1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
4 INTEROPERATE with 1.4.x versions automatically. In many cases a
5 special upgrade step is needed. Please read the user documentation
6 before upgrading any part of a 1.4.x system.
7 * WARNING: Lustre configuration and startup changes are required with
8 1.6.x releases. See https://mail.clusterfs.com/wikis/lustre/MountConf
10 * Support for kernels:
11 2.6.9-42.0.10.EL (RHEL 4)
13 2.4.21-47.0.1.EL (RHEL 3)
14 2.6.12.6 vanilla (kernel.org)
15 2.6.16.27-0.9 (SLES 10)
16 * Client support for unpatched kernels:
17 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
18 2.6.16 - 2.6.19 vanilla (kernel.org)
19 2.6.9-42.0.8EL (RHEL 4)
20 * Recommended e2fsprogs version: 1.39.cfs6
22 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
25 Severity : enhancement
27 Description: lfs setstripe use optional parameters instead of postional
30 Severity : enhancement
32 Description: Nanosecond timestamp support for ldiskfs
33 Details : The on-disk ldiskfs filesystem has added support for nanosecond
34 resolution timestamps. There is not yet support for this at
35 the Lustre filesystem level.
38 Frequency : during server recovery
40 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
41 Details : request with rq_no_resend flag not awake l_wait_event if they get a
45 Frequency : nfs export on patchless client
47 Description: connectathon hang when test nfs export over patchless client
48 Details : Disconnected dentry cannot be found with lookup, so we do not need
49 to unhash it or make it invalid
53 Description: ENOENT returned for valid filehandle during dbench.
54 Details: : Check if a directory has children when invalidating dentries
55 associated with an inode during lock cancellation. This fixes
56 an incorrect ENOENT sometimes seen for valid filehandles during
59 --------------------------------------------------------------------------------
61 tbd Cluster File Systems, Inc. <info@clusterfs.com>
63 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
64 INTEROPERATE with older versions automatically. In many cases a
65 special upgrade step is needed. Please read the
66 user documentation before upgrading any part of a 1.4.x system.
67 * WARNING: Lustre configuration and startup changes are required with
68 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
70 * Support for kernels:
71 2.6.9-42.0.10.EL (RHEL 4)
73 2.4.21-47.0.1.EL (RHEL 3)
74 2.6.12.6 vanilla (kernel.org)
75 2.6.16.27-0.9 (SLES 10)
76 * Client support for unpatched kernels:
77 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
78 2.6.16 - 2.6.19 vanilla (kernel.org)
79 2.6.9-42.0.8EL (RHEL 4)
80 * Recommended e2fsprogs version: 1.39.cfs5
82 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
85 Severity : enhancement
87 Description: MountConf
88 Details : Lustre configuration is now managed via mkfs and mount
89 commands instead of lmc and lconf. New obd types (MGS, MGC)
90 are added for dynamic configuration management. See
91 https://mail.clusterfs.com/wikis/lustre/MountConf for
94 Severity : enhancement
96 Description: dynamic OST addition
97 Details : OSTs can now be added to a live filesystem
99 Severity : enhancement
101 Description: startup order invariance
102 Details : MDTs and OSTs can be started in any order. Clients only
103 require the MDT to complete startup.
105 Severity : enhancement
107 Description: parallel, asynchronous orphan cleanup
108 Details : orphan cleanup is now performed in separate threads for each
109 OST, allowing parallel non-blocking operation.
111 Severity : enhancement
113 Description: optimized stripe assignment
114 Details : stripe assignments are now made based on ost space available,
115 ost previous usage, and OSS previous usage, in order to try
116 to optimize storage space and networking resources.
118 Severity : enhancement
120 Description: Permanently set tunables
121 Details : All writable /proc/fs/lustre tunables can now be permanently
122 set on a per-server basis, at mkfs time or on a live
125 Severity : enhancement
127 Description: Lustre message v2
128 Details : Add lustre message format v2.
130 Severity : enhancement
132 Description: client OST exclusion list
133 Details : Clients can be started with a list of OSTs that should be
134 declared "inactive" for known non-responsive OSTs.
137 Frequency : SFS test only (otherwise harmless)
139 Description: SPEC SFS validation failure on NFS v2 over lustre.
140 Details : Changes the blocksize for regular files to be 2x RPC size,
141 and not depend on stripe size.
143 Severity : enhancement
145 Description: Multiple MD RPCs in flight.
146 Details : Further unserialise some read-only MDT RPCs - learn about intents.
147 To avoid overly-overloading MDT, introduce a limit on number of
148 MDT RPCs in flight for a single client and add /proc controls
149 to adjust this limit.
151 Severity : enhancement
153 Description: client read/write statistics
154 Details : Add client read/write call usage stats for performance
155 analysis of user processes.
156 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
157 file access. extents_stats shows chunk size distribution.
158 extents_stats_per_process show chunk size distribution per
161 Severity : enhancement
163 Description: per-client statistics on server
164 Details : Add ldlm and operations statistics for each client in
165 /proc/fs/lustre/mds|obdfilter/*/exports/
167 Severity : enhancement
169 Description: improved MDT statistics
170 Details : Add detailed MDT operations statistics in
171 /proc/fs/lustre/mds/*/stats
173 Severity : enhancement
175 Description: VFS operations stats
176 Details : Add client VFS call stats, trackable by pid, ppid, or gid
177 /proc/fs/lustre/llite/*/vfs_ops_stats
178 /proc/fs/lustre/llite/*/vfs_track_[pid|ppid|gid]
183 Description: Fix client-side osc byte counters
184 Details : The osc read/write byte counters in
185 /proc/fs/lustre/osc/*/stats are now working
188 Frequency : always as root on SLES
190 Description: Failure of copying files with lustre special EAs.
191 Details : Client side always return success for setxattr call for lustre
192 special xattr (currently only "trusted.lov").
197 Description: Refcount LNET uuids
198 Details : The global LNET uuid list grew linearly with every startup;
199 refcount repeated list entries instead of always adding to
202 Severity : enhancement
204 Description: Dynamic service threads
205 Details : Within a small range, start extra service threads
206 automatically when the request queue builds up.
209 Frequency : mixed-endian client/server environments
211 Description: mixed-endian crashes
212 Details : The new msg_v2 system had some failures in mixed-endian
215 Severity : enhancement
217 Description: Easy OST removal
218 Details : OSTs can be permanently deactivated with e.g. 'lctl
219 conf_param lustre-OST0001.osc.active=0'
221 Severity : enhancement
223 Description: MGS proc entries
224 Details : Added basic proc entries for the MGS showing what filesystems
227 Severity : enhancement
229 Description: provide MGS failover
230 Details : Added config lock reacquisition after MGS server failover.
232 Severity : enhancement
234 Description: add Linux 2.4 support
235 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
239 Description: a large application tries to do I/O to the same resource and dies
241 Details : Check the req->rq_arrival time after the call to
242 ost_brw_lock_get(), but before we do anything about
243 processing it & sending the BULK transfer request. This
244 should help move old stale pending locks off the queue as
245 quickly as obd_timeout.
248 Frequency : when an incorrect nid is specified during startup
250 Description: ptlrpc connect to non-existant node causes kernel crash
251 Details : LNET can't be re-entered from an event callback, which
252 happened when we expire a message after the export has been
253 cleaned up. Instead, hand the zombie cleanup off to another
256 Severity : enhancement
258 Description: plain/inodebits lock performance improvement
259 Details : Grouping plain/inodebits in granted list by their request modes
260 and bits policy, thus improving the performance of search through
264 Frequency : only for kernels with patches from Lustre below 1.4.3
266 Description: Remove old rdonly API
267 Details : Remove old rdonly API which unsed from at least lustre 1.4.3
270 Frequency : only for devices with external journals
272 Description: Set external device read-only also
273 Details : During a commanded failover stop, we set the disk device
274 read-only while the server shuts down. We now also set any
275 external journal device read-only at the same time.
278 Frequency : when upgrading from 1.4 while trying to change parameters
280 Description: The wrong (new) MDC name was used when setting parameters for
281 upgraded MDT's. Also allows changing of OSC (and MDC)
282 parameters if --writeconf is specified at tunefs upgrade time.
285 Frequency : when setting specific ost indicies
287 Description: QOS code breaks on skipped indicies
288 Details : Add checks for missing OST indicies in the QOS code, so OSTs
289 created with --index need not be sequential.
291 Severity : enhancement
293 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
294 Details : The uninit_groups feature works in conjunction with the kernel
295 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
296 up the pass1 processing of e2fsck. This is a read-only feature
297 in ldiskfs2 only, so older kernels and current ldiskfs cannot
298 mount filesystems that have had this feature enabled.
300 Severity : enhancement
302 Description: Improve multi-block allocation algorithm to avoid fragmentation
303 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
304 allocation locality and avoid filesystem fragmentation.
306 ------------------------------------------------------------------------------
308 TBD Cluster File Systems, Inc. <info@clusterfs.com>
310 * Support for kernels:
311 2.4.21-47.0.1.EL (RHEL 3)
313 2.6.9-42.0.10.EL (RHEL 4)
314 2.6.12.6 vanilla (kernel.org)
315 2.6.16.27-0.9 (SLES 10)
316 * Recommended e2fsprogs version: 1.39.cfs6
317 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
320 ------------------------------------------------------------------------------
322 2007-04-30 Cluster File Systems, Inc. <info@clusterfs.com>
324 * Support for kernels:
325 2.4.21-47.0.1.EL (RHEL 3)
327 2.6.9-42.0.10.EL (RHEL 4)
328 2.6.12.6 vanilla (kernel.org)
329 2.6.16.27-0.9 (SLES 10)
330 * Recommended e2fsprogs version: 1.39.cfs6
331 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
335 Frequency : occasional, depends on client load and configuration
336 Bugzilla : 12181, 12203
337 Description: data loss for recently-modified files
339 Details : In some cases it is possible that recently written or created
340 files may not be written to disk in a timely manner (this should
341 normally be within 30s unless client IO load is very high).
342 The problem appears as zero-length files or files that are a
343 multiple of 1MB in size after a client crash or client eviction
344 that are missing data at the end of the file.
346 This problem is more likely to be hit on clients where files are
347 repeatedly created and unlinked in the same directory, clients
348 have a large amount of RAM, have many CPUs, the filesystem has
349 many OSTs, the clients are rebooted frequently, and/or the files
350 are not accessed by other nodes after being written.
352 The presence of the problem can be detected by looking at
353 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
354 smaller than the second (nr_unused) then dirty files will not
355 be flushed automatically to disk. "sync; sleep 10" should be
356 run several times on the node before unmounting it to update
357 Lustre (this is also safe to run on nodes without this problem).
359 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
360 that can cause this same problem, so customers using that kernel
361 also need to update the kernel in addition to Lustre. In order
362 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
364 It is normal that files written just before a client crash (less
365 than 30s) may not yet have been flushed to disk, even for local
369 Frequency : frequent on thin XT3 nodes
371 Description: UUID collision on thin XT3 Linux nodes
372 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
373 randomly, since we relied on an insufficiently-seeded PRNG.
378 Description: OSS hangs after "All ost request buffers busy"
379 Details : A deadlock between quota and journal operations caused OSS
380 hangs after printing "All ost request buffers busy."
383 Frequency : always on liblustre builds
385 Description: Cleanup compiler warnings on liblustre
387 Severity : enhancement
389 Description: Move random uuid functions to prng.c.
391 ------------------------------------------------------------------------------
393 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
395 * Support for kernels:
396 2.6.16.21-0.8 (SLES10)
397 2.6.9-42.0.8EL (RHEL 4)
399 2.4.21-47.0.1.EL (RHEL 3)
400 2.6.12.6 vanilla (kernel.org)
401 * Recommended e2fsprogs version: 1.39.cfs5
404 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
406 Description: System hang on startup
407 Details : This bug allowed the liblustre (e.g. catamount) client to
408 return to the app before handling all startup RPCs. This
409 could leave the node unresponsive to lustre network traffic
410 and manifested as a server ptllnd timeout.
412 Severity : enhancement
414 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
415 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
416 boolean to control whether to print peer debug info when a
417 client's RPC times out.
422 Description: make O_SYNC working on 2.6 kernels
423 Details : 2.6 kernels use different method for mark pages for write,
424 so need add a code to lustre for O_SYNC work.
429 Description: Failure to close file and release space on NFS
430 Details : Put inode details into lock acquired in ll_intent_file_open.
431 Use mdc_intent_lock in ll_intent_open to properly
432 detect all kind of errors unhandled by mdc_enqueue
437 Description: proc file read during shutdown sometimes raced obd removal,
439 Details : Add lock to prevent obd access after proc file removal.
442 Frequency : Only for files larger than 4GB on 32-bit clients.
444 Description: improperly doing page alignment of locks
445 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
446 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
451 Description: RPCs being resent when they shouldn't be
452 Details : Some RPCs that should not be resent are being resent. This
453 can cause inconsistencies in the RPC state machine. Do not
454 resend such requests.
457 Frequency : rare, only with NFS export
459 Description: Crash on NFS re-export node
460 Details : under very unusual load conditions an assertion is hit in
461 ll_intent_file_open()
464 Frequency : only if OST filesystem is corrupted
466 Description: client incorrectly hits assertion in ptlrpc_replay_req()
467 Details : for a short time RPCs with bulk IO are in the replay list,
468 but replay of bulk IOs is unimplemented. If the OST filesystem
469 is corrupted due to disk cache incoherency and then replay is
470 started it is possible to trip an assertion. Avoid putting
471 committed RPCs into the replay list at all to avoid this issue.
476 Description: large O_DIRECT requests fail under memory pressure/fragmentation
477 Details : Large single O_DIRECT read and write calls can fail to allocate
478 a sufficiently large buffer to process the request. In case of
479 allocation failure the allocation is retried with a smaller
480 buffer and broken into smaller requests.
485 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
486 > 32000 subdirectories
487 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
488 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
489 subdirectory count crosses 32000. This will aid e2fsck to
490 correctly handle more than 32000 subdirectories.
495 Description: versioning check is incomplete
496 Details : Checking the version difference of client vs. server, report
497 error if the gap is too big.
502 Description: add support PG_writeback bit
503 Details : add support for PG_writeback bit for Lustre, for more carefull
504 work with page cache in 2.6 kernel. This also fix some deadlocks
505 and remove hack for work O_SYNC with 2.6 kernel.
507 ------------------------------------------------------------------------------
509 2006-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
511 * Support for kernels:
512 2.6.16.21-0.8 (SLES10)
513 2.6.9-42.0.3EL (RHEL 4)
515 2.4.21-47.0.1.EL (RHEL 3)
516 2.6.12.6 vanilla (kernel.org)
519 * The backwards-compatible /proc/sys/portals symlink has been removed
520 in this release. Before upgrading, please ensure that you change
521 any configuration scripts or /etc/sysctl.conf files that access
522 /proc/sys/portals/* or sysctl portals.* to use the corresponding
523 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
524 in advance of the upgrade on any system running Lustre 1.4.6 or
525 newer, since /proc/sys/lnet was added in that version.
526 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
530 Frequency : MDS failover only, very rarely
532 Description: "went back in time" messages on mds failover
533 Details : The greatest transno may be lost when the current operation
534 finishes with an error (transno==0) and the client's last_rcvd
535 record is over-written. Save the greatest transno in the
536 mds_last_transno for this case.
539 Frequency : always for specific kernels and striping counts
541 Description: client may get "Matching packet too big" without ACL support
542 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
543 when trying to access files in certain configurations. The
544 clients should in fact be denied when mounting because they do
548 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
550 Description: many clients connecting with IO in progress causes connect timeouts
551 Details : Avoid synchronous journal commits to avoid delays caused by many
552 clients connecting/disconnecting when bulk IO is in progress.
553 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
554 OST_IO_PORTAL to avoid delays behind potentially many pending
558 Frequency : occasionally with multiple writers to a single file
560 Description: shared writes to file may result in wrong size reported by stat()
561 Details : Allow growing of kms when extent lock is cancelled
564 Frequency : always with random mmap IO to multi-striped file
566 Description: mmap write might be lost if we are writing to a 'hole' in stripe
567 Details : Only if the hole is at the end of OST object so that kms is too
568 small. Fix is to increase kms accordingly in ll_nopage.
571 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
573 Description: writes to a missing object would leak memory on the OST
574 Details : If there is an inconsistency between the MDS and OST filesystems,
575 such that the MDS references an object that doesn't exist, writes
576 to that object will leak memory due to incorrect cleanup in the
577 error handling path, eventually running out of memory on the OST.
582 Description: Creating too long symlink causes lustre errors
583 Details : Check symlink and name lengths before sending requests to MDS.
586 Frequency : only if flock is enabled (not on by default)
588 Description: posix locks not released on fd closure on 2.6.9+
589 Details : We failed to add posix locks to list of inode locks on 2.6.9+
590 kernels, this caused such locks not to be released on fd close and
591 then assertions on fs unmount about still used locks.
594 Frequency : MDS failover only, very rarely
596 Description: clients may get ASSERTION(granted_lock != NULL)
597 Details : When request was taking a long time, and a client was resending
598 a getattr by name lock request. The were multiple lock requests
599 with the same client lock handle and
600 mds_getattr_name->fixup_handle_for_resent_request found one of the
601 lock handles but later failed with ASSERTION(granted_lock != NULL).
606 Description: handle->h_buffer_credits > 0, assertion failure
607 Details : h_buffer_credits is zero after truncate, causing assertion
608 failure. This patch extends the transaction or creates a new
612 Frequency : NFS re-export or patchless client
613 Bugzilla : 11179, 10796
614 Description: Crash on NFS re-export node (__d_move)
615 Details : We do not want to hash the dentry if we don't have a lock.
616 But if this dentry is later used in d_move, we'd hit uninitialised
617 list head d_hash, so we just do this to init d_hash field but
618 leave dentry unhashed.
621 Frequency : NFS re-export or patchless client
623 Description: NFS exports has problem with symbolic link
624 Details : lustre client didn't properly install dentry when re-exported
625 to NFS or running patchless client.
628 Frequency : NFS re-export or patchless client
630 Description: Various nfs/patchless fixes.
631 Details : fixes reuse disconected alias for lookup process - this fixes
632 warning "find_exported_dentry: npd != pd",
633 fix permission error with open files at nfs.
634 fix apply umaks when do revalidate.
637 Frequency : occasional
639 Description: Crash on NFS re-export node
640 Details : calling clear_page() on the wrong pointer triggered oops in
641 generic_mapping_read().
644 Frequency : rarely, using O_DIRECT IO
646 Description: unaligned directio crashes client with LASSERT
647 Details : check for unaligned buffers before trying any requests.
650 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
652 Description: stale data returned from RAID cache
653 Details : If only a small amount of IO is done to the RAID device before
654 reading it again it is possible to get stale data from the RAID
655 cache instead of reading it from disk.
658 Frequency : depends on arch, kernel and compiler version, always on sles10
661 Description: recursive or deep enough symlinks cause stack overflow
662 Details : getting rid of large stack-allocated variable in
666 Frequency : depends on hardware
668 Description: lustre write performance loss in the SLES10 kernel
669 Details : the performance loss is caused by using of write barriers in the
670 ext3 code. The SLES10 kernel turns barrier support on by
671 default. The fix is to undo that change for ldiskfs.
673 ------------------------------------------------------------------------------
675 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
677 * Support for kernels:
678 2.6.9-42.0.3EL (RHEL 4)
680 2.4.21-47.0.1.EL (RHEL 3)
681 2.6.12.6 vanilla (kernel.org)
685 Frequency : quota enabled and large files being deleted
687 Description: releasing more than 4GB of quota at once hangs OST
688 Details : If a user deletes more than 4GB of files on a single OST it
689 will cause the OST to spin in an infinite loop. Release
690 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
695 Description: statfs data retrieved from /proc may be stale or zero
696 Details : When reading per-device statfs data from /proc, in the
697 {kbytes,files}_{total,free,avail} files, it may appear
698 as zero or be out of date.
701 Frequency : systems with MD RAID1 external journal devices
703 Description: lconf's call to blkid is confused by RAID1 journal devices
704 Details : Use the "blkid -l" flag to locate the MD RAID device instead
705 of returning all block devices that match the journal UUID.
708 Frequency : always, for aggregate stripe size over 4GB
710 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
711 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
712 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
715 Frequency : always if "lfs find" used on a local file/directory
717 Description: "lfs find" segfaults if used on a local file/directory
718 Details : The case where a directory component was not specified wasn't
719 handled correctly. Handle this properly.
722 Frequency : always on ppc64
724 Description: the write to an ext3 filesystem mounted with mballoc got stuck
725 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
726 perform endianness conversion.
729 Frequency : rarely (truncate to non-zero file size after write under load)
730 Bugzilla : 10730, 10687
731 Description: Files padded with zeros to next 4K multiple
732 Details : With filesystems mounted using the "extents" option (2.6 kernels)
733 it is possible that files that are truncated to a non-zero size
734 immediately after being written are filled with zero bytes beyond
735 the truncated size. No file data is lost.
737 Severity : enhancement
739 Description: Allow recovery/failover for liblustre clients.
740 Details : liblustre clients were unaware of failover configurations until
743 Severity : enhancement
745 Description: user file locks should fail when not mounting with flock option
746 Details : Set up an error-returning stub in ll_file_operations.lock field
747 to prevent incorrect behaviour when client is mounted without
748 flock option. Also, set up properly f_op->flock field for
752 Frequency : always on ia64
754 Description: "lfs df" loops on printing out MDS statfs information
755 Details : The obd_ioctl_data was not initialized and in some systems
756 this caused a failure during the ioctl that did not return
757 an error. Initialize the struct and return an error on failure.
760 Frequency : SLES 9 only
762 Description: Error of copying files with lustre special EAs as root
763 Details : Client side always return success for setxattr call for lustre
764 special xattr (currently only "trusted.lov").
767 Frequency : rarely on clusters with both ia64+i386 clients
769 Description: ia64+i686 clients doing shared IO on the same file may LBUG
770 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
771 clients are doing concurrent writes to the same file it is
772 possible that the ia64 clients may LASSERT because the OST
773 extent locks are not PAGE_SIZE aligned. Ensure that grown
774 locks are always aligned on the request boundary.
777 Frequency : specific use, occasional
779 Description: Overwriting in use executable truncates on-disk binary image
780 Details : If one node attempts to overwrite an executable in use by
781 another node, we now correctly return ETXTBSY instead of
787 Description: chmod on Lustre root is propagated to other clients
788 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
789 invalid by the follow_mount time.
792 Frequency : liblustre clients only
794 Description: Race in 'instant cancel' lock handling could lead to such locks
795 never to be granted in case of SMP MDS
796 Details : Do not destroy not yet granted but cbpending locks in
800 Frequency : replay/resend of open
802 Description: non null lock assetion failure in mds_intent_policy
803 Details : Trying to replay/resend lockless open requests resulted in
804 mds_open() returning 0 with no lock. Now it sets a flag if
805 a lock is going to be returned.
807 Severity : enhancement
809 Description: Checksum enhancements
810 Details : New checksum enhancements allow for resending RPCs that failed
813 Severity : enhancement
815 Description: Tunables on number of dirty pages in cacche
816 Details : Allow to set limit on number of dirty pages cached.
821 Description: client crash on unmount - lock still has references
822 Details : In some error handling cases it was possible to leak a lock
823 reference on a client while accessing a file. This was not
824 harmful to the client during operation, but would cause the
825 client to crash when the filesystem is unmounted.
828 Frequency : specific case, rare
830 Description: ETXTBSY on mds though file not in use by client
831 Details : ETXTBSY is no longer incorrectly returned when attempting to
832 chmod or chown a directory that the user previously tried to
833 execute or a currently-executing binary.
836 Frequency : extremely rare except on liblustre-based clients
838 Description: Lustre space not freed when files are deleted
839 Details : Clean up open-unlinked files after client eviction. Previously
840 the unlink was skipped and the files remained as orphans.
845 Description: OST failure "would be an LBUG" in waiting_locks_callback()
846 Details : In some cases it was possible to send a blocking callback to
847 a client doing a glimpse, even though that client didn't get
848 a lock granted. When the glimpse lock is cancelled on the OST
849 the freed lock is left on the waiting list and corrupted the list.
852 Frequency : all core dumps
854 Description: Broke core dumps to lustre
855 Details : Negative dentry may be unhashed if parent does not have UPDATE
856 lock, but some callers, e.g. do_coredump, expect dentry to be
857 hashed after successful create, hash it in ll_create_it.
859 ------------------------------------------------------------------------------
861 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
863 * Support for kernels:
864 2.6.9-42.0.2.EL (RHEL 4)
866 2.4.21-40.EL (RHEL 3)
867 2.6.12.6 vanilla (kernel.org)
871 Frequency : always on RHEL 3
873 Description: Number of open files grows over time
874 Details : The number of open files grows over time, whether or not
875 Lustre is started. This was due to a filp leak introduced
876 by one of our kernel patches.
878 ------------------------------------------------------------------------------
880 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
882 * Support for kernels:
885 2.4.21-40.EL (RHEL 3)
886 2.6.12.6 vanilla (kernel.org)
891 Bugzilla : 5719, 9635, 9792, 9684
892 Description: OST (or MDS) trips assertions in (re)connection under heavy load
893 Details : If a server is under heavy load and cannot reply to new
894 connection requests before the client resends the (re)connect,
895 the connection handling code can behave badly if two service
896 threads are concurrently handing separate (re)connections from
897 the same client. Add better locking to the connection handling
898 code, and ensure that only a single connection will be processed
899 for a given client UUID, even if the lock is dropped.
901 Severity : enhancement
903 Description: add TCP zero-copy support to kernel
904 Details : Add support to the kernel TCP stack to allow zero-copy bulk
905 sends if the hardware supports scatter-gather and checksumming.
906 This allows socklnd to do client-write and server-read more
907 efficiently and reduce CPU utilization from skbuf copying.
910 Frequency : only if NFS exporting from client
912 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
913 Details : If setting attributes on a file created under NFS that had
914 never been opened it would be possible to oops the client
915 if the file had no objects.
919 Bugzilla : 9326, 10402, 10897
920 Description: client crash in ptlrpcd_wake() thread when sending async RPC
921 Details : It is possible that ptlrpcd_wake() dereferences a freed async
922 RPC. In rare cases the ptlrpcd thread alread processed the RPC
923 before ptlrpcd_wake() was called and the request was freed.
926 Frequency : always for liblustre
928 Description: liblustre client does MDS+OSTs setattr RPC for each write
929 Details : When doing a write from a liblustre client, the client
930 incorrectly issued an RPC to the MDS and each OST the file was
931 striped over in order to update the timestamps. When writing
932 with small chunks and many clients this could overwhelm the MDS
933 with RPCs. In all cases it would slow down the write because
934 these RPCs are unnecessary.
936 Severity : enhancement
938 Description: allow number of MDS service threads to be changed at module load
939 Details : It is now possible to change the number of MDS service threads
940 running. Adding "options mds mds_num_threads={N}" to the MDS's
941 /etc/modprobe.conf will set the number of threads for the next
942 time Lustre is restarted (assuming the "mds" module is also
943 reloaded at that time). The default number of threads will
944 stay the same, 32 for most systems.
949 Description: OST crash if filesystem is unformatted or corrupt
950 Details : If an OST is started on a device that has never been formatted
951 or if the filesystem is corrupt and cannot even mount then the
952 error handling cleanup routines would dereference a NULL pointer.
957 Description: NULL pointer deref in llap_from_page.
958 Details : get_cache_page_nowait can return a page with NULL (or otherwise
959 incorrect) mapping if the page was truncated/reclaimed while it was
960 searched for. Check for this condition and skip such pages when
961 doing readahead. Introduce extra check to llap_from_page() to
962 verify page->mapping->host is non-NULL (so page is not anonymous).
965 Frequency : Sometimes when using sys_sendfile
967 Description: "page not covered by a lock" warnings from ll_readpage
968 Details : sendfile called ll_readpage without right page locks present.
969 Now we introduced ll_file_sendfile that does necessary locking
970 around call to generic_file_sendfile() much like we do in
974 Frequency : with certain MDS communication failures at client mount time
976 Description: NULL pointer deref after failed client mount
977 Details : a client connection request may delayed by the network layer
978 and not be sent until after the PTLRPC layer has timed out the
979 request. If the client fails the mount immediately it will try
980 to clean up before the network times out the request. Add a
981 reference from the request import to the obd device and delay
982 the cleanup until the network drops the request.
985 Frequency : occasionally during client (re)connect
987 Description: assertion failure during client (re)connect
988 Details : processing a client connection request may be delayed by the
989 client or server longer than the client connect timeout. This
990 causes the client to resend the connection request. If the
991 original connection request is replied in this interval, the
992 client may trip an assertion failure in ptlrpc_connect_interpret()
993 which thought it would be the only running connect process.
996 Frequency : only with obd_echo servers and clients that are rebooted
998 Description: kernel BUG accessing uninitialized data structure
999 Details : When running an obd_echo server it did not start the ping_evictor
1000 thread, and when a client was evicted an uninitialized data
1001 structure was accessed. Start the ping_evictor in the RPC
1002 service startup instead of the OBD startup.
1004 Severity : enhancement
1005 Bugzilla : 10193 (patchless)
1006 Description: Remove dependency on various unexported kernel interfaces.
1007 Details : No longer need reparent_to_init, exit_mm, exit_files,
1008 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1011 Frequency : rare (only users of deprecated and unsupported LDAP config)
1013 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1014 Details : LDAP apparently contains 'lustreName' attributes instead of
1015 'name'. A simple remapping of the name is sufficient.
1018 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1020 Description: waiting_locks_callback trips kernel BUG if client is evicted
1021 Details : Running with the dump_on_timeout debug flag turned on makes
1022 it possible that the waiting_locks_callback() can try to dump
1023 the Lustre kernel debug logs from an interrupt handler. Defer
1024 this log dumping to the expired_lock_main() thread.
1026 Severity : enhancement
1028 Description: Support NFS exporting on 2.6 kernels.
1029 Details : Implement non-rawops metadata methods for NFS server to use without
1030 changing NFS server code.
1033 Frequency : very rare (synthetic metadata workload only)
1035 Description: two racing renames might cause an MDS thread to deadlock
1036 Details : Running the "racer" program may cause one MDS thread to rename
1037 a file from being the source of a rename to being the target of
1038 a rename at exactly the same time that another thread is doing
1039 so, and the second thread has already enqueued these locks after
1040 doing a lookup of the target and is trying to relock them in
1041 order. Ensure that we don't try to re-lock the same resource.
1044 Frequency : only very large systems with liblustre clients
1046 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1047 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1048 all OSTs in parallel. This allows the eviction of stale liblustre
1049 clients to proceed much faster than if they were done in series,
1050 and also offers similar improvements for other set_info RPCs.
1055 Description: excessive CPU usage during initial read phase on client
1056 Details : During the initial read phase on a client, it would agressively
1057 retry readahead on the file, consuming too much CPU and impacting
1058 performance (since 1.4.5.8). Improve the readahead algorithm
1059 to avoid this, and also improve some other common cases (read
1060 of small files in particular, where "small" is files smaller than
1061 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1066 Description: MDS crash when receiving packet with unknown intent.
1067 Details : Do not LBUG in unknown intent case, just return -EFAULT
1069 Severity : enhancement
1070 Bugzilla : 9293, 9385
1071 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1072 Details : Do not serialize getattr (non-intent version) and statfs.
1075 Frequency : occasional, when OST network is overloaded/intermittent
1077 Description: client evicted by OST after bulk IO timeout
1078 Details : If a client sends a bulk IO request (read or write) the OST
1079 may evict the client if it is unresposive to its data GET/PUT
1080 request. This is incorrect if the network is overloaded (takes
1081 too long to transfer the RPC data) or dropped the OST GET/PUT
1082 request. There is no need to evict the client at all, since
1083 the pinger and/or lock callbacks will handle this, and the
1084 client can restart the bulk request.
1087 Frequency : Always when mmapping file with no objects
1089 Description: client crashes when mmapping file with no objects
1090 Details : Check that we actually have objects in a file before doing any
1091 operations on objects in ll_vm_open, ll_vm_close and
1097 Description: Request leak when working with deleted CWD
1098 Details : Introduce advanced request refcount tracking for requests
1099 referenced from lustre intent.
1101 Severity : Enhancement
1103 Description: Cache open file handles on client.
1104 Details : MDS now will return special lock along with openhandle, if
1105 requested and client is allowed to hold openhandle, even if unused,
1106 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1107 closing files for every read/write openration.
1109 Severity : Enhancement
1111 Description: Cache open negative dentries on client when possible.
1112 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1113 negative dentries on lock revocation.
1118 Description: Remounting a client read-only wasn't possible with a zconf mount
1119 Details : It wasn't possible to remount a client read-only with llmount.
1121 Severity : enhancement
1122 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1123 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1124 MPICH is included to add Lustre-specific ADIO interfaces.
1125 This is based closely on the UFS ADIO layer and only differs
1126 in file creation, in order to allow the OST striping to be set.
1127 This is user-contributed code and not supported by CFS.
1132 Description: extended inode attributes (immutable, append-only) work improperly
1133 when 2.4 and 2.6 kernels are used on client/server or vice versa
1134 Details : Introduce kernel-independent values for these flags.
1136 Severity : enhancement
1139 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1140 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1141 granularity. Now, Fractional MB granularity is be supported,
1142 this is very useful for low memory system.
1144 Severity : enhancement
1146 Description: Getattr by fid
1147 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1148 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1149 have locks and avoids getattr rpc requests for it on every stat.
1152 Frequency : Always, for filesystems larger than 2TB
1154 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1155 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1156 filesystems larger than 2TB to be mounted with mballoc enabled.
1159 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1161 Description: filesystem corruption for non-standard kernels and very large OSTs
1162 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1163 filesystems larger than 2TB is mounted then the kernel will
1164 silently corrupt the start of the filesystem. CONFIG_LBD is
1165 enabled for all CFS-supported kernels, but the possibility of
1166 this happening with a modified kernel config exists.
1168 Severity : enhancement
1170 Description: add client O_DIRECT support for 2.6 kernels
1171 Details : It is now possible to do O_DIRECT reads and writes to files
1172 in the Lustre client mountpoint on 2.6 kernel clients.
1174 Severity : enhancement
1176 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1177 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1178 parallel, not waiting for response from every OST before sending
1179 a rpc to the next OST.
1184 Description: setattr vs write race when updating file timestamps
1185 Details : Client processes that update a file timestamp into the past
1186 right after writing to the file (e.g. tar) it is possible that
1187 the updated file modification time can be reset to the current
1188 time due to a race between processing the setattr and write RPC.
1190 Severity : enhancement
1192 Description: Bring 'lfs find' closer in line with regular Linux find.
1193 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1194 -print0 options and obtains all the needed info through the lustre
1197 Severity : enhancement
1199 Description: support up to 1024 configured devices on one node
1200 Details : change obd_dev array from statically allocated to dynamically
1201 allocated structs as they are first used to reduce memory usage
1206 Description: Flush dirty partially truncated pages during truncate
1207 Details : Immediatelly flush partially truncated pages in filter_setattr,
1208 this way we completely avoid having any pages in page cache on OST
1209 and can retire ugly workarounds during writes to flush such pages.
1214 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1215 Details : protect inode from truncation within vfs_unlink() context
1216 just take a reference before calling vfs_unlink() and release it
1217 when parent's i_sem is free.
1220 Frequency : always, if extents are used on OSTs
1222 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1223 Details : OSTs using the extents format would not zero the high 16 bits of
1224 the index physical block number. This is not a problem for any
1225 OST filesystems smaller than 16TB, and no kernels support ext3
1226 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1227 new/modified files) and can be fixed for existing filesystems
1228 with e2fsprogs-1.39-cfs1.
1233 Description: import connection selection may be incorrect if timer wraps
1234 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1235 import connections to be ignored if the 32-bit jiffies counter
1236 wraps. Use a 64-bit jiffies counter.
1239 Frequency : during server recovery
1241 Description: crash after server is denying duplicate export
1242 Details : If clients are resending connect requests to the server, the
1243 server refuses to allow a client to connect multiple times.
1244 Fixed a bug in the handling of this case.
1247 Frequency : very large clusters immediately after boot
1249 Description: LNET request buffers exhausted under heavy short-term load
1250 Details : If a large number of client requests are generated on a service
1251 that has previously never seen so many requests it is possible
1252 that the request buffer growth cannot keep up with the spike in
1253 demand. Instead of dropping incoming requests, they are held in
1254 the LND until the RPC service can accept more requests.
1257 Frequency : Sometimes during replay
1259 Description: Assertion failure in ll_local_open after replay.
1260 Details : If replay happened on an open request reply before we were able
1261 to set replay handler, reply will become not swabbed tripping the
1262 assertion in ll_local_open. Now we set the handler right after
1263 recognising of open request
1266 Frequency : very rare
1268 Description: kernel reports "badness in vsnprintf"
1269 Details : Reading from the "recovery_status" /proc file in small chunks
1270 may cause a negative length in lprocfs_obd_rd_recovery_status()
1271 call to vsnprintf() (which is otherwise harmless). Exit early
1272 if there is no more space in the output buffer.
1274 Severity : enhancement
1276 Description: clear OBD RPC statistics by writing to them
1277 Details : It is now possible to clear the OBD RPC statistics by writing
1278 to the "stats" file.
1283 Description: Client mtime is not the same on different clients after utimes
1284 Details : In some cases, the client was using the utimes() syscall on
1285 a file cached on another node. The clients now validate the
1286 ctime from the MDS + OSTs to determine which one is right.
1291 Description: Inability to activate failout mode
1292 Details : lconf script incorrectly assumed that in python string's numeric
1293 value is used in comparisons.
1296 Frequency : always with multiple stripes per file
1298 Description: Inefficient object allocation for mutli-stripe files
1299 Details : When selecting which OSTs to stripe files over, for files with
1300 a stripe count that divides evenly into the number of OSTs,
1301 the MDS is always picking the same starting OST for each file.
1302 Return the OST selection heuristic to the original design.
1307 Description: mount failures may take full timeout to return an error
1308 Details : Under some heavy load conditions it is possible that a
1309 failed mount can wait for the full obd_timeout interval,
1310 possibly several minutes, before reporting an error.
1311 Instead return an error as soon as the status is known.
1313 ------------------------------------------------------------------------------
1315 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
1317 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1318 INTEROPERATE with older versions automatically. Please read the
1319 user documentation before upgrading any part of a live system.
1320 * WARNING: Lustre networking configuration changes are required with
1321 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1324 * Support for kernels:
1325 2.6.9-22.0.2.EL (RHEL 4)
1326 2.6.5-7.244 (SLES 9)
1327 2.6.12.6 vanilla (kernel.org)
1330 Severity : enhancement
1331 Bugzilla : 7981/8208
1332 Description: Introduced Lustre Networking (LNET)
1333 Details : LNET is new networking infrastructure for Lustre, it includes
1334 a reorganized network configuration mode (see the user
1335 documentation for full details) as well as support for routing
1336 between different network fabrics. Lustre Networking Devices
1337 (LNDs) for the supported network fabrics have also been
1338 created for this new infrastructure.
1340 Severity : enhancement
1341 Description: Introduced Access control lists
1342 Details : clients can set ACLs on files and directories in order to have
1343 more fine-grained permissions than the standard Unix UGO+RWX.
1344 The MDS must be started with the "-o acl" mount option.
1346 Severity : enhancement
1347 Description: Introduced filesystem quotas
1348 Details : Administrators may now establish per-user quotas on the
1351 Severity : enhancement
1353 Description: Configuration change for the XT3
1354 The PTLLND is now used to run Lustre over Portals on the XT3
1355 The configure option(s) --with-cray-portals are no longer used.
1356 Rather --with-portals=<path-to-portals-includes> is used to
1357 enable building on the XT3. In addition to enable XT3 specific
1358 features the option --enable-cray-xt3 must be used.
1363 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1364 Details : A race between allocating a new llcd and re-getting the llcd_lock
1365 allowed another thread to grab newly-allocated llcd.
1367 Severity : enhancement
1369 Description: 2.6 OST async journal commit and locking fix to improve performance
1370 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1371 2.6 kernels are now async as they already were in 2.4 kernels so
1372 that they can commit concurrently with the network bulk transfer.
1373 For block-allocated files the filter allocation semaphore is held
1374 to avoid filesystem fragmentation during allocation. BKL lock
1375 removed for 2.6 xattr operations where it is no longer needed.
1380 Description: lconf incorrectly determined whether two IP networks could talk
1381 Details : In some more complicated routing and multiple-network
1382 configurations, lconf will avoid trying to make a network
1383 connection to a disjoint part of the IP space. It was doing the
1384 math incorrectly for one set of cases.
1389 Description: Fix for potential infinite loop processing records in an llog.
1390 Details : If an llog record is corrupted/zeroed, it is possible to loop
1391 forever in llog_process(). Validate the llog record length
1392 and skip the remainder of the block on error.
1395 Frequency : occasional (liblustre only)
1397 Description: liblustre could not open files whose last component is a symlink
1398 Details : sysio_path_walk() would incorrectly pass the open intent to
1399 intermediate path components.
1402 Frequency : rare (liblustre only with non-standard tuning)
1403 Bugzilla : 7201 (7350)
1404 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1405 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1406 after it has been released, only do this for referenced locks
1408 Severity : enhancement
1410 Description: specifying an (invalid) directory default stripe_size of -1
1411 would reset the directory default striping
1412 Details : stripe_size -1 was used internally to signal directory stripe
1413 removal, now use "all default" to signal dir stripe removal
1414 as a directory striping of "all default" is not useful
1417 Frequency : common for large clusters running liblustre clients
1419 Description: doing an ls when liblustre clients are running is slow
1420 Details : sending a glimpse AST to a liblustre client waits for every AST
1421 to time out, as liblustre clients will not respond. Since they
1422 cannot cache data we refresh the OST lock LVB from disk instead.
1424 Severity : enhancement
1426 Description: doing an ls at the same time as file IO can be slow
1427 Details : enqueue and other "small" requests can be blocked behind many
1428 large IO requests. Create a new OST IO portal for non-IO
1429 requests so they can be processed faster.
1432 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1434 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1435 created file with mode 000
1436 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1437 allow an ftruncate() as a non-root user to a file with mode 000.
1438 Lustre now respects this flag to disable mode checks when
1439 truncating a file owned by the user
1442 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1445 Description: Revoking locks from clients that went dead or catatonic might take
1447 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1448 cancellation of such locks instant on servers without waiting for
1449 any reply from clients. Clients drops these locks when cancel
1450 notification from server is received without replying.
1453 Frequency : liblustre-only, when liblustre client dies or becomes busy
1455 Description: Doing ls on Linux clients can take a long time with active
1457 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1458 granting such locks to it in the first place if possible. Locks
1459 are taken by proxy on the OST during the read or write and
1460 dropped immediately afterward. Add connect flags handling, do
1461 not grant locks to liblustre clients for glimpse ASTs.
1463 Severity : enhancement
1465 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1466 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1467 sequential reads and behaves badly with applications doing random
1468 reads. Improve it by reading ahead at least read region, and
1469 avoiding excessive large RPC for small reads.
1471 Severity : enhancement
1473 Description: Creating more than 1000 files for a single job may cause a load
1474 imbalance on the OSTs if there are also a large number of OSTs.
1475 Details : qos_prep_create() uses an OST index reseed value that is an
1476 even multiple of the number of available OSTs so that if the
1477 reseed happens in the middle of the object allocation it will
1478 still utilize the OSTs as uniformly as possible.
1483 Description: OST or MDS may oops in ping_evictor_main()
1484 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1485 but doesn't restart at beginning of obd_exports_timed list
1488 Severity : enhancement
1490 Description: improve by-nid export eviction on the MDS and OST
1491 Details : allow multiple exports with the same NID to be evicted at one
1492 time without re-searching the exports list.
1495 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1497 Description: MDS may oops in groups_free()
1498 Details : in rare race conditions a newly allocated group_info struct is
1499 freed again, and this can be NULL. The 2.4 compatibility code
1500 for groups_free() checked for a NULL pointer, but 2.6 did not.
1503 Frequency : common for liblustre clients doing little filesystem IO
1504 Bugzilla : 9352, 7313
1505 Description: server may evict liblustre clients accessing contended locks
1506 Details : if a client is granted a lock or receives a completion AST
1507 with a blocking AST already set it would not reply to the AST
1508 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1511 Frequency : lfs setstripe, only systems with more than 160 OSTs
1513 Description: unable to set striping with a starting offset beyond OST 160
1514 Details : llapi_create_file() incorrectly limited the starting stripe
1515 index to the maximum single-file stripe count.
1518 Frequency : LDAP users only
1520 Description: lconf did not handle in-kernel recovery with LDAP properly
1521 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1523 Severity : enhancement
1525 Description: bind OST threads to NUMA nodes to improve performance
1526 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1527 node and do their allocations there to localize memory access
1529 Severity : enhancement
1531 Description: llmount can determine client NID directly from Myrinet (GM)
1532 Details : the client NID code from gmnalnid was moved directly into
1533 llmount, removing the need to use this or specifying the
1534 client NID explicitly when mounting GM clients with zeroconf
1537 Frequency : if client is started with down MDS
1539 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1540 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1541 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1542 ptlrpc_import_delayed_req() because we didn't check if the
1543 request was interrupted, and we also didn't break out of the
1544 event loop if there was no timeout
1549 Description: data loss during non-page-aligned writes to a single file from
1550 both multiple nodes and multiple threads on one node at same time
1551 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1552 inconsistency led to false short-reads, that were cached and later
1553 used by ->prepare_write() to fill in partially written page,
1554 leading to data loss.
1557 Frequency : always, if lconf --abort_recovery used
1559 Description: lconf --abort_recovery fails with 'Operation not supported'
1560 Details : lconf was attempting to abort recovery on the MDT device and not
1563 Severity : enhancement
1565 Description: remove cleanup logs
1566 Details : replace lconf-generated cleanup logs with lustre internal
1567 cleanup routines. Eliminates the need for client-cleanup and
1570 Severity : enhancement
1572 Description: add support for EAs (user and system) on lustre filesystems
1573 Details : it is now possible to store extended attributes in the Lustre
1574 client filesystem, and with the user_xattr mount option it
1575 is possible to allow users to store EAs on their files also
1577 Severity : enhancement
1579 Description: Add possibility (config option) to show minimal available OST free
1581 Details : When compiled with --enable-mindf configure option, statfs(2)
1582 (and so, df) will return least minimal free space available from
1583 all OSTs as amount of free space on FS, instead of summary of
1584 free spaces of all OSTs.
1586 Severity : enhancement
1588 Description: do not expand extent locks acquired on OST-side
1589 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1590 by server: they are not cached anyway.
1593 Frequency : when mmap is used/binaries executed from Lustre
1595 Description: Unmmap pages before throwing them away from read cache.
1596 Details : llap_shrink cache now attempts to unmap pages before discarding
1597 them (if unmapping failed - do not discard). SLES9 kernel has
1598 extra checks that trigger if this unmapping is not done first.
1603 Description: lconf didn't resolve symlinks before checking to see whether a
1604 given mountpoint was already in use
1607 Frequency : when migrating failover services
1608 Bugzilla : 6395, 9514
1609 Description: When migrating a subset of services from a node (e.g. failback
1610 from a failover service node) the remaining services would
1611 time out and evict clients.
1612 Details : lconf --force (implied by --failover) sets the global obd_timeout
1613 to 5 seconds in order to quickly disconnect, but this caused
1614 other RPCs to time out too quickly. Do not change the global
1615 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1617 Severity : enhancement
1618 Frequency : if MDS is started with down OST
1619 Bugzilla : 9439,5706
1620 Description: Allow startup/shutdown of an MDS without depending on the
1621 availability of the OSTs.
1622 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1623 Add appropriate locking and lov-osc refcounts for safe
1624 cleaning. Add osc abort_inflight calls in case the
1625 synchronize never started.
1628 Frequency : occasional (Cray XT3 only)
1630 Description: root not authorized to access files in CRAY_PORTALS environment
1631 Details : The client process capabilities were not honoured on the MDS in
1632 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1633 been accessed by an authorized user then root was able to access
1634 the file on the local client also. The root user capabilities
1635 are now allowed on the MDS, as this environment has secure UID.
1638 Frequency : occasional
1640 Description: ldiskfs "too long searching" message happens too often
1641 Details : A debugging message (otherwise harmless) prints too often on
1642 the OST console. This has been reduced to only happen when
1643 there are fragmentation problems on the filesystem.
1648 Description: Division by zero in statfs when all OSCs are inactive
1649 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
1650 lov_statfs divides by value returned by lov_get_stripecnt().
1654 Bugzilla : 9489, 3273
1655 Description: First write from each client to each OST was only 4kB in size,
1656 to initialize client writeback cache, which caused sub-optimal
1657 RPCs and poor layout on disk for the first writen file.
1658 Details : Clients now request an initial cache grant at (re)connect time
1659 and so that they can start streaming writes to the cache right
1660 away and always do full-sized RPCs if there is enough data.
1661 If the OST is rebooted the client also re-establishes its grant
1662 so that client cached writes will be honoured under the grant.
1667 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
1668 Details : Now I/O RPCs go to different portal number and (presumably) fast
1669 lock requests (and glimses) and other RPCs get their own service
1670 threads pool that should be able to service those RPCs
1673 Severity : enhancement
1675 Description: Ability to exchange lustre version between client and servers and
1676 issue warnings at client side if client is too old. Also for
1677 liblustre clients there is ability to refuse connection of too old
1679 Details : New 'version' field is added to connect data structure that is
1680 filled with version info. That info is later checked by server and
1684 Frequency : rare, liblustre only.
1685 Bugzilla : 9296, 9581
1686 Description: Two simultaneous writes from liblustre at offset within same page
1687 might proceed at the same time overwriting eachother with stale
1689 Details : I/O lock withing llu_file_prwv was released too early, before data
1690 actually was hitting the wire. Extended lock-holding time until
1691 server acknowledges receiving data.
1694 Frequency : extremely rare. Never observed in practice.
1696 Description: avoid generating lustre_handle cookie of 0.
1697 Details : class_handle_hash() generates handle cookies by incrementing
1698 global counter, and can hit 0 occasionaly (this is unlikely, but
1699 not impossible, because initial value of cookie counter is
1700 selected randonly). Value of 0 is used as a sentinel meaning
1701 "unassigned handle" --- avoid it. Also coalesce two critical
1702 sections in this function into one.
1704 Severity : enhancement
1706 Description: allow liblustre clients to delegate truncate locking to OST
1707 Details : To avoid overhead of locking, liblustre client instructs OST to
1708 take extent lock in ost_punch() on client's behalf. New connection
1709 flag is added to handle backward compatibility.
1711 Severity : enhancement
1712 Bugzilla : 4928, 7341, 9758
1713 Description: allow number of OST service threads to be specified
1714 Details : a module parameter allows the number of OST service threads
1715 to be specified via "options ost ost_num_threads={N}" in the
1716 OSS's /etc/modules.conf or /etc/modprobe.conf.
1720 Bugzilla : 6146, 9635, 9895
1721 Description: servers crash with bad pointer in target_handle_connect()
1722 Details : In rare cases when a client is reconnecting it was possible that
1723 the connection request was the last reference for that export.
1724 We would temporarily drop the export reference and get a new
1725 one, but this may have been the last reference and the export
1726 was just destroyed. Get new reference before dropping old one.
1728 Severity : enhancement
1729 Frequency : if client is started with failover MDS
1731 Description: Allow multiple MDS hostnames in the mount command
1732 Details : Try to read the configuration from all specified MDS
1733 hostnames during a client mount in case the "primary"
1736 Severity : enhancement
1738 Description: Stop sending data to evicted clients as soon as possible.
1739 Details : Check if the client we are about to send or are sending data to
1740 was evicted already. (Check is done every second of waiting,
1741 for which l_wait_event interface was extended to allow checking
1742 of exit condition at specified intervals).
1745 Frequency : rare, normally only when NFS exporting is done from client
1747 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
1749 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
1752 Frequency : always, if srand() is called before liblustre initialization
1754 Description: Liblustre uses system PRNG disturbing its usage by user application
1755 Details : Introduce internal to lustre fast and high-quality PRNG for
1756 lustre usage and make liblustre and some other places in generic
1757 lustre code to use it.
1759 Severity : enhancement
1760 Bugzilla : 9477, 9557, 9870
1761 Description: Verify that the MDS configuration logs are updated when xml is
1762 Details : Check if the .xml configuration logs are newer than the config
1763 logs stored on the MDS and report an error if this is the case.
1764 Request --write-conf, or allow starting with --old_conf.
1766 Severity : enhancement
1768 Description: Handle symlinks in the path when checking if Lustre is mounted.
1769 Details : Resolve intermediate symlinks when checking if a client has
1770 mounted a filesystem to avoid duplicate client mounts.
1775 Description: lconf can hit an error exception but still return success.
1776 Details : The lconf command catches the Command error exception at the top
1777 level script context and will exit with the associated exit
1778 status, but doesn't ensure that this exit status is non-zero.
1783 Description: failure of ptlrpc thread startup can cause oops
1784 Details : Starting a ptlrpc service thread can fail if there are a large
1785 number of threads or the server memory is very fragmented.
1786 Handle this without oopsing.
1789 Frequency : always, only if liblustre and non-default acceptor port was used
1791 Description: liblustre cannot connect to servers with non-default acceptor port
1792 Details : tcpnal_set_default_params() was not called and was therefore
1793 ignoring the environment varaible TCPNAL_PORT, as well as other
1794 TCPNAL_ environment variables
1799 Description: two objects could be created on the same OST for a single file
1800 Details : If an OST is down, in some cases it was possible to create two
1801 objects on a single OST for a single file. No problems other
1802 than potential performance impact and spurious error messages.
1806 Bugzilla : 5681, 9562
1807 Description: Client may oops in ll_unhash_aliases
1808 Details : Client dcache may become inconsistent in race condition.
1809 In some cases "getcwd" can fail if the current directory is
1815 Description: Inode refcounting problems in NFS export code
1816 Details : link_raw functions used to call d_instantiate without obtaining
1817 extra inode reference first.
1821 Bugzilla : 9942, 9903
1822 Description: Referencing freed requests leading to crash, memleaks with NFS.
1823 Details : We used to require that call to ll_revalidate_it was always
1824 followed by ll_lookup_it. Also with revalidate_special() it is
1825 possible to call ll_revalidate_it() twice for the same dentry
1826 even if first occurence returned success. This fix changes semantic
1827 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
1828 reference on a request referred from the intent.
1829 ll_intent_release() then releases such a request.
1832 Frequency : rare, normally benchmark loads only
1834 Description: unlinked inodes were kept in memory on the client
1835 Details : If a client is repeatedly creating and unlinking files it
1836 can accumulate a lot of stale inodes in the inode slab cache.
1837 If there is no other client load running this can cause the
1838 client node to run out of memory. Instead flush old inodes
1839 from client cache that have the same inode number as a new inode.
1842 Frequency : SLES9 2.6.5 kernel and long filenames only
1843 Bugzilla : 9969, 10379
1844 Description: utime reports stale NFS file handle
1845 Details : SLES9 uses out-of-dentry names in some cases, which confused
1846 the lustre dentry revalidation. Change it to always use the
1850 Frequency : rare, unless heavy write-truncate concurrency is continuous
1851 Bugzilla : 4180, 6984, 7171, 9963, 9331
1852 Description: OST becomes very slow and/or deadlocked during object unlink
1853 Details : filter_destroy() was holding onto the parent directory lock
1854 while truncating+unlinking objects. For very large objects this
1855 may block other threads for a long time and slow overall OST
1856 responsiveness. It may also be possible to get a lock ordering
1857 deadlock in this case, or run out of journal credits because of
1858 the combined truncate+unlink. Solution is to do object truncate
1859 first in one transaction without parent lock, and then do the
1860 final unlink in a new transaction with the parent lock. This
1861 reduces the lock hold time dramatically.
1864 Frequency : rare, 2.4 kernels only
1866 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
1867 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
1868 before unmounting their filesystem in order to avoid deadlock.
1869 The kernel_locked() function in 2.4 kernels only checks whether
1870 the kernel lock is held, not whether it is this process that is
1871 holding it as 2.6 kernels do.
1876 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
1877 Details : The client ptlrpc code may be trying to reconnect to a down
1878 server before a previous connection attempt has timed out.
1879 Increase the reconnect interval to be longer than the connection
1880 timeout interval to avoid sending duplicate connections to
1884 Frequency : echo_client brw_test command
1886 Description: fix echo_client to work with OST preallocated code
1887 Details : OST preallocation code (5137) didn't take echo_client IO path
1888 into account: echo_client calls filter methods outside of any
1889 OST thread and, hence, there is no per-thread preallocated
1890 pages and buffers to use. Solution: hijack pga pages for IO. As
1891 a byproduct, this avoids unnecessary data copying.
1895 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
1896 Description: Client can oops in mdc_commit_close() after open replay
1897 Details : It was possible for the MDS to return an open request with no
1898 transaction number in mds_finish_transno() if the client was
1899 evicted, but without actually returning an error. Clients
1900 would later try to replay that open and may trip an assertion
1901 Simplify the client close codepath, and always return an error
1902 from the MDS in case the open is not successful.
1905 Frequency : rare, 2.6 OSTs only
1907 Description: OST may deadlock under high load on fragmented files
1908 Details : If there was a heavy load and highly-fragmented OST filesystems
1909 it was possible to have all the OST threads deadlock waiting on
1910 allocation of biovecs, because the biovecs were not released
1911 until the entire RPC IO was completed. Instead, release biovecs
1912 as soon as they are complete to ensure forward IO progress.
1914 Severity : enhancement
1916 Description: Support for specifying external journal device at mount
1917 Details : If an OST or MDS device is formatted with an external journal
1918 device, this device major/minor is stored in the ext3 superblock
1919 and may not be valid for failover. Allow detecting and
1920 specifying the external journal at mount time.
1925 Description: Mounting an MDS with pending unlinked files may cause oops
1926 Details : target_finish_recovery() calls mds_postrecov() which returned
1927 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
1928 considers this an error and immediately begins cleaning up the
1929 lov, just after starting the mds_lov process
1931 Severity : enhancement
1933 Description: Implement 'lfs df' to report actual free space on per-OST basis
1934 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
1935 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
1936 usage of objects; '-h' to report in human readable format.
1938 ------------------------------------------------------------------------------
1940 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
1947 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
1948 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
1949 at startup during orphan unlinking. Instead, skip buddy update
1950 before setup, as it will be regenerated after recovery is complete.
1955 Description: If an OST is inactive, its locks might reference stale inodes.
1956 Details : lov_change_cbdata() must iterate over all namespaces, even if
1957 they are inactive to clear inode references from the lock.
1959 Severity : enhancement
1960 Frequency : occasional, if non-standard max_dirty_mb used
1962 Description: Client will block write RPCs if not enough grant
1963 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
1964 then the client will block writes while waiting for another RPC
1965 to complete instead of consuming its dirty limit. With change
1966 we get improved performance when max_dirty_mb is small.
1968 Severity : enhancement
1969 Bugzilla : 3389, 6253
1970 Description: Add support for supplementary groups on the MDS.
1971 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
1972 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
1973 lookups for user supplementary groups into a cache.
1977 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
1978 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
1979 for liblustre O_EXCL is enforced by MDS.
1983 Description: readdir never returns NULL in liblustre.
1984 Details : Corrected llu_iop_getdirentries logic, to return offset of next
1985 dentry in struct dirent.
1989 Frequency : liblustre only, depends on application IO pattern
1990 Description: liblustre clients evicted if not contacting servers
1991 Details : Don't put liblustre clients into the ping_evictor list, so
1992 they will not be evicted by the pinger ever.
1994 Severity : enhancement
1996 Description: Add ability to evict clients by NID from MDS.
1997 Details : By echoing "nid:$NID" string into
1998 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
1999 $NID would be instantly evicted from this MDS and from all active
2000 OSTs connected to it.
2004 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2005 Details : lookup_it_finish() used to query file size from OSTs that was not
2010 Description: service threads change working directory to that of init
2011 Details : Starting lustre service threads may pin the working directory
2012 of the parent thread, making that filesystem busy. Threads
2013 now change to the working directory of init to avoid this.
2017 Frequency : during shutdown only
2018 Description: shutdown with a failed MDS or OST can cause unmount to hang
2019 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2024 Frequency : 2.6 only
2025 Description: chmod/chown may include an extra supplementary group
2026 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2027 supplementary group and if none is specified this is used.
2030 Bugzilla : 5479 (6816)
2032 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2033 Details : If lookup is in progress on a file that is unlinked we might try
2034 to revalidate the inode and fail in revalidate after lookup is
2035 complete and ll_file_open() enqueues the open again but
2036 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2039 Frequency : always, if lconf --abort_recovery used
2041 Description: lconf --abort_recovery fails with 'Operation not supported'
2042 Details : lconf was attempting to abort recovery on the MDT device and not
2045 ------------------------------------------------------------------------------
2047 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2052 Frequency : rare (only unsupported configurations with a node running as an
2054 Bugzilla : 6514, 5137
2055 Description: Mounting a Lustre file system on a node running as an OST could
2057 Details : OSTs now preallocates memory needed to write out data at
2058 startup, instead of when needed, to avoid having to
2059 allocate memory in possibly low memory situations.
2060 Specifically, if the file system is mounted on on OST,
2061 memory pressure could force it to try to write out data,
2062 which it needed to allocate memory to do. Due to the low
2063 memory, it would be unable to do so and the node would
2064 become unresponsive.
2066 Severity : enhancement
2068 Description: Addition of lconf --service command line option
2069 Details : lconf now accepts a '--service <arg>' option, which is
2070 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2072 Severity : enhancement
2074 Description: Failover mode is now the default for OSTs.
2075 Details : By default, OSTs will now run in failover mode. To return to
2076 the old behaviour, add '--failout' to the lmc line for OSTs.
2078 Severity : enhancement
2080 Description: Health checks are now provided for MDS and OSTs
2081 Details : Additional detailed health check information on MSD and OSTs
2082 is now provided through the procfs health_check value.
2085 Frequency : occasional, depends on IO load
2087 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2088 after numerous create/delete cycles
2089 Details : The ext3 inode allocation policy would not allocate new inodes
2090 very well on the OSTs because there are no new directories
2091 being created. Instead we look for groups with free space if
2092 the parent directories are nearly full.
2097 Description: Network or server problems during mount may cause partially
2098 mounted clients instead of returning an error.
2099 Details : The config llog parsing code may overwrite the error return
2100 code during mount error handling, returning success instead
2106 Description: MDS can fail to allocate large reply buffers
2107 Details : After long uptimes the MDS can fail to allocate large reply
2108 buffers (e.g. zconf client mount config records) due to memory
2109 fragmentation or consumption by the buffer cache. Preallocate
2110 some large reply buffers so that these replies can be sent even
2111 under memory pressure.
2115 Frequency : rare (liblustre)
2116 Description: fsx running with liblustre complained that using truncate() to
2117 extend the file doesn't work. This patch corrects that issue.
2118 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2119 ATTR_SIZE and lsm use in llu_setattr_raw.
2123 Frequency : rare, only 2.6 kernels
2124 Description: Unusual file access patterns on the MDS may result in inode
2125 data being lost in very rare circumstances.
2126 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2127 code in the 2.6 kernel caused the inode and/or EA data not to
2128 be read from disk, causing single-file corruption.
2132 Frequency : rare, only 2.6 filesystems using extents
2133 Description: Heavy concurrent write and delete load may cause data corruption.
2134 Details : It was possible under high-load situations to have an extent
2135 metadata block in the block device cache from a just-unlinked
2136 file overwrite a newly-allocated data block. We now unmap any
2137 metadata buffers that alias just-allocated data blocks.
2141 Frequency : filesystems with default stripe_count larger than 77
2142 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2143 Details : lconf specifies an inode size of 4096 bytes when the default
2144 stripe_count is larger than 77. This conflicts with the default
2145 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2146 this case to avoid pinning too much memory for large EAs.
2148 ------------------------------------------------------------------------------
2150 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2155 Frequency : rare (extremely heavy IO load with hundreds of clients)
2157 Description: Client is evicted, gets IO error writing to file
2158 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2159 caused clients to be evicted for AST timeouts. The fixes in
2160 bug 5192 mean we no longer need to have such short AST timeouts
2161 so ldlm_timeout has been increased.
2164 Frequency : occasional during --force or --failover shutdown under load
2165 Bugzilla : 5949, 4834
2166 Description: Server oops/LBUG if stopped with --force or --failover under load
2167 Details : a collection of import/export refcount and cleanup ordering
2168 issues fixed for safer force cleanup
2171 Frequency : only filesystems larger than 120 OSTs
2172 Bugzilla : 5990, 6223
2173 Description: lfs getstripe would oops on a very large filesystem
2174 Details : lov_getconfig used kfree on vmalloc'd memory
2177 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2178 Bugzilla : 6242, 6243
2179 Description: reading from files that had been truncated to a non-zero size
2180 but never opened returned no data
2181 Details : ll_file_read() reads zeros from no-object files to EOF
2186 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2187 Details : There's little harm in aborting MDS/OSS recovery and letting it
2188 try again, so I removed the LASSERT and return an error instead.
2190 Severity : enhancement
2192 Description: New debugging infrastructure for tracking down data corruption
2193 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2194 (b) cover more of the client-side code path, and (c) try to narrow
2195 down where problems occurred
2199 Bugzilla : 3819, 4364, 4397, 6313
2200 Description: Racing close and eviction MDS could cause assertion in mds_close
2201 Details : It was possible to get multiple mfd references during close and
2202 client eviction, leading to one thread referencing a freed mfd.
2204 Severity: : enhancement
2205 Bugzilla : 3262, 6359
2206 Description: Attempts to reconnect to servers are now more aggressive.
2207 Details : This builds on the enhanced upcall-less recovery that was added
2208 in 1.4.2. When trying to reconnect to servers, clients will
2209 now try each server in the failover group every 10 seconds. By
2210 default, clients would previously try one server every 25 seconds.
2215 Description: After recovery, certain operations trigger a failed
2216 assertion on a client.
2217 Details : Failing over an mds, using lconf -d --failover, while a
2218 client was doing a readdir() call would cause the client to
2219 LBUG after recovery completed and the readdir() was resent.
2221 Severity : enhancement
2223 Description: Default groups are now added by lconf
2224 Details : You can now run lconf --group <servicename> without having to
2225 manually add groups with lmc.
2228 Frequency : occasional
2230 Description: Nodes with an elan id of 0 trigger a failed assertion
2233 Frequency : always when accessing e.g. tty/console device nodes
2235 Description: tty and some other devices nodes cannot be used on lustre
2236 Details : file's private_data field is used by device data and lustre
2237 values in there got lost. New field was added to struct file to
2238 store fs-specific private data.
2241 Frequency : when exporting Lustre via NFS
2243 Description: NFSD failed occasionally when looking up a path component
2244 Details : NFSD is looking up ".." which was broken in ext3 directories
2245 that had grown large enough to become hashed.
2248 Frequency : Clusters with multiple interfaces not on the same subnet
2250 Description: Nodes will repeatedly try to reconnect to an interface which it
2251 cannot reach and report an error to the log.
2252 Details : Extra peer list entries will be created by lconf with some peers
2253 unreachable. lconf now validates the peer before adding it.
2256 Frequency : Only if a default stripe is set on the filesystem root.
2258 Description: Setting a default stripe on the filesystem root prevented the
2259 filesystem from being remounted.
2260 Details : The client was sending extra request flags in the root getattr
2261 request and did not allocate a reply buffer for the dir EA.
2264 Frequency : occasional, higher if lots of files are accessed by one client
2265 Bugzilla : 6159, 6097
2266 Description: Client trips assertion regarding lsm mismatch/magic
2267 Details : While revalidating inodes the VFS looks up inodes with ifind()
2268 and in rare cases can find an inode that is being freed.
2269 The ll_test_inode() code will free the lsm during ifind()
2270 when it finds an existing inode and then the VFS later attaches
2271 this free lsm to a new inode.
2275 Bugzilla : 6422, 7030
2276 Description: MDS deadlock between mkdir and client eviction
2277 Details : Creating a new file via mkdir or mknod (starting a transaction
2278 and getting the ns lock) can deadlock with client eviction
2279 (gets ns lock and trying to finish a synchronous transaction).
2282 Frequency : occasional
2283 Description: While starting a server, the fsfilt_ext3 module could not be
2285 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2286 kernels. Previously, lconf would still use the ext3 name
2287 when trying to load modules. Now, it will correctly use
2288 ext3 on 2.4 and ldiskfs on 2.6.
2290 Severity : enhancement
2291 Description: The default stripe count has been changed to 1
2292 Details : The interpretation of the default stripe count (0, to lfs
2293 or lmc) has been changed to mean striping across a single
2294 OST, rather than all available. For general usage we have
2295 found a stripe count of 1 or 2 works best.
2297 Severity : enhancement
2298 Description: Add support for compiling against Cray portals.
2299 Details : Conditional compiling for some areas that are different
2303 Frequency : occasional
2304 Bugzilla : 6409, 6834
2305 Description: Creating files with an explicit stripe count may lead to
2306 a failed assertion on the MDS
2307 Details : If some OSTs are full or unavailable, creating files may
2308 trigger a failed assertion on the MDS. Now, Lustre will
2309 try to use other servers or return an error to the
2313 Frequency : occasional
2315 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2316 caused lock timeout during readahead (since 1.4.2).
2317 Details : Processes doing readahead might match a lock that hasn't been
2318 granted yet if there are overlapping and conflicting lock
2319 requests. The readahead process waits on ungranted lock
2320 (original lock is CBPENDING), while OST waits for that process
2321 to cancel CBPENDING read lock and eventually evicts client.
2323 Severity : enhancement
2325 Description: Initial enabling of flock support for clients
2326 Details : Implements fcntl advisory locking and file status functions.
2327 This feature is provided as an optional mount flag (default
2328 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2329 locking are implemented yet, and those that are are not guaranteed
2330 to be completely correct in production environments.
2331 mount -t lustre -o [flock|noflock] ...
2334 Frequency : occasional
2336 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2337 assertion in the ext3 JBD (journaling) layer.
2338 Details : The b_committed_data struct is protected by the big kernel lock
2339 in 2.4 kernels, serializing journal_commit_transaction() and
2340 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2341 access to this struct. In 2.6 kernels there is finer grained
2342 locking to improve SMP performance of the JBD layer.
2346 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2348 -----------------------------------------------------------------------------
2350 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2352 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2353 versions of Lustre. Please update all servers and clients to
2354 version 1.4.2 or later at the same time. You must also run
2355 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2356 to update the configuration logs.
2358 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2359 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2360 - swab llog records of type '0' so we get proper header size/idx (5861)
2361 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2362 - fix rename of one directory over another leaking an inode (5953)
2363 - avoid SetPageDirty on 2.6 (5981)
2364 - don't re-add just-being-destroyed locks to the waiting list (5653)
2365 - when creating new directories, inherit the parent's custom
2366 striping settings if present parent (3048)
2367 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2368 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2369 - don't hold client locks on temporary worklist from l_lru (5666)
2370 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2371 - automatically evict dead clients (5921)
2372 - Update file size properly in create+truncate+fstat case (6196)
2373 - Do not unhash mountpoint dentries, do not allow removal of
2375 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2376 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2377 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2378 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2379 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2380 - Correctly handle reads of files with no objects (6243)
2381 - lctl recover will also mark a device active if deactivate used (5933)
2383 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2384 - allow --write-conf on an MDS with different nettype than client (5619)
2385 - don't write config llogs to MDS for mounts not from that MDS (5617)
2386 - lconf should create multiple TCP connections from a client (5201)
2387 - init scripts are now turned off by default; run chkconfig --on
2388 lustre and chkconfig --on lustrefs to use them
2389 - upcalls are no longer needed for clients to recover to failover
2391 - add --abort-recovery option to lconf to abort recovery on device
2393 - add support for an arbitrary number of OSTs (3026)
2394 - Quota support protocol changes.
2395 - forward compatibility changes to wire structs (6007)
2396 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2397 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2398 - improved "lustre status" script
2399 - initialize blocksize for non-regular files (6062)
2400 - added --disable-server and --disable-client configure options (5782)
2401 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2402 - Vanilla 2.4.29 support
2403 - increase maximum number of obd devices to 520 (6242)
2404 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2405 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2406 SLES 9 2.6.5 kernels
2407 - limit stripes per file to 160 (the maximum EA size) (6093)
2409 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2412 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2413 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2414 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2415 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2416 - small ext3 extents cleanups and fixes (5733)
2417 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2418 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2419 - avoid needless client->OST connect, fix handle mismatch (5317)
2420 - fix DLM error path that led to out-of-sync client, long delays (5779)
2421 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2422 - fix several locking issues related to i_size (5492,5624,5654,5672)
2423 - don't move pending lock onto export if it is already evicted (5683)
2424 - fix kernel oops when creating .foo in unlinked directory (5548)
2425 - fix deadlock in obdfilter statistics vs. object create (5811)
2426 - use time_{before,after} to avoid timer jiffies wrap (5882)
2427 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2428 - Do not leak request if server was not able to process it (5154)
2429 - If mds_open unable to find parent dir, make that negative lookup(5154)
2430 - don't create new directories with extent-mapping (5909, 5936)
2432 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2433 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2434 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2435 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2436 - support for RHEL4 kernel on i686 (5773)
2437 - provide error messages when incompatible logs are encountered (5898)
2439 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2440 * version 1.4.0.10 (1.4.1 release candidate 1)
2442 - don't keep a lock reference when lock is not granted (4238)
2443 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2444 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2445 - drop import inflight refcount on signal_completed_replay error (5255)
2446 - unlock page after async write error during send (3677)
2447 - handle missing objects in filter_preprw_read properly (5265)
2448 - no transno return for symlink open, don't save no-trasno open (3440)
2449 - don't try to complete elan receive that already failed (4012)
2450 - free RPC server reply state on error (5406)
2451 - clean up thread from ptlrpc_start_thread() on error (5160)
2452 - readahead could read extra page into cache that wasn't ejected (5388)
2453 - prevent races in class_attach/setup/cleanup/detach (5260)
2454 - don't dereference de->d_inode after l_dput of de (5458)
2455 - use "int" for stripe value returned from lock_to_stripe (5544)
2456 - mballoc allocation and error-checking fixes in 2.6 (5504)
2457 - block device patches to fix I/O request sizes in 2.6 (5482)
2458 - look up hostnames for IB nals (5602)
2459 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2460 - don't start multiple acceptors for the same port (5277)
2461 - fix incorrect LASSERT in mds_getattr_name (5635)
2462 - export a proc file for general "ping" checking (5628)
2463 - fix "lfs check" to not block when the MDS is down (5628)
2465 - service request history (4965)
2466 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2467 - create an "evict_client" /proc entry on OSTs, like the MDS has
2468 - fix mount usage message, return errors per mount(8) (5168)
2469 - change grep [] to grep "[]" in tests so they work in more UMLs
2470 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2471 - remove ancient LOV_MAGIC_V0 EA support (5047)
2472 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2473 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2474 - allow readahead allocations to fail when low on memory (5383)
2475 - mmap locking landed again, after considerable improvement (2828)
2476 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2478 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2481 - send OST transaction number in read/write reply to free req (4966)
2482 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2483 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2484 - the watchdog thread now runs as interruptible (5246)
2485 - flock/lockf fixes (but it's still disabled, pending 5135)
2486 - don't use EXT3 constants in llite code (5094)
2487 - memory shortage at startup could cause assertion (5176)
2489 - reorganization of lov code
2490 - single portals codebase
2492 - add extents/mballoc support (5025)
2493 - direct I/O reads in the obdfilter (4048)
2494 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2496 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2499 - send OST transaction number in read/write reply to free req (4966)
2500 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2501 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2502 - the watchdog thread now runs as interruptible (5246)
2503 - handle missing objects in filter_preprw_read properly (5265)
2504 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2505 - drop import inflight refcount on signal_completed_replay error (5255)
2506 - unlock page after async write error during send (3677)
2507 - return original error code on reconstructed replies (3761)
2508 - no transno return for symlink open, don't save no-trasno open (3440)
2510 - add pid to ldlm debugging output (4922)
2511 - bump the watchdog timeouts -- we can't handle 30sec yet
2512 - extra debugging for orphan dentry/inode bug (5259)
2514 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2517 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2518 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2519 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2520 - fix NULL dereference in /proc/sys/portals/routes (4827)
2521 - allow failed mdc_close() operations to be interrupted (4561)
2522 - stop precreate on OST before MDS would time out on it (4778)
2523 - don't send partial-page writes before EOF from client (4410)
2524 - discard client grant for sub-page writes on large-page clients (4520)
2525 - don't free dentries not owned by NFS code, check generation (4806)
2526 - fix lsm leak if mds_create_objects() fails (4801)
2527 - limit debug_daemon file size, always print CERROR messages (4789)
2528 - use transno after validating reply (3892)
2529 - process timed out requests if import state changes (3754)
2530 - update mtime on OST during writes, return in glimpse (4829)
2531 - add mkfsoptions to LDAP (4679)
2532 - use ->max_readahead method instead of zapping global ra (5039)
2533 - don't interrupt __l_wait_event() during strace
2535 - add software watchdogs to catch hung threads quickly (4941)
2536 - make lustrefs init script start after nfs is mounted
2537 - fix CWARN/ERROR duplication (4930)
2538 - return async write errors to application if possible (2248)
2539 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2540 - print NAL number in %x format (4645)
2541 - update barely-supported suse-2.4.21-171 series (4842)
2542 - support for sles 9 %post scripts
2543 - support for building 2.6 kernel-source packages
2544 - support for sles km_* packages
2546 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2549 - ignore -ENOENT errors in osc_destroy (3639)
2550 - notify osc create thread that OSC is being cleaned up (4600)
2551 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2552 - reconstruct ost_handle() like mds_handle() (4657)
2553 - create a new thread to do import eviction to avoid deadlock (3969)
2554 - let lconf resolve symlinked-to devices (4629)
2555 - don't unlink "objects" from directory with default EA (4554)
2556 - hold socknal file ref over connect in case target is down (4394)
2557 - allow more than 32000 subdirectories in a single directory (3244)
2558 - fix blocks count for O_DIRECT writes (3751)
2559 - OST returns ENOSPC from object create when no space left (4539)
2560 - don't send truncate RPC if file size isn't changing (4410)
2561 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2562 - bind to privileged port in socknal and tcpnal (3689)
2564 - rate limit CERROR/CWARN console message to avoid overload (4519)
2565 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2566 - basic mmap support (3918)
2567 - kernel patch series update from b1_4 (4711)
2569 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2572 - avoid crash during MDS cleanup with OST shut down (2775)
2573 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2574 - don't use bad inodes on the MDS (3744)
2575 - dynamic object preallocation to improve recovery speed (4236)
2576 - don't hold spinlock over lock dumping or change debug flags (4401)
2577 - don't zero obd_dev when it is force cleaned (3651)
2578 - print grants to console if they go negative (4431)
2579 - "lctl deactivate" will stop automatic recovery attempts (3406)
2580 - look for existing locks in ldlm_handle_enqueue() (3764)
2581 - don't resolve lock handle twice in recovery avoiding race (4401)
2582 - revalidate should check working dir is a directory (4134)
2584 - don't always mark "slow" obdfilter messages as errors (4418)
2586 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2589 - don't close LustreDB during write_conf until it is done (3860)
2590 - fix typo in lconf for_each_profile (3821)
2591 - allow dumping logs from multiple threads at one time (3820)
2592 - don't allow multiple threads in OSC recovery (3812)
2593 - fix debug_size parameters (3864)
2594 - fix mds_postrecov to initialize import for llog ctxt (3121)
2595 - replace config semaphore with spinlock (3306)
2596 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2597 - don't allow enqueue to complete on a destroyed export (3822)
2598 - down write_lock before checking llog header bitmap (3825)
2599 - recover from lock replay timeout (3764)
2600 - up llog sem before sending rpc (3652)
2601 - reduce ns lock hold times when setting kms (3267)
2602 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2603 - fix NULL deref and obd_dev leak on setup error (3312)
2604 - replace some LBUG about llog ops with error handling (3841)
2605 - don't match INVALID dentries from d_lookup and spin (3784)
2606 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2607 - fix invalid assertion in ptlrpc_set_wait (3880)
2609 - add libwrap support for the TCP acceptor (3996)
2610 - add /proc/sys/portals/routes for non-root route listing (3994)
2611 - allow setting MDS UUID in .xml (2580)
2612 - print the stack of a process that LBUGs (4228)
2614 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2617 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2618 - make sure to unset replay flag from failed open requests (3440)
2619 - if default stripe count is 0, use OST count for inode size (3636)
2620 - update parent mtime/ctime on client for create/unlink (2611)
2621 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2622 - free recovery state on server during a forced cleanup (3571)
2623 - unregister_reply for resent reqs (3063)
2624 - loop back devices mounting and status check on 2.6 (3563)
2625 - fix resource-creation race that can provoke i_size == 0 (3513)
2626 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2627 - more debugging for page-accounting assertion (3746)
2628 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2629 - avoid deadlock after precreation failure (3758)
2630 - fix race and lock order deadlock in orphan handling (3450, 3750)
2631 - add validity checks when grabbing inodes from l_ast_data (3599)
2633 - add /proc/.../recovery_status to obdfilter (3428)
2634 - lightweight CDEBUG infrastructure, debug daemon (3668)
2635 - change default OSC RPC parameters to be better on small clusters
2636 - turn off OST read cache for files smaller than 32MB
2637 - install man pages and include them in rpms (3100)
2638 - add new init script for (un)mounting lustre filesystems (2593)
2639 - run chkconfig in %post for init scripts (3701)
2640 - drop scimac NAL (unmaintained)
2642 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
2645 - clean kiobufs before and after use (3485)
2646 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
2647 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
2648 - change init script to fail more clearly if not run as root (1528)
2649 - allow clients to reconnect during replay (1742)
2650 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
2651 - don't do DNS lookups on NIDs too small for IP addresses (3442)
2652 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
2653 - fix cond_resched (3554)
2654 - only evict unfinished clients after recovery (3515)
2655 - allow bulk resend, prevent data loss (3570)
2656 - dynamic ptlrpc request buffer allocation (2102)
2657 - don't allow unlinking open directory if it isn't empty (2904)
2658 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
2659 - remove extraneous obd dereference causing LASSERT failure (3334)
2660 - don't use get_cycles() when creating temp. files on the mds (3156)
2661 - hold i_sem when setting i_size in ll_extent_lock() (3564)
2662 - handle EEXIST for set-stripe, set proper directory name (3336)
2664 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
2665 - fix ksocknal_fmb_callback() error messages (2918)
2667 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
2670 - don't copy lvb into (possibly NULL) reply on error (2983)
2671 - don't deref dentry after dput, don't free lvb on error (2922)
2672 - use the kms to determine writeback rpc length (2947)
2673 - increment oti_logcookies when osc is inactive (2948)
2674 - update client's i_blocks count via lvb messages (2543)
2675 - handle intent open/close of special files properly (1557)
2676 - mount MDS with errors=remount-ro, like obdfilter (2009)
2677 - initialize lock handle to avoid ASSERT on error cleanup (3057)
2678 - don't use cancelling-locks' kms values (2947)
2679 - use highest lock extent for kms, not last one (2925)
2680 - don't dereference ERR_PTR() dentry in error handling path (3107)
2681 - fix thread race in portals_debug_dumplog() (3122)
2682 - create lprocfs device entries at setup instead of at attach (1519)
2683 - common AST error handler, don't evict client on completion race (3145)
2684 - zero nameidata in detach_mnt in 2.6 (3118)
2685 - verify d_inode after revalidate_special is valid in 2.6 (3116)
2686 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
2687 - initialize RPC timeout timer earlier for 2.6 (3219)
2688 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
2689 - print nal/nid for unknown nid (3258)
2690 - additional checks for oscc recovery before doing precreate (3284)
2691 - fix ll_extent_lock() error return code for 64-bit systems (3043)
2692 - don't crash in mdc_close for bad permissions on open (3285)
2693 - zero i_rdev for non-device files (3147)
2694 - clear page->private before handing to FS, better assertion (3119)
2695 - tune the read pipeline (3236)
2696 - fix incorrect decref of invalidated dentry (2350)
2697 - provide read-ahead stats and refine rpc in flight stats (3328)
2698 - don't hold journal transaction open across create RPC (3313)
2699 - update atime on MDS at close time (3265)
2700 - close LDAP connection when recovering to avoid server load (3315)
2701 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
2702 - don't leak open file on MDS after open resend (3325)
2703 - serialize filter_precreate and filter_destroy_precreated (3329)
2704 - loop device shouldn't call sync_dev() for nul device (3092)
2705 - clear page cache after eviction (2766)
2706 - resynchronize MDS->OST in background (2824)
2707 - refuse to mount the same filesystem twice on same mountpoint (3394)
2708 - allow llmount to create routes for mounting behind routers (3320)
2709 - push lock cancellation to blocking thread for glimpse ASTs (3409)
2710 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
2711 - fix rare problem with rename on htree directories (3417)
2713 - allow default OST striping configuration per directory (1414)
2714 - fix compilation for qswnal for 2.6 kernels (3125)
2715 - increase maximum number of MDS request buffers for large systems
2716 - change liblustreapi to be useful for external progs like lfsck (3098)
2717 - increase local configuration timeout for slow disks (3353)
2718 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
2720 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2723 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
2724 - don't overwrite extent policy data in reply if lock was blocked (2901)
2725 - drop filter export grants atomically with removal from device (2663)
2726 - del obd_self_export from work_list in class_disconnect_exports (2908)
2727 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
2728 - swab reply message in mdc_close, other PPC fixes (2464)
2729 - fix destroying of named logs (2325)
2730 - overwrite old logs when running lconf --write_conf (2264)
2731 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
2732 - fix race in target_handle_connect (2898)
2733 - mds_reint_create() should take same inode create lock (2926)
2734 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
2735 - don't close files for self_export to avoid uninitialized obd (2936)
2736 - allow MDS with the same name as client node (2939)
2737 - hold dentry reference for closed log files for unlink (2325)
2738 - reserve space for all logs during transactions (2059)
2739 - don't evict page beyond end of stripe extent (2925)
2740 - don't oops on a deleted current working directory (2399)
2741 - handle hard links to targets without a parent properly (2517)
2742 - don't dereference NULL lock when racing during eviction (2867)
2743 - don't grow lock extents when lots of conflicting locks (2919)
2745 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
2748 - account for cache space usage on clients to avoid data loss (974)
2749 - lfsck support in lustre kernel code (2349)
2750 - reduce journal credits needed for BRW writes (2370)
2751 - orphan handling to avoid losing space on client/server crashes
2752 - ptlrpcd can be blocked, stopping ALL progress (2477)
2753 - use lock value blocks to assist in proper KMS, faster stat (1021)
2754 - takes i_sem instead of DLM locks internally on obdfilter (2720)
2755 - recovery for initial connections (2355)
2756 - fixes for mds_cleanup_orphans (1934)
2757 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
2758 - block all file creations until orphan recovery completes (1901)
2759 - client remove rq_connection from request struct (2423)
2760 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
2761 - recovery timer race (2670)
2762 - mdc_close recovey bug (2532)
2763 - ptlrpc cleanup bug (2710)
2764 - mds timeout on local locks (2588)
2765 - namespace lock held during RPCs (2431)
2766 - handle interrupted sync write properly (2503)
2767 - don't try to handle a message that hasn't been replied to (2699)
2768 - client assert failure during cleanup after abort recovery (2701)
2769 - leak mdc device after failed mount (2712)
2770 - ptlrpc_check_set allows timedout requests to complete (2714)
2771 - wait for inflight reqs when ptlrpcd finishes (2710)
2772 - make sure unregistered services are removed from the srv_list
2773 - reset bulk XID's when resending them (caught by 1138 test)
2774 - unregister_bulk after timeout
2775 - fix lconf error (2694)
2776 - handle write after unfinished setstripe, stripe-only getstripe (2388)
2777 - readahead locks pages, leaves pending causing memory pressure (2673)
2778 - increase OST request buffers to 4096 on large machines (2729)
2779 - fix up permission of existing directories in simple_mkdir (2661)
2780 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
2781 - don't assign transno to errored transactions (2742)
2782 - don't delete objects on OST if given a bogus objid from MDS (2751)
2783 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
2784 - if rq_no_resend, then timeout request after recovery (2432)
2785 - fix MDS llog_logid record size, 64-bit array alignment (2733)
2786 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
2787 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
2788 - MDS recovery shouldn't delete objects that it has given out (2730)
2789 - if enqueue arrives after completion, don't clobber LVB (2819)
2790 - don't unlock pages twice when trigger_group_io returns error (2814)
2791 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
2792 - don't write pages to disk if there was an error (1450)
2793 - don't ping imports that have recovery disabled (2676)
2794 - take buffered bytes into account when balancing socknal conn (2817)
2795 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
2796 - reconnect unlink llog connection after MDS reconnects to OST (2816)
2797 - remove little-endian swabbing of llog records (1987)
2798 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
2799 - retry reposting request buffers if they fail (1191)
2800 - grow extent at grant time to avoid granting a revoked lock (2809)
2801 - lock revoke doesn't evict page if covered by a second lock (2765)
2802 - disable VM readahead to avoid reading outside lock extents (2805)
2804 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
2805 - updated kernel patches for hp-2.4.20 kernel (2681)
2807 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
2810 - fix truncated write corruption (2366)
2811 - fix for failed assertion in iopen_connect_dentry (1792,2517)
2813 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
2814 - ensure len(uuid) < 37 in lmc (1171)
2815 - fix ia64 OOPS in llog_test (2255)
2816 - zero end of page at obdfilter for partial page writes (2648)
2817 - don't leave stale dentries around after renames (bug 2428)
2818 - fix timeouts when evicting a client with a single lock held (2642)
2819 - set deadline for the initial HELLO message to drain (2634)
2820 - print out dotted-quad IP addresses in the socknal (2302)
2822 - additional debugging for MDS client eviction problem (2443)
2823 - fix mkfsoptions support for osts (2603, 2604)
2825 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
2828 - add series for the vanilla 2.6.0 kernel
2829 - add series for the vanilla 2.4.24 kernel
2830 - add series for a cray x86/64 UL kernel drop
2831 - fix xattr patches for the vanilla 2.4.19 series
2833 - generate true UUIDs in lmc (1171)
2834 - have portals stack dumping break in UML (2466)
2835 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
2836 - allocate with _NOFS in ldlm to avoid deadlock (1933)
2837 - wake callback waiting threads on client eviction (2460)
2838 - Add --ptldebug and --subsystem to lmc (1719)
2839 - update assertion to allow safe interrupt allocation
2840 - set rq_no_resend for cancel requests (2432)
2841 - recalculate ptlrpcd timeout after resend (2494)
2842 - call vfs_rmdir when removing pending directories (2368)
2843 - fix renaming a file to itself (2429)
2844 - lmc creates a default one-stripe lov (2454)
2845 - expand procfs space to handle large clusters (2326)
2846 - increase UML stack to avoid overflow
2847 - update lconf's list of debug and subsystem masks
2848 - fix lfs find --obd (2510)
2849 - /proc tunable for disabling filter read caching (2591)
2850 - stop rpm packages from altering slapd.conf (2301)
2851 - disable nagle in the socknal under 0conf (2578)
2852 - choose mds inode size based on stripe count (2572)
2853 - fix kernel-source rpm problems (2516)
2855 - add --disable-doc to avoid pdf generation (2421)
2856 - update documentation, tests, type-os, comments
2857 - avoid format warnings on ia64
2858 - remove the TOE NAL
2859 - tiny code cleanups by removing unused fields
2861 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
2864 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
2865 - remove the most blatant lies from BUILDING file (2371)
2866 - change default debug level to reasonable production setting
2867 - reduce client side cache size to reduce cache flush time
2868 - reduce max RPCs in flight to avoid unnecessary file fragmentation
2869 - make TCP zerocopy and pinger support enabled by default (2476)
2870 - sync writes completed after process exits caused crashes (2319)
2871 - maintain correct mount count on the MDS (2356)
2872 - backout 1557, because 2316 wasn't really fixed
2873 - better file I/O statistics gathering in /proc
2874 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
2875 - another kernel patch to fix zero-copy TCP function export
2876 - don't take duplicate lock when processing re-sent getattr (2420)
2877 - lctl uses obd_self_export instead of creating new conn (2353)
2878 - MDS/OST recovery case which requires object creation asserted (2425)
2879 - move lfs from /usr/sbin to /usr/bin in packages
2880 - fix race between mds_client_add and mds_client_free (2417)
2881 - use kmalloc instead of slabs in portals (2430)
2882 - don't create duplicate records when a failover MDS is present (2442)
2883 - remove unnecessary mount age check (2332)
2884 - don't remove directory inodes from locks prematurely (2451)
2885 - don't break if MDS service name is the same as hostname (2103)
2886 - fix races in client write RPC generation when cache full (2482)
2888 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
2891 - remove now-unused request->rq_obd (278)
2892 - if an allocation fails, print out how much memory we've used (1933)
2893 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
2894 - add the "configurable stack size" patch to most series files (1256)
2895 - ability to write large log records, for 100+ OST configs (2306)
2896 - fix NULL deref when filter_prep fails (2314)
2897 - fix operator precedence error in filter_sync
2898 - dynamic allocation of socknal TX descriptors (2315)
2899 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
2900 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
2901 - ensure that CWARN messages in Portals always get to the syslog
2902 - __init/__exit are not for prototype decls (ldlm_init/exit)
2903 - x86-64 compile warning fixes
2904 - fix gateway LMC keyword conflict (2318)
2905 - fix MDS lock inversions in getattr/reint paths (1844)
2906 - fix a rare lock re-ordering bug, which caused deadlock (2322)
2907 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
2908 - DLM race condition prevented some lock evictions (2328)
2909 - ENOMEM detection and retry on socknal sends (2230)
2910 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
2911 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
2912 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
2913 - better error messages when a client is rejected during recovery (1505)
2914 - avoid cancelling locks which were never granted, after failure (2330)
2915 - fix i_sem/journal inversion in mds_client_add (2333)
2916 - fix truncate/getattr lock cycle deadlock (2334)
2917 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
2918 - fix two rare exit paths which could leak an l_lock() ref (2321)
2919 - fencepost error in MDS/OST orphan recovery (2226)
2920 - make log record alignment 8 bytes (1988)
2921 - lstripe now fails when requested offset > ost_count (2237)
2922 - ensure that all kernel series have a complete list.h (1607)
2923 - fix crashes in special-file operations (2316)
2924 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
2926 - allow configurable automake binary, for testing new versions
2927 - small update to the lfs documentation
2929 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
2931 * fix negative export reference count in fsfilt_sync (2312)
2933 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
2934 * release candidate 0.9.1
2936 - orphans are moved into the PENDING directory for possible recovery
2937 - replayed opens now open by fid for orphan/rename safety (1042)
2938 - last close of an orphan inode generates a transno (683)
2939 - chdir() and mount() now pin the directory entry (1020)
2940 - avoid CERROR in normal ll_setattr_raw() error case (1500)
2941 - discard very old requests without processing them (1502)
2942 - remove some common, well-understood CERRORs (1505)
2943 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
2944 - clear "grant" flags in OST replies until OST grant code lands (1644)
2945 - fix read performance by not clobbering i_blksize on client (1598)
2946 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
2947 - make LRU size a /proc tunable, clears locks when reduced (707)
2948 - fix some lprocfs rot that prevented ptlbd from loading (1732)
2949 - server locks take references on exports now (1558)
2950 - build fixes for 2.4.20-rh trees (1663)
2951 - return an error from lov_create if all OSCs are inactive (1751)
2952 - fix import levels when a reconnect happens without a timeout (1597)
2953 - exit early from mds_open if we get a lookup error (1749)
2954 - partial page read at EOF wouldn't wait for disk before sending (1642)
2955 - avoid NULL deref in obdfilter when reading page past EOF (1592)
2956 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
2957 - fix LBUG in ll_it_open_error with rc = -2 (1861)
2958 - write/truncate lock inversion (1639)
2959 - Don't auto-load obdclass, portals modules during cleanup (1495)
2960 - fix timestamps from jumping to "now" (1763)
2961 - extra journal assertions (1648)
2962 - add an extra multiunlink test (1771)
2963 - fix read_record/write_record API (1776)
2964 - fix leak of offset_extent, possible incorrect i_size later (1772)
2965 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
2966 - Debugging for the kqswnal_get_idle_tx problems (1820)
2967 - Allow recovery to be attempted multiple times (1536)
2968 - Write out MDS last_rcvd file after it is first created (1600)
2969 - Fix tx_descriptor leak in failed transmit situations (1827)
2970 - ext3 journaling fixes for assertion failure after IO error (1871)
2971 - class_export_put() on freed export after completion AST error (1896)
2972 - Fix revalidate looping in VFS (1322)
2973 - Don't access a freed export during MDS_REINT timeout (1521)
2974 - Add open-unlink recovery support on the MDS (1673,1764)
2975 - Return an error if no MDS data was read from last_rcvd (1946)
2976 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
2977 - Don't setup a disk device that doesn't match exported UUID (317)
2978 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
2979 - avoid committing NULL handle in force close
2980 - local.sh is now a one-stripe LOV configuration
2981 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
2982 - fix ext3 htree duplicate directory entry corruption (1516)
2983 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
2984 - update to new LOV EA format (2097)
2985 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
2987 2003-06-15 Phil Schwan <phil@clusterfs.com>
2990 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
2991 - per-import/export recovery handling (958, 931, 959)
2992 - multiple last-rcvd slots, for serving multiple FSes (949)
2993 - connections are again shared between multiple imp/exports (963, 964)
2994 - "umount -f" would hang if any requests needed to be sent (393, 978)
2995 - avoid pinning large req buffer by copying for queued messages (989)
2996 - add "uuid" to "lctl device" command to help upcalls (991)
2997 - "open" RPCs with transnos would confuse recovery counters (1037)
2998 - do proper endian conversion of all wire messages (288, 340, 891)
2999 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3000 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3001 - fix ptlrpc_connection leak in target_handle_connect (1174)
3002 - fix import refcounting bug in OST and MDS cleanup (1134)
3003 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3004 - fix very unlikely obd_types race condition (501)
3005 - remove osc_open hack for echo_client (1187)
3006 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3007 - a failure in read_inode2 leads to deadlock (1139)
3008 - cancel ack-locks as soon as transaction is committed (1072)
3009 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3010 - make sure to commitrw after any preprw to avoid deadlock (1162)
3011 - failing to execute a file in a lustre FS would lock inode (1203)
3012 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3013 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3014 - obd_uuid2tgt would walk off the end of the list (1255)
3015 - on IA64 the osc would give portals incorrect bulk size (1258)
3016 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3017 - fix lock inversion caused by new llite matching code (1282)
3018 - limit the number of dirty pages on a client to 10MB (1286)
3019 - timed out locks were not being corrected cancelled (1289)
3020 - fix O_DIRECT above 4GB on IA-32 (1292)
3021 * major user-visible changes
3022 - fail out/fail over policy now controlled by the upcall (993)
3024 - add OBD_PING to check server availability and failure (954)
3025 - lustre messages are now sent in sending host order (288, 340, 891)
3026 - add eadatalen to MDS getattr reply (340)
3027 - OST read replies may contain second buffer, with per-page status (593)
3029 2003-03-11 Phil Schwan <phil@clusterfs.com>
3032 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3033 - failed open()s could cause deadlock; fixed (867, 869)
3034 - stop cancelling OST locks when files are closed (481)
3035 - overlapping XID spaces caused network corruption (851, 853)
3036 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3037 - setattr_raw added, to do single-RPC, server-side setattrs
3038 - lmc/lconf syntax change for OST UUIDs
3039 - fix crashy race condition between ptlrpc_free_req and osc_close
3040 - don't use request in mdc_enqueue if we hit a timeout (889)
3041 - don't set the inode i_size for regular files from the MDS (896)
3042 - handle out of order completion AST (842)
3043 - don't LBUG if a lock request times out after receiving AST (913)
3044 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3045 - if a bad lock AST arrives, send an error instead of dropping entirely
3046 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3047 - fix leak in bulk IO when only partially completed (899, 900, 926)
3048 - fix O_DIRECT for ia64 (55)
3049 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3050 - C-z after timeout could hang a process forever; fixed (977)
3052 - client-side I/O cache (678, 924, 929, 941, 970)
3054 - READPAGE and SETATTRs which don't take server-side locks get
3057 2003-02-11 Phil Schwan <phil@clusterfs.com>
3060 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3061 - Fix fsfilt_extN_readpage() to read a full page of directory
3062 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3063 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3064 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3065 - Most intent operations were removed, in favour of a new RPC mode
3066 that does a single RPC to the server and bypasses most of the VFS
3067 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3068 - Aggressively cancel local locks on DLM servers
3069 - mds_reint_unlink sends EA to the client if it's the last nlink.
3070 client uses that EA to unlink OST objects.
3071 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3072 - recursive symlinks were fixed (439)
3073 - fixed NULL deref in DEBUG_REQ
3074 - filter_update_lastobjid no longer calls sync, which annoyed extN
3075 - fixed multi-client small-writes to a single file problem (445)
3076 - fixed mtime updates during file writes (607)
3077 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3078 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3079 - fixed Linux OST size reporting problem (444, 656)
3080 - OST now updates object mtime with writes or setattr (607, 619)
3081 - client verifies file size before zeroing page past EOF (445)
3082 - OST now writes last allocated objid to disk with allocation (108)
3083 - LOV on echo now works (409)
3085 - mds_reint_unlink sends a new buffer, with the EA included. this
3086 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3087 set if a regular file was being unlinked, and it was the last link
3088 - use PtlGet from the target for bulk writes (315)
3089 - OST now updates object mtime with writes or setattr (607, 619)
3090 - LDLM now has a grant-time callback to revalidate locked items, if
3092 - Many MDS operations were reorganized to combat race conditions
3094 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3095 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3097 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3100 - Fully reactivate OST imports after reconnection (512, others)
3101 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3102 - More graceful error handling for truncating on dead OST (515)
3103 - Don't error out unless we're actually accessing dead stripes (474)
3104 - Fix garbage sizes when stripes are missing (410)
3105 - LRU counters were broken, causing constant lock purge (433, 432)
3106 - garbage on read from stripes with failed OSTs (441)
3107 - mark OSCs as active before reconnecting during recovery (438)
3108 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3109 - lfind did not preserve OST order in output (443)
3110 - symlinks cause hung clients, incorrect data (439)
3111 - stop dereferencing request after dropping refcount (457)
3112 - don't LASSERT(spin_is_locked) on non-SMP (455)
3113 - fixes for many rename() bugs
3114 - fstat didn't correctly synchronize attributes (399)
3115 - server must handle lock cancellation during blocking AST prep (487)
3116 - bulk descriptors were free()d too soon (511)
3117 - fix paths in lconf, which would load incorrect modules (451, 507)
3118 - fix confusing lconf 'host not found' error message (386)
3119 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3120 - fix race condition in mdc_blocking_ast() for inode access (526)
3121 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3122 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3123 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3124 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3125 - fix lstripe in several ways (kernel side) (527)
3126 - fix request leak in ldlm_cli_enqueue (262)
3127 - incorrect OSC was marked inactive after OST failure
3128 - call mds_fs_cleanup before unmounting filesystem (524)
3129 - fix races between taking ns_lock and ldlm_lock_change_resource
3130 - fix races updating LOV export open file list
3131 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3132 - fix recovery NULL deref in ldlm_cli_cancel_unused
3133 - fix some DLM races by using new hash table for lock handles (419)
3134 - permit the client to specify desired inodes, at replay
3135 - duplicate requests when we queue them for replay reintegration
3136 - fix last_rcvd offset calculation
3137 - sync after each recovered transaction, so we always make progress
3138 - never, not always, ERESTART requests without transnos
3139 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3141 - skip replay if the MDS says that the client is already connected
3142 - don't check for a recovery-enabled export to match lctl's UUID
3143 - don't INC_USE_COUNT for phantom exports
3144 - don't crash when cleaning up phantom exports (567)
3145 - don't double-finish or set replay data for errored mdc_open requests
3146 - abort requests when they time out, so we don't get old replies
3147 - send/receive replies for AST messages again
3148 - if the client says that it doesn't have the lock, cancel it on the
3150 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3151 mark it as destroyed, it will all work out when decref is called
3152 - fix module use counts (22, 581)
3154 - ASTs now expect a reply (server cancels lock on error reply)
3156 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3159 - fix many simultaneous client startup (392)
3160 - fix dentry->d_it clobbering
3161 - credentials weren't being shipped for readdir/getattr operations
3162 - remove invalid assertions triggered during some concurrent MD
3164 - proper Lustre versions added (336, 389)
3165 - fix memory leak for create error case (398)
3166 - fix LOV locking bug that would get cli/srv out of sync
3167 - fix echo client over LOV (409)
3168 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3169 - fix double-O_EXCL intent crash (424)
3170 - avoid sending multiple lock CANCELs (352)
3172 - MDS can do multi-client recovery (modulo bugs in new code)
3174 - many updates, edits, cleanups
3176 2002-11-18 Phil Schwan <phil@clusterfs.com>
3179 - fix null d_it dereference (346)
3180 - fix full OST/dbench hang (333)
3181 - fix permission problem with file removal (286)
3182 - fix removal of OSCs from LOV when they fail
3183 - fix NULL deref during bulk timeout (214)
3184 - fix problems related to multiple filesystems on one MDS (241)
3185 - fixed serious subtle metadata locking bugs
3186 - free locks on clients when inodes are removed due to memory
3188 - fix inode pointer in lock data (285)
3189 - partial support for multiple MDS on a single host (241)
3190 - data locks weren't cancelled at clear_inode time (290, 311)
3191 - intent locks could lead to unbounded lock growth (205)
3192 - added a maximum lock count, an LRU list, and a flusher
3193 - fix multiple rename (365)
3194 - properly abstracted the echo client
3195 - OSC locked 1 byte too many; fixed
3196 - rewrote brw callback code:
3197 - fixed recovery bugs related to LOVs (306)
3198 - fixed too-many-pages-in-one-write crash (191)
3199 - fixed (again) crash in sync_io_timeout (214)
3200 - probably fixed callback-related race (385)
3202 - Add capability to MDS protocol
3203 - LDLM cancellations and callbacks on different portals
3205 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3208 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3209 - defer open object destruction to close time (601981, 138)
3210 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3211 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3212 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3213 - don't call obd_setattr in ll_file_release for destroyed objects
3214 * protocol change to lustre_msg: move |version| and add |flags|
3215 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3216 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3217 * added replay of create, unlink, link and rename operations during
3218 MDS failover; recovery should be much more robust now
3219 * remove failed OSCs from LOVs (only lov_create uses this so far)
3220 * the lustre-HOWTO was brought (more) up to date (582544)
3222 2002-10-23 Phil Schwan <phil@clusterfs.com>
3225 - in-use dentries weren't being reused properly (617851)
3226 - prevent multiple LDLM setup (599178)
3227 - fix LOV size calculations for truncate (617853)
3228 - fix client handling of MDS intent errors (POSIX)
3229 - fix permission bug in lovstripe.c test (624321)
3230 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3231 - truncate past end of file could corrupt data
3232 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3233 - a race in recovery could return ETIMEDOUT to apps (623947)
3234 - building outside the source directory was fixed
3235 * the lustre-HOWTO was brought (more) up to date (582544)
3236 * major progress was made on recovery functionality
3238 2002-10-10 Phil Schwan <phil@clusterfs.com>
3241 - recovery deadlock fix
3242 - rm -rf causes LBUG fix (617817)
3243 - file open by multiple tasks fix (618962)
3244 - directory permissions bugs (602707 and 620007)
3245 - journal_stop fixed with locking (611313)
3246 - O_APPEND failures resolved (618273, perhaps 614459)
3247 - lconf PATH fix (619770)
3248 - IA64 build fix (621450)
3249 - RPC buffer sizes scale with amount of memory
3251 2002-10-01 Phil Schwan <phil@clusterfs.com>
3254 - locks would be cancelled without throwing away data pages,
3255 resulting in inconsistent data (605627)
3256 - inode attributes were not always being refreshed (605627, 612449)
3257 - lconf now continues to cleanup after lctl reports an error
3258 - MDS now enforces user permissions (602707)
3259 - lprocfs cleanup fixed, but not yet enabled (614157)
3260 - fixed infinite server hang, should a client not respond to an AST
3261 - avoid going into recovery if user calls readlink() with a buffer
3262 that's too small (613941)
3263 - AST RPCs no longer require replies (614867) -- this may be changed
3264 - don't crash server if client sends an IOV that's too big (611336)
3265 - fixed lock conversion deadlock (611892)
3266 - fixed the following of symlinks (614622)
3267 * recovery: the server can remove locks from a client that dies, other
3268 clients can make progress
3269 * more extN patch fixes
3270 * compile-time configurable ptlrpc buffer allocations
3272 - collaborative read cache document
3273 - Lustre Lite Performance CDR document-in-progress
3275 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3278 - fix typo in patch-2.4.18
3280 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3283 - clear ptlrpc request each time in handle_incoming_request()
3284 - unlink of files now destroys the object on the OST
3286 2002-09-19 Peter Braam <braam@clusterfs.com>
3288 * add hard link support
3289 * change obdfile creation method
3290 * kernel patch changed
3292 2002-09-19 Peter Braam <braam@clusterfs.com>
3295 - stack overflow bug in extN fixed
3297 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3299 * documentation updates
3300 - add man pages for config tools
3301 - update tests/README to describe testing with new config tools
3302 - finish metadata API descriptions
3303 * bug fixes and cleanups
3304 - statfs workaround for 16TB limit
3305 - LOV stripe allocation improved, can stripe on subset of OSTs
3306 - LOV file size/IO offset was wrong for files > 4GB in size
3307 - object EA data was being dropped, caused files to be unreadable
3308 - memory overflow with non-LOV OST caused memory corruption
3309 - fixed regression tests to work with new config tools, obdfilter
3310 - fixed bug when directory size became larger than 1 block
3311 - fixed bug (for single client case) when PWD was deleted
3312 - invalidate local directory pages when doing intent-based ops
3313 - avoid LDLM oops when lock callback contained bad data
3315 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3317 * documentation updates
3318 * bug fixes and cleanups
3319 - configuration tools
3322 - 64-bit compile warnings
3323 - 64-bit internal statfs data
3325 * test_brw on persistent OST devices
3327 * lprocfs (disabled)
3329 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3331 * documentation updates
3332 * bug fixes and cleanups
3333 * configuration tools
3335 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3338 * many small fixes to 0_5_4
3339 * io/network handling
3340 * thinkos in MDS operations
3342 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3345 * crucial basic fixes to 0.5.3
3346 * IOR, Iozone work over Elan
3349 2002-08-07 Phil Schwan <phil@clusterfs.com>
3350 * version 0_5_3, our first alpha
3351 * we use the new Portals iovs
3352 * documentation updates
3353 * bug fixes and cleanups
3354 * small changes in the DLM wire protocol
3356 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3357 * version 0_5_1 with some initial stability,
3358 * locking on MD and file I/O.
3359 * documentation updates
3360 * several bug fixes since 0.5.0
3361 * small changes in wire protocol
3363 2002-07-18 Phil Schwan <phil@clusterfs.com>
3365 * delivered as Lustre Light Alpha
3366 * fixed a crash after handling invalid MDS requests
3367 * fixed directory pages for architectures with non-4k pages sizes
3369 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3370 * release version v0_4_4
3371 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3373 * Fixes a number of interruption problems with OST operations.
3374 * Update documentation for portals header changes
3375 * Move all wire protocol structs/defines to lustre_idl.h
3376 * Fixes symlink length bug.
3377 * Add tcpdump to repository.
3379 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3380 * release version v0_4_3
3381 * Fixes statfs for inodes on extN.
3382 * Fixes bug in runtests which would delete /etc/hosts.
3383 * Use 64-bit object IDs wherever possible (not into VFS though)
3384 Remove ost_get_info, which is unused by lustre, and out of date.
3386 2002-07-03 Peter Braam <braam@clusterfs.com>
3387 * release version v0_4_2 Fixes a lookup error (type not passed)
3388 * move forward to head of Portals
3389 * move forward to latest Lustre kernel
3391 2002-06-25 Peter Braam <braam@clusterfs.com>
3392 * release version v0_4_1. Hopefully stable on single node use.