1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4 2.4.21-47.0.1.EL (RHEL 3)
6 2.6.9-42.0.10.EL (RHEL 4)
7 2.6.12.6 vanilla (kernel.org)
8 2.6.16.27-0.9 (SLES 10)
9 * Client support for unpatched kernels:
10 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
11 2.6.9-42.0.10.EL (RHEL 4)
12 2.6.16 - 2.6.21 vanilla (kernel.org)
13 * Recommended e2fsprogs version: 1.39.cfs7
14 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
17 Severity : enhancement
19 Description: add optional extra BUILD_VERSION info
20 Details : add a new environment variable (namely LUSTRE_VERS) which allows
21 to override the lustre version.
24 Frequency : 2.6.18 servers only
26 Description: ll_kern_mount() doesn't release the module reference
27 Details : The ldiskfs module reference count never drops down to 0
28 because ll_kern_mount() doesn't release the module reference.
33 Description: server LBUG when using old ost_num_threads parameter
34 Details : Accept the old ost_num_threads parameter but warn that it
35 is deprecated, and fix an off-by-one error that caused an LBUG.
40 Description: Transient SCSI error results in persistent IO issue
41 Details : iobuf->dr_error is not reinitialized to 0 between two
45 Frequency : sometimes when underlying device returns I/O errors
47 Description: OSTs not going read-only during write failures
48 Details : OSTs are not remounted read-only when the journal commit threads
49 get I/O errors because fsfilt_ext3 calls journal_start/stop()
50 instead of the ext3 wrappers.
54 Description: poor connect scaling with increasing client count
55 Details : Don't run filter_grant_sanity_check for more than 100 exports
56 to improve scaling for large numbers of clients.
59 Frequency : SLES10 only
61 Description: sanity-quota.sh quotacheck failed: rc = -22
62 Details : Quotas cannot be enabled on SLES10.
65 Frequency : liblustre clients only
67 Description: getdirentries does not give error when run on compute nodes
68 Details : getdirentries does not fail when the size specified as an argument
69 is too small to contain at least one entry
71 Severity : enhancement
73 Description: Add LNET router traceability for debug purposes
74 Details : If a checksum failure occurs with a router as part of the
75 IO path, the NID of the last router that forwarded the bulk data
76 is printed so it can be identified.
81 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
82 Details : Due to a race condition, liblustre clients were occasionally
85 Severity : enhancement
87 Description: lfs setstripe use optional parameters instead of postional
90 Severity : enhancement
92 Description: Nanosecond timestamp support for ldiskfs
93 Details : The on-disk ldiskfs filesystem has added support for nanosecond
94 resolution timestamps. There is not yet support for this at
95 the Lustre filesystem level.
98 Frequency : during server recovery
100 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
101 Details : request with rq_no_resend flag not awake l_wait_event if they get a
105 Frequency : nfs export on patchless client
107 Description: connectathon hang when test nfs export over patchless client
108 Details : Disconnected dentry cannot be found with lookup, so we do not need
109 to unhash it or make it invalid
112 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
113 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
114 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
119 Description: open req refcounting wrong on reconnect
120 Details : If reconnect happened between getting open reply from server and
121 call to mdc_set_replay_data in ll_file_open, we will schedule
122 replay for unreferenced request that we are about to free.
123 Subsequent close will crash in variety of ways.
124 Check that request is still eligible for replay in
125 mdc_set_replay_data().
130 Description: disable writes to filesystem when reading health_check file
131 Details : the default for reading the health_check proc file has changed
132 to NOT do a journal transaction and write to disk, because this
133 can cause reads of the /proc file to hang and block HA state
134 checking on a healthy but otherwise heavily loaded system. It
135 is possible to return to the previous behaviour during configure
136 with --enable-health-write.
138 Severity : enhancement
140 Description: 64-bit inode version
141 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
142 to the inode. This will be required for version-based recovery.
147 Description: MDS fails to start if a duplicate client export is detected
148 Details : in some rare cases it was possible for a client to connect to
149 an MDS multiple times. Upon recovery the MDS would detect this
150 and fail during startup. Handle this more gracefully.
152 Severity : enhancement
154 Description: Add -o localflock option to simulate old noflock
156 Details : This will achieve local-only flock/fcntl locks
162 Description: log_commit_thread vs filter_destroy race leads to crash
163 Details : Take import reference before releasing llog record semaphore
168 Description: Wrong request locking in request set processing
169 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
170 imp_delayed_list, in this place should be used imp_lock.
173 Frequency : when reconnection
175 Description: Grant Leak when osc reconnect to OST
176 Details : When osc reconnect ost, OST(filter) should check whether it
177 should grant more space to client by comparing fed_grant and
178 cl_avail_grant, and return the granted space to client instead
179 of "new granted" space, because client will call osc_init_grant
180 to update the client grant space info.
183 Frequency : when client reconnect to OST
185 Description: Grant Leak when osc do resend and replay bulk write
186 Details : When osc reconnect to OST, OST(filter)should clear grant info of
187 bulk write request, because the grant info while be sync between
188 OSC and OST when reconnect, and we should ignore the grant info
189 these of resend/replay write req.
194 Description: Grant space more than avaiable left space sometimes.
195 Details : When then OST is about to be full, if two bulk writing from
196 different clients came to OST. Accord the avaliable space of the
197 OST, the first req should be permitted, and the second one
198 should be denied by ENOSPC. But if the seconde arrived before
199 the first one is commited. The OST might wrongly permit second
200 writing, which will cause grant space > avaiable space.
203 Frequency : when client is evicted
205 Description: Grant might be wrongly erased when osc is evicted by OST
206 Details : when the import is evicted by server, it will fork another
207 thread ptlrpc_invalidate_import_thread to invalidate the
208 import, where the grant will be set to 0. While the original
209 thread will update the grant it got when connecting. So if
210 the former happened latter, the grant will be wrongly errased
211 because of this race.
216 Description: Checking Stale with correct fid
217 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
218 to check whether it is stale, because sometimes, we want the
219 enqueue happened anyway, and op_data.fid2 will not be initialized.
221 Severity : enhancement
223 Description: update patchless client
224 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
227 Frequency : only with 2.4 kernel
229 Description: random memory corruption
230 Details : size of struct ll_inode_info is to big for union inode.u and this
231 can be cause of random memory corruption.
236 Description: Memory leak in recovery
237 Details : Lov_mds_md was not free in an error handler in mds_create_object.
238 It should also check obd_fail before fsfilt_start, otherwise if
239 fsfilt_start return -EROFS,(failover mds during mds recovery).
240 then the req will return with repmsg->transno = 0 and rc = EROFS.
241 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
242 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
243 be freed no matter whether fsfilt_commit success or not.
246 Frequency : only with huge count clients
248 Description: Prevents from taking the superblock lock in llap_from_page for
250 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
251 for llap_from_page call in ll_removepage prevents from taking the
252 superblock lock for a soon died page.
257 Description: Not check open intent error before release open handle
258 Details : in some rare cases, the open intent error is not checked before
259 release open handle, which may cause
260 ASSERTION(open_req->rq_transno != 0), because it tries to release
261 the failed open handle.
266 Description: Set cat log bitmap only after create log success.
267 Details : in some rare cases, the cat log bitmap is set too early. and it
268 should be set only after create log success.
272 Description: Accessing a block bevice can re-enable I/O when Lustre is
273 tearing down a device.
274 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
279 Description: service threads may hog cpus when there are a lot of requests
281 Details : Insert cond_resched to give other threads a chance to use some of
287 Description: the cat log was not initialized in recovery
288 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
289 unlink log on mds will not be initialized until mds post
290 recovery. And also in mds post recovery, the unlink log will
291 initialization will be done asynchronausly, so there will be race
292 between add unlink log and unlink log initialization.
296 Description: brw_stats were being printed incorrectly
297 Details : brw_stats were being printed as log2 but all of them were not
298 recorded as log2. Also remove some code duplication arising from
299 filter_tally_{read,write}.
303 Frequency : rare, only in recovery.
304 Description: ASSERTION(req->rq_type != LI_POISON) failed
305 Details : imp_lock should be held while iterating over imp_sending_list for
306 prevent destroy request after get timeout in ptlrpc_queue_wait.
310 Description: replay-single.sh test 52 fails
311 Details : A lock's skiplist need to be cleanup when it being unlinked
312 from its resource list.
316 Description: Short directio read returns full requested size rather than
318 Details : Direct I/O operations should return actual amount of bytes
319 transferred rather than requested size.
321 --------------------------------------------------------------------------------
323 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
328 Frequency : on some architectures
330 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
331 Details : Uninitialized flags sometimes cause configuration commands to
335 Frequency : patchless clients only
337 Description: missing __iget() symbol export
338 Details : The __iget() symbol export is missing. To avoid the need for
339 this on patchless clients the deathrow inode reaper is turned
340 off, and we depend on the VM to clean up old inodes. This
341 dependency was during via the fix for bug 12181.
343 --------------------------------------------------------------------------------
345 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
347 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
348 INTEROPERATE with older versions automatically. In many cases a
349 special upgrade step is needed. Please read the
350 user documentation before upgrading any part of a live system.
351 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
352 version will not interoperate with 1.6 betas before beta5 (1.5.95).
353 * WARNING: Lustre configuration and startup changes are required with
354 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
359 Severity : enhancement
361 Description: MountConf
362 Details : Lustre configuration is now managed via mkfs and mount
363 commands instead of lmc and lconf. New obd types (MGS, MGC)
364 are added for dynamic configuration management. See
365 https://mail.clusterfs.com/wikis/lustre/MountConf for
368 Severity : enhancement
370 Description: dynamic OST addition
371 Details : OSTs can now be added to a live filesystem
373 Severity : enhancement
375 Description: startup order invariance
376 Details : MDTs and OSTs can be started in any order. Clients only
377 require the MDT to complete startup.
379 Severity : enhancement
381 Description: parallel, asynchronous orphan cleanup
382 Details : orphan cleanup is now performed in separate threads for each
383 OST, allowing parallel non-blocking operation.
385 Severity : enhancement
387 Description: optimized stripe assignment
388 Details : stripe assignments are now made based on ost space available,
389 ost previous usage, and OSS previous usage, in order to try
390 to optimize storage space and networking resources.
392 Severity : enhancement
394 Description: Permanently set tunables
395 Details : All writable /proc/fs/lustre tunables can now be permanently
396 set on a per-server basis, at mkfs time or on a live
399 Severity : enhancement
401 Description: Lustre message v2
402 Details : Add lustre message format v2.
404 Severity : enhancement
406 Description: client OST exclusion list
407 Details : Clients can be started with a list of OSTs that should be
408 declared "inactive" for known non-responsive OSTs.
412 Description: SPEC SFS validation failure on NFS v2 over lustre.
413 Details : Changes the blocksize for regular files to be 2x RPC size,
414 and not depend on stripe size.
416 Severity : enhancement
418 Description: Multiple MD RPCs in flight.
419 Details : Further unserialise some read-only MDS RPCs - learn about intents.
420 To avoid overly-overloading MDS, introduce a limit on number of
421 MDS RPCs in flight for a single client and add /proc controls
422 to adjust this limit.
424 Severity : enhancement
426 Description: client read/write statistics
427 Details : Add client read/write call usage stats for performance
428 analysis of user processes.
429 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
430 file access. extents_stats shows chunk size distribution.
431 extents_stats_per_process show chunk size distribution per
434 Severity : enhancement
436 Description: mds statistics
437 Details : Add detailed mds operations statistics in
438 /proc/fs/lustre/mds/*/stats.
442 Description: Failure of copying files with lustre special EAs.
443 Details : Client side always return success for setxattr call for lustre
444 special xattr (currently only "trusted.lov").
449 Description: Refcount LNET uuids
450 Details : The global LNET uuid list grew linearly with every startup;
451 refcount repeated list entries instead of always adding to
454 Severity : enhancement
456 Description: Dynamic service threads
457 Details : Within a small range, start extra service threads
458 automatically when the request queue builds up.
461 Frequency : mixed-endian client/server environments
463 Description: mixed-endian crashes
464 Details : The new msg_v2 system had some failures in mixed-endian
467 Severity : enhancement
469 Description: Easy OST removal
470 Details : OSTs can be permanently deactivated with e.g. 'lctl
471 conf_param lustre-OST0001.osc.active=0'
473 Severity : enhancement
475 Description: MGS proc entries
476 Details : Added basic proc entries for the MGS showing what filesystems
479 Severity : enhancement
481 Description: provide MGS failover
482 Details : Added config lock reacquisition after MGS server failover.
484 Severity : enhancement
486 Description: add Linux 2.4 support
487 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
491 Description: a large application tries to do I/O to the same resource and dies
493 Details : Check the req->rq_arrival time after the call to
494 ost_brw_lock_get(), but before we do anything about
495 processing it & sending the BULK transfer request. This
496 should help move old stale pending locks off the queue as
497 quickly as obd_timeout.
500 Frequency : when an incorrect nid is specified during startup
502 Description: ptlrpc connect to non-existant node causes kernel crash
503 Details : LNET can't be re-entered from an event callback, which
504 happened when we expire a message after the export has been
505 cleaned up. Instead, hand the zombie cleanup off to another
508 Severity : enhancement
510 Description: plain/inodebits lock performance improvement
511 Details : Grouping plain/inodebits in granted list by their request modes
512 and bits policy, thus improving the performance of search through
516 Frequency : only if OST filesystem is corrupted
518 Description: client incorrectly hits assertion in ptlrpc_replay_req()
519 Details : for a short time RPCs with bulk IO are in the replay list,
520 but replay of bulk IOs is unimplemented. If the OST filesystem
521 is corrupted due to disk cache incoherency and then replay is
522 started it is possible to trip an assertion. Avoid putting
523 committed RPCs into the replay list at all to avoid this issue.
526 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
528 Description: System hang on startup
529 Details : This bug allowed the liblustre (e.g. catamount) client to
530 return to the app before handling all startup RPCs. This
531 could leave the node unresponsive to lustre network traffic
532 and manifested as a server ptllnd timeout.
534 Severity : enhancement
536 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
537 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
538 boolean to control whether to print peer debug info when a
539 client's RPC times out.
542 Frequency : only for kernels with patches from Lustre below 1.4.3
544 Description: Remove old rdonly API
545 Details : Remove old rdonly API which unused from at least lustre 1.4.3
548 Frequency : only for devices with external journals
550 Description: Set external device read-only also
551 Details : During a commanded failover stop, we set the disk device
552 read-only while the server shuts down. We now also set any
553 external journal device read-only at the same time.
556 Frequency : when upgrading from 1.4 while trying to change parameters
558 Description: The wrong (new) MDC name was used when setting parameters for
559 upgraded MDT's. Also allows changing of OSC (and MDC)
560 parameters if --writeconf is specified at tunefs upgrade time.
563 Frequency : when setting specific ost indicies
565 Description: QOS code breaks on skipped indicies
566 Details : Add checks for missing OST indicies in the QOS code, so OSTs
567 created with --index need not be sequential.
569 Severity : enhancement
571 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
572 Details : The uninit_groups feature works in conjunction with the kernel
573 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
574 up the pass1 processing of e2fsck. This is a read-only feature
575 in ldiskfs2 only, so older kernels and current ldiskfs cannot
576 mount filesystems that have had this feature enabled.
578 Severity : enhancement
580 Description: Improve multi-block allocation algorithm to avoid fragmentation
581 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
582 allocation locality and avoid filesystem fragmentation.
584 ------------------------------------------------------------------------------
586 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
588 * Support for kernels:
589 2.4.21-47.0.1.EL (RHEL 3)
591 2.6.9-42.0.10.EL (RHEL 4)
592 2.6.12.6 vanilla (kernel.org)
593 2.6.16.27-0.9 (SLES 10)
594 * Recommended e2fsprogs version: 1.39.cfs5
596 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
600 Frequency : occasional, depends on client load and configuration
601 Bugzilla : 12181, 12203
602 Description: data loss for recently-modified files
604 Details : In some cases it is possible that recently written or created
605 files may not be written to disk in a timely manner (this should
606 normally be within 30s unless client IO load is very high).
607 The problem appears as zero-length files or files that are a
608 multiple of 1MB in size after a client crash or client eviction
609 that are missing data at the end of the file.
611 This problem is more likely to be hit on clients where files are
612 repeatedly created and unlinked in the same directory, clients
613 have a large amount of RAM, have many CPUs, the filesystem has
614 many OSTs, the clients are rebooted frequently, and/or the files
615 are not accessed by other nodes after being written.
617 The presence of the problem can be detected by looking at
618 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
619 smaller than the second (nr_unused) then dirty files will not
620 be flushed automatically to disk. "sync; sleep 10" should be
621 run several times on the node before unmounting it to update
622 Lustre (this is also safe to run on nodes without this problem).
624 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
625 that can cause this same problem, so customers using that kernel
626 also need to update the kernel in addition to Lustre. In order
627 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
629 It is normal that files written just before a client crash (less
630 than 30s) may not yet have been flushed to disk, even for local
634 Frequency : frequent on thin XT3 nodes
636 Description: UUID collision on thin XT3 Linux nodes
637 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
638 randomly, since we relied on an insufficiently-seeded PRNG.
643 Description: OSS hangs after "All ost request buffers busy"
644 Details : A deadlock between quota and journal operations caused OSS
645 hangs after printing "All ost request buffers busy."
648 Frequency : always on liblustre builds
650 Description: Cleanup compiler warnings on liblustre
653 Frequency : always on liblustre builds on XT3
655 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
659 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
660 > 32000 subdirectories
661 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
662 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
663 subdirectory count crosses 32000. This will aid e2fsck to
664 correctly handle more than 32000 subdirectories.
667 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
669 Description: System hang on startup
670 Details : This bug allowed the liblustre (e.g. catamount) client to
671 return to the app before handling all startup RPCs. This
672 could leave the node unresponsive to lustre network traffic
673 and manifested as a server ptllnd timeout.
675 Severity : enhancement
677 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
678 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
679 boolean to control whether to print peer debug info when a
680 client's RPC times out.
685 Description: make O_SYNC working on 2.6 kernels
686 Details : 2.6 kernels use different method for mark pages for write,
687 so need add a code to lustre for O_SYNC work.
692 Description: Failure to close file and release space on NFS
693 Details : Put inode details into lock acquired in ll_intent_file_open.
694 Use mdc_intent_lock in ll_intent_open to properly
695 detect all kind of errors unhandled by mdc_enqueue.
700 Description: proc file read during shutdown sometimes raced obd removal,
702 Details : Add lock to prevent obd access after proc file removal.
705 Frequency : Only for files larger than 4GB on 32-bit clients.
707 Description: improperly doing page alignment of locks
708 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
709 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
714 Description: RPCs being resent when they shouldn't be
715 Details : Some RPCs that should not be resent are being resent. This
716 can cause inconsistencies in the RPC state machine. Do not
717 resend such requests.
720 Frequency : rare, only with NFS export
722 Description: Crash on NFS re-export node
723 Details : under very unusual load conditions an assertion is hit in
724 ll_intent_file_open()
727 Frequency : only if OST filesystem is corrupted
729 Description: client incorrectly hits assertion in ptlrpc_replay_req()
730 Details : for a short time RPCs with bulk IO are in the replay list,
731 but replay of bulk IOs is unimplemented. If the OST filesystem
732 is corrupted due to disk cache incoherency and then replay is
733 started it is possible to trip an assertion. Avoid putting
734 committed RPCs into the replay list at all to avoid this issue.
739 Description: large O_DIRECT requests fail under memory pressure/fragmentation
740 Details : Large single O_DIRECT read and write calls can fail to allocate
741 a sufficiently large buffer to process the request. In case of
742 allocation failure the allocation is retried with a smaller
743 buffer and broken into smaller requests.
745 Severity : enhancement
747 Description: Add -o localflock option to simulate old noflock behaviour.
748 Details : This will achieve local-only flock/fcntl locks coherentness.
753 Description: versioning check is incomplete
754 Details : Checking the version difference of client vs. server, report
755 error if the gap is too big.
760 Description: add support PG_writeback bit
761 Details : add support for PG_writeback bit for Lustre, for more carefull
762 work with page cache in 2.6 kernel. This also fix some deadlocks
763 and remove hack for work O_SYNC with 2.6 kernel.
765 Severity : enhancement
767 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
768 Details : The uninit_groups feature works in conjunction with the kernel
769 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
770 up the pass1 processing of e2fsck. This is a read-only feature
771 in ldiskfs2 only, so older kernels and current ldiskfs cannot
772 mount filesystems that have had this feature enabled.
774 Severity : enhancement
776 Description: Improve multi-block allocation algorithm to avoid fragmentation
777 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
778 allocation locality and avoid filesystem fragmentation.
780 ------------------------------------------------------------------------------
782 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
784 * Support for kernels:
785 2.6.9-42.0.3.EL (RHEL 4)
787 2.4.21-47.0.1.EL (RHEL 3)
788 2.6.12.6 vanilla (kernel.org)
789 2.6.16.21-0.8 (SLES10)
790 * Recommended e2fsprogs version: 1.39.cfs2-0
792 * The backwards-compatible /proc/sys/portals symlink has been removed
793 in this release. Before upgrading, please ensure that you change
794 any configuration scripts or /etc/sysctl.conf files that access
795 /proc/sys/portals/* or sysctl portals.* to use the corresponding
796 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
797 in advance of the upgrade on any system running Lustre 1.4.6 or
798 newer, since /proc/sys/lnet was added in that version.
799 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
803 Frequency : only when quota is used
805 Description: avoid scanning export list for quota master
806 Details : Change the algorithms to avoid scanning export list in order
807 to improve the efficiency.
810 Frequency : MDS failover only, very rarely
812 Description: "went back in time" messages on mds failover
813 Details : The greatest transno may be lost when the current operation
814 finishes with an error (transno==0) and the client's last_rcvd
815 record is over-written. Save the greatest transno in the
816 mds_last_transno for this case.
819 Frequency : always for specific kernels and striping counts
821 Description: client may get "Matching packet too big" without ACL support
822 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
823 when trying to access files in certain configurations. The
824 clients should in fact be denied when mounting because they do
828 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
830 Description: many clients connecting with IO in progress causes connect timeouts
831 Details : Avoid synchronous journal commits to avoid delays caused by many
832 clients connecting/disconnecting when bulk IO is in progress.
833 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
834 OST_IO_PORTAL to avoid delays behind potentially many pending
838 Frequency : occasionally with multiple writers to a single file
840 Description: shared writes to file may result in wrong size reported by stat()
841 Details : Allow growing of kms when extent lock is cancelled
844 Frequency : always with random mmap IO to multi-striped file
846 Description: mmap write might be lost if we are writing to a 'hole' in stripe
847 Details : Only if the hole is at the end of OST object so that kms is too
848 small. Fix is to increase kms accordingly in ll_nopage.
851 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
853 Description: writes to a missing object would leak memory on the OST
854 Details : If there is an inconsistency between the MDS and OST filesystems,
855 such that the MDS references an object that doesn't exist, writes
856 to that object will leak memory due to incorrect cleanup in the
857 error handling path, eventually running out of memory on the OST.
862 Description: Creating too long symlink causes lustre errors
863 Details : Check symlink and name lengths before sending requests to MDS.
866 Frequency : only if flock is enabled (not on by default)
868 Description: posix locks not released on fd closure on 2.6.9+
869 Details : We failed to add posix locks to list of inode locks on 2.6.9+
870 kernels, this caused such locks not to be released on fd close and
871 then assertions on fs unmount about still used locks.
874 Frequency : MDS failover only, very rarely
876 Description: clients may get ASSERTION(granted_lock != NULL)
877 Details : When request was taking a long time, and a client was resending
878 a getattr by name lock request. The were multiple lock requests
879 with the same client lock handle and
880 mds_getattr_name->fixup_handle_for_resent_request found one of the
881 lock handles but later failed with ASSERTION(granted_lock != NULL).
886 Description: handle->h_buffer_credits > 0, assertion failure
887 Details : h_buffer_credits is zero after truncate, causing assertion
888 failure. This patch extends the transaction or creates a new
892 Frequency : NFS re-export or patchless client
893 Bugzilla : 11179, 10796
894 Description: Crash on NFS re-export node (__d_move)
895 Details : We do not want to hash the dentry if we don't have a lock.
896 But if this dentry is later used in d_move, we'd hit uninitialised
897 list head d_hash, so we just do this to init d_hash field but
898 leave dentry unhashed.
901 Frequency : NFS re-export or patchless client
903 Description: NFS exports has problem with symbolic link
904 Details : lustre client didn't properly install dentry when re-exported
905 to NFS or running patchless client.
908 Frequency : NFS re-export or patchless client
910 Description: Various nfs/patchless fixes.
911 Details : fixes reuse disconected alias for lookup process - this fixes
912 warning "find_exported_dentry: npd != pd",
913 fix permission error with open files at nfs.
914 fix apply umask when do revalidate.
917 Frequency : occasional
919 Description: Crash on NFS re-export node
920 Details : calling clear_page() on the wrong pointer triggered oops in
921 generic_mapping_read().
924 Frequency : rarely, using O_DIRECT IO
926 Description: unaligned directio crashes client with LASSERT
927 Details : check for unaligned buffers before trying any requests.
930 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
932 Description: stale data returned from RAID cache
933 Details : If only a small amount of IO is done to the RAID device before
934 reading it again it is possible to get stale data from the RAID
935 cache instead of reading it from disk.
938 Frequency : always for sles10 kernel
940 Description: sles10 support
941 Details : ll_follow_link: compile fixes and using of nd_set_link
945 Frequency : depends on arch, kernel and compiler version, always on sles10
948 Description: recursive or deep enough symlinks cause stack overflow
949 Details : getting rid of large stack-allocated variable in
953 Frequency : depends on hardware
955 Description: lustre write performance loss in the SLES10 kernel
956 Details : the performance loss is caused by using of write barriers in the
957 ext3 code. The SLES10 kernel turns barrier support on by
958 default. The fix is to undo that change for ldiskfs.
960 ------------------------------------------------------------------------------
962 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
964 * Support for kernels:
965 2.6.9-42.0.3EL (RHEL 4)
967 2.4.21-47.0.1.EL (RHEL 3)
968 2.6.12.6 vanilla (kernel.org)
972 Frequency : quota enabled and large files being deleted
974 Description: releasing more than 4GB of quota at once hangs OST
975 Details : If a user deletes more than 4GB of files on a single OST it
976 will cause the OST to spin in an infinite loop. Release
977 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
982 Description: statfs data retrieved from /proc may be stale or zero
983 Details : When reading per-device statfs data from /proc, in the
984 {kbytes,files}_{total,free,avail} files, it may appear
985 as zero or be out of date.
988 Frequency : systems with MD RAID1 external journal devices
990 Description: lconf's call to blkid is confused by RAID1 journal devices
991 Details : Use the "blkid -l" flag to locate the MD RAID device instead
992 of returning all block devices that match the journal UUID.
995 Frequency : always, for aggregate stripe size over 4GB
997 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
998 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
999 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1002 Frequency : always if "lfs find" used on a local file/directory
1004 Description: "lfs find" segfaults if used on a local file/directory
1005 Details : The case where a directory component was not specified wasn't
1006 handled correctly. Handle this properly.
1009 Frequency : always on ppc64
1011 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1012 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1013 perform endianness conversion.
1016 Frequency : rarely (truncate to non-zero file size after write under load)
1017 Bugzilla : 10730, 10687
1018 Description: Files padded with zeros to next 4K multiple
1019 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1020 it is possible that files that are truncated to a non-zero size
1021 immediately after being written are filled with zero bytes beyond
1022 the truncated size. No file data is lost.
1024 Severity : enhancement
1026 Description: Allow recovery/failover for liblustre clients.
1027 Details : liblustre clients were unaware of failover configurations until
1030 Severity : enhancement
1032 Description: user file locks should fail when not mounting with flock option
1033 Details : Set up an error-returning stub in ll_file_operations.lock field
1034 to prevent incorrect behaviour when client is mounted without
1035 flock option. Also, set up properly f_op->flock field for
1039 Frequency : always on ia64
1041 Description: "lfs df" loops on printing out MDS statfs information
1042 Details : The obd_ioctl_data was not initialized and in some systems
1043 this caused a failure during the ioctl that did not return
1044 an error. Initialize the struct and return an error on failure.
1047 Frequency : SLES 9 only
1049 Description: Error of copying files with lustre special EAs as root
1050 Details : Client side always return success for setxattr call for lustre
1051 special xattr (currently only "trusted.lov").
1054 Frequency : rarely on clusters with both ia64+i386 clients
1056 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1057 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1058 clients are doing concurrent writes to the same file it is
1059 possible that the ia64 clients may LASSERT because the OST
1060 extent locks are not PAGE_SIZE aligned. Ensure that grown
1061 locks are always aligned on the request boundary.
1064 Frequency : specific use, occasional
1066 Description: Overwriting in use executable truncates on-disk binary image
1067 Details : If one node attempts to overwrite an executable in use by
1068 another node, we now correctly return ETXTBSY instead of
1069 truncating the file.
1071 Severity : enhancement
1073 Description: Async OSC create to avoid the blocking unnecessarily.
1074 Details : If a OST has no remain object, system will block on the creating
1075 when need to create a new object on this OST. Now, ways use
1076 pre-created objects when available, instead of blocking on an
1077 empty osc while others are not empty. If we must block, we block
1078 for the shortest possible period of time.
1083 Description: chmod on Lustre root is propagated to other clients
1084 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1085 invalid by the follow_mount time.
1088 Frequency : liblustre clients only
1090 Description: Race in 'instant cancel' lock handling could lead to such locks
1091 never to be granted in case of SMP MDS
1092 Details : Do not destroy not yet granted but cbpending locks in
1096 Frequency : replay/resend of open
1098 Description: non null lock assetion failure in mds_intent_policy
1099 Details : Trying to replay/resend lockless open requests resulted in
1100 mds_open() returning 0 with no lock. Now it sets a flag if
1101 a lock is going to be returned.
1103 Severity : enhancement
1105 Description: Checksum enhancements
1106 Details : New checksum enhancements allow for resending RPCs that failed
1109 Severity : enhancement
1111 Description: Tunables on number of dirty pages in cacche
1112 Details : Allow to set limit on number of dirty pages cached.
1117 Description: client crash on unmount - lock still has references
1118 Details : In some error handling cases it was possible to leak a lock
1119 reference on a client while accessing a file. This was not
1120 harmful to the client during operation, but would cause the
1121 client to crash when the filesystem is unmounted.
1124 Frequency : specific case, rare
1126 Description: ETXTBSY on mds though file not in use by client
1127 Details : ETXTBSY is no longer incorrectly returned when attempting to
1128 chmod or chown a directory that the user previously tried to
1129 execute or a currently-executing binary.
1132 Frequency : extremely rare except on liblustre-based clients
1134 Description: Lustre space not freed when files are deleted
1135 Details : Clean up open-unlinked files after client eviction. Previously
1136 the unlink was skipped and the files remained as orphans.
1141 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1142 Details : In some cases it was possible to send a blocking callback to
1143 a client doing a glimpse, even though that client didn't get
1144 a lock granted. When the glimpse lock is cancelled on the OST
1145 the freed lock is left on the waiting list and corrupted the list.
1148 Frequency : all core dumps
1150 Description: Broke core dumps to lustre
1151 Details : Negative dentry may be unhashed if parent does not have UPDATE
1152 lock, but some callers, e.g. do_coredump, expect dentry to be
1153 hashed after successful create, hash it in ll_create_it.
1155 ------------------------------------------------------------------------------
1157 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1159 * Support for kernels:
1160 2.6.9-42.0.2.EL (RHEL 4)
1161 2.6.5-7.276 (SLES 9)
1162 2.4.21-40.EL (RHEL 3)
1163 2.6.12.6 vanilla (kernel.org)
1167 Frequency : always on RHEL 3
1169 Description: Number of open files grows over time
1170 Details : The number of open files grows over time, whether or not
1171 Lustre is started. This was due to a filp leak introduced
1172 by one of our kernel patches.
1174 ------------------------------------------------------------------------------
1176 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1178 * Support for kernels:
1179 2.6.9-42.EL (RHEL 4)
1180 2.6.5-7.276 (SLES 9)
1181 2.4.21-40.EL (RHEL 3)
1182 2.6.12.6 vanilla (kernel.org)
1187 Bugzilla : 5719, 9635, 9792, 9684
1188 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1189 Details : If a server is under heavy load and cannot reply to new
1190 connection requests before the client resends the (re)connect,
1191 the connection handling code can behave badly if two service
1192 threads are concurrently handing separate (re)connections from
1193 the same client. Add better locking to the connection handling
1194 code, and ensure that only a single connection will be processed
1195 for a given client UUID, even if the lock is dropped.
1197 Severity : enhancement
1199 Description: add TCP zero-copy support to kernel
1200 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1201 sends if the hardware supports scatter-gather and checksumming.
1202 This allows socklnd to do client-write and server-read more
1203 efficiently and reduce CPU utilization from skbuf copying.
1206 Frequency : only if NFS exporting from client
1208 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1209 Details : If setting attributes on a file created under NFS that had
1210 never been opened it would be possible to oops the client
1211 if the file had no objects.
1214 Frequency : always for liblustre
1216 Description: liblustre client does MDS+OSTs setattr RPC for each write
1217 Details : When doing a write from a liblustre client, the client
1218 incorrectly issued an RPC to the MDS and each OST the file was
1219 striped over in order to update the timestamps. When writing
1220 with small chunks and many clients this could overwhelm the MDS
1221 with RPCs. In all cases it would slow down the write because
1222 these RPCs are unnecessary.
1224 Severity : enhancement
1226 Description: allow number of MDS service threads to be changed at module load
1227 Details : It is now possible to change the number of MDS service threads
1228 running. Adding "options mds mds_num_threads={N}" to the MDS's
1229 /etc/modprobe.conf will set the number of threads for the next
1230 time Lustre is restarted (assuming the "mds" module is also
1231 reloaded at that time). The default number of threads will
1232 stay the same, 32 for most systems.
1237 Description: OST crash if filesystem is unformatted or corrupt
1238 Details : If an OST is started on a device that has never been formatted
1239 or if the filesystem is corrupt and cannot even mount then the
1240 error handling cleanup routines would dereference a NULL pointer.
1245 Description: NULL pointer deref in llap_from_page.
1246 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1247 incorrect) mapping if the page was truncated/reclaimed while it was
1248 searched for. Check for this condition and skip such pages when
1249 doing readahead. Introduce extra check to llap_from_page() to
1250 verify page->mapping->host is non-NULL (so page is not anonymous).
1253 Frequency : Sometimes when using sys_sendfile
1255 Description: "page not covered by a lock" warnings from ll_readpage
1256 Details : sendfile called ll_readpage without right page locks present.
1257 Now we introduced ll_file_sendfile that does necessary locking
1258 around call to generic_file_sendfile() much like we do in
1262 Frequency : with certain MDS communication failures at client mount time
1264 Description: NULL pointer deref after failed client mount
1265 Details : a client connection request may delayed by the network layer
1266 and not be sent until after the PTLRPC layer has timed out the
1267 request. If the client fails the mount immediately it will try
1268 to clean up before the network times out the request. Add a
1269 reference from the request import to the obd device and delay
1270 the cleanup until the network drops the request.
1273 Frequency : occasionally during client (re)connect
1275 Description: assertion failure during client (re)connect
1276 Details : processing a client connection request may be delayed by the
1277 client or server longer than the client connect timeout. This
1278 causes the client to resend the connection request. If the
1279 original connection request is replied in this interval, the
1280 client may trip an assertion failure in ptlrpc_connect_interpret()
1281 which thought it would be the only running connect process.
1284 Frequency : only with obd_echo servers and clients that are rebooted
1286 Description: kernel BUG accessing uninitialized data structure
1287 Details : When running an obd_echo server it did not start the ping_evictor
1288 thread, and when a client was evicted an uninitialized data
1289 structure was accessed. Start the ping_evictor in the RPC
1290 service startup instead of the OBD startup.
1292 Severity : enhancement
1293 Bugzilla : 10193 (patchless)
1294 Description: Remove dependency on various unexported kernel interfaces.
1295 Details : No longer need reparent_to_init, exit_mm, exit_files,
1296 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1299 Frequency : rare (only users of deprecated and unsupported LDAP config)
1301 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1302 Details : LDAP apparently contains 'lustreName' attributes instead of
1303 'name'. A simple remapping of the name is sufficient.
1306 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1308 Description: waiting_locks_callback trips kernel BUG if client is evicted
1309 Details : Running with the dump_on_timeout debug flag turned on makes
1310 it possible that the waiting_locks_callback() can try to dump
1311 the Lustre kernel debug logs from an interrupt handler. Defer
1312 this log dumping to the expired_lock_main() thread.
1314 Severity : enhancement
1316 Description: Support NFS exporting on 2.6 kernels.
1317 Details : Implement non-rawops metadata methods for NFS server to use without
1318 changing NFS server code.
1321 Frequency : very rare (synthetic metadata workload only)
1323 Description: two racing renames might cause an MDS thread to deadlock
1324 Details : Running the "racer" program may cause one MDS thread to rename
1325 a file from being the source of a rename to being the target of
1326 a rename at exactly the same time that another thread is doing
1327 so, and the second thread has already enqueued these locks after
1328 doing a lookup of the target and is trying to relock them in
1329 order. Ensure that we don't try to re-lock the same resource.
1332 Frequency : only very large systems with liblustre clients
1334 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1335 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1336 all OSTs in parallel. This allows the eviction of stale liblustre
1337 clients to proceed much faster than if they were done in series,
1338 and also offers similar improvements for other set_info RPCs.
1343 Description: excessive CPU usage during initial read phase on client
1344 Details : During the initial read phase on a client, it would agressively
1345 retry readahead on the file, consuming too much CPU and impacting
1346 performance (since 1.4.5.8). Improve the readahead algorithm
1347 to avoid this, and also improve some other common cases (read
1348 of small files in particular, where "small" is files smaller than
1349 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1354 Description: MDS crash when receiving packet with unknown intent.
1355 Details : Do not LBUG in unknown intent case, just return -EFAULT
1357 Severity : enhancement
1358 Bugzilla : 9293, 9385
1359 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1360 Details : Do not serialize getattr (non-intent version) and statfs.
1363 Frequency : occasional, when OST network is overloaded/intermittent
1365 Description: client evicted by OST after bulk IO timeout
1366 Details : If a client sends a bulk IO request (read or write) the OST
1367 may evict the client if it is unresposive to its data GET/PUT
1368 request. This is incorrect if the network is overloaded (takes
1369 too long to transfer the RPC data) or dropped the OST GET/PUT
1370 request. There is no need to evict the client at all, since
1371 the pinger and/or lock callbacks will handle this, and the
1372 client can restart the bulk request.
1375 Frequency : Always when mmapping file with no objects
1377 Description: client crashes when mmapping file with no objects
1378 Details : Check that we actually have objects in a file before doing any
1379 operations on objects in ll_vm_open, ll_vm_close and
1385 Description: Request leak when working with deleted CWD
1386 Details : Introduce advanced request refcount tracking for requests
1387 referenced from lustre intent.
1389 Severity : Enhancement
1391 Description: Cache open file handles on client.
1392 Details : MDS now will return special lock along with openhandle, if
1393 requested and client is allowed to hold openhandle, even if unused,
1394 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1395 closing files for every read/write openration.
1397 Severity : Enhancement
1399 Description: Cache open negative dentries on client when possible.
1400 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1401 negative dentries on lock revocation.
1406 Description: Remounting a client read-only wasn't possible with a zconf mount
1407 Details : It wasn't possible to remount a client read-only with llmount.
1409 Severity : enhancement
1410 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1411 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1412 MPICH is included to add Lustre-specific ADIO interfaces.
1413 This is based closely on the UFS ADIO layer and only differs
1414 in file creation, in order to allow the OST striping to be set.
1415 This is user-contributed code and not supported by CFS.
1420 Description: extended inode attributes (immutable, append-only) work improperly
1421 when 2.4 and 2.6 kernels are used on client/server or vice versa
1422 Details : Introduce kernel-independent values for these flags.
1424 Severity : enhancement
1427 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1428 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1429 granularity. Now, Fractional MB granularity is be supported,
1430 this is very useful for low memory system.
1432 Severity : enhancement
1434 Description: Getattr by fid
1435 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1436 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1437 have locks and avoids getattr rpc requests for it on every stat.
1440 Frequency : Always, for filesystems larger than 2TB
1442 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1443 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1444 filesystems larger than 2TB to be mounted with mballoc enabled.
1447 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1449 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1450 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1451 filesystems larger than 2TB is mounted then the kernel will
1452 silently corrupt the start of the filesystem. CONFIG_LBD is
1453 enabled for all CFS-supported kernels, but the possibility of
1454 this happening with a modified kernel config exists.
1456 Severity : enhancement
1458 Description: add client O_DIRECT support for 2.6 kernels
1459 Details : It is now possible to do O_DIRECT reads and writes to files
1460 in the Lustre client mountpoint on 2.6 kernel clients.
1462 Severity : enhancement
1464 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1465 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1466 parallel, not waiting for response from every OST before sending
1467 a rpc to the next OST.
1472 Description: setattr vs write race when updating file timestamps
1473 Details : Client processes that update a file timestamp into the past
1474 right after writing to the file (e.g. tar) it is possible that
1475 the updated file modification time can be reset to the current
1476 time due to a race between processing the setattr and write RPC.
1478 Severity : enhancement
1480 Description: Bring 'lfs find' closer in line with regular Linux find.
1481 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1482 -print0 options and obtains all the needed info through the lustre
1485 Severity : enhancement
1487 Description: support up to 1024 configured devices on one node
1488 Details : change obd_dev array from statically allocated to dynamically
1489 allocated structs as they are first used to reduce memory usage
1494 Description: Flush dirty partially truncated pages during truncate
1495 Details : Immediatelly flush partially truncated pages in filter_setattr,
1496 this way we completely avoid having any pages in page cache on OST
1497 and can retire ugly workarounds during writes to flush such pages.
1502 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1503 Details : protect inode from truncation within vfs_unlink() context
1504 just take a reference before calling vfs_unlink() and release it
1505 when parent's i_sem is free.
1510 Description: last_id value checked outside lock on OST caused LASSERT failure
1511 Details : If there were multiple MDS->OST object precreate requests in
1512 flight, it was possible that the OST's last object id was checked
1513 outside a lock and incorrectly tripped an assertion. Move checks
1514 inside locks, and discard old precreate requests.
1517 Frequency : always, if extents are used on OSTs
1519 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1520 Details : OSTs using the extents format would not zero the high 16 bits of
1521 the index physical block number. This is not a problem for any
1522 OST filesystems smaller than 16TB, and no kernels support ext3
1523 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1524 new/modified files) and can be fixed for existing filesystems
1525 with e2fsprogs-1.39-cfs1.
1530 Description: import connection selection may be incorrect if timer wraps
1531 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1532 import connections to be ignored if the 32-bit jiffies counter
1533 wraps. Use a 64-bit jiffies counter.
1536 Frequency : very large clusters immediately after boot
1538 Description: LNET request buffers exhausted under heavy short-term load
1539 Details : If a large number of client requests are generated on a service
1540 that has previously never seen so many requests it is possible
1541 that the request buffer growth cannot keep up with the spike in
1542 demand. Instead of dropping incoming requests, they are held in
1543 the LND until the RPC service can accept more requests.
1546 Frequency : Sometimes during replay
1548 Description: Assertion failure in ll_local_open after replay.
1549 Details : If replay happened on an open request reply before we were able
1550 to set replay handler, reply will become not swabbed tripping the
1551 assertion in ll_local_open. Now we set the handler right after
1552 recognising of open request
1555 Frequency : very rare
1557 Description: kernel reports "badness in vsnprintf"
1558 Details : Reading from the "recovery_status" /proc file in small chunks
1559 may cause a negative length in lprocfs_obd_rd_recovery_status()
1560 call to vsnprintf() (which is otherwise harmless). Exit early
1561 if there is no more space in the output buffer.
1563 Severity : enhancement
1565 Description: clear OBD RPC statistics by writing to them
1566 Details : It is now possible to clear the OBD RPC statistics by writing
1567 to the "stats" file.
1572 Description: Client mtime is not the same on different clients after utimes
1573 Details : In some cases, the client was using the utimes() syscall on
1574 a file cached on another node. The clients now validate the
1575 ctime from the MDS + OSTs to determine which one is right.
1580 Description: Inability to activate failout mode
1581 Details : lconf script incorrectly assumed that in pythong string's numeric
1582 value is used in comparisons.
1585 Frequency : always with multiple stripes per file
1587 Description: Inefficient object allocation for mutli-stripe files
1588 Details : When selecting which OSTs to stripe files over, for files with
1589 a stripe count that divides evenly into the number of OSTs,
1590 the MDS is always picking the same starting OST for each file.
1591 Return the OST selection heuristic to the original design.
1596 Description: mount failures may take full timeout to return an error
1597 Details : Under some heavy load conditions it is possible that a
1598 failed mount can wait for the full obd_timeout interval,
1599 possibly several minutes, before reporting an error.
1600 Instead return an error as soon as the status is known.
1602 Frequency : quota enabled and large files being deleted
1604 Description: releasing more than 4GB of quota at once hangs OST
1605 Details : If a user deletes more than 4GB of files on a single OST it
1606 will cause the OST to spin in an infinite loop. Release
1607 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1612 Description: statfs data retrieved from /proc may be stale or zero
1613 Details : When reading per-device statfs data from /proc, in the
1614 {kbytes,files}_{total,free,avail} files, it may appear
1615 as zero or be out of date.
1618 Frequency : systems with MD RAID1 external journal devices
1620 Description: lconf's call to blkid is confused by RAID1 journal devices
1621 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1622 of returning all block devices that match the journal UUID.
1625 Frequency : always, for aggregate stripe size over 4GB
1627 Description: assertion fails when trying to use 4GB stripe size
1628 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
1629 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
1632 Frequency : always on ppc64
1634 Description: the first write on an ext3 filesystem with mballoc got stuck
1635 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1636 perform endianness conversion.
1638 ------------------------------------------------------------------------------
1640 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1642 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1643 INTEROPERATE with older versions automatically. Please read the
1644 user documentation before upgrading any part of a live system.
1645 * WARNING: Lustre networking configuration changes are required with
1646 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1649 * Support for kernels:
1650 2.6.9-22.0.2.EL (RHEL 4)
1651 2.6.5-7.244 (SLES 9)
1652 2.6.12.6 vanilla (kernel.org)
1655 Severity : enhancement
1656 Bugzilla : 7981/8208
1657 Description: Introduced Lustre Networking (LNET)
1658 Details : LNET is new networking infrastructure for Lustre, it includes
1659 a reorganized network configuration mode (see the user
1660 documentation for full details) as well as support for routing
1661 between different network fabrics. Lustre Networking Devices
1662 (LNDs) for the supported network fabrics have also been
1663 created for this new infrastructure.
1665 Severity : enhancement
1666 Description: Introduced Access control lists
1667 Details : clients can set ACLs on files and directories in order to have
1668 more fine-grained permissions than the standard Unix UGO+RWX.
1669 The MDS must be started with the "-o acl" mount option.
1671 Severity : enhancement
1672 Description: Introduced filesystem quotas
1673 Details : Administrators may now establish per-user quotas on the
1676 Severity : enhancement
1678 Description: Configuration change for the XT3
1679 The PTLLND is now used to run Lustre over Portals on the XT3
1680 The configure option(s) --with-cray-portals are no longer used.
1681 Rather --with-portals=<path-to-portals-includes> is used to
1682 enable building on the XT3. In addition to enable XT3 specific
1683 features the option --enable-cray-xt3 must be used.
1688 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1689 Details : A race between allocating a new llcd and re-getting the llcd_lock
1690 allowed another thread to grab newly-allocated llcd.
1692 Severity : enhancement
1694 Description: 2.6 OST async journal commit and locking fix to improve performance
1695 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1696 2.6 kernels are now async as they already were in 2.4 kernels so
1697 that they can commit concurrently with the network bulk transfer.
1698 For block-allocated files the filter allocation semaphore is held
1699 to avoid filesystem fragmentation during allocation. BKL lock
1700 removed for 2.6 xattr operations where it is no longer needed.
1705 Description: lconf incorrectly determined whether two IP networks could talk
1706 Details : In some more complicated routing and multiple-network
1707 configurations, lconf will avoid trying to make a network
1708 connection to a disjoint part of the IP space. It was doing the
1709 math incorrectly for one set of cases.
1714 Description: Fix for potential infinite loop processing records in an llog.
1715 Details : If an llog record is corrupted/zeroed, it is possible to loop
1716 forever in llog_process(). Validate the llog record length
1717 and skip the remainder of the block on error.
1720 Frequency : occasional (liblustre only)
1722 Description: liblustre could not open files whose last component is a symlink
1723 Details : sysio_path_walk() would incorrectly pass the open intent to
1724 intermediate path components.
1727 Frequency : rare (liblustre only with non-standard tuning)
1728 Bugzilla : 7201 (7350)
1729 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1730 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1731 after it has been released, only do this for referenced locks
1733 Severity : enhancement
1735 Description: specifying an (invalid) directory default stripe_size of -1
1736 would reset the directory default striping
1737 Details : stripe_size -1 was used internally to signal directory stripe
1738 removal, now use "all default" to signal dir stripe removal
1739 as a directory striping of "all default" is not useful
1742 Frequency : common for large clusters running liblustre clients
1744 Description: doing an ls when liblustre clients are running is slow
1745 Details : sending a glimpse AST to a liblustre client waits for every AST
1746 to time out, as liblustre clients will not respond. Since they
1747 cannot cache data we refresh the OST lock LVB from disk instead.
1749 Severity : enhancement
1751 Description: doing an ls at the same time as file IO can be slow
1752 Details : enqueue and other "small" requests can be blocked behind many
1753 large IO requests. Create a new OST IO portal for non-IO
1754 requests so they can be processed faster.
1757 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1759 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1760 created file with mode 000
1761 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1762 allow an ftruncate() as a non-root user to a file with mode 000.
1763 Lustre now respects this flag to disable mode checks when
1764 truncating a file owned by the user
1767 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1770 Description: Revoking locks from clients that went dead or catatonic might take
1772 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1773 cancellation of such locks instant on servers without waiting for
1774 any reply from clients. Clients drops these locks when cancel
1775 notification from server is received without replying.
1778 Frequency : liblustre-only, when liblustre client dies or becomes busy
1780 Description: Doing ls on Linux clients can take a long time with active
1782 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1783 granting such locks to it in the first place if possible. Locks
1784 are taken by proxy on the OST during the read or write and
1785 dropped immediately afterward. Add connect flags handling, do
1786 not grant locks to liblustre clients for glimpse ASTs.
1788 Severity : enhancement
1790 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1791 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1792 sequential reads and behaves badly with applications doing random
1793 reads. Improve it by reading ahead at least read region, and
1794 avoiding excessive large RPC for small reads.
1796 Severity : enhancement
1798 Description: Creating more than 1000 files for a single job may cause a load
1799 imbalance on the OSTs if there are also a large number of OSTs.
1800 Details : qos_prep_create() uses an OST index reseed value that is an
1801 even multiple of the number of available OSTs so that if the
1802 reseed happens in the middle of the object allocation it will
1803 still utilize the OSTs as uniformly as possible.
1808 Description: OST or MDS may oops in ping_evictor_main()
1809 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1810 but doesn't restart at beginning of obd_exports_timed list
1813 Severity : enhancement
1815 Description: improve by-nid export eviction on the MDS and OST
1816 Details : allow multiple exports with the same NID to be evicted at one
1817 time without re-searching the exports list.
1820 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1822 Description: MDS may oops in groups_free()
1823 Details : in rare race conditions a newly allocated group_info struct is
1824 freed again, and this can be NULL. The 2.4 compatibility code
1825 for groups_free() checked for a NULL pointer, but 2.6 did not.
1828 Frequency : common for liblustre clients doing little filesystem IO
1829 Bugzilla : 9352, 7313
1830 Description: server may evict liblustre clients accessing contended locks
1831 Details : if a client is granted a lock or receives a completion AST
1832 with a blocking AST already set it would not reply to the AST
1833 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1836 Frequency : lfs setstripe, only systems with more than 160 OSTs
1838 Description: unable to set striping with a starting offset beyond OST 160
1839 Details : llapi_create_file() incorrectly limited the starting stripe
1840 index to the maximum single-file stripe count.
1843 Frequency : LDAP users only
1845 Description: lconf did not handle in-kernel recovery with LDAP properly
1846 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1848 Severity : enhancement
1850 Description: bind OST threads to NUMA nodes to improve performance
1851 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1852 node and do their allocations there to localize memory access
1854 Severity : enhancement
1856 Description: llmount can determine client NID directly from Myrinet (GM)
1857 Details : the client NID code from gmnalnid was moved directly into
1858 llmount, removing the need to use this or specifying the
1859 client NID explicitly when mounting GM clients with zeroconf
1862 Frequency : if client is started with down MDS
1864 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1865 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1866 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1867 ptlrpc_import_delayed_req() because we didn't check if the
1868 request was interrupted, and we also didn't break out of the
1869 event loop if there was no timeout
1874 Description: data loss during non-page-aligned writes to a single file from
1875 both multiple nodes and multiple threads on one node at same time
1876 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1877 inconsistency led to false short-reads, that were cached and later
1878 used by ->prepare_write() to fill in partially written page,
1879 leading to data loss.
1882 Frequency : always, if lconf --abort_recovery used
1884 Description: lconf --abort_recovery fails with 'Operation not supported'
1885 Details : lconf was attempting to abort recovery on the MDT device and not
1888 Severity : enhancement
1890 Description: remove cleanup logs
1891 Details : replace lconf-generated cleanup logs with lustre internal
1892 cleanup routines. Eliminates the need for client-cleanup and
1895 Severity : enhancement
1897 Description: add support for EAs (user and system) on lustre filesystems
1898 Details : it is now possible to store extended attributes in the Lustre
1899 client filesystem, and with the user_xattr mount option it
1900 is possible to allow users to store EAs on their files also
1902 Severity : enhancement
1904 Description: Add possibility (config option) to show minimal available OST free
1906 Details : When compiled with --enable-mindf configure option, statfs(2)
1907 (and so, df) will return least minimal free space available from
1908 all OSTs as amount of free space on FS, instead of summary of
1909 free spaces of all OSTs.
1911 Severity : enhancement
1913 Description: do not expand extent locks acquired on OST-side
1914 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1915 by server: they are not cached anyway.
1918 Frequency : when mmap is used/binaries executed from Lustre
1920 Description: Unmmap pages before throwing them away from read cache.
1921 Details : llap_shrink cache now attempts to unmap pages before discarding
1922 them (if unmapping failed - do not discard). SLES9 kernel has
1923 extra checks that trigger if this unmapping is not done first.
1928 Description: lconf didn't resolve symlinks before checking to see whether a
1929 given mountpoint was already in use
1932 Frequency : when migrating failover services
1933 Bugzilla : 6395, 9514
1934 Description: When migrating a subset of services from a node (e.g. failback
1935 from a failover service node) the remaining services would
1936 time out and evict clients.
1937 Details : lconf --force (implied by --failover) sets the global obd_timeout
1938 to 5 seconds in order to quickly disconnect, but this caused
1939 other RPCs to time out too quickly. Do not change the global
1940 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1942 Severity : enhancement
1943 Frequency : if MDS is started with down OST
1944 Bugzilla : 9439,5706
1945 Description: Allow startup/shutdown of an MDS without depending on the
1946 availability of the OSTs.
1947 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1948 Add appropriate locking and lov-osc refcounts for safe
1949 cleaning. Add osc abort_inflight calls in case the
1950 synchronize never started.
1953 Frequency : occasional (Cray XT3 only)
1955 Description: root not authorized to access files in CRAY_PORTALS environment
1956 Details : The client process capabilities were not honoured on the MDS in
1957 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1958 been accessed by an authorized user then root was able to access
1959 the file on the local client also. The root user capabilities
1960 are now allowed on the MDS, as this environment has secure UID.
1963 Frequency : occasional
1965 Description: ldiskfs "too long searching" message happens too often
1966 Details : A debugging message (otherwise harmless) prints too often on
1967 the OST console. This has been reduced to only happen when
1968 there are fragmentation problems on the filesystem.
1973 Description: Division by zero in statfs when all OSCs are inactive
1974 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
1975 lov_statfs divides by value returned by lov_get_stripecnt().
1979 Bugzilla : 9489, 3273
1980 Description: First write from each client to each OST was only 4kB in size,
1981 to initialize client writeback cache, which caused sub-optimal
1982 RPCs and poor layout on disk for the first writen file.
1983 Details : Clients now request an initial cache grant at (re)connect time
1984 and so that they can start streaming writes to the cache right
1985 away and always do full-sized RPCs if there is enough data.
1986 If the OST is rebooted the client also re-establishes its grant
1987 so that client cached writes will be honoured under the grant.
1992 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
1993 Details : Now I/O RPCs go to different portal number and (presumably) fast
1994 lock requests (and glimses) and other RPCs get their own service
1995 threads pool that should be able to service those RPCs
1998 Severity : enhancement
2000 Description: Ability to exchange lustre version between client and servers and
2001 issue warnings at client side if client is too old. Also for
2002 liblustre clients there is ability to refuse connection of too old
2004 Details : New 'version' field is added to connect data structure that is
2005 filled with version info. That info is later checked by server and
2009 Frequency : rare, liblustre only.
2010 Bugzilla : 9296, 9581
2011 Description: Two simultaneous writes from liblustre at offset within same page
2012 might proceed at the same time overwriting eachother with stale
2014 Details : I/O lock withing llu_file_prwv was released too early, before data
2015 actually was hitting the wire. Extended lock-holding time until
2016 server acknowledges receiving data.
2019 Frequency : extremely rare. Never observed in practice.
2021 Description: avoid generating lustre_handle cookie of 0.
2022 Details : class_handle_hash() generates handle cookies by incrementing
2023 global counter, and can hit 0 occasionaly (this is unlikely, but
2024 not impossible, because initial value of cookie counter is
2025 selected randonly). Value of 0 is used as a sentinel meaning
2026 "unassigned handle" --- avoid it. Also coalesce two critical
2027 sections in this function into one.
2029 Severity : enhancement
2031 Description: allow liblustre clients to delegate truncate locking to OST
2032 Details : To avoid overhead of locking, liblustre client instructs OST to
2033 take extent lock in ost_punch() on client's behalf. New connection
2034 flag is added to handle backward compatibility.
2036 Severity : enhancement
2037 Bugzilla : 4928, 7341, 9758
2038 Description: allow number of OST service threads to be specified
2039 Details : a module parameter allows the number of OST service threads
2040 to be specified via "options ost ost_num_threads={N}" in the
2041 OSS's /etc/modules.conf or /etc/modprobe.conf.
2045 Bugzilla : 6146, 9635, 9895
2046 Description: servers crash with bad pointer in target_handle_connect()
2047 Details : In rare cases when a client is reconnecting it was possible that
2048 the connection request was the last reference for that export.
2049 We would temporarily drop the export reference and get a new
2050 one, but this may have been the last reference and the export
2051 was just destroyed. Get new reference before dropping old one.
2053 Severity : enhancement
2054 Frequency : if client is started with failover MDS
2056 Description: Allow multiple MDS hostnames in the mount command
2057 Details : Try to read the configuration from all specified MDS
2058 hostnames during a client mount in case the "primary"
2061 Severity : enhancement
2063 Description: Stop sending data to evicted clients as soon as possible.
2064 Details : Check if the client we are about to send or are sending data to
2065 was evicted already. (Check is done every second of waiting,
2066 for which l_wait_event interface was extended to allow checking
2067 of exit condition at specified intervals).
2070 Frequency : rare, normally only when NFS exporting is done from client
2072 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2074 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2077 Frequency : always, if srand() is called before liblustre initialization
2079 Description: Liblustre uses system PRNG disturbing its usage by user application
2080 Details : Introduce internal to lustre fast and high-quality PRNG for
2081 lustre usage and make liblustre and some other places in generic
2082 lustre code to use it.
2084 Severity : enhancement
2085 Bugzilla : 9477, 9557, 9870
2086 Description: Verify that the MDS configuration logs are updated when xml is
2087 Details : Check if the .xml configuration logs are newer than the config
2088 logs stored on the MDS and report an error if this is the case.
2089 Request --write-conf, or allow starting with --old_conf.
2091 Severity : enhancement
2093 Description: Handle symlinks in the path when checking if Lustre is mounted.
2094 Details : Resolve intermediate symlinks when checking if a client has
2095 mounted a filesystem to avoid duplicate client mounts.
2100 Description: lconf can hit an error exception but still return success.
2101 Details : The lconf command catches the Command error exception at the top
2102 level script context and will exit with the associated exit
2103 status, but doesn't ensure that this exit status is non-zero.
2108 Description: failure of ptlrpc thread startup can cause oops
2109 Details : Starting a ptlrpc service thread can fail if there are a large
2110 number of threads or the server memory is very fragmented.
2111 Handle this without oopsing.
2114 Frequency : always, only if liblustre and non-default acceptor port was used
2116 Description: liblustre cannot connect to servers with non-default acceptor port
2117 Details : tcpnal_set_default_params() was not called and was therefore
2118 ignoring the environment varaible TCPNAL_PORT, as well as other
2119 TCPNAL_ environment variables
2124 Description: two objects could be created on the same OST for a single file
2125 Details : If an OST is down, in some cases it was possible to create two
2126 objects on a single OST for a single file. No problems other
2127 than potential performance impact and spurious error messages.
2131 Bugzilla : 5681, 9562
2132 Description: Client may oops in ll_unhash_aliases
2133 Details : Client dcache may become inconsistent in race condition.
2134 In some cases "getcwd" can fail if the current directory is
2140 Description: Inode refcounting problems in NFS export code
2141 Details : link_raw functions used to call d_instantiate without obtaining
2142 extra inode reference first.
2146 Bugzilla : 9942, 9903
2147 Description: Referencing freed requests leading to crash, memleaks with NFS.
2148 Details : We used to require that call to ll_revalidate_it was always
2149 followed by ll_lookup_it. Also with revalidate_special() it is
2150 possible to call ll_revalidate_it() twice for the same dentry
2151 even if first occurence returned success. This fix changes semantic
2152 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2153 reference on a request referred from the intent.
2154 ll_intent_release() then releases such a request.
2157 Frequency : rare, normally benchmark loads only
2159 Description: unlinked inodes were kept in memory on the client
2160 Details : If a client is repeatedly creating and unlinking files it
2161 can accumulate a lot of stale inodes in the inode slab cache.
2162 If there is no other client load running this can cause the
2163 client node to run out of memory. Instead flush old inodes
2164 from client cache that have the same inode number as a new inode.
2167 Frequency : SLES9 2.6.5 kernel and long filenames only
2168 Bugzilla : 9969, 10379
2169 Description: utime reports stale NFS file handle
2170 Details : SLES9 uses out-of-dentry names in some cases, which confused
2171 the lustre dentry revalidation. Change it to always use the
2175 Frequency : rare, unless heavy write-truncate concurrency is continuous
2176 Bugzilla : 4180, 6984, 7171, 9963, 9331
2177 Description: OST becomes very slow and/or deadlocked during object unlink
2178 Details : filter_destroy() was holding onto the parent directory lock
2179 while truncating+unlinking objects. For very large objects this
2180 may block other threads for a long time and slow overall OST
2181 responsiveness. It may also be possible to get a lock ordering
2182 deadlock in this case, or run out of journal credits because of
2183 the combined truncate+unlink. Solution is to do object truncate
2184 first in one transaction without parent lock, and then do the
2185 final unlink in a new transaction with the parent lock. This
2186 reduces the lock hold time dramatically.
2189 Frequency : rare, 2.4 kernels only
2191 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2192 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2193 before unmounting their filesystem in order to avoid deadlock.
2194 The kernel_locked() function in 2.4 kernels only checks whether
2195 the kernel lock is held, not whether it is this process that is
2196 holding it as 2.6 kernels do.
2201 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2202 Details : The client ptlrpc code may be trying to reconnect to a down
2203 server before a previous connection attempt has timed out.
2204 Increase the reconnect interval to be longer than the connection
2205 timeout interval to avoid sending duplicate connections to
2209 Frequency : echo_client brw_test command
2211 Description: fix echo_client to work with OST preallocated code
2212 Details : OST preallocation code (5137) didn't take echo_client IO path
2213 into account: echo_client calls filter methods outside of any
2214 OST thread and, hence, there is no per-thread preallocated
2215 pages and buffers to use. Solution: hijack pga pages for IO. As
2216 a byproduct, this avoids unnecessary data copying.
2220 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2221 Description: Client can oops in mdc_commit_close() after open replay
2222 Details : It was possible for the MDS to return an open request with no
2223 transaction number in mds_finish_transno() if the client was
2224 evicted, but without actually returning an error. Clients
2225 would later try to replay that open and may trip an assertion
2226 Simplify the client close codepath, and always return an error
2227 from the MDS in case the open is not successful.
2230 Frequency : rare, 2.6 OSTs only
2232 Description: OST may deadlock under high load on fragmented files
2233 Details : If there was a heavy load and highly-fragmented OST filesystems
2234 it was possible to have all the OST threads deadlock waiting on
2235 allocation of biovecs, because the biovecs were not released
2236 until the entire RPC IO was completed. Instead, release biovecs
2237 as soon as they are complete to ensure forward IO progress.
2239 Severity : enhancement
2241 Description: Support for specifying external journal device at mount
2242 Details : If an OST or MDS device is formatted with an external journal
2243 device, this device major/minor is stored in the ext3 superblock
2244 and may not be valid for failover. Allow detecting and
2245 specifying the external journal at mount time.
2250 Description: Mounting an MDS with pending unlinked files may cause oops
2251 Details : target_finish_recovery() calls mds_postrecov() which returned
2252 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2253 considers this an error and immediately begins cleaning up the
2254 lov, just after starting the mds_lov process
2256 Severity : enhancement
2258 Description: Implement 'lfs df' to report actual free space on per-OST basis
2259 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2260 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2261 usage of objects; '-h' to report in human readable format.
2263 ------------------------------------------------------------------------------
2265 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2272 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2273 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2274 at startup during orphan unlinking. Instead, skip buddy update
2275 before setup, as it will be regenerated after recovery is complete.
2280 Description: If an OST is inactive, its locks might reference stale inodes.
2281 Details : lov_change_cbdata() must iterate over all namespaces, even if
2282 they are inactive to clear inode references from the lock.
2284 Severity : enhancement
2285 Frequency : occasional, if non-standard max_dirty_mb used
2287 Description: Client will block write RPCs if not enough grant
2288 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2289 then the client will block writes while waiting for another RPC
2290 to complete instead of consuming its dirty limit. With change
2291 we get improved performance when max_dirty_mb is small.
2293 Severity : enhancement
2294 Bugzilla : 3389, 6253
2295 Description: Add support for supplementary groups on the MDS.
2296 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2297 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2298 lookups for user supplementary groups into a cache.
2302 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2303 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2304 for liblustre O_EXCL is enforced by MDS.
2308 Description: readdir never returns NULL in liblustre.
2309 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2310 dentry in struct dirent.
2314 Frequency : liblustre only, depends on application IO pattern
2315 Description: liblustre clients evicted if not contacting servers
2316 Details : Don't put liblustre clients into the ping_evictor list, so
2317 they will not be evicted by the pinger ever.
2319 Severity : enhancement
2321 Description: Add ability to evict clients by NID from MDS.
2322 Details : By echoing "nid:$NID" string into
2323 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2324 $NID would be instantly evicted from this MDS and from all active
2325 OSTs connected to it.
2329 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2330 Details : lookup_it_finish() used to query file size from OSTs that was not
2335 Description: service threads change working directory to that of init
2336 Details : Starting lustre service threads may pin the working directory
2337 of the parent thread, making that filesystem busy. Threads
2338 now change to the working directory of init to avoid this.
2342 Frequency : during shutdown only
2343 Description: shutdown with a failed MDS or OST can cause unmount to hang
2344 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2349 Frequency : 2.6 only
2350 Description: chmod/chown may include an extra supplementary group
2351 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2352 supplementary group and if none is specified this is used.
2355 Bugzilla : 5479 (6816)
2357 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2358 Details : If lookup is in progress on a file that is unlinked we might try
2359 to revalidate the inode and fail in revalidate after lookup is
2360 complete and ll_file_open() enqueues the open again but
2361 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2364 Frequency : always, if lconf --abort_recovery used
2366 Description: lconf --abort_recovery fails with 'Operation not supported'
2367 Details : lconf was attempting to abort recovery on the MDT device and not
2370 ------------------------------------------------------------------------------
2372 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2377 Frequency : rare (only unsupported configurations with a node running as an
2379 Bugzilla : 6514, 5137
2380 Description: Mounting a Lustre file system on a node running as an OST could
2382 Details : OSTs now preallocates memory needed to write out data at
2383 startup, instead of when needed, to avoid having to
2384 allocate memory in possibly low memory situations.
2385 Specifically, if the file system is mounted on on OST,
2386 memory pressure could force it to try to write out data,
2387 which it needed to allocate memory to do. Due to the low
2388 memory, it would be unable to do so and the node would
2389 become unresponsive.
2391 Severity : enhancement
2393 Description: Addition of lconf --service command line option
2394 Details : lconf now accepts a '--service <arg>' option, which is
2395 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2397 Severity : enhancement
2399 Description: Failover mode is now the default for OSTs.
2400 Details : By default, OSTs will now run in failover mode. To return to
2401 the old behaviour, add '--failout' to the lmc line for OSTs.
2403 Severity : enhancement
2405 Description: Health checks are now provided for MDS and OSTs
2406 Details : Additional detailed health check information on MSD and OSTs
2407 is now provided through the procfs health_check value.
2410 Frequency : occasional, depends on IO load
2412 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2413 after numerous create/delete cycles
2414 Details : The ext3 inode allocation policy would not allocate new inodes
2415 very well on the OSTs because there are no new directories
2416 being created. Instead we look for groups with free space if
2417 the parent directories are nearly full.
2422 Description: Network or server problems during mount may cause partially
2423 mounted clients instead of returning an error.
2424 Details : The config llog parsing code may overwrite the error return
2425 code during mount error handling, returning success instead
2431 Description: MDS can fail to allocate large reply buffers
2432 Details : After long uptimes the MDS can fail to allocate large reply
2433 buffers (e.g. zconf client mount config records) due to memory
2434 fragmentation or consumption by the buffer cache. Preallocate
2435 some large reply buffers so that these replies can be sent even
2436 under memory pressure.
2440 Frequency : rare (liblustre)
2441 Description: fsx running with liblustre complained that using truncate() to
2442 extend the file doesn't work. This patch corrects that issue.
2443 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2444 ATTR_SIZE and lsm use in llu_setattr_raw.
2448 Frequency : rare, only 2.6 kernels
2449 Description: Unusual file access patterns on the MDS may result in inode
2450 data being lost in very rare circumstances.
2451 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2452 code in the 2.6 kernel caused the inode and/or EA data not to
2453 be read from disk, causing single-file corruption.
2457 Frequency : rare, only 2.6 filesystems using extents
2458 Description: Heavy concurrent write and delete load may cause data corruption.
2459 Details : It was possible under high-load situations to have an extent
2460 metadata block in the block device cache from a just-unlinked
2461 file overwrite a newly-allocated data block. We now unmap any
2462 metadata buffers that alias just-allocated data blocks.
2466 Frequency : filesystems with default stripe_count larger than 77
2467 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2468 Details : lconf specifies an inode size of 4096 bytes when the default
2469 stripe_count is larger than 77. This conflicts with the default
2470 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2471 this case to avoid pinning too much memory for large EAs.
2473 ------------------------------------------------------------------------------
2475 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2480 Frequency : rare (extremely heavy IO load with hundreds of clients)
2482 Description: Client is evicted, gets IO error writing to file
2483 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2484 caused clients to be evicted for AST timeouts. The fixes in
2485 bug 5192 mean we no longer need to have such short AST timeouts
2486 so ldlm_timeout has been increased.
2489 Frequency : occasional during --force or --failover shutdown under load
2490 Bugzilla : 5949, 4834
2491 Description: Server oops/LBUG if stopped with --force or --failover under load
2492 Details : a collection of import/export refcount and cleanup ordering
2493 issues fixed for safer force cleanup
2496 Frequency : only filesystems larger than 120 OSTs
2497 Bugzilla : 5990, 6223
2498 Description: lfs getstripe would oops on a very large filesystem
2499 Details : lov_getconfig used kfree on vmalloc'd memory
2502 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2503 Bugzilla : 6242, 6243
2504 Description: reading from files that had been truncated to a non-zero size
2505 but never opened returned no data
2506 Details : ll_file_read() reads zeros from no-object files to EOF
2511 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2512 Details : There's little harm in aborting MDS/OSS recovery and letting it
2513 try again, so I removed the LASSERT and return an error instead.
2515 Severity : enhancement
2517 Description: New debugging infrastructure for tracking down data corruption
2518 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2519 (b) cover more of the client-side code path, and (c) try to narrow
2520 down where problems occurred
2524 Bugzilla : 3819, 4364, 4397, 6313
2525 Description: Racing close and eviction MDS could cause assertion in mds_close
2526 Details : It was possible to get multiple mfd references during close and
2527 client eviction, leading to one thread referencing a freed mfd.
2529 Severity: : enhancement
2530 Bugzilla : 3262, 6359
2531 Description: Attempts to reconnect to servers are now more aggressive.
2532 Details : This builds on the enhanced upcall-less recovery that was added
2533 in 1.4.2. When trying to reconnect to servers, clients will
2534 now try each server in the failover group every 10 seconds. By
2535 default, clients would previously try one server every 25 seconds.
2540 Description: After recovery, certain operations trigger a failed
2541 assertion on a client.
2542 Details : Failing over an mds, using lconf -d --failover, while a
2543 client was doing a readdir() call would cause the client to
2544 LBUG after recovery completed and the readdir() was resent.
2546 Severity : enhancement
2548 Description: Default groups are now added by lconf
2549 Details : You can now run lconf --group <servicename> without having to
2550 manually add groups with lmc.
2553 Frequency : occasional
2555 Description: Nodes with an elan id of 0 trigger a failed assertion
2558 Frequency : always when accessing e.g. tty/console device nodes
2560 Description: tty and some other devices nodes cannot be used on lustre
2561 Details : file's private_data field is used by device data and lustre
2562 values in there got lost. New field was added to struct file to
2563 store fs-specific private data.
2566 Frequency : when exporting Lustre via NFS
2568 Description: NFSD failed occasionally when looking up a path component
2569 Details : NFSD is looking up ".." which was broken in ext3 directories
2570 that had grown large enough to become hashed.
2573 Frequency : Clusters with multiple interfaces not on the same subnet
2575 Description: Nodes will repeatedly try to reconnect to an interface which it
2576 cannot reach and report an error to the log.
2577 Details : Extra peer list entries will be created by lconf with some peers
2578 unreachable. lconf now validates the peer before adding it.
2581 Frequency : Only if a default stripe is set on the filesystem root.
2583 Description: Setting a default stripe on the filesystem root prevented the
2584 filesystem from being remounted.
2585 Details : The client was sending extra request flags in the root getattr
2586 request and did not allocate a reply buffer for the dir EA.
2589 Frequency : occasional, higher if lots of files are accessed by one client
2590 Bugzilla : 6159, 6097
2591 Description: Client trips assertion regarding lsm mismatch/magic
2592 Details : While revalidating inodes the VFS looks up inodes with ifind()
2593 and in rare cases can find an inode that is being freed.
2594 The ll_test_inode() code will free the lsm during ifind()
2595 when it finds an existing inode and then the VFS later attaches
2596 this free lsm to a new inode.
2600 Bugzilla : 6422, 7030
2601 Description: MDS deadlock between mkdir and client eviction
2602 Details : Creating a new file via mkdir or mknod (starting a transaction
2603 and getting the ns lock) can deadlock with client eviction
2604 (gets ns lock and trying to finish a synchronous transaction).
2607 Frequency : occasional
2608 Description: While starting a server, the fsfilt_ext3 module could not be
2610 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2611 kernels. Previously, lconf would still use the ext3 name
2612 when trying to load modules. Now, it will correctly use
2613 ext3 on 2.4 and ldiskfs on 2.6.
2615 Severity : enhancement
2616 Description: The default stripe count has been changed to 1
2617 Details : The interpretation of the default stripe count (0, to lfs
2618 or lmc) has been changed to mean striping across a single
2619 OST, rather than all available. For general usage we have
2620 found a stripe count of 1 or 2 works best.
2622 Severity : enhancement
2623 Description: Add support for compiling against Cray portals.
2624 Details : Conditional compiling for some areas that are different
2628 Frequency : occasional
2629 Bugzilla : 6409, 6834
2630 Description: Creating files with an explicit stripe count may lead to
2631 a failed assertion on the MDS
2632 Details : If some OSTs are full or unavailable, creating files may
2633 trigger a failed assertion on the MDS. Now, Lustre will
2634 try to use other servers or return an error to the
2638 Frequency : occasional
2640 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2641 caused lock timeout during readahead (since 1.4.2).
2642 Details : Processes doing readahead might match a lock that hasn't been
2643 granted yet if there are overlapping and conflicting lock
2644 requests. The readahead process waits on ungranted lock
2645 (original lock is CBPENDING), while OST waits for that process
2646 to cancel CBPENDING read lock and eventually evicts client.
2648 Severity : enhancement
2650 Description: Initial enabling of flock support for clients
2651 Details : Implements fcntl advisory locking and file status functions.
2652 This feature is provided as an optional mount flag (default
2653 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2654 locking are implemented yet, and those that are are not guaranteed
2655 to be completely correct in production environments.
2656 mount -t lustre -o [flock|noflock] ...
2659 Frequency : occasional
2661 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2662 assertion in the ext3 JBD (journaling) layer.
2663 Details : The b_committed_data struct is protected by the big kernel lock
2664 in 2.4 kernels, serializing journal_commit_transaction() and
2665 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2666 access to this struct. In 2.6 kernels there is finer grained
2667 locking to improve SMP performance of the JBD layer.
2671 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2673 -----------------------------------------------------------------------------
2675 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2677 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2678 versions of Lustre. Please update all servers and clients to
2679 version 1.4.2 or later at the same time. You must also run
2680 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2681 to update the configuration logs.
2683 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2684 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2685 - swab llog records of type '0' so we get proper header size/idx (5861)
2686 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2687 - fix rename of one directory over another leaking an inode (5953)
2688 - avoid SetPageDirty on 2.6 (5981)
2689 - don't re-add just-being-destroyed locks to the waiting list (5653)
2690 - when creating new directories, inherit the parent's custom
2691 striping settings if present parent (3048)
2692 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2693 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2694 - don't hold client locks on temporary worklist from l_lru (5666)
2695 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2696 - automatically evict dead clients (5921)
2697 - Update file size properly in create+truncate+fstat case (6196)
2698 - Do not unhash mountpoint dentries, do not allow removal of
2700 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2701 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2702 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2703 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2704 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2705 - Correctly handle reads of files with no objects (6243)
2706 - lctl recover will also mark a device active if deactivate used (5933)
2708 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2709 - allow --write-conf on an MDS with different nettype than client (5619)
2710 - don't write config llogs to MDS for mounts not from that MDS (5617)
2711 - lconf should create multiple TCP connections from a client (5201)
2712 - init scripts are now turned off by default; run chkconfig --on
2713 lustre and chkconfig --on lustrefs to use them
2714 - upcalls are no longer needed for clients to recover to failover
2716 - add --abort-recovery option to lconf to abort recovery on device
2718 - add support for an arbitrary number of OSTs (3026)
2719 - Quota support protocol changes.
2720 - forward compatibility changes to wire structs (6007)
2721 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2722 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2723 - improved "lustre status" script
2724 - initialize blocksize for non-regular files (6062)
2725 - added --disable-server and --disable-client configure options (5782)
2726 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2727 - Vanilla 2.4.29 support
2728 - increase maximum number of obd devices to 520 (6242)
2729 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2730 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2731 SLES 9 2.6.5 kernels
2732 - limit stripes per file to 160 (the maximum EA size) (6093)
2734 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2737 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2738 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2739 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2740 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2741 - small ext3 extents cleanups and fixes (5733)
2742 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2743 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2744 - avoid needless client->OST connect, fix handle mismatch (5317)
2745 - fix DLM error path that led to out-of-sync client, long delays (5779)
2746 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2747 - fix several locking issues related to i_size (5492,5624,5654,5672)
2748 - don't move pending lock onto export if it is already evicted (5683)
2749 - fix kernel oops when creating .foo in unlinked directory (5548)
2750 - fix deadlock in obdfilter statistics vs. object create (5811)
2751 - use time_{before,after} to avoid timer jiffies wrap (5882)
2752 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2753 - Do not leak request if server was not able to process it (5154)
2754 - If mds_open unable to find parent dir, make that negative lookup(5154)
2755 - don't create new directories with extent-mapping (5909, 5936)
2757 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2758 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2759 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2760 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2761 - support for RHEL4 kernel on i686 (5773)
2762 - provide error messages when incompatible logs are encountered (5898)
2764 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2765 * version 1.4.0.10 (1.4.1 release candidate 1)
2767 - don't keep a lock reference when lock is not granted (4238)
2768 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2769 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2770 - drop import inflight refcount on signal_completed_replay error (5255)
2771 - unlock page after async write error during send (3677)
2772 - handle missing objects in filter_preprw_read properly (5265)
2773 - no transno return for symlink open, don't save no-trasno open (3440)
2774 - don't try to complete elan receive that already failed (4012)
2775 - free RPC server reply state on error (5406)
2776 - clean up thread from ptlrpc_start_thread() on error (5160)
2777 - readahead could read extra page into cache that wasn't ejected (5388)
2778 - prevent races in class_attach/setup/cleanup/detach (5260)
2779 - don't dereference de->d_inode after l_dput of de (5458)
2780 - use "int" for stripe value returned from lock_to_stripe (5544)
2781 - mballoc allocation and error-checking fixes in 2.6 (5504)
2782 - block device patches to fix I/O request sizes in 2.6 (5482)
2783 - look up hostnames for IB nals (5602)
2784 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2785 - don't start multiple acceptors for the same port (5277)
2786 - fix incorrect LASSERT in mds_getattr_name (5635)
2787 - export a proc file for general "ping" checking (5628)
2788 - fix "lfs check" to not block when the MDS is down (5628)
2790 - service request history (4965)
2791 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2792 - create an "evict_client" /proc entry on OSTs, like the MDS has
2793 - fix mount usage message, return errors per mount(8) (5168)
2794 - change grep [] to grep "[]" in tests so they work in more UMLs
2795 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2796 - remove ancient LOV_MAGIC_V0 EA support (5047)
2797 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2798 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2799 - allow readahead allocations to fail when low on memory (5383)
2800 - mmap locking landed again, after considerable improvement (2828)
2801 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2803 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2806 - send OST transaction number in read/write reply to free req (4966)
2807 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2808 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2809 - the watchdog thread now runs as interruptible (5246)
2810 - flock/lockf fixes (but it's still disabled, pending 5135)
2811 - don't use EXT3 constants in llite code (5094)
2812 - memory shortage at startup could cause assertion (5176)
2814 - reorganization of lov code
2815 - single portals codebase
2817 - add extents/mballoc support (5025)
2818 - direct I/O reads in the obdfilter (4048)
2819 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2821 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2824 - send OST transaction number in read/write reply to free req (4966)
2825 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2826 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2827 - the watchdog thread now runs as interruptible (5246)
2828 - handle missing objects in filter_preprw_read properly (5265)
2829 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2830 - drop import inflight refcount on signal_completed_replay error (5255)
2831 - unlock page after async write error during send (3677)
2832 - return original error code on reconstructed replies (3761)
2833 - no transno return for symlink open, don't save no-trasno open (3440)
2835 - add pid to ldlm debugging output (4922)
2836 - bump the watchdog timeouts -- we can't handle 30sec yet
2837 - extra debugging for orphan dentry/inode bug (5259)
2839 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2842 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2843 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2844 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2845 - fix NULL dereference in /proc/sys/portals/routes (4827)
2846 - allow failed mdc_close() operations to be interrupted (4561)
2847 - stop precreate on OST before MDS would time out on it (4778)
2848 - don't send partial-page writes before EOF from client (4410)
2849 - discard client grant for sub-page writes on large-page clients (4520)
2850 - don't free dentries not owned by NFS code, check generation (4806)
2851 - fix lsm leak if mds_create_objects() fails (4801)
2852 - limit debug_daemon file size, always print CERROR messages (4789)
2853 - use transno after validating reply (3892)
2854 - process timed out requests if import state changes (3754)
2855 - update mtime on OST during writes, return in glimpse (4829)
2856 - add mkfsoptions to LDAP (4679)
2857 - use ->max_readahead method instead of zapping global ra (5039)
2858 - don't interrupt __l_wait_event() during strace
2860 - add software watchdogs to catch hung threads quickly (4941)
2861 - make lustrefs init script start after nfs is mounted
2862 - fix CWARN/ERROR duplication (4930)
2863 - return async write errors to application if possible (2248)
2864 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2865 - print NAL number in %x format (4645)
2866 - update barely-supported suse-2.4.21-171 series (4842)
2867 - support for sles 9 %post scripts
2868 - support for building 2.6 kernel-source packages
2869 - support for sles km_* packages
2871 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2874 - ignore -ENOENT errors in osc_destroy (3639)
2875 - notify osc create thread that OSC is being cleaned up (4600)
2876 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2877 - reconstruct ost_handle() like mds_handle() (4657)
2878 - create a new thread to do import eviction to avoid deadlock (3969)
2879 - let lconf resolve symlinked-to devices (4629)
2880 - don't unlink "objects" from directory with default EA (4554)
2881 - hold socknal file ref over connect in case target is down (4394)
2882 - allow more than 32000 subdirectories in a single directory (3244)
2883 - fix blocks count for O_DIRECT writes (3751)
2884 - OST returns ENOSPC from object create when no space left (4539)
2885 - don't send truncate RPC if file size isn't changing (4410)
2886 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2887 - bind to privileged port in socknal and tcpnal (3689)
2889 - rate limit CERROR/CWARN console message to avoid overload (4519)
2890 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2891 - basic mmap support (3918)
2892 - kernel patch series update from b1_4 (4711)
2894 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2897 - avoid crash during MDS cleanup with OST shut down (2775)
2898 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2899 - don't use bad inodes on the MDS (3744)
2900 - dynamic object preallocation to improve recovery speed (4236)
2901 - don't hold spinlock over lock dumping or change debug flags (4401)
2902 - don't zero obd_dev when it is force cleaned (3651)
2903 - print grants to console if they go negative (4431)
2904 - "lctl deactivate" will stop automatic recovery attempts (3406)
2905 - look for existing locks in ldlm_handle_enqueue() (3764)
2906 - don't resolve lock handle twice in recovery avoiding race (4401)
2907 - revalidate should check working dir is a directory (4134)
2909 - don't always mark "slow" obdfilter messages as errors (4418)
2911 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2914 - don't close LustreDB during write_conf until it is done (3860)
2915 - fix typo in lconf for_each_profile (3821)
2916 - allow dumping logs from multiple threads at one time (3820)
2917 - don't allow multiple threads in OSC recovery (3812)
2918 - fix debug_size parameters (3864)
2919 - fix mds_postrecov to initialize import for llog ctxt (3121)
2920 - replace config semaphore with spinlock (3306)
2921 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2922 - don't allow enqueue to complete on a destroyed export (3822)
2923 - down write_lock before checking llog header bitmap (3825)
2924 - recover from lock replay timeout (3764)
2925 - up llog sem before sending rpc (3652)
2926 - reduce ns lock hold times when setting kms (3267)
2927 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2928 - fix NULL deref and obd_dev leak on setup error (3312)
2929 - replace some LBUG about llog ops with error handling (3841)
2930 - don't match INVALID dentries from d_lookup and spin (3784)
2931 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2932 - fix invalid assertion in ptlrpc_set_wait (3880)
2934 - add libwrap support for the TCP acceptor (3996)
2935 - add /proc/sys/portals/routes for non-root route listing (3994)
2936 - allow setting MDS UUID in .xml (2580)
2937 - print the stack of a process that LBUGs (4228)
2939 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2942 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2943 - make sure to unset replay flag from failed open requests (3440)
2944 - if default stripe count is 0, use OST count for inode size (3636)
2945 - update parent mtime/ctime on client for create/unlink (2611)
2946 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2947 - free recovery state on server during a forced cleanup (3571)
2948 - unregister_reply for resent reqs (3063)
2949 - loop back devices mounting and status check on 2.6 (3563)
2950 - fix resource-creation race that can provoke i_size == 0 (3513)
2951 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2952 - more debugging for page-accounting assertion (3746)
2953 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2954 - avoid deadlock after precreation failure (3758)
2955 - fix race and lock order deadlock in orphan handling (3450, 3750)
2956 - add validity checks when grabbing inodes from l_ast_data (3599)
2958 - add /proc/.../recovery_status to obdfilter (3428)
2959 - lightweight CDEBUG infrastructure, debug daemon (3668)
2960 - change default OSC RPC parameters to be better on small clusters
2961 - turn off OST read cache for files smaller than 32MB
2962 - install man pages and include them in rpms (3100)
2963 - add new init script for (un)mounting lustre filesystems (2593)
2964 - run chkconfig in %post for init scripts (3701)
2965 - drop scimac NAL (unmaintained)
2967 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
2970 - clean kiobufs before and after use (3485)
2971 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
2972 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
2973 - change init script to fail more clearly if not run as root (1528)
2974 - allow clients to reconnect during replay (1742)
2975 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
2976 - don't do DNS lookups on NIDs too small for IP addresses (3442)
2977 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
2978 - fix cond_resched (3554)
2979 - only evict unfinished clients after recovery (3515)
2980 - allow bulk resend, prevent data loss (3570)
2981 - dynamic ptlrpc request buffer allocation (2102)
2982 - don't allow unlinking open directory if it isn't empty (2904)
2983 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
2984 - remove extraneous obd dereference causing LASSERT failure (3334)
2985 - don't use get_cycles() when creating temp. files on the mds (3156)
2986 - hold i_sem when setting i_size in ll_extent_lock() (3564)
2987 - handle EEXIST for set-stripe, set proper directory name (3336)
2989 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
2990 - fix ksocknal_fmb_callback() error messages (2918)
2992 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
2995 - don't copy lvb into (possibly NULL) reply on error (2983)
2996 - don't deref dentry after dput, don't free lvb on error (2922)
2997 - use the kms to determine writeback rpc length (2947)
2998 - increment oti_logcookies when osc is inactive (2948)
2999 - update client's i_blocks count via lvb messages (2543)
3000 - handle intent open/close of special files properly (1557)
3001 - mount MDS with errors=remount-ro, like obdfilter (2009)
3002 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3003 - don't use cancelling-locks' kms values (2947)
3004 - use highest lock extent for kms, not last one (2925)
3005 - don't dereference ERR_PTR() dentry in error handling path (3107)
3006 - fix thread race in portals_debug_dumplog() (3122)
3007 - create lprocfs device entries at setup instead of at attach (1519)
3008 - common AST error handler, don't evict client on completion race (3145)
3009 - zero nameidata in detach_mnt in 2.6 (3118)
3010 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3011 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3012 - initialize RPC timeout timer earlier for 2.6 (3219)
3013 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3014 - print nal/nid for unknown nid (3258)
3015 - additional checks for oscc recovery before doing precreate (3284)
3016 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3017 - don't crash in mdc_close for bad permissions on open (3285)
3018 - zero i_rdev for non-device files (3147)
3019 - clear page->private before handing to FS, better assertion (3119)
3020 - tune the read pipeline (3236)
3021 - fix incorrect decref of invalidated dentry (2350)
3022 - provide read-ahead stats and refine rpc in flight stats (3328)
3023 - don't hold journal transaction open across create RPC (3313)
3024 - update atime on MDS at close time (3265)
3025 - close LDAP connection when recovering to avoid server load (3315)
3026 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3027 - don't leak open file on MDS after open resend (3325)
3028 - serialize filter_precreate and filter_destroy_precreated (3329)
3029 - loop device shouldn't call sync_dev() for nul device (3092)
3030 - clear page cache after eviction (2766)
3031 - resynchronize MDS->OST in background (2824)
3032 - refuse to mount the same filesystem twice on same mountpoint (3394)
3033 - allow llmount to create routes for mounting behind routers (3320)
3034 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3035 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3036 - fix rare problem with rename on htree directories (3417)
3038 - allow default OST striping configuration per directory (1414)
3039 - fix compilation for qswnal for 2.6 kernels (3125)
3040 - increase maximum number of MDS request buffers for large systems
3041 - change liblustreapi to be useful for external progs like lfsck (3098)
3042 - increase local configuration timeout for slow disks (3353)
3043 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3045 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3048 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3049 - don't overwrite extent policy data in reply if lock was blocked (2901)
3050 - drop filter export grants atomically with removal from device (2663)
3051 - del obd_self_export from work_list in class_disconnect_exports (2908)
3052 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3053 - swab reply message in mdc_close, other PPC fixes (2464)
3054 - fix destroying of named logs (2325)
3055 - overwrite old logs when running lconf --write_conf (2264)
3056 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3057 - fix race in target_handle_connect (2898)
3058 - mds_reint_create() should take same inode create lock (2926)
3059 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3060 - don't close files for self_export to avoid uninitialized obd (2936)
3061 - allow MDS with the same name as client node (2939)
3062 - hold dentry reference for closed log files for unlink (2325)
3063 - reserve space for all logs during transactions (2059)
3064 - don't evict page beyond end of stripe extent (2925)
3065 - don't oops on a deleted current working directory (2399)
3066 - handle hard links to targets without a parent properly (2517)
3067 - don't dereference NULL lock when racing during eviction (2867)
3068 - don't grow lock extents when lots of conflicting locks (2919)
3070 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3073 - account for cache space usage on clients to avoid data loss (974)
3074 - lfsck support in lustre kernel code (2349)
3075 - reduce journal credits needed for BRW writes (2370)
3076 - orphan handling to avoid losing space on client/server crashes
3077 - ptlrpcd can be blocked, stopping ALL progress (2477)
3078 - use lock value blocks to assist in proper KMS, faster stat (1021)
3079 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3080 - recovery for initial connections (2355)
3081 - fixes for mds_cleanup_orphans (1934)
3082 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3083 - block all file creations until orphan recovery completes (1901)
3084 - client remove rq_connection from request struct (2423)
3085 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3086 - recovery timer race (2670)
3087 - mdc_close recovey bug (2532)
3088 - ptlrpc cleanup bug (2710)
3089 - mds timeout on local locks (2588)
3090 - namespace lock held during RPCs (2431)
3091 - handle interrupted sync write properly (2503)
3092 - don't try to handle a message that hasn't been replied to (2699)
3093 - client assert failure during cleanup after abort recovery (2701)
3094 - leak mdc device after failed mount (2712)
3095 - ptlrpc_check_set allows timedout requests to complete (2714)
3096 - wait for inflight reqs when ptlrpcd finishes (2710)
3097 - make sure unregistered services are removed from the srv_list
3098 - reset bulk XID's when resending them (caught by 1138 test)
3099 - unregister_bulk after timeout
3100 - fix lconf error (2694)
3101 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3102 - readahead locks pages, leaves pending causing memory pressure (2673)
3103 - increase OST request buffers to 4096 on large machines (2729)
3104 - fix up permission of existing directories in simple_mkdir (2661)
3105 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3106 - don't assign transno to errored transactions (2742)
3107 - don't delete objects on OST if given a bogus objid from MDS (2751)
3108 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3109 - if rq_no_resend, then timeout request after recovery (2432)
3110 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3111 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3112 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3113 - MDS recovery shouldn't delete objects that it has given out (2730)
3114 - if enqueue arrives after completion, don't clobber LVB (2819)
3115 - don't unlock pages twice when trigger_group_io returns error (2814)
3116 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3117 - don't write pages to disk if there was an error (1450)
3118 - don't ping imports that have recovery disabled (2676)
3119 - take buffered bytes into account when balancing socknal conn (2817)
3120 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3121 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3122 - remove little-endian swabbing of llog records (1987)
3123 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3124 - retry reposting request buffers if they fail (1191)
3125 - grow extent at grant time to avoid granting a revoked lock (2809)
3126 - lock revoke doesn't evict page if covered by a second lock (2765)
3127 - disable VM readahead to avoid reading outside lock extents (2805)
3129 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3130 - updated kernel patches for hp-2.4.20 kernel (2681)
3132 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3135 - fix truncated write corruption (2366)
3136 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3138 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3139 - ensure len(uuid) < 37 in lmc (1171)
3140 - fix ia64 OOPS in llog_test (2255)
3141 - zero end of page at obdfilter for partial page writes (2648)
3142 - don't leave stale dentries around after renames (bug 2428)
3143 - fix timeouts when evicting a client with a single lock held (2642)
3144 - set deadline for the initial HELLO message to drain (2634)
3145 - print out dotted-quad IP addresses in the socknal (2302)
3147 - additional debugging for MDS client eviction problem (2443)
3148 - fix mkfsoptions support for osts (2603, 2604)
3150 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3153 - add series for the vanilla 2.6.0 kernel
3154 - add series for the vanilla 2.4.24 kernel
3155 - add series for a cray x86/64 UL kernel drop
3156 - fix xattr patches for the vanilla 2.4.19 series
3158 - generate true UUIDs in lmc (1171)
3159 - have portals stack dumping break in UML (2466)
3160 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3161 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3162 - wake callback waiting threads on client eviction (2460)
3163 - Add --ptldebug and --subsystem to lmc (1719)
3164 - update assertion to allow safe interrupt allocation
3165 - set rq_no_resend for cancel requests (2432)
3166 - recalculate ptlrpcd timeout after resend (2494)
3167 - call vfs_rmdir when removing pending directories (2368)
3168 - fix renaming a file to itself (2429)
3169 - lmc creates a default one-stripe lov (2454)
3170 - expand procfs space to handle large clusters (2326)
3171 - increase UML stack to avoid overflow
3172 - update lconf's list of debug and subsystem masks
3173 - fix lfs find --obd (2510)
3174 - /proc tunable for disabling filter read caching (2591)
3175 - stop rpm packages from altering slapd.conf (2301)
3176 - disable nagle in the socknal under 0conf (2578)
3177 - choose mds inode size based on stripe count (2572)
3178 - fix kernel-source rpm problems (2516)
3180 - add --disable-doc to avoid pdf generation (2421)
3181 - update documentation, tests, type-os, comments
3182 - avoid format warnings on ia64
3183 - remove the TOE NAL
3184 - tiny code cleanups by removing unused fields
3186 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3189 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3190 - remove the most blatant lies from BUILDING file (2371)
3191 - change default debug level to reasonable production setting
3192 - reduce client side cache size to reduce cache flush time
3193 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3194 - make TCP zerocopy and pinger support enabled by default (2476)
3195 - sync writes completed after process exits caused crashes (2319)
3196 - maintain correct mount count on the MDS (2356)
3197 - backout 1557, because 2316 wasn't really fixed
3198 - better file I/O statistics gathering in /proc
3199 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3200 - another kernel patch to fix zero-copy TCP function export
3201 - don't take duplicate lock when processing re-sent getattr (2420)
3202 - lctl uses obd_self_export instead of creating new conn (2353)
3203 - MDS/OST recovery case which requires object creation asserted (2425)
3204 - move lfs from /usr/sbin to /usr/bin in packages
3205 - fix race between mds_client_add and mds_client_free (2417)
3206 - use kmalloc instead of slabs in portals (2430)
3207 - don't create duplicate records when a failover MDS is present (2442)
3208 - remove unnecessary mount age check (2332)
3209 - don't remove directory inodes from locks prematurely (2451)
3210 - don't break if MDS service name is the same as hostname (2103)
3211 - fix races in client write RPC generation when cache full (2482)
3213 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3216 - remove now-unused request->rq_obd (278)
3217 - if an allocation fails, print out how much memory we've used (1933)
3218 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3219 - add the "configurable stack size" patch to most series files (1256)
3220 - ability to write large log records, for 100+ OST configs (2306)
3221 - fix NULL deref when filter_prep fails (2314)
3222 - fix operator precedence error in filter_sync
3223 - dynamic allocation of socknal TX descriptors (2315)
3224 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3225 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3226 - ensure that CWARN messages in Portals always get to the syslog
3227 - __init/__exit are not for prototype decls (ldlm_init/exit)
3228 - x86-64 compile warning fixes
3229 - fix gateway LMC keyword conflict (2318)
3230 - fix MDS lock inversions in getattr/reint paths (1844)
3231 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3232 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3233 - DLM race condition prevented some lock evictions (2328)
3234 - ENOMEM detection and retry on socknal sends (2230)
3235 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3236 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3237 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3238 - better error messages when a client is rejected during recovery (1505)
3239 - avoid cancelling locks which were never granted, after failure (2330)
3240 - fix i_sem/journal inversion in mds_client_add (2333)
3241 - fix truncate/getattr lock cycle deadlock (2334)
3242 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3243 - fix two rare exit paths which could leak an l_lock() ref (2321)
3244 - fencepost error in MDS/OST orphan recovery (2226)
3245 - make log record alignment 8 bytes (1988)
3246 - lstripe now fails when requested offset > ost_count (2237)
3247 - ensure that all kernel series have a complete list.h (1607)
3248 - fix crashes in special-file operations (2316)
3249 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3251 - allow configurable automake binary, for testing new versions
3252 - small update to the lfs documentation
3254 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3256 * fix negative export reference count in fsfilt_sync (2312)
3258 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3259 * release candidate 0.9.1
3261 - orphans are moved into the PENDING directory for possible recovery
3262 - replayed opens now open by fid for orphan/rename safety (1042)
3263 - last close of an orphan inode generates a transno (683)
3264 - chdir() and mount() now pin the directory entry (1020)
3265 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3266 - discard very old requests without processing them (1502)
3267 - remove some common, well-understood CERRORs (1505)
3268 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3269 - clear "grant" flags in OST replies until OST grant code lands (1644)
3270 - fix read performance by not clobbering i_blksize on client (1598)
3271 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3272 - make LRU size a /proc tunable, clears locks when reduced (707)
3273 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3274 - server locks take references on exports now (1558)
3275 - build fixes for 2.4.20-rh trees (1663)
3276 - return an error from lov_create if all OSCs are inactive (1751)
3277 - fix import levels when a reconnect happens without a timeout (1597)
3278 - exit early from mds_open if we get a lookup error (1749)
3279 - partial page read at EOF wouldn't wait for disk before sending (1642)
3280 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3281 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3282 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3283 - write/truncate lock inversion (1639)
3284 - Don't auto-load obdclass, portals modules during cleanup (1495)
3285 - fix timestamps from jumping to "now" (1763)
3286 - extra journal assertions (1648)
3287 - add an extra multiunlink test (1771)
3288 - fix read_record/write_record API (1776)
3289 - fix leak of offset_extent, possible incorrect i_size later (1772)
3290 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3291 - Debugging for the kqswnal_get_idle_tx problems (1820)
3292 - Allow recovery to be attempted multiple times (1536)
3293 - Write out MDS last_rcvd file after it is first created (1600)
3294 - Fix tx_descriptor leak in failed transmit situations (1827)
3295 - ext3 journaling fixes for assertion failure after IO error (1871)
3296 - class_export_put() on freed export after completion AST error (1896)
3297 - Fix revalidate looping in VFS (1322)
3298 - Don't access a freed export during MDS_REINT timeout (1521)
3299 - Add open-unlink recovery support on the MDS (1673,1764)
3300 - Return an error if no MDS data was read from last_rcvd (1946)
3301 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3302 - Don't setup a disk device that doesn't match exported UUID (317)
3303 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3304 - avoid committing NULL handle in force close
3305 - local.sh is now a one-stripe LOV configuration
3306 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3307 - fix ext3 htree duplicate directory entry corruption (1516)
3308 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3309 - update to new LOV EA format (2097)
3310 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3312 2003-06-15 Phil Schwan <phil@clusterfs.com>
3315 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3316 - per-import/export recovery handling (958, 931, 959)
3317 - multiple last-rcvd slots, for serving multiple FSes (949)
3318 - connections are again shared between multiple imp/exports (963, 964)
3319 - "umount -f" would hang if any requests needed to be sent (393, 978)
3320 - avoid pinning large req buffer by copying for queued messages (989)
3321 - add "uuid" to "lctl device" command to help upcalls (991)
3322 - "open" RPCs with transnos would confuse recovery counters (1037)
3323 - do proper endian conversion of all wire messages (288, 340, 891)
3324 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3325 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3326 - fix ptlrpc_connection leak in target_handle_connect (1174)
3327 - fix import refcounting bug in OST and MDS cleanup (1134)
3328 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3329 - fix very unlikely obd_types race condition (501)
3330 - remove osc_open hack for echo_client (1187)
3331 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3332 - a failure in read_inode2 leads to deadlock (1139)
3333 - cancel ack-locks as soon as transaction is committed (1072)
3334 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3335 - make sure to commitrw after any preprw to avoid deadlock (1162)
3336 - failing to execute a file in a lustre FS would lock inode (1203)
3337 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3338 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3339 - obd_uuid2tgt would walk off the end of the list (1255)
3340 - on IA64 the osc would give portals incorrect bulk size (1258)
3341 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3342 - fix lock inversion caused by new llite matching code (1282)
3343 - limit the number of dirty pages on a client to 10MB (1286)
3344 - timed out locks were not being corrected cancelled (1289)
3345 - fix O_DIRECT above 4GB on IA-32 (1292)
3346 * major user-visible changes
3347 - fail out/fail over policy now controlled by the upcall (993)
3349 - add OBD_PING to check server availability and failure (954)
3350 - lustre messages are now sent in sending host order (288, 340, 891)
3351 - add eadatalen to MDS getattr reply (340)
3352 - OST read replies may contain second buffer, with per-page status (593)
3354 2003-03-11 Phil Schwan <phil@clusterfs.com>
3357 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3358 - failed open()s could cause deadlock; fixed (867, 869)
3359 - stop cancelling OST locks when files are closed (481)
3360 - overlapping XID spaces caused network corruption (851, 853)
3361 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3362 - setattr_raw added, to do single-RPC, server-side setattrs
3363 - lmc/lconf syntax change for OST UUIDs
3364 - fix crashy race condition between ptlrpc_free_req and osc_close
3365 - don't use request in mdc_enqueue if we hit a timeout (889)
3366 - don't set the inode i_size for regular files from the MDS (896)
3367 - handle out of order completion AST (842)
3368 - don't LBUG if a lock request times out after receiving AST (913)
3369 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3370 - if a bad lock AST arrives, send an error instead of dropping entirely
3371 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3372 - fix leak in bulk IO when only partially completed (899, 900, 926)
3373 - fix O_DIRECT for ia64 (55)
3374 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3375 - C-z after timeout could hang a process forever; fixed (977)
3377 - client-side I/O cache (678, 924, 929, 941, 970)
3379 - READPAGE and SETATTRs which don't take server-side locks get
3382 2003-02-11 Phil Schwan <phil@clusterfs.com>
3385 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3386 - Fix fsfilt_extN_readpage() to read a full page of directory
3387 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3388 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3389 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3390 - Most intent operations were removed, in favour of a new RPC mode
3391 that does a single RPC to the server and bypasses most of the VFS
3392 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3393 - Aggressively cancel local locks on DLM servers
3394 - mds_reint_unlink sends EA to the client if it's the last nlink.
3395 client uses that EA to unlink OST objects.
3396 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3397 - recursive symlinks were fixed (439)
3398 - fixed NULL deref in DEBUG_REQ
3399 - filter_update_lastobjid no longer calls sync, which annoyed extN
3400 - fixed multi-client small-writes to a single file problem (445)
3401 - fixed mtime updates during file writes (607)
3402 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3403 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3404 - fixed Linux OST size reporting problem (444, 656)
3405 - OST now updates object mtime with writes or setattr (607, 619)
3406 - client verifies file size before zeroing page past EOF (445)
3407 - OST now writes last allocated objid to disk with allocation (108)
3408 - LOV on echo now works (409)
3410 - mds_reint_unlink sends a new buffer, with the EA included. this
3411 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3412 set if a regular file was being unlinked, and it was the last link
3413 - use PtlGet from the target for bulk writes (315)
3414 - OST now updates object mtime with writes or setattr (607, 619)
3415 - LDLM now has a grant-time callback to revalidate locked items, if
3417 - Many MDS operations were reorganized to combat race conditions
3419 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3420 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3422 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3425 - Fully reactivate OST imports after reconnection (512, others)
3426 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3427 - More graceful error handling for truncating on dead OST (515)
3428 - Don't error out unless we're actually accessing dead stripes (474)
3429 - Fix garbage sizes when stripes are missing (410)
3430 - LRU counters were broken, causing constant lock purge (433, 432)
3431 - garbage on read from stripes with failed OSTs (441)
3432 - mark OSCs as active before reconnecting during recovery (438)
3433 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3434 - lfind did not preserve OST order in output (443)
3435 - symlinks cause hung clients, incorrect data (439)
3436 - stop dereferencing request after dropping refcount (457)
3437 - don't LASSERT(spin_is_locked) on non-SMP (455)
3438 - fixes for many rename() bugs
3439 - fstat didn't correctly synchronize attributes (399)
3440 - server must handle lock cancellation during blocking AST prep (487)
3441 - bulk descriptors were free()d too soon (511)
3442 - fix paths in lconf, which would load incorrect modules (451, 507)
3443 - fix confusing lconf 'host not found' error message (386)
3444 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3445 - fix race condition in mdc_blocking_ast() for inode access (526)
3446 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3447 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3448 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3449 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3450 - fix lstripe in several ways (kernel side) (527)
3451 - fix request leak in ldlm_cli_enqueue (262)
3452 - incorrect OSC was marked inactive after OST failure
3453 - call mds_fs_cleanup before unmounting filesystem (524)
3454 - fix races between taking ns_lock and ldlm_lock_change_resource
3455 - fix races updating LOV export open file list
3456 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3457 - fix recovery NULL deref in ldlm_cli_cancel_unused
3458 - fix some DLM races by using new hash table for lock handles (419)
3459 - permit the client to specify desired inodes, at replay
3460 - duplicate requests when we queue them for replay reintegration
3461 - fix last_rcvd offset calculation
3462 - sync after each recovered transaction, so we always make progress
3463 - never, not always, ERESTART requests without transnos
3464 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3466 - skip replay if the MDS says that the client is already connected
3467 - don't check for a recovery-enabled export to match lctl's UUID
3468 - don't INC_USE_COUNT for phantom exports
3469 - don't crash when cleaning up phantom exports (567)
3470 - don't double-finish or set replay data for errored mdc_open requests
3471 - abort requests when they time out, so we don't get old replies
3472 - send/receive replies for AST messages again
3473 - if the client says that it doesn't have the lock, cancel it on the
3475 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3476 mark it as destroyed, it will all work out when decref is called
3477 - fix module use counts (22, 581)
3479 - ASTs now expect a reply (server cancels lock on error reply)
3481 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3484 - fix many simultaneous client startup (392)
3485 - fix dentry->d_it clobbering
3486 - credentials weren't being shipped for readdir/getattr operations
3487 - remove invalid assertions triggered during some concurrent MD
3489 - proper Lustre versions added (336, 389)
3490 - fix memory leak for create error case (398)
3491 - fix LOV locking bug that would get cli/srv out of sync
3492 - fix echo client over LOV (409)
3493 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3494 - fix double-O_EXCL intent crash (424)
3495 - avoid sending multiple lock CANCELs (352)
3497 - MDS can do multi-client recovery (modulo bugs in new code)
3499 - many updates, edits, cleanups
3501 2002-11-18 Phil Schwan <phil@clusterfs.com>
3504 - fix null d_it dereference (346)
3505 - fix full OST/dbench hang (333)
3506 - fix permission problem with file removal (286)
3507 - fix removal of OSCs from LOV when they fail
3508 - fix NULL deref during bulk timeout (214)
3509 - fix problems related to multiple filesystems on one MDS (241)
3510 - fixed serious subtle metadata locking bugs
3511 - free locks on clients when inodes are removed due to memory
3513 - fix inode pointer in lock data (285)
3514 - partial support for multiple MDS on a single host (241)
3515 - data locks weren't cancelled at clear_inode time (290, 311)
3516 - intent locks could lead to unbounded lock growth (205)
3517 - added a maximum lock count, an LRU list, and a flusher
3518 - fix multiple rename (365)
3519 - properly abstracted the echo client
3520 - OSC locked 1 byte too many; fixed
3521 - rewrote brw callback code:
3522 - fixed recovery bugs related to LOVs (306)
3523 - fixed too-many-pages-in-one-write crash (191)
3524 - fixed (again) crash in sync_io_timeout (214)
3525 - probably fixed callback-related race (385)
3527 - Add capability to MDS protocol
3528 - LDLM cancellations and callbacks on different portals
3530 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3533 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3534 - defer open object destruction to close time (601981, 138)
3535 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3536 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3537 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3538 - don't call obd_setattr in ll_file_release for destroyed objects
3539 * protocol change to lustre_msg: move |version| and add |flags|
3540 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3541 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3542 * added replay of create, unlink, link and rename operations during
3543 MDS failover; recovery should be much more robust now
3544 * remove failed OSCs from LOVs (only lov_create uses this so far)
3545 * the lustre-HOWTO was brought (more) up to date (582544)
3547 2002-10-23 Phil Schwan <phil@clusterfs.com>
3550 - in-use dentries weren't being reused properly (617851)
3551 - prevent multiple LDLM setup (599178)
3552 - fix LOV size calculations for truncate (617853)
3553 - fix client handling of MDS intent errors (POSIX)
3554 - fix permission bug in lovstripe.c test (624321)
3555 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3556 - truncate past end of file could corrupt data
3557 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3558 - a race in recovery could return ETIMEDOUT to apps (623947)
3559 - building outside the source directory was fixed
3560 * the lustre-HOWTO was brought (more) up to date (582544)
3561 * major progress was made on recovery functionality
3563 2002-10-10 Phil Schwan <phil@clusterfs.com>
3566 - recovery deadlock fix
3567 - rm -rf causes LBUG fix (617817)
3568 - file open by multiple tasks fix (618962)
3569 - directory permissions bugs (602707 and 620007)
3570 - journal_stop fixed with locking (611313)
3571 - O_APPEND failures resolved (618273, perhaps 614459)
3572 - lconf PATH fix (619770)
3573 - IA64 build fix (621450)
3574 - RPC buffer sizes scale with amount of memory
3576 2002-10-01 Phil Schwan <phil@clusterfs.com>
3579 - locks would be cancelled without throwing away data pages,
3580 resulting in inconsistent data (605627)
3581 - inode attributes were not always being refreshed (605627, 612449)
3582 - lconf now continues to cleanup after lctl reports an error
3583 - MDS now enforces user permissions (602707)
3584 - lprocfs cleanup fixed, but not yet enabled (614157)
3585 - fixed infinite server hang, should a client not respond to an AST
3586 - avoid going into recovery if user calls readlink() with a buffer
3587 that's too small (613941)
3588 - AST RPCs no longer require replies (614867) -- this may be changed
3589 - don't crash server if client sends an IOV that's too big (611336)
3590 - fixed lock conversion deadlock (611892)
3591 - fixed the following of symlinks (614622)
3592 * recovery: the server can remove locks from a client that dies, other
3593 clients can make progress
3594 * more extN patch fixes
3595 * compile-time configurable ptlrpc buffer allocations
3597 - collaborative read cache document
3598 - Lustre Lite Performance CDR document-in-progress
3600 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3603 - fix typo in patch-2.4.18
3605 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3608 - clear ptlrpc request each time in handle_incoming_request()
3609 - unlink of files now destroys the object on the OST
3611 2002-09-19 Peter Braam <braam@clusterfs.com>
3613 * add hard link support
3614 * change obdfile creation method
3615 * kernel patch changed
3617 2002-09-19 Peter Braam <braam@clusterfs.com>
3620 - stack overflow bug in extN fixed
3622 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3624 * documentation updates
3625 - add man pages for config tools
3626 - update tests/README to describe testing with new config tools
3627 - finish metadata API descriptions
3628 * bug fixes and cleanups
3629 - statfs workaround for 16TB limit
3630 - LOV stripe allocation improved, can stripe on subset of OSTs
3631 - LOV file size/IO offset was wrong for files > 4GB in size
3632 - object EA data was being dropped, caused files to be unreadable
3633 - memory overflow with non-LOV OST caused memory corruption
3634 - fixed regression tests to work with new config tools, obdfilter
3635 - fixed bug when directory size became larger than 1 block
3636 - fixed bug (for single client case) when PWD was deleted
3637 - invalidate local directory pages when doing intent-based ops
3638 - avoid LDLM oops when lock callback contained bad data
3640 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3642 * documentation updates
3643 * bug fixes and cleanups
3644 - configuration tools
3647 - 64-bit compile warnings
3648 - 64-bit internal statfs data
3650 * test_brw on persistent OST devices
3652 * lprocfs (disabled)
3654 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3656 * documentation updates
3657 * bug fixes and cleanups
3658 * configuration tools
3660 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3663 * many small fixes to 0_5_4
3664 * io/network handling
3665 * thinkos in MDS operations
3667 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3670 * crucial basic fixes to 0.5.3
3671 * IOR, Iozone work over Elan
3674 2002-08-07 Phil Schwan <phil@clusterfs.com>
3675 * version 0_5_3, our first alpha
3676 * we use the new Portals iovs
3677 * documentation updates
3678 * bug fixes and cleanups
3679 * small changes in the DLM wire protocol
3681 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3682 * version 0_5_1 with some initial stability,
3683 * locking on MD and file I/O.
3684 * documentation updates
3685 * several bug fixes since 0.5.0
3686 * small changes in wire protocol
3688 2002-07-18 Phil Schwan <phil@clusterfs.com>
3690 * delivered as Lustre Light Alpha
3691 * fixed a crash after handling invalid MDS requests
3692 * fixed directory pages for architectures with non-4k pages sizes
3694 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3695 * release version v0_4_4
3696 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3698 * Fixes a number of interruption problems with OST operations.
3699 * Update documentation for portals header changes
3700 * Move all wire protocol structs/defines to lustre_idl.h
3701 * Fixes symlink length bug.
3702 * Add tcpdump to repository.
3704 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3705 * release version v0_4_3
3706 * Fixes statfs for inodes on extN.
3707 * Fixes bug in runtests which would delete /etc/hosts.
3708 * Use 64-bit object IDs wherever possible (not into VFS though)
3709 Remove ost_get_info, which is unused by lustre, and out of date.
3711 2002-07-03 Peter Braam <braam@clusterfs.com>
3712 * release version v0_4_2 Fixes a lookup error (type not passed)
3713 * move forward to head of Portals
3714 * move forward to latest Lustre kernel
3716 2002-06-25 Peter Braam <braam@clusterfs.com>
3717 * release version v0_4_1. Hopefully stable on single node use.