak if mds_create_objects() fails (4801)
2881 - limit debug_daemon file size, always print CERROR messages (4789)
2882 - use transno after validating reply (3892)
2883 - process timed out requests if import state changes (3754)
2884 - update mtime on OST during writes, return in glimpse (4829)
2885 - add mkfsoptions to LDAP (4679)
2886 - use ->max_readahead method instead of zapping global ra (5039)
2887 - don't interrupt __l_wait_event() during strace
2889 - add software watchdogs to catch hung threads quickly (4941)
2890 - make lustrefs init script start after nfs is mounted
2891 - fix CWARN/ERROR duplication (4930)
2892 - return async write errors to application if possible (2248)
2893 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2894 - print NAL number in %x format (4645)
2895 - update barely-supported suse-2.4.21-171 series (4842)
2896 - support for sles 9 %post scripts
2897 - support for building 2.6 kernel-source packages
2898 - support for sles km_* packages
2900 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2903 - ignore -ENOENT errors in osc_destroy (3639)
2904 - notify osc create thread that OSC is being cleaned up (4600)
2905 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2906 - reconstruct ost_handle() like mds_handle() (4657)
2907 - create a new thread to do import eviction to avoid deadlock (3969)
2908 - let lconf resolve symlinked-to devices (4629)
2909 - don't unlink "objects" from directory with default EA (4554)
2910 - hold socknal file ref over connect in case target is down (4394)
2911 - allow more than 32000 subdirectories in a single directory (3244)
2912 - fix blocks count for O_DIRECT writes (3751)
2913 - OST returns ENOSPC from object create when no space left (4539)
2914 - don't send truncate RPC if file size isn't changing (4410)
2915 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2916 - bind to privileged port in socknal and tcpnal (3689)
2918 - rate limit CERROR/CWARN console message to avoid overload (4519)
2919 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2920 - basic mmap support (3918)
2921 - kernel patch series update from b1_4 (4711)
2923 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2926 - avoid crash during MDS cleanup with OST shut down (2775)
2927 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2928 - don't use bad inodes on the MDS (3744)
2929 - dynamic object preallocation to improve recovery speed (4236)
2930 - don't hold spinlock over lock dumping or change debug flags (4401)
2931 - don't zero obd_dev when it is force cleaned (3651)
2932 - print grants to console if they go negative (4431)
2933 - "lctl deactivate" will stop automatic recovery attempts (3406)
2934 - look for existing locks in ldlm_handle_enqueue() (3764)
2935 - don't resolve lock handle twice in recovery avoiding race (4401)
2936 - revalidate should check working dir is a directory (4134)
2938 - don't always mark "slow" obdfilter messages as errors (4418)
2940 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2943 - don't close LustreDB during write_conf until it is done (3860)
2944 - fix typo in lconf for_each_profile (3821)
2945 - allow dumping logs from multiple threads at one time (3820)
2946 - don't allow multiple threads in OSC recovery (3812)
2947 - fix debug_size parameters (3864)
2948 - fix mds_postrecov to initialize import for llog ctxt (3121)
2949 - replace config semaphore with spinlock (3306)
2950 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2951 - don't allow enqueue to complete on a destroyed export (3822)
2952 - down write_lock before checking llog header bitmap (3825)
2953 - recover from lock replay timeout (3764)
2954 - up llog sem before sending rpc (3652)
2955 - reduce ns lock hold times when setting kms (3267)
2956 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2957 - fix NULL deref and obd_dev leak on setup error (3312)
2958 - replace some LBUG about llog ops with error handling (3841)
2959 - don't match INVALID dentries from d_lookup and spin (3784)
2960 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2961 - fix invalid assertion in ptlrpc_set_wait (3880)
2963 - add libwrap support for the TCP acceptor (3996)
2964 - add /proc/sys/portals/routes for non-root route listing (3994)
2965 - allow setting MDS UUID in .xml (2580)
2966 - print the stack of a process that LBUGs (4228)
2968 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2971 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2972 - make sure to unset replay flag from failed open requests (3440)
2973 - if default stripe count is 0, use OST count for inode size (3636)
2974 - update parent mtime/ctime on client for create/unlink (2611)
2975 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2976 - free recovery state on server during a forced cleanup (3571)
2977 - unregister_reply for resent reqs (3063)
2978 - loop back devices mounting and status check on 2.6 (3563)
2979 - fix resource-creation race that can provoke i_size == 0 (3513)
2980 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2981 - more debugging for page-accounting assertion (3746)
2982 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2983 - avoid deadlock after precreation failure (3758)
2984 - fix race and lock order deadlock in orphan handling (3450, 3750)
2985 - add validity checks when grabbing inodes from l_ast_data (3599)
2987 - add /proc/.../recovery_status to obdfilter (3428)
2988 - lightweight CDEBUG infrastructure, debug daemon (3668)
2989 - change default OSC RPC parameters to be better on small clusters
2990 - turn off OST read cache for files smaller than 32MB
2991 - install man pages and include them in rpms (3100)
2992 - add new init script for (un)mounting lustre filesystems (2593)
2993 - run chkconfig in %post for init scripts (3701)
2994 - drop scimac NAL (unmaintained)
2996 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
2999 - clean kiobufs before and after use (3485)
3000 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3001 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3002 - change init script to fail more clearly if not run as root (1528)
3003 - allow clients to reconnect during replay (1742)
3004 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3005 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3006 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3007 - fix cond_resched (3554)
3008 - only evict unfinished clients after recovery (3515)
3009 - allow bulk resend, prevent data loss (3570)
3010 - dynamic ptlrpc request buffer allocation (2102)
3011 - don't allow unlinking open directory if it isn't empty (2904)
3012 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3013 - remove extraneous obd dereference causing LASSERT failure (3334)
3014 - don't use get_cycles() when creating temp. files on the mds (3156)
3015 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3016 - handle EEXIST for set-stripe, set proper directory name (3336)
3018 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3019 - fix ksocknal_fmb_callback() error messages (2918)
3021 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3024 - don't copy lvb into (possibly NULL) reply on error (2983)
3025 - don't deref dentry after dput, don't free lvb on error (2922)
3026 - use the kms to determine writeback rpc length (2947)
3027 - increment oti_logcookies when osc is inactive (2948)
3028 - update client's i_blocks count via lvb messages (2543)
3029 - handle intent open/close of special files properly (1557)
3030 - mount MDS with errors=remount-ro, like obdfilter (2009)
3031 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3032 - don't use cancelling-locks' kms values (2947)
3033 - use highest lock extent for kms, not last one (2925)
3034 - don't dereference ERR_PTR() dentry in error handling path (3107)
3035 - fix thread race in portals_debug_dumplog() (3122)
3036 - create lprocfs device entries at setup instead of at attach (1519)
3037 - common AST error handler, don't evict client on completion race (3145)
3038 - zero nameidata in detach_mnt in 2.6 (3118)
3039 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3040 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3041 - initialize RPC timeout timer earlier for 2.6 (3219)
3042 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3043 - print nal/nid for unknown nid (3258)
3044 - additional checks for oscc recovery before doing precreate (3284)
3045 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3046 - don't crash in mdc_close for bad permissions on open (3285)
3047 - zero i_rdev for non-device files (3147)
3048 - clear page->private before handing to FS, better assertion (3119)
3049 - tune the read pipeline (3236)
3050 - fix incorrect decref of invalidated dentry (2350)
3051 - provide read-ahead stats and refine rpc in flight stats (3328)
3052 - don't hold journal transaction open across create RPC (3313)
3053 - update atime on MDS at close time (3265)
3054 - close LDAP connection when recovering to avoid server load (3315)
3055 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3056 - don't leak open file on MDS after open resend (3325)
3057 - serialize filter_precreate and filter_destroy_precreated (3329)
3058 - loop device shouldn't call sync_dev() for nul device (3092)
3059 - clear page cache after eviction (2766)
3060 - resynchronize MDS->OST in background (2824)
3061 - refuse to mount the same filesystem twice on same mountpoint (3394)
3062 - allow llmount to create routes for mounting behind routers (3320)
3063 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3064 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3065 - fix rare problem with rename on htree directories (3417)
3067 - allow default OST striping configuration per directory (1414)
3068 - fix compilation for qswnal for 2.6 kernels (3125)
3069 - increase maximum number of MDS request buffers for large systems
3070 - change liblustreapi to be useful for external progs like lfsck (3098)
3071 - increase local configuration timeout for slow disks (3353)
3072 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3074 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3077 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3078 - don't overwrite extent policy data in reply if lock was blocked (2901)
3079 - drop filter export grants atomically with removal from device (2663)
3080 - del obd_self_export from work_list in class_disconnect_exports (2908)
3081 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3082 - swab reply message in mdc_close, other PPC fixes (2464)
3083 - fix destroying of named logs (2325)
3084 - overwrite old logs when running lconf --write_conf (2264)
3085 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3086 - fix race in target_handle_connect (2898)
3087 - mds_reint_create() should take same inode create lock (2926)
3088 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3089 - don't close files for self_export to avoid uninitialized obd (2936)
3090 - allow MDS with the same name as client node (2939)
3091 - hold dentry reference for closed log files for unlink (2325)
3092 - reserve space for all logs during transactions (2059)
3093 - don't evict page beyond end of stripe extent (2925)
3094 - don't oops on a deleted current working directory (2399)
3095 - handle hard links to targets without a parent properly (2517)
3096 - don't dereference NULL lock when racing during eviction (2867)
3097 - don't grow lock extents when lots of conflicting locks (2919)
3099 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3102 - account for cache space usage on clients to avoid data loss (974)
3103 - lfsck support in lustre kernel code (2349)
3104 - reduce journal credits needed for BRW writes (2370)
3105 - orphan handling to avoid losing space on client/server crashes
3106 - ptlrpcd can be blocked, stopping ALL progress (2477)
3107 - use lock value blocks to assist in proper KMS, faster stat (1021)
3108 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3109 - recovery for initial connections (2355)
3110 - fixes for mds_cleanup_orphans (1934)
3111 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3112 - block all file creations until orphan recovery completes (1901)
3113 - client remove rq_connection from request struct (2423)
3114 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3115 - recovery timer race (2670)
3116 - mdc_close recovey bug (2532)
3117 - ptlrpc cleanup bug (2710)
3118 - mds timeout on local locks (2588)
3119 - namespace lock held during RPCs (2431)
3120 - handle interrupted sync write properly (2503)
3121 - don't try to handle a message that hasn't been replied to (2699)
3122 - client assert failure during cleanup after abort recovery (2701)
3123 - leak mdc device after failed mount (2712)
3124 - ptlrpc_check_set allows timedout requests to complete (2714)
3125 - wait for inflight reqs when ptlrpcd finishes (2710)
3126 - make sure unregistered services are removed from the srv_list
3127 - reset bulk XID's when resending them (caught by 1138 test)
3128 - unregister_bulk after timeout
3129 - fix lconf error (2694)
3130 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3131 - readahead locks pages, leaves pending causing memory pressure (2673)
3132 - increase OST request buffers to 4096 on large machines (2729)
3133 - fix up permission of existing directories in simple_mkdir (2661)
3134 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3135 - don't assign transno to errored transactions (2742)
3136 - don't delete objects on OST if given a bogus objid from MDS (2751)
3137 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3138 - if rq_no_resend, then timeout request after recovery (2432)
3139 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3140 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3141 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3142 - MDS recovery shouldn't delete objects that it has given out (2730)
3143 - if enqueue arrives after completion, don't clobber LVB (2819)
3144 - don't unlock pages twice when trigger_group_io returns error (2814)
3145 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3146 - don't write pages to disk if there was an error (1450)
3147 - don't ping imports that have recovery disabled (2676)
3148 - take buffered bytes into account when balancing socknal conn (2817)
3149 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3150 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3151 - remove little-endian swabbing of llog records (1987)
3152 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3153 - retry reposting request buffers if they fail (1191)
3154 - grow extent at grant time to avoid granting a revoked lock (2809)
3155 - lock revoke doesn't evict page if covered by a second lock (2765)
3156 - disable VM readahead to avoid reading outside lock extents (2805)
3158 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3159 - updated kernel patches for hp-2.4.20 kernel (2681)
3161 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3164 - fix truncated write corruption (2366)
3165 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3167 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3168 - ensure len(uuid) < 37 in lmc (1171)
3169 - fix ia64 OOPS in llog_test (2255)
3170 - zero end of page at obdfilter for partial page writes (2648)
3171 - don't leave stale dentries around after renames (bug 2428)
3172 - fix timeouts when evicting a client with a single lock held (2642)
3173 - set deadline for the initial HELLO message to drain (2634)
3174 - print out dotted-quad IP addresses in the socknal (2302)
3176 - additional debugging for MDS client eviction problem (2443)
3177 - fix mkfsoptions support for osts (2603, 2604)
3179 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3182 - add series for the vanilla 2.6.0 kernel
3183 - add series for the vanilla 2.4.24 kernel
3184 - add series for a cray x86/64 UL kernel drop
3185 - fix xattr patches for the vanilla 2.4.19 series
3187 - generate true UUIDs in lmc (1171)
3188 - have portals stack dumping break in UML (2466)
3189 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3190 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3191 - wake callback waiting threads on client eviction (2460)
3192 - Add --ptldebug and --subsystem to lmc (1719)
3193 - update assertion to allow safe interrupt allocation
3194 - set rq_no_resend for cancel requests (2432)
3195 - recalculate ptlrpcd timeout after resend (2494)
3196 - call vfs_rmdir when removing pending directories (2368)
3197 - fix renaming a file to itself (2429)
3198 - lmc creates a default one-stripe lov (2454)
3199 - expand procfs space to handle large clusters (2326)
3200 - increase UML stack to avoid overflow
3201 - update lconf's list of debug and subsystem masks
3202 - fix lfs find --obd (2510)
3203 - /proc tunable for disabling filter read caching (2591)
3204 - stop rpm packages from altering slapd.conf (2301)
3205 - disable nagle in the socknal under 0conf (2578)
3206 - choose mds inode size based on stripe count (2572)
3207 - fix kernel-source rpm problems (2516)
3209 - add --disable-doc to avoid pdf generation (2421)
3210 - update documentation, tests, type-os, comments
3211 - avoid format warnings on ia64
3212 - remove the TOE NAL
3213 - tiny code cleanups by removing unused fields
3215 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3218 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3219 - remove the most blatant lies from BUILDING file (2371)
3220 - change default debug level to reasonable production setting
3221 - reduce client side cache size to reduce cache flush time
3222 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3223 - make TCP zerocopy and pinger support enabled by default (2476)
3224 - sync writes completed after process exits caused crashes (2319)
3225 - maintain correct mount count on the MDS (2356)
3226 - backout 1557, because 2316 wasn't really fixed
3227 - better file I/O statistics gathering in /proc
3228 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3229 - another kernel patch to fix zero-copy TCP function export
3230 - don't take duplicate lock when processing re-sent getattr (2420)
3231 - lctl uses obd_self_export instead of creating new conn (2353)
3232 - MDS/OST recovery case which requires object creation asserted (2425)
3233 - move lfs from /usr/sbin to /usr/bin in packages
3234 - fix race between mds_client_add and mds_client_free (2417)
3235 - use kmalloc instead of slabs in portals (2430)
3236 - don't create duplicate records when a failover MDS is present (2442)
3237 - remove unnecessary mount age check (2332)
3238 - don't remove directory inodes from locks prematurely (2451)
3239 - don't break if MDS service name is the same as hostname (2103)
3240 - fix races in client write RPC generation when cache full (2482)
3242 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3245 - remove now-unused request->rq_obd (278)
3246 - if an allocation fails, print out how much memory we've used (1933)
3247 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3248 - add the "configurable stack size" patch to most series files (1256)
3249 - ability to write large log records, for 100+ OST configs (2306)
3250 - fix NULL deref when filter_prep fails (2314)
3251 - fix operator precedence error in filter_sync
3252 - dynamic allocation of socknal TX descriptors (2315)
3253 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3254 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3255 - ensure that CWARN messages in Portals always get to the syslog
3256 - __init/__exit are not for prototype decls (ldlm_init/exit)
3257 - x86-64 compile warning fixes
3258 - fix gateway LMC keyword conflict (2318)
3259 - fix MDS lock inversions in getattr/reint paths (1844)
3260 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3261 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3262 - DLM race condition prevented some lock evictions (2328)
3263 - ENOMEM detection and retry on socknal sends (2230)
3264 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3265 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3266 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3267 - better error messages when a client is rejected during recovery (1505)
3268 - avoid cancelling locks which were never granted, after failure (2330)
3269 - fix i_sem/journal inversion in mds_client_add (2333)
3270 - fix truncate/getattr lock cycle deadlock (2334)
3271 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3272 - fix two rare exit paths which could leak an l_lock() ref (2321)
3273 - fencepost error in MDS/OST orphan recovery (2226)
3274 - make log record alignment 8 bytes (1988)
3275 - lstripe now fails when requested offset > ost_count (2237)
3276 - ensure that all kernel series have a complete list.h (1607)
3277 - fix crashes in special-file operations (2316)
3278 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3280 - allow configurable automake binary, for testing new versions
3281 - small update to the lfs documentation
3283 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3285 * fix negative export reference count in fsfilt_sync (2312)
3287 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3288 * release candidate 0.9.1
3290 - orphans are moved into the PENDING directory for possible recovery
3291 - replayed opens now open by fid for orphan/rename safety (1042)
3292 - last close of an orphan inode generates a transno (683)
3293 - chdir() and mount() now pin the directory entry (1020)
3294 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3295 - discard very old requests without processing them (1502)
3296 - remove some common, well-understood CERRORs (1505)
3297 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3298 - clear "grant" flags in OST replies until OST grant code lands (1644)
3299 - fix read performance by not clobbering i_blksize on client (1598)
3300 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3301 - make LRU size a /proc tunable, clears locks when reduced (707)
3302 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3303 - server locks take references on exports now (1558)
3304 - build fixes for 2.4.20-rh trees (1663)
3305 - return an error from lov_create if all OSCs are inactive (1751)
3306 - fix import levels when a reconnect happens without a timeout (1597)
3307 - exit early from mds_open if we get a lookup error (1749)
3308 - partial page read at EOF wouldn't wait for disk before sending (1642)
3309 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3310 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3311 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3312 - write/truncate lock inversion (1639)
3313 - Don't auto-load obdclass, portals modules during cleanup (1495)
3314 - fix timestamps from jumping to "now" (1763)
3315 - extra journal assertions (1648)
3316 - add an extra multiunlink test (1771)
3317 - fix read_record/write_record API (1776)
3318 - fix leak of offset_extent, possible incorrect i_size later (1772)
3319 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3320 - Debugging for the kqswnal_get_idle_tx problems (1820)
3321 - Allow recovery to be attempted multiple times (1536)
3322 - Write out MDS last_rcvd file after it is first created (1600)
3323 - Fix tx_descriptor leak in failed transmit situations (1827)
3324 - ext3 journaling fixes for assertion failure after IO error (1871)
3325 - class_export_put() on freed export after completion AST error (1896)
3326 - Fix revalidate looping in VFS (1322)
3327 - Don't access a freed export during MDS_REINT timeout (1521)
3328 - Add open-unlink recovery support on the MDS (1673,1764)
3329 - Return an error if no MDS data was read from last_rcvd (1946)
3330 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3331 - Don't setup a disk device that doesn't match exported UUID (317)
3332 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3333 - avoid committing NULL handle in force close
3334 - local.sh is now a one-stripe LOV configuration
3335 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3336 - fix ext3 htree duplicate directory entry corruption (1516)
3337 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3338 - update to new LOV EA format (2097)
3339 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3341 2003-06-15 Phil Schwan <phil@clusterfs.com>
3344 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3345 - per-import/export recovery handling (958, 931, 959)
3346 - multiple last-rcvd slots, for serving multiple FSes (949)
3347 - connections are again shared between multiple imp/exports (963, 964)
3348 - "umount -f" would hang if any requests needed to be sent (393, 978)
3349 - avoid pinning large req buffer by copying for queued messages (989)
3350 - add "uuid" to "lctl device" command to help upcalls (991)
3351 - "open" RPCs with transnos would confuse recovery counters (1037)
3352 - do proper endian conversion of all wire messages (288, 340, 891)
3353 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3354 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3355 - fix ptlrpc_connection leak in target_handle_connect (1174)
3356 - fix import refcounting bug in OST and MDS cleanup (1134)
3357 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3358 - fix very unlikely obd_types race condition (501)
3359 - remove osc_open hack for echo_client (1187)
3360 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3361 - a failure in read_inode2 leads to deadlock (1139)
3362 - cancel ack-locks as soon as transaction is committed (1072)
3363 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3364 - make sure to commitrw after any preprw to avoid deadlock (1162)
3365 - failing to execute a file in a lustre FS would lock inode (1203)
3366 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3367 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3368 - obd_uuid2tgt would walk off the end of the list (1255)
3369 - on IA64 the osc would give portals incorrect bulk size (1258)
3370 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3371 - fix lock inversion caused by new llite matching code (1282)
3372 - limit the number of dirty pages on a client to 10MB (1286)
3373 - timed out locks were not being corrected cancelled (1289)
3374 - fix O_DIRECT above 4GB on IA-32 (1292)
3375 * major user-visible changes
3376 - fail out/fail over policy now controlled by the upcall (993)
3378 - add OBD_PING to check server availability and failure (954)
3379 - lustre messages are now sent in sending host order (288, 340, 891)
3380 - add eadatalen to MDS getattr reply (340)
3381 - OST read replies may contain second buffer, with per-page status (593)
3383 2003-03-11 Phil Schwan <phil@clusterfs.com>
3386 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3387 - failed open()s could cause deadlock; fixed (867, 869)
3388 - stop cancelling OST locks when files are closed (481)
3389 - overlapping XID spaces caused network corruption (851, 853)
3390 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3391 - setattr_raw added, to do single-RPC, server-side setattrs
3392 - lmc/lconf syntax change for OST UUIDs
3393 - fix crashy race condition between ptlrpc_free_req and osc_close
3394 - don't use request in mdc_enqueue if we hit a timeout (889)
3395 - don't set the inode i_size for regular files from the MDS (896)
3396 - handle out of order completion AST (842)
3397 - don't LBUG if a lock request times out after receiving AST (913)
3398 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3399 - if a bad lock AST arrives, send an error instead of dropping entirely
3400 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3401 - fix leak in bulk IO when only partially completed (899, 900, 926)
3402 - fix O_DIRECT for ia64 (55)
3403 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3404 - C-z after timeout could hang a process forever; fixed (977)
3406 - client-side I/O cache (678, 924, 929, 941, 970)
3408 - READPAGE and SETATTRs which don't take server-side locks get
3411 2003-02-11 Phil Schwan <phil@clusterfs.com>
3414 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3415 - Fix fsfilt_extN_readpage() to read a full page of directory
3416 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3417 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3418 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3419 - Most intent operations were removed, in favour of a new RPC mode
3420 that does a single RPC to the server and bypasses most of the VFS
3421 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3422 - Aggressively cancel local locks on DLM servers
3423 - mds_reint_unlink sends EA to the client if it's the last nlink.
3424 client uses that EA to unlink OST objects.
3425 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3426 - recursive symlinks were fixed (439)
3427 - fixed NULL deref in DEBUG_REQ
3428 - filter_update_lastobjid no longer calls sync, which annoyed extN
3429 - fixed multi-client small-writes to a single file problem (445)
3430 - fixed mtime updates during file writes (607)
3431 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3432 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3433 - fixed Linux OST size reporting problem (444, 656)
3434 - OST now updates object mtime with writes or setattr (607, 619)
3435 - client verifies file size before zeroing page past EOF (445)
3436 - OST now writes last allocated objid to disk with allocation (108)
3437 - LOV on echo now works (409)
3439 - mds_reint_unlink sends a new buffer, with the EA included. this
3440 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3441 set if a regular file was being unlinked, and it was the last link
3442 - use PtlGet from the target for bulk writes (315)
3443 - OST now updates object mtime with writes or setattr (607, 619)
3444 - LDLM now has a grant-time callback to revalidate locked items, if
3446 - Many MDS operations were reorganized to combat race conditions
3448 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3449 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3451 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3454 - Fully reactivate OST imports after reconnection (512, others)
3455 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3456 - More graceful error handling for truncating on dead OST (515)
3457 - Don't error out unless we're actually accessing dead stripes (474)
3458 - Fix garbage sizes when stripes are missing (410)
3459 - LRU counters were broken, causing constant lock purge (433, 432)
3460 - garbage on read from stripes with failed OSTs (441)
3461 - mark OSCs as active before reconnecting during recovery (438)
3462 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3463 - lfind did not preserve OST order in output (443)
3464 - symlinks cause hung clients, incorrect data (439)
3465 - stop dereferencing request after dropping refcount (457)
3466 - don't LASSERT(spin_is_locked) on non-SMP (455)
3467 - fixes for many rename() bugs
3468 - fstat didn't correctly synchronize attributes (399)
3469 - server must handle lock cancellation during blocking AST prep (487)
3470 - bulk descriptors were free()d too soon (511)
3471 - fix paths in lconf, which would load incorrect modules (451, 507)
3472 - fix confusing lconf 'host not found' error message (386)
3473 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3474 - fix race condition in mdc_blocking_ast() for inode access (526)
3475 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3476 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3477 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3478 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3479 - fix lstripe in several ways (kernel side) (527)
3480 - fix request leak in ldlm_cli_enqueue (262)
3481 - incorrect OSC was marked inactive after OST failure
3482 - call mds_fs_cleanup before unmounting filesystem (524)
3483 - fix races between taking ns_lock and ldlm_lock_change_resource
3484 - fix races updating LOV export open file list
3485 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3486 - fix recovery NULL deref in ldlm_cli_cancel_unused
3487 - fix some DLM races by using new hash table for lock handles (419)
3488 - permit the client to specify desired inodes, at replay
3489 - duplicate requests when we queue them for replay reintegration
3490 - fix last_rcvd offset calculation
3491 - sync after each recovered transaction, so we always make progress
3492 - never, not always, ERESTART requests without transnos
3493 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3495 - skip replay if the MDS says that the client is already connected
3496 - don't check for a recovery-enabled export to match lctl's UUID
3497 - don't INC_USE_COUNT for phantom exports
3498 - don't crash when cleaning up phantom exports (567)
3499 - don't double-finish or set replay data for errored mdc_open requests
3500 - abort requests when they time out, so we don't get old replies
3501 - send/receive replies for AST messages again
3502 - if the client says that it doesn't have the lock, cancel it on the
3504 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3505 mark it as destroyed, it will all work out when decref is called
3506 - fix module use counts (22, 581)
3508 - ASTs now expect a reply (server cancels lock on error reply)
3510 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3513 - fix many simultaneous client startup (392)
3514 - fix dentry->d_it clobbering
3515 - credentials weren't being shipped for readdir/getattr operations
3516 - remove invalid assertions triggered during some concurrent MD
3518 - proper Lustre versions added (336, 389)
3519 - fix memory leak for create error case (398)
3520 - fix LOV locking bug that would get cli/srv out of sync
3521 - fix echo client over LOV (409)
3522 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3523 - fix double-O_EXCL intent crash (424)
3524 - avoid sending multiple lock CANCELs (352)
3526 - MDS can do multi-client recovery (modulo bugs in new code)
3528 - many updates, edits, cleanups
3530 2002-11-18 Phil Schwan <phil@clusterfs.com>
3533 - fix null d_it dereference (346)
3534 - fix full OST/dbench hang (333)
3535 - fix permission problem with file removal (286)
3536 - fix removal of OSCs from LOV when they fail
3537 - fix NULL deref during bulk timeout (214)
3538 - fix problems related to multiple filesystems on one MDS (241)
3539 - fixed serious subtle metadata locking bugs
3540 - free locks on clients when inodes are removed due to memory
3542 - fix inode pointer in lock data (285)
3543 - partial support for multiple MDS on a single host (241)
3544 - data locks weren't cancelled at clear_inode time (290, 311)
3545 - intent locks could lead to unbounded lock growth (205)
3546 - added a maximum lock count, an LRU list, and a flusher
3547 - fix multiple rename (365)
3548 - properly abstracted the echo client
3549 - OSC locked 1 byte too many; fixed
3550 - rewrote brw callback code:
3551 - fixed recovery bugs related to LOVs (306)
3552 - fixed too-many-pages-in-one-write crash (191)
3553 - fixed (again) crash in sync_io_timeout (214)
3554 - probably fixed callback-related race (385)
3556 - Add capability to MDS protocol
3557 - LDLM cancellations and callbacks on different portals
3559 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3562 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3563 - defer open object destruction to close time (601981, 138)
3564 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3565 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3566 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3567 - don't call obd_setattr in ll_file_release for destroyed objects
3568 * protocol change to lustre_msg: move |version| and add |flags|
3569 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3570 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3571 * added replay of create, unlink, link and rename operations during
3572 MDS failover; recovery should be much more robust now
3573 * remove failed OSCs from LOVs (only lov_create uses this so far)
3574 * the lustre-HOWTO was brought (more) up to date (582544)
3576 2002-10-23 Phil Schwan <phil@clusterfs.com>
3579 - in-use dentries weren't being reused properly (617851)
3580 - prevent multiple LDLM setup (599178)
3581 - fix LOV size calculations for truncate (617853)
3582 - fix client handling of MDS intent errors (POSIX)
3583 - fix permission bug in lovstripe.c test (624321)
3584 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3585 - truncate past end of file could corrupt data
3586 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3587 - a race in recovery could return ETIMEDOUT to apps (623947)
3588 - building outside the source directory was fixed
3589 * the lustre-HOWTO was brought (more) up to date (582544)
3590 * major progress was made on recovery functionality
3592 2002-10-10 Phil Schwan <phil@clusterfs.com>
3595 - recovery deadlock fix
3596 - rm -rf causes LBUG fix (617817)
3597 - file open by multiple tasks fix (618962)
3598 - directory permissions bugs (602707 and 620007)
3599 - journal_stop fixed with locking (611313)
3600 - O_APPEND failures resolved (618273, perhaps 614459)
3601 - lconf PATH fix (619770)
3602 - IA64 build fix (621450)
3603 - RPC buffer sizes scale with amount of memory
3605 2002-10-01 Phil Schwan <phil@clusterfs.com>
3608 - locks would be cancelled without throwing away data pages,
3609 resulting in inconsistent data (605627)
3610 - inode attributes were not always being refreshed (605627, 612449)
3611 - lconf now continues to cleanup after lctl reports an error
3612 - MDS now enforces user permissions (602707)
3613 - lprocfs cleanup fixed, but not yet enabled (614157)
3614 - fixed infinite server hang, should a client not respond to an AST
3615 - avoid going into recovery if user calls readlink() with a buffer
3616 that's too small (613941)
3617 - AST RPCs no longer require replies (614867) -- this may be changed
3618 - don't crash server if client sends an IOV that's too big (611336)
3619 - fixed lock conversion deadlock (611892)
3620 - fixed the following of symlinks (614622)
3621 * recovery: the server can remove locks from a client that dies, other
3622 clients can make progress
3623 * more extN patch fixes
3624 * compile-time configurable ptlrpc buffer allocations
3626 - collaborative read cache document
3627 - Lustre Lite Performance CDR document-in-progress
3629 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3632 - fix typo in patch-2.4.18
3634 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3637 - clear ptlrpc request each time in handle_incoming_request()
3638 - unlink of files now destroys the object on the OST
3640 2002-09-19 Peter Braam <braam@clusterfs.com>
3642 * add hard link support
3643 * change obdfile creation method
3644 * kernel patch changed
3646 2002-09-19 Peter Braam <braam@clusterfs.com>
3649 - stack overflow bug in extN fixed
3651 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3653 * documentation updates
3654 - add man pages for config tools
3655 - update tests/README to describe testing with new config tools
3656 - finish metadata API descriptions
3657 * bug fixes and cleanups
3658 - statfs workaround for 16TB limit
3659 - LOV stripe allocation improved, can stripe on subset of OSTs
3660 - LOV file size/IO offset was wrong for files > 4GB in size
3661 - object EA data was being dropped, caused files to be unreadable
3662 - memory overflow with non-LOV OST caused memory corruption
3663 - fixed regression tests to work with new config tools, obdfilter
3664 - fixed bug when directory size became larger than 1 block
3665 - fixed bug (for single client case) when PWD was deleted
3666 - invalidate local directory pages when doing intent-based ops
3667 - avoid LDLM oops when lock callback contained bad data
3669 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3671 * documentation updates
3672 * bug fixes and cleanups
3673 - configuration tools
3676 - 64-bit compile warnings
3677 - 64-bit internal statfs data
3679 * test_brw on persistent OST devices
3681 * lprocfs (disabled)
3683 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3685 * documentation updates
3686 * bug fixes and cleanups
3687 * configuration tools
3689 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3692 * many small fixes to 0_5_4
3693 * io/network handling
3694 * thinkos in MDS operations
3696 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3699 * crucial basic fixes to 0.5.3
3700 * IOR, Iozone work over Elan
3703 2002-08-07 Phil Schwan <phil@clusterfs.com>
3704 * version 0_5_3, our first alpha
3705 * we use the new Portals iovs
3706 * documentation updates
3707 * bug fixes and cleanups
3708 * small changes in the DLM wire protocol
3710 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3711 * version 0_5_1 with some initial stability,
3712 * locking on MD and file I/O.
3713 * documentation updates
3714 * several bug fixes since 0.5.0
3715 * small changes in wire protocol
3717 2002-07-18 Phil Schwan <phil@clusterfs.com>
3719 * delivered as Lustre Light Alpha
3720 * fixed a crash after handling invalid MDS requests
3721 * fixed directory pages for architectures with non-4k pages sizes
3723 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3724 * release version v0_4_4
3725 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3727 * Fixes a number of interruption problems with OST operations.
3728 * Update documentation for portals header changes
3729 * Move all wire protocol structs/defines to lustre_idl.h
3730 * Fixes symlink length bug.
3731 * Add tcpdump to repository.
3733 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3734 * release version v0_4_3
3735 * Fixes statfs for inodes on extN.
3736 * Fixes bug in runtests which would delete /etc/hosts.
3737 * Use 64-bit object IDs wherever possible (not into VFS though)
3738 Remove ost_get_info, which is unused by lustre, and out of date.
3740 2002-07-03 Peter Braam <braam@clusterfs.com>
3741 * release version v0_4_2 Fixes a lookup error (type not passed)
3742 * move forward to head of Portals
3743 * move forward to latest Lustre kernel
3745 2002-06-25 Peter Braam <braam@clusterfs.com>
3746 * release version v0_4_1. Hopefully stable on single node use.