1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4 2.6.9-42.0.10.EL (RHEL 4)
6 2.6.12.6 vanilla (kernel.org)
7 2.6.16.27-0.9 (SLES 10)
8 * Client support for unpatched kernels:
9 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
10 2.6.16 - 2.6.19 vanilla (kernel.org)
11 2.6.9-42.0.8EL (RHEL 4)
12 * Recommended e2fsprogs version: 1.39.cfs6
14 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
17 Severity : enhancement
19 Description: Add LNET router traceability for debug purposes
20 Details : If a checksum failure occurs with a router as part of the
21 IO path, the NID of the last router that forwarded the bulk data
22 is printed so it can be identified.
27 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
28 Details : Due to a race condition, liblustre clients were occasionally
31 Severity : enhancement
33 Description: lfs setstripe use optional parameters instead of postional
36 Severity : enhancement
38 Description: Nanosecond timestamp support for ldiskfs
39 Details : The on-disk ldiskfs filesystem has added support for nanosecond
40 resolution timestamps. There is not yet support for this at
41 the Lustre filesystem level.
44 Frequency : during server recovery
46 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
47 Details : request with rq_no_resend flag not awake l_wait_event if they get a
51 Frequency : nfs export on patchless client
53 Description: connectathon hang when test nfs export over patchless client
54 Details : Disconnected dentry cannot be found with lookup, so we do not need
55 to unhash it or make it invalid
58 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
59 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
60 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
65 Description: open req refcounting wrong on reconnect
66 Details : If reconnect happened between getting open reply from server and
67 call to mdc_set_replay_data in ll_file_open, we will schedule
68 replay for unreferenced request that we are about to free.
69 Subsequent close will crash in variety of ways.
70 Check that request is still eligible for replay in
71 mdc_set_replay_data().
73 --------------------------------------------------------------------------------
75 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
80 Frequency : on some architectures
82 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
83 Details : Uninitialized flags sometimes cause configuration commands to
86 --------------------------------------------------------------------------------
88 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
90 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
91 INTEROPERATE with older versions automatically. In many cases a
92 special upgrade step is needed. Please read the
93 user documentation before upgrading any part of a 1.4.x system.
94 * WARNING: Lustre configuration and startup changes are required with
95 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
97 * Support for kernels:
98 2.4.21-47.0.1.EL (RHEL 3)
100 2.6.9-42.0.10.EL (RHEL 4)
101 2.6.12.6 vanilla (kernel.org)
102 2.6.16.27-0.9 (SLES10)
103 * Client support for unpatched kernels:
104 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
105 2.6.16 - 2.6.19 vanilla (kernel.org)
106 2.6.9-42.0.8EL (RHEL 4)
107 * Recommended e2fsprogs version: 1.39.cfs6
108 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
111 Severity : enhancement
113 Description: MountConf
114 Details : Lustre configuration is now managed via mkfs and mount
115 commands instead of lmc and lconf. New obd types (MGS, MGC)
116 are added for dynamic configuration management. See
117 https://mail.clusterfs.com/wikis/lustre/MountConf for
120 Severity : enhancement
122 Description: dynamic OST addition
123 Details : OSTs can now be added to a live filesystem
125 Severity : enhancement
127 Description: startup order invariance
128 Details : MDTs and OSTs can be started in any order. Clients only
129 require the MDT to complete startup.
131 Severity : enhancement
133 Description: parallel, asynchronous orphan cleanup
134 Details : orphan cleanup is now performed in separate threads for each
135 OST, allowing parallel non-blocking operation.
137 Severity : enhancement
139 Description: optimized stripe assignment
140 Details : stripe assignments are now made based on ost space available,
141 ost previous usage, and OSS previous usage, in order to try
142 to optimize storage space and networking resources.
144 Severity : enhancement
146 Description: Permanently set tunables
147 Details : All writable /proc/fs/lustre tunables can now be permanently
148 set on a per-server basis, at mkfs time or on a live
151 Severity : enhancement
153 Description: Lustre message v2
154 Details : Add lustre message format v2.
156 Severity : enhancement
158 Description: client OST exclusion list
159 Details : Clients can be started with a list of OSTs that should be
160 declared "inactive" for known non-responsive OSTs.
164 Description: ENOENT returned for valid filehandle during dbench.
165 Details : Check if a directory has children when invalidating dentries
166 associated with an inode during lock cancellation. This fixes
167 an incorrect ENOENT sometimes seen for valid filehandles during
171 Frequency : SFS test only (otherwise harmless)
173 Description: SPEC SFS validation failure on NFS v2 over lustre.
174 Details : Changes the blocksize for regular files to be 2x RPC size,
175 and not depend on stripe size.
177 Severity : enhancement
179 Description: fine-grained SMP locking inside DLM
180 Details : Improve DLM performance on SMP systems by removing the single
181 per-namespace lock and replace it with per-resource locks.
183 Severity : enhancement
185 Description: don't hold multiple extent locks at one time
186 Details : To avoid client eviction during large writes, locks are not
187 held on multiple stripes at one time or for very large writes.
188 Otherwise, clients can block waiting for a lock on a failed OST
189 while holding locks on other OSTs and be evicted.
191 Severity : enhancement
193 Description: Multiple MD RPCs in flight.
194 Details : Further unserialise some read-only MDT RPCs - learn about intents.
195 To avoid overly-overloading MDT, introduce a limit on number of
196 MDT RPCs in flight for a single client and add /proc controls
197 to adjust this limit.
199 Severity : enhancement
201 Description: client read/write statistics
202 Details : Add client read/write call usage stats for performance
203 analysis of user processes.
204 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
205 file access. extents_stats shows chunk size distribution.
206 extents_stats_per_process show chunk size distribution per
209 Severity : enhancement
211 Description: per-client statistics on server
212 Details : Add ldlm and operations statistics for each client in
213 /proc/fs/lustre/mds|obdfilter/*/exports/
215 Severity : enhancement
217 Description: improved MDT statistics
218 Details : Add detailed MDT operations statistics in
219 /proc/fs/lustre/mds/*/stats
221 Severity : enhancement
223 Description: VFS operations stats
224 Details : Add client VFS call stats, trackable by pid, ppid, or gid
225 /proc/fs/lustre/llite/*/vfs_ops_stats
226 /proc/fs/lustre/llite/*/vfs_track_[pid|ppid|gid]
231 Description: Fix client-side osc byte counters
232 Details : The osc read/write byte counters in
233 /proc/fs/lustre/osc/*/stats are now working
236 Frequency : always as root on SLES
238 Description: Failure of copying files with lustre special EAs.
239 Details : Client side always return success for setxattr call for lustre
240 special xattr (currently only "trusted.lov").
245 Description: Refcount LNET uuids
246 Details : The global LNET uuid list grew linearly with every startup;
247 refcount repeated list entries instead of always adding to
250 Severity : enhancement
252 Description: Dynamic service threads
253 Details : Within a small range, start extra service threads
254 automatically when the request queue builds up.
257 Frequency : mixed-endian client/server environments
259 Description: mixed-endian crashes
260 Details : The new msg_v2 system had some failures in mixed-endian
263 Severity : enhancement
265 Description: Easy OST removal
266 Details : OSTs can be permanently deactivated with e.g. 'lctl
267 conf_param lustre-OST0001.osc.active=0'
269 Severity : enhancement
271 Description: MGS proc entries
272 Details : Added basic proc entries for the MGS showing what filesystems
275 Severity : enhancement
277 Description: provide MGS failover
278 Details : Added config lock reacquisition after MGS server failover.
280 Severity : enhancement
282 Description: add Linux 2.4 support
283 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
287 Description: a large application tries to do I/O to the same resource and dies
289 Details : Check the req->rq_arrival time after the call to
290 ost_brw_lock_get(), but before we do anything about
291 processing it & sending the BULK transfer request. This
292 should help move old stale pending locks off the queue as
293 quickly as obd_timeout.
296 Frequency : when an incorrect nid is specified during startup
298 Description: ptlrpc connect to non-existant node causes kernel crash
299 Details : LNET can't be re-entered from an event callback, which
300 happened when we expire a message after the export has been
301 cleaned up. Instead, hand the zombie cleanup off to another
304 Severity : enhancement
306 Description: plain/inodebits lock performance improvement
307 Details : Grouping plain/inodebits in granted list by their request modes
308 and bits policy, thus improving the performance of search through
312 Frequency : only if OST filesystem is corrupted
314 Description: client incorrectly hits assertion in ptlrpc_replay_req()
315 Details : for a short time RPCs with bulk IO are in the replay list,
316 but replay of bulk IOs is unimplemented. If the OST filesystem
317 is corrupted due to disk cache incoherency and then replay is
318 started it is possible to trip an assertion. Avoid putting
319 committed RPCs into the replay list at all to avoid this issue.
322 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
324 Description: System hang on startup
325 Details : This bug allowed the liblustre (e.g. catamount) client to
326 return to the app before handling all startup RPCs. This
327 could leave the node unresponsive to lustre network traffic
328 and manifested as a server ptllnd timeout.
330 Severity : enhancement
332 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
333 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
334 boolean to control whether to print peer debug info when a
335 client's RPC times out.
338 Frequency : only for kernels with patches from Lustre below 1.4.3
340 Description: Remove old rdonly API
341 Details : Remove old rdonly API which unused from at least lustre 1.4.3
344 Frequency : only for devices with external journals
346 Description: Set external device read-only also
347 Details : During a commanded failover stop, we set the disk device
348 read-only while the server shuts down. We now also set any
349 external journal device read-only at the same time.
352 Frequency : when upgrading from 1.4 while trying to change parameters
354 Description: The wrong (new) MDC name was used when setting parameters for
355 upgraded MDT's. Also allows changing of OSC (and MDC)
356 parameters if --writeconf is specified at tunefs upgrade time.
359 Frequency : when setting specific ost indicies
361 Description: QOS code breaks on skipped indicies
362 Details : Add checks for missing OST indicies in the QOS code, so OSTs
363 created with --index need not be sequential.
365 Severity : enhancement
367 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
368 Details : The uninit_groups feature works in conjunction with the kernel
369 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
370 up the pass1 processing of e2fsck. This is a read-only feature
371 in ldiskfs2 only, so older kernels and current ldiskfs cannot
372 mount filesystems that have had this feature enabled.
374 Severity : enhancement
376 Description: Improve multi-block allocation algorithm to avoid fragmentation
377 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
378 allocation locality and avoid filesystem fragmentation.
380 ------------------------------------------------------------------------------
382 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
384 * Support for kernels:
385 2.4.21-47.0.1.EL (RHEL 3)
387 2.6.9-42.0.10.EL (RHEL 4)
388 2.6.12.6 vanilla (kernel.org)
389 2.6.16.27-0.9 (SLES 10)
390 * Recommended e2fsprogs version: 1.39.cfs5
392 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
398 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
399 > 32000 subdirectories
400 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
401 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
402 subdirectory count crosses 32000. This will aid e2fsck to
403 correctly handle more than 32000 subdirectories.
406 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
408 Description: System hang on startup
409 Details : This bug allowed the liblustre (e.g. catamount) client to
410 return to the app before handling all startup RPCs. This
411 could leave the node unresponsive to lustre network traffic
412 and manifested as a server ptllnd timeout.
414 Severity : enhancement
416 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
417 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
418 boolean to control whether to print peer debug info when a
419 client's RPC times out.
424 Description: make O_SYNC working on 2.6 kernels
425 Details : 2.6 kernels use different method for mark pages for write,
426 so need add a code to lustre for O_SYNC work.
431 Description: Failure to close file and release space on NFS
432 Details : Put inode details into lock acquired in ll_intent_file_open.
433 Use mdc_intent_lock in ll_intent_open to properly
434 detect all kind of errors unhandled by mdc_enqueue.
439 Description: proc file read during shutdown sometimes raced obd removal,
441 Details : Add lock to prevent obd access after proc file removal.
444 Frequency : Only for files larger than 4GB on 32-bit clients.
446 Description: improperly doing page alignment of locks
447 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
448 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
453 Description: RPCs being resent when they shouldn't be
454 Details : Some RPCs that should not be resent are being resent. This
455 can cause inconsistencies in the RPC state machine. Do not
456 resend such requests.
459 Frequency : rare, only with NFS export
461 Description: Crash on NFS re-export node
462 Details : under very unusual load conditions an assertion is hit in
463 ll_intent_file_open()
466 Frequency : only if OST filesystem is corrupted
468 Description: client incorrectly hits assertion in ptlrpc_replay_req()
469 Details : for a short time RPCs with bulk IO are in the replay list,
470 but replay of bulk IOs is unimplemented. If the OST filesystem
471 is corrupted due to disk cache incoherency and then replay is
472 started it is possible to trip an assertion. Avoid putting
473 committed RPCs into the replay list at all to avoid this issue.
478 Description: large O_DIRECT requests fail under memory pressure/fragmentation
479 Details : Large single O_DIRECT read and write calls can fail to allocate
480 a sufficiently large buffer to process the request. In case of
481 allocation failure the allocation is retried with a smaller
482 buffer and broken into smaller requests.
484 Severity : enhancement
486 Description: Add -o localflock option to simulate old noflock behaviour.
487 Details : This will achieve local-only flock/fcntl locks coherentness.
492 Description: versioning check is incomplete
493 Details : Checking the version difference of client vs. server, report
494 error if the gap is too big.
499 Description: add support PG_writeback bit
500 Details : add support for PG_writeback bit for Lustre, for more carefull
501 work with page cache in 2.6 kernel. This also fix some deadlocks
502 and remove hack for work O_SYNC with 2.6 kernel.
504 Severity : enhancement
506 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
507 Details : The uninit_groups feature works in conjunction with the kernel
508 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
509 up the pass1 processing of e2fsck. This is a read-only feature
510 in ldiskfs2 only, so older kernels and current ldiskfs cannot
511 mount filesystems that have had this feature enabled.
513 Severity : enhancement
515 Description: Improve multi-block allocation algorithm to avoid fragmentation
516 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
517 allocation locality and avoid filesystem fragmentation.
519 ------------------------------------------------------------------------------
521 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
523 * Support for kernels:
524 2.6.9-42.0.3EL (RHEL 4)
526 2.4.21-47.0.1.EL (RHEL 3)
527 2.6.12.6 vanilla (kernel.org)
528 * Recommended e2fsprogs version: 1.39.cfs2-0
530 * The backwards-compatible /proc/sys/portals symlink has been removed
531 in this release. Before upgrading, please ensure that you change
532 any configuration scripts or /etc/sysctl.conf files that access
533 /proc/sys/portals/* or sysctl portals.* to use the corresponding
534 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
535 in advance of the upgrade on any system running Lustre 1.4.6 or
536 newer, since /proc/sys/lnet was added in that version.
537 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
541 Frequency : only when quota is used
543 Description: avoid scanning export list for quota master
544 Details : Change the algorithms to avoid scanning export list in order
545 to improve the efficiency.
548 Frequency : MDS failover only, very rarely
550 Description: "went back in time" messages on mds failover
551 Details : The greatest transno may be lost when the current operation
552 finishes with an error (transno==0) and the client's last_rcvd
553 record is over-written. Save the greatest transno in the
554 mds_last_transno for this case.
557 Frequency : always for specific kernels and striping counts
559 Description: client may get "Matching packet too big" without ACL support
560 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
561 when trying to access files in certain configurations. The
562 clients should in fact be denied when mounting because they do
566 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
568 Description: many clients connecting with IO in progress causes connect timeouts
569 Details : Avoid synchronous journal commits to avoid delays caused by many
570 clients connecting/disconnecting when bulk IO is in progress.
571 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
572 OST_IO_PORTAL to avoid delays behind potentially many pending
576 Frequency : occasionally with multiple writers to a single file
578 Description: shared writes to file may result in wrong size reported by stat()
579 Details : Allow growing of kms when extent lock is cancelled
582 Frequency : always with random mmap IO to multi-striped file
584 Description: mmap write might be lost if we are writing to a 'hole' in stripe
585 Details : Only if the hole is at the end of OST object so that kms is too
586 small. Fix is to increase kms accordingly in ll_nopage.
589 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
591 Description: writes to a missing object would leak memory on the OST
592 Details : If there is an inconsistency between the MDS and OST filesystems,
593 such that the MDS references an object that doesn't exist, writes
594 to that object will leak memory due to incorrect cleanup in the
595 error handling path, eventually running out of memory on the OST.
600 Description: Creating too long symlink causes lustre errors
601 Details : Check symlink and name lengths before sending requests to MDS.
604 Frequency : only if flock is enabled (not on by default)
606 Description: posix locks not released on fd closure on 2.6.9+
607 Details : We failed to add posix locks to list of inode locks on 2.6.9+
608 kernels, this caused such locks not to be released on fd close and
609 then assertions on fs unmount about still used locks.
612 Frequency : MDS failover only, very rarely
614 Description: clients may get ASSERTION(granted_lock != NULL)
615 Details : When request was taking a long time, and a client was resending
616 a getattr by name lock request. The were multiple lock requests
617 with the same client lock handle and
618 mds_getattr_name->fixup_handle_for_resent_request found one of the
619 lock handles but later failed with ASSERTION(granted_lock != NULL).
624 Description: handle->h_buffer_credits > 0, assertion failure
625 Details : h_buffer_credits is zero after truncate, causing assertion
626 failure. This patch extends the transaction or creates a new
630 Frequency : NFS re-export or patchless client
631 Bugzilla : 11179, 10796
632 Description: Crash on NFS re-export node (__d_move)
633 Details : We do not want to hash the dentry if we don't have a lock.
634 But if this dentry is later used in d_move, we'd hit uninitialised
635 list head d_hash, so we just do this to init d_hash field but
636 leave dentry unhashed.
639 Frequency : NFS re-export or patchless client
641 Description: NFS exports has problem with symbolic link
642 Details : lustre client didn't properly install dentry when re-exported
643 to NFS or running patchless client.
646 Frequency : NFS re-export or patchless client
648 Description: Various nfs/patchless fixes.
649 Details : fixes reuse disconected alias for lookup process - this fixes
650 warning "find_exported_dentry: npd != pd",
651 fix permission error with open files at nfs.
652 fix apply umask when do revalidate.
655 Frequency : occasional
657 Description: Crash on NFS re-export node
658 Details : calling clear_page() on the wrong pointer triggered oops in
659 generic_mapping_read().
662 Frequency : rarely, using O_DIRECT IO
664 Description: unaligned directio crashes client with LASSERT
665 Details : check for unaligned buffers before trying any requests.
668 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
670 Description: stale data returned from RAID cache
671 Details : If only a small amount of IO is done to the RAID device before
672 reading it again it is possible to get stale data from the RAID
673 cache instead of reading it from disk.
676 Frequency : always for sles10 kernel
678 Description: sles10 support
679 Details : ll_follow_link: compile fixes and using of nd_set_link
683 Frequency : depends on arch, kernel and compiler version, always on sles10
686 Description: recursive or deep enough symlinks cause stack overflow
687 Details : getting rid of large stack-allocated variable in
691 Frequency : depends on hardware
693 Description: lustre write performance loss in the SLES10 kernel
694 Details : the performance loss is caused by using of write barriers in the
695 ext3 code. The SLES10 kernel turns barrier support on by
696 default. The fix is to undo that change for ldiskfs.
699 ------------------------------------------------------------------------------
701 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
703 * Support for kernels:
704 2.6.9-42.0.3EL (RHEL 4)
706 2.4.21-47.0.1.EL (RHEL 3)
707 2.6.12.6 vanilla (kernel.org)
711 Frequency : quota enabled and large files being deleted
713 Description: releasing more than 4GB of quota at once hangs OST
714 Details : If a user deletes more than 4GB of files on a single OST it
715 will cause the OST to spin in an infinite loop. Release
716 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
721 Description: statfs data retrieved from /proc may be stale or zero
722 Details : When reading per-device statfs data from /proc, in the
723 {kbytes,files}_{total,free,avail} files, it may appear
724 as zero or be out of date.
727 Frequency : systems with MD RAID1 external journal devices
729 Description: lconf's call to blkid is confused by RAID1 journal devices
730 Details : Use the "blkid -l" flag to locate the MD RAID device instead
731 of returning all block devices that match the journal UUID.
734 Frequency : always, for aggregate stripe size over 4GB
736 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
737 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
738 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
741 Frequency : always if "lfs find" used on a local file/directory
743 Description: "lfs find" segfaults if used on a local file/directory
744 Details : The case where a directory component was not specified wasn't
745 handled correctly. Handle this properly.
748 Frequency : always on ppc64
750 Description: the write to an ext3 filesystem mounted with mballoc got stuck
751 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
752 perform endianness conversion.
755 Frequency : rarely (truncate to non-zero file size after write under load)
756 Bugzilla : 10730, 10687
757 Description: Files padded with zeros to next 4K multiple
758 Details : With filesystems mounted using the "extents" option (2.6 kernels)
759 it is possible that files that are truncated to a non-zero size
760 immediately after being written are filled with zero bytes beyond
761 the truncated size. No file data is lost.
763 Severity : enhancement
765 Description: Allow recovery/failover for liblustre clients.
766 Details : liblustre clients were unaware of failover configurations until
769 Severity : enhancement
771 Description: user file locks should fail when not mounting with flock option
772 Details : Set up an error-returning stub in ll_file_operations.lock field
773 to prevent incorrect behaviour when client is mounted without
774 flock option. Also, set up properly f_op->flock field for
778 Frequency : always on ia64
780 Description: "lfs df" loops on printing out MDS statfs information
781 Details : The obd_ioctl_data was not initialized and in some systems
782 this caused a failure during the ioctl that did not return
783 an error. Initialize the struct and return an error on failure.
786 Frequency : SLES 9 only
788 Description: Error of copying files with lustre special EAs as root
789 Details : Client side always return success for setxattr call for lustre
790 special xattr (currently only "trusted.lov").
793 Frequency : rarely on clusters with both ia64+i386 clients
795 Description: ia64+i686 clients doing shared IO on the same file may LBUG
796 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
797 clients are doing concurrent writes to the same file it is
798 possible that the ia64 clients may LASSERT because the OST
799 extent locks are not PAGE_SIZE aligned. Ensure that grown
800 locks are always aligned on the request boundary.
803 Frequency : specific use, occasional
805 Description: Overwriting in use executable truncates on-disk binary image
806 Details : If one node attempts to overwrite an executable in use by
807 another node, we now correctly return ETXTBSY instead of
810 Severity : enhancement
812 Description: Async OSC create to avoid the blocking unnecessarily.
813 Details : If a OST has no remain object, system will block on the creating
814 when need to create a new object on this OST. Now, ways use
815 pre-created objects when available, instead of blocking on an
816 empty osc while others are not empty. If we must block, we block
817 for the shortest possible period of time.
822 Description: chmod on Lustre root is propagated to other clients
823 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
824 invalid by the follow_mount time.
827 Frequency : liblustre clients only
829 Description: Race in 'instant cancel' lock handling could lead to such locks
830 never to be granted in case of SMP MDS
831 Details : Do not destroy not yet granted but cbpending locks in
835 Frequency : replay/resend of open
837 Description: non null lock assetion failure in mds_intent_policy
838 Details : Trying to replay/resend lockless open requests resulted in
839 mds_open() returning 0 with no lock. Now it sets a flag if
840 a lock is going to be returned.
842 Severity : enhancement
844 Description: Checksum enhancements
845 Details : New checksum enhancements allow for resending RPCs that failed
848 Severity : enhancement
850 Description: Tunables on number of dirty pages in cacche
851 Details : Allow to set limit on number of dirty pages cached.
856 Description: client crash on unmount - lock still has references
857 Details : In some error handling cases it was possible to leak a lock
858 reference on a client while accessing a file. This was not
859 harmful to the client during operation, but would cause the
860 client to crash when the filesystem is unmounted.
863 Frequency : specific case, rare
865 Description: ETXTBSY on mds though file not in use by client
866 Details : ETXTBSY is no longer incorrectly returned when attempting to
867 chmod or chown a directory that the user previously tried to
868 execute or a currently-executing binary.
871 Frequency : extremely rare except on liblustre-based clients
873 Description: Lustre space not freed when files are deleted
874 Details : Clean up open-unlinked files after client eviction. Previously
875 the unlink was skipped and the files remained as orphans.
880 Description: OST failure "would be an LBUG" in waiting_locks_callback()
881 Details : In some cases it was possible to send a blocking callback to
882 a client doing a glimpse, even though that client didn't get
883 a lock granted. When the glimpse lock is cancelled on the OST
884 the freed lock is left on the waiting list and corrupted the list.
887 Frequency : all core dumps
889 Description: Broke core dumps to lustre
890 Details : Negative dentry may be unhashed if parent does not have UPDATE
891 lock, but some callers, e.g. do_coredump, expect dentry to be
892 hashed after successful create, hash it in ll_create_it.
894 ------------------------------------------------------------------------------
896 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
898 * Support for kernels:
899 2.6.9-42.0.2.EL (RHEL 4)
901 2.4.21-40.EL (RHEL 3)
902 2.6.12.6 vanilla (kernel.org)
906 Frequency : always on RHEL 3
908 Description: Number of open files grows over time
909 Details : The number of open files grows over time, whether or not
910 Lustre is started. This was due to a filp leak introduced
911 by one of our kernel patches.
913 ------------------------------------------------------------------------------
915 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
917 * Support for kernels:
920 2.4.21-40.EL (RHEL 3)
921 2.6.12.6 vanilla (kernel.org)
926 Bugzilla : 5719, 9635, 9792, 9684
927 Description: OST (or MDS) trips assertions in (re)connection under heavy load
928 Details : If a server is under heavy load and cannot reply to new
929 connection requests before the client resends the (re)connect,
930 the connection handling code can behave badly if two service
931 threads are concurrently handing separate (re)connections from
932 the same client. Add better locking to the connection handling
933 code, and ensure that only a single connection will be processed
934 for a given client UUID, even if the lock is dropped.
936 Severity : enhancement
938 Description: add TCP zero-copy support to kernel
939 Details : Add support to the kernel TCP stack to allow zero-copy bulk
940 sends if the hardware supports scatter-gather and checksumming.
941 This allows socklnd to do client-write and server-read more
942 efficiently and reduce CPU utilization from skbuf copying.
945 Frequency : only if NFS exporting from client
947 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
948 Details : If setting attributes on a file created under NFS that had
949 never been opened it would be possible to oops the client
950 if the file had no objects.
954 Bugzilla : 9326, 10402, 10897
955 Description: client crash in ptlrpcd_wake() thread when sending async RPC
956 Details : It is possible that ptlrpcd_wake() dereferences a freed async
957 RPC. In rare cases the ptlrpcd thread alread processed the RPC
958 before ptlrpcd_wake() was called and the request was freed.
961 Frequency : always for liblustre
963 Description: liblustre client does MDS+OSTs setattr RPC for each write
964 Details : When doing a write from a liblustre client, the client
965 incorrectly issued an RPC to the MDS and each OST the file was
966 striped over in order to update the timestamps. When writing
967 with small chunks and many clients this could overwhelm the MDS
968 with RPCs. In all cases it would slow down the write because
969 these RPCs are unnecessary.
971 Severity : enhancement
973 Description: allow number of MDS service threads to be changed at module load
974 Details : It is now possible to change the number of MDS service threads
975 running. Adding "options mds mds_num_threads={N}" to the MDS's
976 /etc/modprobe.conf will set the number of threads for the next
977 time Lustre is restarted (assuming the "mds" module is also
978 reloaded at that time). The default number of threads will
979 stay the same, 32 for most systems.
984 Description: OST crash if filesystem is unformatted or corrupt
985 Details : If an OST is started on a device that has never been formatted
986 or if the filesystem is corrupt and cannot even mount then the
987 error handling cleanup routines would dereference a NULL pointer.
992 Description: NULL pointer deref in llap_from_page.
993 Details : get_cache_page_nowait can return a page with NULL (or otherwise
994 incorrect) mapping if the page was truncated/reclaimed while it was
995 searched for. Check for this condition and skip such pages when
996 doing readahead. Introduce extra check to llap_from_page() to
997 verify page->mapping->host is non-NULL (so page is not anonymous).
1000 Frequency : Sometimes when using sys_sendfile
1002 Description: "page not covered by a lock" warnings from ll_readpage
1003 Details : sendfile called ll_readpage without right page locks present.
1004 Now we introduced ll_file_sendfile that does necessary locking
1005 around call to generic_file_sendfile() much like we do in
1009 Frequency : with certain MDS communication failures at client mount time
1011 Description: NULL pointer deref after failed client mount
1012 Details : a client connection request may delayed by the network layer
1013 and not be sent until after the PTLRPC layer has timed out the
1014 request. If the client fails the mount immediately it will try
1015 to clean up before the network times out the request. Add a
1016 reference from the request import to the obd device and delay
1017 the cleanup until the network drops the request.
1020 Frequency : occasionally during client (re)connect
1022 Description: assertion failure during client (re)connect
1023 Details : processing a client connection request may be delayed by the
1024 client or server longer than the client connect timeout. This
1025 causes the client to resend the connection request. If the
1026 original connection request is replied in this interval, the
1027 client may trip an assertion failure in ptlrpc_connect_interpret()
1028 which thought it would be the only running connect process.
1031 Frequency : only with obd_echo servers and clients that are rebooted
1033 Description: kernel BUG accessing uninitialized data structure
1034 Details : When running an obd_echo server it did not start the ping_evictor
1035 thread, and when a client was evicted an uninitialized data
1036 structure was accessed. Start the ping_evictor in the RPC
1037 service startup instead of the OBD startup.
1039 Severity : enhancement
1040 Bugzilla : 10193 (patchless)
1041 Description: Remove dependency on various unexported kernel interfaces.
1042 Details : No longer need reparent_to_init, exit_mm, exit_files,
1043 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1046 Frequency : rare (only users of deprecated and unsupported LDAP config)
1048 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1049 Details : LDAP apparently contains 'lustreName' attributes instead of
1050 'name'. A simple remapping of the name is sufficient.
1053 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1055 Description: waiting_locks_callback trips kernel BUG if client is evicted
1056 Details : Running with the dump_on_timeout debug flag turned on makes
1057 it possible that the waiting_locks_callback() can try to dump
1058 the Lustre kernel debug logs from an interrupt handler. Defer
1059 this log dumping to the expired_lock_main() thread.
1061 Severity : enhancement
1063 Description: Support NFS exporting on 2.6 kernels.
1064 Details : Implement non-rawops metadata methods for NFS server to use without
1065 changing NFS server code.
1068 Frequency : very rare (synthetic metadata workload only)
1070 Description: two racing renames might cause an MDS thread to deadlock
1071 Details : Running the "racer" program may cause one MDS thread to rename
1072 a file from being the source of a rename to being the target of
1073 a rename at exactly the same time that another thread is doing
1074 so, and the second thread has already enqueued these locks after
1075 doing a lookup of the target and is trying to relock them in
1076 order. Ensure that we don't try to re-lock the same resource.
1079 Frequency : only very large systems with liblustre clients
1081 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1082 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1083 all OSTs in parallel. This allows the eviction of stale liblustre
1084 clients to proceed much faster than if they were done in series,
1085 and also offers similar improvements for other set_info RPCs.
1090 Description: excessive CPU usage during initial read phase on client
1091 Details : During the initial read phase on a client, it would agressively
1092 retry readahead on the file, consuming too much CPU and impacting
1093 performance (since 1.4.5.8). Improve the readahead algorithm
1094 to avoid this, and also improve some other common cases (read
1095 of small files in particular, where "small" is files smaller than
1096 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1101 Description: MDS crash when receiving packet with unknown intent.
1102 Details : Do not LBUG in unknown intent case, just return -EFAULT
1104 Severity : enhancement
1105 Bugzilla : 9293, 9385
1106 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1107 Details : Do not serialize getattr (non-intent version) and statfs.
1110 Frequency : occasional, when OST network is overloaded/intermittent
1112 Description: client evicted by OST after bulk IO timeout
1113 Details : If a client sends a bulk IO request (read or write) the OST
1114 may evict the client if it is unresposive to its data GET/PUT
1115 request. This is incorrect if the network is overloaded (takes
1116 too long to transfer the RPC data) or dropped the OST GET/PUT
1117 request. There is no need to evict the client at all, since
1118 the pinger and/or lock callbacks will handle this, and the
1119 client can restart the bulk request.
1122 Frequency : Always when mmapping file with no objects
1124 Description: client crashes when mmapping file with no objects
1125 Details : Check that we actually have objects in a file before doing any
1126 operations on objects in ll_vm_open, ll_vm_close and
1132 Description: Request leak when working with deleted CWD
1133 Details : Introduce advanced request refcount tracking for requests
1134 referenced from lustre intent.
1136 Severity : Enhancement
1138 Description: Cache open file handles on client.
1139 Details : MDS now will return special lock along with openhandle, if
1140 requested and client is allowed to hold openhandle, even if unused,
1141 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1142 closing files for every read/write openration.
1144 Severity : Enhancement
1146 Description: Cache open negative dentries on client when possible.
1147 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1148 negative dentries on lock revocation.
1153 Description: Remounting a client read-only wasn't possible with a zconf mount
1154 Details : It wasn't possible to remount a client read-only with llmount.
1156 Severity : enhancement
1157 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1158 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1159 MPICH is included to add Lustre-specific ADIO interfaces.
1160 This is based closely on the UFS ADIO layer and only differs
1161 in file creation, in order to allow the OST striping to be set.
1162 This is user-contributed code and not supported by CFS.
1167 Description: extended inode attributes (immutable, append-only) work improperly
1168 when 2.4 and 2.6 kernels are used on client/server or vice versa
1169 Details : Introduce kernel-independent values for these flags.
1171 Severity : enhancement
1174 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1175 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1176 granularity. Now, Fractional MB granularity is be supported,
1177 this is very useful for low memory system.
1179 Severity : enhancement
1181 Description: Getattr by fid
1182 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1183 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1184 have locks and avoids getattr rpc requests for it on every stat.
1187 Frequency : Always, for filesystems larger than 2TB
1189 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1190 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1191 filesystems larger than 2TB to be mounted with mballoc enabled.
1194 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1196 Description: filesystem corruption for non-standard kernels and very large OSTs
1197 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1198 filesystems larger than 2TB is mounted then the kernel will
1199 silently corrupt the start of the filesystem. CONFIG_LBD is
1200 enabled for all CFS-supported kernels, but the possibility of
1201 this happening with a modified kernel config exists.
1203 Severity : enhancement
1205 Description: add client O_DIRECT support for 2.6 kernels
1206 Details : It is now possible to do O_DIRECT reads and writes to files
1207 in the Lustre client mountpoint on 2.6 kernel clients.
1209 Severity : enhancement
1211 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1212 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1213 parallel, not waiting for response from every OST before sending
1214 a rpc to the next OST.
1219 Description: setattr vs write race when updating file timestamps
1220 Details : Client processes that update a file timestamp into the past
1221 right after writing to the file (e.g. tar) it is possible that
1222 the updated file modification time can be reset to the current
1223 time due to a race between processing the setattr and write RPC.
1225 Severity : enhancement
1227 Description: Bring 'lfs find' closer in line with regular Linux find.
1228 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1229 -print0 options and obtains all the needed info through the lustre
1232 Severity : enhancement
1234 Description: support up to 1024 configured devices on one node
1235 Details : change obd_dev array from statically allocated to dynamically
1236 allocated structs as they are first used to reduce memory usage
1241 Description: Flush dirty partially truncated pages during truncate
1242 Details : Immediatelly flush partially truncated pages in filter_setattr,
1243 this way we completely avoid having any pages in page cache on OST
1244 and can retire ugly workarounds during writes to flush such pages.
1249 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1250 Details : protect inode from truncation within vfs_unlink() context
1251 just take a reference before calling vfs_unlink() and release it
1252 when parent's i_sem is free.
1255 Frequency : always, if extents are used on OSTs
1257 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1258 Details : OSTs using the extents format would not zero the high 16 bits of
1259 the index physical block number. This is not a problem for any
1260 OST filesystems smaller than 16TB, and no kernels support ext3
1261 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1262 new/modified files) and can be fixed for existing filesystems
1263 with e2fsprogs-1.39-cfs1.
1268 Description: import connection selection may be incorrect if timer wraps
1269 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1270 import connections to be ignored if the 32-bit jiffies counter
1271 wraps. Use a 64-bit jiffies counter.
1274 Frequency : during server recovery
1276 Description: crash after server is denying duplicate export
1277 Details : If clients are resending connect requests to the server, the
1278 server refuses to allow a client to connect multiple times.
1279 Fixed a bug in the handling of this case.
1282 Frequency : very large clusters immediately after boot
1284 Description: LNET request buffers exhausted under heavy short-term load
1285 Details : If a large number of client requests are generated on a service
1286 that has previously never seen so many requests it is possible
1287 that the request buffer growth cannot keep up with the spike in
1288 demand. Instead of dropping incoming requests, they are held in
1289 the LND until the RPC service can accept more requests.
1292 Frequency : Sometimes during replay
1294 Description: Assertion failure in ll_local_open after replay.
1295 Details : If replay happened on an open request reply before we were able
1296 to set replay handler, reply will become not swabbed tripping the
1297 assertion in ll_local_open. Now we set the handler right after
1298 recognising of open request
1301 Frequency : very rare
1303 Description: kernel reports "badness in vsnprintf"
1304 Details : Reading from the "recovery_status" /proc file in small chunks
1305 may cause a negative length in lprocfs_obd_rd_recovery_status()
1306 call to vsnprintf() (which is otherwise harmless). Exit early
1307 if there is no more space in the output buffer.
1309 Severity : enhancement
1311 Description: clear OBD RPC statistics by writing to them
1312 Details : It is now possible to clear the OBD RPC statistics by writing
1313 to the "stats" file.
1318 Description: Client mtime is not the same on different clients after utimes
1319 Details : In some cases, the client was using the utimes() syscall on
1320 a file cached on another node. The clients now validate the
1321 ctime from the MDS + OSTs to determine which one is right.
1326 Description: Inability to activate failout mode
1327 Details : lconf script incorrectly assumed that in python string's numeric
1328 value is used in comparisons.
1331 Frequency : always with multiple stripes per file
1333 Description: Inefficient object allocation for mutli-stripe files
1334 Details : When selecting which OSTs to stripe files over, for files with
1335 a stripe count that divides evenly into the number of OSTs,
1336 the MDS is always picking the same starting OST for each file.
1337 Return the OST selection heuristic to the original design.
1342 Description: mount failures may take full timeout to return an error
1343 Details : Under some heavy load conditions it is possible that a
1344 failed mount can wait for the full obd_timeout interval,
1345 possibly several minutes, before reporting an error.
1346 Instead return an error as soon as the status is known.
1348 ------------------------------------------------------------------------------
1350 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
1352 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1353 INTEROPERATE with older versions automatically. Please read the
1354 user documentation before upgrading any part of a live system.
1355 * WARNING: Lustre networking configuration changes are required with
1356 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1359 * Support for kernels:
1360 2.6.9-22.0.2.EL (RHEL 4)
1361 2.6.5-7.244 (SLES 9)
1362 2.6.12.6 vanilla (kernel.org)
1365 Severity : enhancement
1366 Bugzilla : 7981/8208
1367 Description: Introduced Lustre Networking (LNET)
1368 Details : LNET is new networking infrastructure for Lustre, it includes
1369 a reorganized network configuration mode (see the user
1370 documentation for full details) as well as support for routing
1371 between different network fabrics. Lustre Networking Devices
1372 (LNDs) for the supported network fabrics have also been
1373 created for this new infrastructure.
1375 Severity : enhancement
1376 Description: Introduced Access control lists
1377 Details : clients can set ACLs on files and directories in order to have
1378 more fine-grained permissions than the standard Unix UGO+RWX.
1379 The MDS must be started with the "-o acl" mount option.
1381 Severity : enhancement
1382 Description: Introduced filesystem quotas
1383 Details : Administrators may now establish per-user quotas on the
1386 Severity : enhancement
1388 Description: Configuration change for the XT3
1389 The PTLLND is now used to run Lustre over Portals on the XT3
1390 The configure option(s) --with-cray-portals are no longer used.
1391 Rather --with-portals=<path-to-portals-includes> is used to
1392 enable building on the XT3. In addition to enable XT3 specific
1393 features the option --enable-cray-xt3 must be used.
1398 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1399 Details : A race between allocating a new llcd and re-getting the llcd_lock
1400 allowed another thread to grab newly-allocated llcd.
1402 Severity : enhancement
1404 Description: 2.6 OST async journal commit and locking fix to improve performance
1405 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1406 2.6 kernels are now async as they already were in 2.4 kernels so
1407 that they can commit concurrently with the network bulk transfer.
1408 For block-allocated files the filter allocation semaphore is held
1409 to avoid filesystem fragmentation during allocation. BKL lock
1410 removed for 2.6 xattr operations where it is no longer needed.
1415 Description: lconf incorrectly determined whether two IP networks could talk
1416 Details : In some more complicated routing and multiple-network
1417 configurations, lconf will avoid trying to make a network
1418 connection to a disjoint part of the IP space. It was doing the
1419 math incorrectly for one set of cases.
1424 Description: Fix for potential infinite loop processing records in an llog.
1425 Details : If an llog record is corrupted/zeroed, it is possible to loop
1426 forever in llog_process(). Validate the llog record length
1427 and skip the remainder of the block on error.
1430 Frequency : occasional (liblustre only)
1432 Description: liblustre could not open files whose last component is a symlink
1433 Details : sysio_path_walk() would incorrectly pass the open intent to
1434 intermediate path components.
1437 Frequency : rare (liblustre only with non-standard tuning)
1438 Bugzilla : 7201 (7350)
1439 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1440 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1441 after it has been released, only do this for referenced locks
1443 Severity : enhancement
1445 Description: specifying an (invalid) directory default stripe_size of -1
1446 would reset the directory default striping
1447 Details : stripe_size -1 was used internally to signal directory stripe
1448 removal, now use "all default" to signal dir stripe removal
1449 as a directory striping of "all default" is not useful
1452 Frequency : common for large clusters running liblustre clients
1454 Description: doing an ls when liblustre clients are running is slow
1455 Details : sending a glimpse AST to a liblustre client waits for every AST
1456 to time out, as liblustre clients will not respond. Since they
1457 cannot cache data we refresh the OST lock LVB from disk instead.
1459 Severity : enhancement
1461 Description: doing an ls at the same time as file IO can be slow
1462 Details : enqueue and other "small" requests can be blocked behind many
1463 large IO requests. Create a new OST IO portal for non-IO
1464 requests so they can be processed faster.
1467 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1469 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1470 created file with mode 000
1471 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1472 allow an ftruncate() as a non-root user to a file with mode 000.
1473 Lustre now respects this flag to disable mode checks when
1474 truncating a file owned by the user
1477 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1480 Description: Revoking locks from clients that went dead or catatonic might take
1482 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1483 cancellation of such locks instant on servers without waiting for
1484 any reply from clients. Clients drops these locks when cancel
1485 notification from server is received without replying.
1488 Frequency : liblustre-only, when liblustre client dies or becomes busy
1490 Description: Doing ls on Linux clients can take a long time with active
1492 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1493 granting such locks to it in the first place if possible. Locks
1494 are taken by proxy on the OST during the read or write and
1495 dropped immediately afterward. Add connect flags handling, do
1496 not grant locks to liblustre clients for glimpse ASTs.
1498 Severity : enhancement
1500 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1501 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1502 sequential reads and behaves badly with applications doing random
1503 reads. Improve it by reading ahead at least read region, and
1504 avoiding excessive large RPC for small reads.
1506 Severity : enhancement
1508 Description: Creating more than 1000 files for a single job may cause a load
1509 imbalance on the OSTs if there are also a large number of OSTs.
1510 Details : qos_prep_create() uses an OST index reseed value that is an
1511 even multiple of the number of available OSTs so that if the
1512 reseed happens in the middle of the object allocation it will
1513 still utilize the OSTs as uniformly as possible.
1518 Description: OST or MDS may oops in ping_evictor_main()
1519 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1520 but doesn't restart at beginning of obd_exports_timed list
1523 Severity : enhancement
1525 Description: improve by-nid export eviction on the MDS and OST
1526 Details : allow multiple exports with the same NID to be evicted at one
1527 time without re-searching the exports list.
1530 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1532 Description: MDS may oops in groups_free()
1533 Details : in rare race conditions a newly allocated group_info struct is
1534 freed again, and this can be NULL. The 2.4 compatibility code
1535 for groups_free() checked for a NULL pointer, but 2.6 did not.
1538 Frequency : common for liblustre clients doing little filesystem IO
1539 Bugzilla : 9352, 7313
1540 Description: server may evict liblustre clients accessing contended locks
1541 Details : if a client is granted a lock or receives a completion AST
1542 with a blocking AST already set it would not reply to the AST
1543 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1546 Frequency : lfs setstripe, only systems with more than 160 OSTs
1548 Description: unable to set striping with a starting offset beyond OST 160
1549 Details : llapi_create_file() incorrectly limited the starting stripe
1550 index to the maximum single-file stripe count.
1553 Frequency : LDAP users only
1555 Description: lconf did not handle in-kernel recovery with LDAP properly
1556 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1558 Severity : enhancement
1560 Description: bind OST threads to NUMA nodes to improve performance
1561 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1562 node and do their allocations there to localize memory access
1564 Severity : enhancement
1566 Description: llmount can determine client NID directly from Myrinet (GM)
1567 Details : the client NID code from gmnalnid was moved directly into
1568 llmount, removing the need to use this or specifying the
1569 client NID explicitly when mounting GM clients with zeroconf
1572 Frequency : if client is started with down MDS
1574 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1575 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1576 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1577 ptlrpc_import_delayed_req() because we didn't check if the
1578 request was interrupted, and we also didn't break out of the
1579 event loop if there was no timeout
1584 Description: data loss during non-page-aligned writes to a single file from
1585 both multiple nodes and multiple threads on one node at same time
1586 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1587 inconsistency led to false short-reads, that were cached and later
1588 used by ->prepare_write() to fill in partially written page,
1589 leading to data loss.
1592 Frequency : always, if lconf --abort_recovery used
1594 Description: lconf --abort_recovery fails with 'Operation not supported'
1595 Details : lconf was attempting to abort recovery on the MDT device and not
1598 Severity : enhancement
1600 Description: remove cleanup logs
1601 Details : replace lconf-generated cleanup logs with lustre internal
1602 cleanup routines. Eliminates the need for client-cleanup and
1605 Severity : enhancement
1607 Description: add support for EAs (user and system) on lustre filesystems
1608 Details : it is now possible to store extended attributes in the Lustre
1609 client filesystem, and with the user_xattr mount option it
1610 is possible to allow users to store EAs on their files also
1612 Severity : enhancement
1614 Description: Add possibility (config option) to show minimal available OST free
1616 Details : When compiled with --enable-mindf configure option, statfs(2)
1617 (and so, df) will return least minimal free space available from
1618 all OSTs as amount of free space on FS, instead of summary of
1619 free spaces of all OSTs.
1621 Severity : enhancement
1623 Description: do not expand extent locks acquired on OST-side
1624 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1625 by server: they are not cached anyway.
1628 Frequency : when mmap is used/binaries executed from Lustre
1630 Description: Unmmap pages before throwing them away from read cache.
1631 Details : llap_shrink cache now attempts to unmap pages before discarding
1632 them (if unmapping failed - do not discard). SLES9 kernel has
1633 extra checks that trigger if this unmapping is not done first.
1638 Description: lconf didn't resolve symlinks before checking to see whether a
1639 given mountpoint was already in use
1642 Frequency : when migrating failover services
1643 Bugzilla : 6395, 9514
1644 Description: When migrating a subset of services from a node (e.g. failback
1645 from a failover service node) the remaining services would
1646 time out and evict clients.
1647 Details : lconf --force (implied by --failover) sets the global obd_timeout
1648 to 5 seconds in order to quickly disconnect, but this caused
1649 other RPCs to time out too quickly. Do not change the global
1650 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1652 Severity : enhancement
1653 Frequency : if MDS is started with down OST
1654 Bugzilla : 9439,5706
1655 Description: Allow startup/shutdown of an MDS without depending on the
1656 availability of the OSTs.
1657 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1658 Add appropriate locking and lov-osc refcounts for safe
1659 cleaning. Add osc abort_inflight calls in case the
1660 synchronize never started.
1663 Frequency : occasional (Cray XT3 only)
1665 Description: root not authorized to access files in CRAY_PORTALS environment
1666 Details : The client process capabilities were not honoured on the MDS in
1667 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1668 been accessed by an authorized user then root was able to access
1669 the file on the local client also. The root user capabilities
1670 are now allowed on the MDS, as this environment has secure UID.
1673 Frequency : occasional
1675 Description: ldiskfs "too long searching" message happens too often
1676 Details : A debugging message (otherwise harmless) prints too often on
1677 the OST console. This has been reduced to only happen when
1678 there are fragmentation problems on the filesystem.
1683 Description: Division by zero in statfs when all OSCs are inactive
1684 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
1685 lov_statfs divides by value returned by lov_get_stripecnt().
1689 Bugzilla : 9489, 3273
1690 Description: First write from each client to each OST was only 4kB in size,
1691 to initialize client writeback cache, which caused sub-optimal
1692 RPCs and poor layout on disk for the first writen file.
1693 Details : Clients now request an initial cache grant at (re)connect time
1694 and so that they can start streaming writes to the cache right
1695 away and always do full-sized RPCs if there is enough data.
1696 If the OST is rebooted the client also re-establishes its grant
1697 so that client cached writes will be honoured under the grant.
1702 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
1703 Details : Now I/O RPCs go to different portal number and (presumably) fast
1704 lock requests (and glimses) and other RPCs get their own service
1705 threads pool that should be able to service those RPCs
1708 Severity : enhancement
1710 Description: Ability to exchange lustre version between client and servers and
1711 issue warnings at client side if client is too old. Also for
1712 liblustre clients there is ability to refuse connection of too old
1714 Details : New 'version' field is added to connect data structure that is
1715 filled with version info. That info is later checked by server and
1719 Frequency : rare, liblustre only.
1720 Bugzilla : 9296, 9581
1721 Description: Two simultaneous writes from liblustre at offset within same page
1722 might proceed at the same time overwriting eachother with stale
1724 Details : I/O lock withing llu_file_prwv was released too early, before data
1725 actually was hitting the wire. Extended lock-holding time until
1726 server acknowledges receiving data.
1729 Frequency : extremely rare. Never observed in practice.
1731 Description: avoid generating lustre_handle cookie of 0.
1732 Details : class_handle_hash() generates handle cookies by incrementing
1733 global counter, and can hit 0 occasionaly (this is unlikely, but
1734 not impossible, because initial value of cookie counter is
1735 selected randonly). Value of 0 is used as a sentinel meaning
1736 "unassigned handle" --- avoid it. Also coalesce two critical
1737 sections in this function into one.
1739 Severity : enhancement
1741 Description: allow liblustre clients to delegate truncate locking to OST
1742 Details : To avoid overhead of locking, liblustre client instructs OST to
1743 take extent lock in ost_punch() on client's behalf. New connection
1744 flag is added to handle backward compatibility.
1746 Severity : enhancement
1747 Bugzilla : 4928, 7341, 9758
1748 Description: allow number of OST service threads to be specified
1749 Details : a module parameter allows the number of OST service threads
1750 to be specified via "options ost ost_num_threads={N}" in the
1751 OSS's /etc/modules.conf or /etc/modprobe.conf.
1755 Bugzilla : 6146, 9635, 9895
1756 Description: servers crash with bad pointer in target_handle_connect()
1757 Details : In rare cases when a client is reconnecting it was possible that
1758 the connection request was the last reference for that export.
1759 We would temporarily drop the export reference and get a new
1760 one, but this may have been the last reference and the export
1761 was just destroyed. Get new reference before dropping old one.
1763 Severity : enhancement
1764 Frequency : if client is started with failover MDS
1766 Description: Allow multiple MDS hostnames in the mount command
1767 Details : Try to read the configuration from all specified MDS
1768 hostnames during a client mount in case the "primary"
1771 Severity : enhancement
1773 Description: Stop sending data to evicted clients as soon as possible.
1774 Details : Check if the client we are about to send or are sending data to
1775 was evicted already. (Check is done every second of waiting,
1776 for which l_wait_event interface was extended to allow checking
1777 of exit condition at specified intervals).
1780 Frequency : rare, normally only when NFS exporting is done from client
1782 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
1784 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
1787 Frequency : always, if srand() is called before liblustre initialization
1789 Description: Liblustre uses system PRNG disturbing its usage by user application
1790 Details : Introduce internal to lustre fast and high-quality PRNG for
1791 lustre usage and make liblustre and some other places in generic
1792 lustre code to use it.
1794 Severity : enhancement
1795 Bugzilla : 9477, 9557, 9870
1796 Description: Verify that the MDS configuration logs are updated when xml is
1797 Details : Check if the .xml configuration logs are newer than the config
1798 logs stored on the MDS and report an error if this is the case.
1799 Request --write-conf, or allow starting with --old_conf.
1801 Severity : enhancement
1803 Description: Handle symlinks in the path when checking if Lustre is mounted.
1804 Details : Resolve intermediate symlinks when checking if a client has
1805 mounted a filesystem to avoid duplicate client mounts.
1810 Description: lconf can hit an error exception but still return success.
1811 Details : The lconf command catches the Command error exception at the top
1812 level script context and will exit with the associated exit
1813 status, but doesn't ensure that this exit status is non-zero.
1818 Description: failure of ptlrpc thread startup can cause oops
1819 Details : Starting a ptlrpc service thread can fail if there are a large
1820 number of threads or the server memory is very fragmented.
1821 Handle this without oopsing.
1824 Frequency : always, only if liblustre and non-default acceptor port was used
1826 Description: liblustre cannot connect to servers with non-default acceptor port
1827 Details : tcpnal_set_default_params() was not called and was therefore
1828 ignoring the environment varaible TCPNAL_PORT, as well as other
1829 TCPNAL_ environment variables
1834 Description: two objects could be created on the same OST for a single file
1835 Details : If an OST is down, in some cases it was possible to create two
1836 objects on a single OST for a single file. No problems other
1837 than potential performance impact and spurious error messages.
1841 Bugzilla : 5681, 9562
1842 Description: Client may oops in ll_unhash_aliases
1843 Details : Client dcache may become inconsistent in race condition.
1844 In some cases "getcwd" can fail if the current directory is
1850 Description: Inode refcounting problems in NFS export code
1851 Details : link_raw functions used to call d_instantiate without obtaining
1852 extra inode reference first.
1856 Bugzilla : 9942, 9903
1857 Description: Referencing freed requests leading to crash, memleaks with NFS.
1858 Details : We used to require that call to ll_revalidate_it was always
1859 followed by ll_lookup_it. Also with revalidate_special() it is
1860 possible to call ll_revalidate_it() twice for the same dentry
1861 even if first occurence returned success. This fix changes semantic
1862 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
1863 reference on a request referred from the intent.
1864 ll_intent_release() then releases such a request.
1867 Frequency : rare, normally benchmark loads only
1869 Description: unlinked inodes were kept in memory on the client
1870 Details : If a client is repeatedly creating and unlinking files it
1871 can accumulate a lot of stale inodes in the inode slab cache.
1872 If there is no other client load running this can cause the
1873 client node to run out of memory. Instead flush old inodes
1874 from client cache that have the same inode number as a new inode.
1877 Frequency : SLES9 2.6.5 kernel and long filenames only
1878 Bugzilla : 9969, 10379
1879 Description: utime reports stale NFS file handle
1880 Details : SLES9 uses out-of-dentry names in some cases, which confused
1881 the lustre dentry revalidation. Change it to always use the
1885 Frequency : rare, unless heavy write-truncate concurrency is continuous
1886 Bugzilla : 4180, 6984, 7171, 9963, 9331
1887 Description: OST becomes very slow and/or deadlocked during object unlink
1888 Details : filter_destroy() was holding onto the parent directory lock
1889 while truncating+unlinking objects. For very large objects this
1890 may block other threads for a long time and slow overall OST
1891 responsiveness. It may also be possible to get a lock ordering
1892 deadlock in this case, or run out of journal credits because of
1893 the combined truncate+unlink. Solution is to do object truncate
1894 first in one transaction without parent lock, and then do the
1895 final unlink in a new transaction with the parent lock. This
1896 reduces the lock hold time dramatically.
1899 Frequency : rare, 2.4 kernels only
1901 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
1902 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
1903 before unmounting their filesystem in order to avoid deadlock.
1904 The kernel_locked() function in 2.4 kernels only checks whether
1905 the kernel lock is held, not whether it is this process that is
1906 holding it as 2.6 kernels do.
1911 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
1912 Details : The client ptlrpc code may be trying to reconnect to a down
1913 server before a previous connection attempt has timed out.
1914 Increase the reconnect interval to be longer than the connection
1915 timeout interval to avoid sending duplicate connections to
1919 Frequency : echo_client brw_test command
1921 Description: fix echo_client to work with OST preallocated code
1922 Details : OST preallocation code (5137) didn't take echo_client IO path
1923 into account: echo_client calls filter methods outside of any
1924 OST thread and, hence, there is no per-thread preallocated
1925 pages and buffers to use. Solution: hijack pga pages for IO. As
1926 a byproduct, this avoids unnecessary data copying.
1930 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
1931 Description: Client can oops in mdc_commit_close() after open replay
1932 Details : It was possible for the MDS to return an open request with no
1933 transaction number in mds_finish_transno() if the client was
1934 evicted, but without actually returning an error. Clients
1935 would later try to replay that open and may trip an assertion
1936 Simplify the client close codepath, and always return an error
1937 from the MDS in case the open is not successful.
1940 Frequency : rare, 2.6 OSTs only
1942 Description: OST may deadlock under high load on fragmented files
1943 Details : If there was a heavy load and highly-fragmented OST filesystems
1944 it was possible to have all the OST threads deadlock waiting on
1945 allocation of biovecs, because the biovecs were not released
1946 until the entire RPC IO was completed. Instead, release biovecs
1947 as soon as they are complete to ensure forward IO progress.
1949 Severity : enhancement
1951 Description: Support for specifying external journal device at mount
1952 Details : If an OST or MDS device is formatted with an external journal
1953 device, this device major/minor is stored in the ext3 superblock
1954 and may not be valid for failover. Allow detecting and
1955 specifying the external journal at mount time.
1960 Description: Mounting an MDS with pending unlinked files may cause oops
1961 Details : target_finish_recovery() calls mds_postrecov() which returned
1962 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
1963 considers this an error and immediately begins cleaning up the
1964 lov, just after starting the mds_lov process
1966 Severity : enhancement
1968 Description: Implement 'lfs df' to report actual free space on per-OST basis
1969 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
1970 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
1971 usage of objects; '-h' to report in human readable format.
1973 ------------------------------------------------------------------------------
1975 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
1982 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
1983 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
1984 at startup during orphan unlinking. Instead, skip buddy update
1985 before setup, as it will be regenerated after recovery is complete.
1990 Description: If an OST is inactive, its locks might reference stale inodes.
1991 Details : lov_change_cbdata() must iterate over all namespaces, even if
1992 they are inactive to clear inode references from the lock.
1994 Severity : enhancement
1995 Frequency : occasional, if non-standard max_dirty_mb used
1997 Description: Client will block write RPCs if not enough grant
1998 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
1999 then the client will block writes while waiting for another RPC
2000 to complete instead of consuming its dirty limit. With change
2001 we get improved performance when max_dirty_mb is small.
2003 Severity : enhancement
2004 Bugzilla : 3389, 6253
2005 Description: Add support for supplementary groups on the MDS.
2006 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2007 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2008 lookups for user supplementary groups into a cache.
2012 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2013 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2014 for liblustre O_EXCL is enforced by MDS.
2018 Description: readdir never returns NULL in liblustre.
2019 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2020 dentry in struct dirent.
2024 Frequency : liblustre only, depends on application IO pattern
2025 Description: liblustre clients evicted if not contacting servers
2026 Details : Don't put liblustre clients into the ping_evictor list, so
2027 they will not be evicted by the pinger ever.
2029 Severity : enhancement
2031 Description: Add ability to evict clients by NID from MDS.
2032 Details : By echoing "nid:$NID" string into
2033 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2034 $NID would be instantly evicted from this MDS and from all active
2035 OSTs connected to it.
2039 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2040 Details : lookup_it_finish() used to query file size from OSTs that was not
2045 Description: service threads change working directory to that of init
2046 Details : Starting lustre service threads may pin the working directory
2047 of the parent thread, making that filesystem busy. Threads
2048 now change to the working directory of init to avoid this.
2052 Frequency : during shutdown only
2053 Description: shutdown with a failed MDS or OST can cause unmount to hang
2054 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2059 Frequency : 2.6 only
2060 Description: chmod/chown may include an extra supplementary group
2061 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2062 supplementary group and if none is specified this is used.
2065 Bugzilla : 5479 (6816)
2067 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2068 Details : If lookup is in progress on a file that is unlinked we might try
2069 to revalidate the inode and fail in revalidate after lookup is
2070 complete and ll_file_open() enqueues the open again but
2071 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2074 Frequency : always, if lconf --abort_recovery used
2076 Description: lconf --abort_recovery fails with 'Operation not supported'
2077 Details : lconf was attempting to abort recovery on the MDT device and not
2080 ------------------------------------------------------------------------------
2082 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2087 Frequency : rare (only unsupported configurations with a node running as an
2089 Bugzilla : 6514, 5137
2090 Description: Mounting a Lustre file system on a node running as an OST could
2092 Details : OSTs now preallocates memory needed to write out data at
2093 startup, instead of when needed, to avoid having to
2094 allocate memory in possibly low memory situations.
2095 Specifically, if the file system is mounted on on OST,
2096 memory pressure could force it to try to write out data,
2097 which it needed to allocate memory to do. Due to the low
2098 memory, it would be unable to do so and the node would
2099 become unresponsive.
2101 Severity : enhancement
2103 Description: Addition of lconf --service command line option
2104 Details : lconf now accepts a '--service <arg>' option, which is
2105 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2107 Severity : enhancement
2109 Description: Failover mode is now the default for OSTs.
2110 Details : By default, OSTs will now run in failover mode. To return to
2111 the old behaviour, add '--failout' to the lmc line for OSTs.
2113 Severity : enhancement
2115 Description: Health checks are now provided for MDS and OSTs
2116 Details : Additional detailed health check information on MSD and OSTs
2117 is now provided through the procfs health_check value.
2120 Frequency : occasional, depends on IO load
2122 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2123 after numerous create/delete cycles
2124 Details : The ext3 inode allocation policy would not allocate new inodes
2125 very well on the OSTs because there are no new directories
2126 being created. Instead we look for groups with free space if
2127 the parent directories are nearly full.
2132 Description: Network or server problems during mount may cause partially
2133 mounted clients instead of returning an error.
2134 Details : The config llog parsing code may overwrite the error return
2135 code during mount error handling, returning success instead
2141 Description: MDS can fail to allocate large reply buffers
2142 Details : After long uptimes the MDS can fail to allocate large reply
2143 buffers (e.g. zconf client mount config records) due to memory
2144 fragmentation or consumption by the buffer cache. Preallocate
2145 some large reply buffers so that these replies can be sent even
2146 under memory pressure.
2150 Frequency : rare (liblustre)
2151 Description: fsx running with liblustre complained that using truncate() to
2152 extend the file doesn't work. This patch corrects that issue.
2153 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2154 ATTR_SIZE and lsm use in llu_setattr_raw.
2158 Frequency : rare, only 2.6 kernels
2159 Description: Unusual file access patterns on the MDS may result in inode
2160 data being lost in very rare circumstances.
2161 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2162 code in the 2.6 kernel caused the inode and/or EA data not to
2163 be read from disk, causing single-file corruption.
2167 Frequency : rare, only 2.6 filesystems using extents
2168 Description: Heavy concurrent write and delete load may cause data corruption.
2169 Details : It was possible under high-load situations to have an extent
2170 metadata block in the block device cache from a just-unlinked
2171 file overwrite a newly-allocated data block. We now unmap any
2172 metadata buffers that alias just-allocated data blocks.
2176 Frequency : filesystems with default stripe_count larger than 77
2177 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2178 Details : lconf specifies an inode size of 4096 bytes when the default
2179 stripe_count is larger than 77. This conflicts with the default
2180 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2181 this case to avoid pinning too much memory for large EAs.
2183 ------------------------------------------------------------------------------
2185 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2190 Frequency : rare (extremely heavy IO load with hundreds of clients)
2192 Description: Client is evicted, gets IO error writing to file
2193 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2194 caused clients to be evicted for AST timeouts. The fixes in
2195 bug 5192 mean we no longer need to have such short AST timeouts
2196 so ldlm_timeout has been increased.
2199 Frequency : occasional during --force or --failover shutdown under load
2200 Bugzilla : 5949, 4834
2201 Description: Server oops/LBUG if stopped with --force or --failover under load
2202 Details : a collection of import/export refcount and cleanup ordering
2203 issues fixed for safer force cleanup
2206 Frequency : only filesystems larger than 120 OSTs
2207 Bugzilla : 5990, 6223
2208 Description: lfs getstripe would oops on a very large filesystem
2209 Details : lov_getconfig used kfree on vmalloc'd memory
2212 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2213 Bugzilla : 6242, 6243
2214 Description: reading from files that had been truncated to a non-zero size
2215 but never opened returned no data
2216 Details : ll_file_read() reads zeros from no-object files to EOF
2221 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2222 Details : There's little harm in aborting MDS/OSS recovery and letting it
2223 try again, so I removed the LASSERT and return an error instead.
2225 Severity : enhancement
2227 Description: New debugging infrastructure for tracking down data corruption
2228 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2229 (b) cover more of the client-side code path, and (c) try to narrow
2230 down where problems occurred
2234 Bugzilla : 3819, 4364, 4397, 6313
2235 Description: Racing close and eviction MDS could cause assertion in mds_close
2236 Details : It was possible to get multiple mfd references during close and
2237 client eviction, leading to one thread referencing a freed mfd.
2239 Severity: : enhancement
2240 Bugzilla : 3262, 6359
2241 Description: Attempts to reconnect to servers are now more aggressive.
2242 Details : This builds on the enhanced upcall-less recovery that was added
2243 in 1.4.2. When trying to reconnect to servers, clients will
2244 now try each server in the failover group every 10 seconds. By
2245 default, clients would previously try one server every 25 seconds.
2250 Description: After recovery, certain operations trigger a failed
2251 assertion on a client.
2252 Details : Failing over an mds, using lconf -d --failover, while a
2253 client was doing a readdir() call would cause the client to
2254 LBUG after recovery completed and the readdir() was resent.
2256 Severity : enhancement
2258 Description: Default groups are now added by lconf
2259 Details : You can now run lconf --group <servicename> without having to
2260 manually add groups with lmc.
2263 Frequency : occasional
2265 Description: Nodes with an elan id of 0 trigger a failed assertion
2268 Frequency : always when accessing e.g. tty/console device nodes
2270 Description: tty and some other devices nodes cannot be used on lustre
2271 Details : file's private_data field is used by device data and lustre
2272 values in there got lost. New field was added to struct file to
2273 store fs-specific private data.
2276 Frequency : when exporting Lustre via NFS
2278 Description: NFSD failed occasionally when looking up a path component
2279 Details : NFSD is looking up ".." which was broken in ext3 directories
2280 that had grown large enough to become hashed.
2283 Frequency : Clusters with multiple interfaces not on the same subnet
2285 Description: Nodes will repeatedly try to reconnect to an interface which it
2286 cannot reach and report an error to the log.
2287 Details : Extra peer list entries will be created by lconf with some peers
2288 unreachable. lconf now validates the peer before adding it.
2291 Frequency : Only if a default stripe is set on the filesystem root.
2293 Description: Setting a default stripe on the filesystem root prevented the
2294 filesystem from being remounted.
2295 Details : The client was sending extra request flags in the root getattr
2296 request and did not allocate a reply buffer for the dir EA.
2299 Frequency : occasional, higher if lots of files are accessed by one client
2300 Bugzilla : 6159, 6097
2301 Description: Client trips assertion regarding lsm mismatch/magic
2302 Details : While revalidating inodes the VFS looks up inodes with ifind()
2303 and in rare cases can find an inode that is being freed.
2304 The ll_test_inode() code will free the lsm during ifind()
2305 when it finds an existing inode and then the VFS later attaches
2306 this free lsm to a new inode.
2310 Bugzilla : 6422, 7030
2311 Description: MDS deadlock between mkdir and client eviction
2312 Details : Creating a new file via mkdir or mknod (starting a transaction
2313 and getting the ns lock) can deadlock with client eviction
2314 (gets ns lock and trying to finish a synchronous transaction).
2317 Frequency : occasional
2318 Description: While starting a server, the fsfilt_ext3 module could not be
2320 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2321 kernels. Previously, lconf would still use the ext3 name
2322 when trying to load modules. Now, it will correctly use
2323 ext3 on 2.4 and ldiskfs on 2.6.
2325 Severity : enhancement
2326 Description: The default stripe count has been changed to 1
2327 Details : The interpretation of the default stripe count (0, to lfs
2328 or lmc) has been changed to mean striping across a single
2329 OST, rather than all available. For general usage we have
2330 found a stripe count of 1 or 2 works best.
2332 Severity : enhancement
2333 Description: Add support for compiling against Cray portals.
2334 Details : Conditional compiling for some areas that are different
2338 Frequency : occasional
2339 Bugzilla : 6409, 6834
2340 Description: Creating files with an explicit stripe count may lead to
2341 a failed assertion on the MDS
2342 Details : If some OSTs are full or unavailable, creating files may
2343 trigger a failed assertion on the MDS. Now, Lustre will
2344 try to use other servers or return an error to the
2348 Frequency : occasional
2350 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2351 caused lock timeout during readahead (since 1.4.2).
2352 Details : Processes doing readahead might match a lock that hasn't been
2353 granted yet if there are overlapping and conflicting lock
2354 requests. The readahead process waits on ungranted lock
2355 (original lock is CBPENDING), while OST waits for that process
2356 to cancel CBPENDING read lock and eventually evicts client.
2358 Severity : enhancement
2360 Description: Initial enabling of flock support for clients
2361 Details : Implements fcntl advisory locking and file status functions.
2362 This feature is provided as an optional mount flag (default
2363 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2364 locking are implemented yet, and those that are are not guaranteed
2365 to be completely correct in production environments.
2366 mount -t lustre -o [flock|noflock] ...
2369 Frequency : occasional
2371 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2372 assertion in the ext3 JBD (journaling) layer.
2373 Details : The b_committed_data struct is protected by the big kernel lock
2374 in 2.4 kernels, serializing journal_commit_transaction() and
2375 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2376 access to this struct. In 2.6 kernels there is finer grained
2377 locking to improve SMP performance of the JBD layer.
2381 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2383 -----------------------------------------------------------------------------
2385 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2387 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2388 versions of Lustre. Please update all servers and clients to
2389 version 1.4.2 or later at the same time. You must also run
2390 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2391 to update the configuration logs.
2393 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2394 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2395 - swab llog records of type '0' so we get proper header size/idx (5861)
2396 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2397 - fix rename of one directory over another leaking an inode (5953)
2398 - avoid SetPageDirty on 2.6 (5981)
2399 - don't re-add just-being-destroyed locks to the waiting list (5653)
2400 - when creating new directories, inherit the parent's custom
2401 striping settings if present parent (3048)
2402 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2403 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2404 - don't hold client locks on temporary worklist from l_lru (5666)
2405 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2406 - automatically evict dead clients (5921)
2407 - Update file size properly in create+truncate+fstat case (6196)
2408 - Do not unhash mountpoint dentries, do not allow removal of
2410 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2411 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2412 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2413 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2414 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2415 - Correctly handle reads of files with no objects (6243)
2416 - lctl recover will also mark a device active if deactivate used (5933)
2418 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2419 - allow --write-conf on an MDS with different nettype than client (5619)
2420 - don't write config llogs to MDS for mounts not from that MDS (5617)
2421 - lconf should create multiple TCP connections from a client (5201)
2422 - init scripts are now turned off by default; run chkconfig --on
2423 lustre and chkconfig --on lustrefs to use them
2424 - upcalls are no longer needed for clients to recover to failover
2426 - add --abort-recovery option to lconf to abort recovery on device
2428 - add support for an arbitrary number of OSTs (3026)
2429 - Quota support protocol changes.
2430 - forward compatibility changes to wire structs (6007)
2431 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2432 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2433 - improved "lustre status" script
2434 - initialize blocksize for non-regular files (6062)
2435 - added --disable-server and --disable-client configure options (5782)
2436 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2437 - Vanilla 2.4.29 support
2438 - increase maximum number of obd devices to 520 (6242)
2439 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2440 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2441 SLES 9 2.6.5 kernels
2442 - limit stripes per file to 160 (the maximum EA size) (6093)
2444 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2447 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2448 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2449 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2450 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2451 - small ext3 extents cleanups and fixes (5733)
2452 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2453 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2454 - avoid needless client->OST connect, fix handle mismatch (5317)
2455 - fix DLM error path that led to out-of-sync client, long delays (5779)
2456 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2457 - fix several locking issues related to i_size (5492,5624,5654,5672)
2458 - don't move pending lock onto export if it is already evicted (5683)
2459 - fix kernel oops when creating .foo in unlinked directory (5548)
2460 - fix deadlock in obdfilter statistics vs. object create (5811)
2461 - use time_{before,after} to avoid timer jiffies wrap (5882)
2462 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2463 - Do not leak request if server was not able to process it (5154)
2464 - If mds_open unable to find parent dir, make that negative lookup(5154)
2465 - don't create new directories with extent-mapping (5909, 5936)
2467 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2468 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2469 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2470 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2471 - support for RHEL4 kernel on i686 (5773)
2472 - provide error messages when incompatible logs are encountered (5898)
2474 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2475 * version 1.4.0.10 (1.4.1 release candidate 1)
2477 - don't keep a lock reference when lock is not granted (4238)
2478 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2479 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2480 - drop import inflight refcount on signal_completed_replay error (5255)
2481 - unlock page after async write error during send (3677)
2482 - handle missing objects in filter_preprw_read properly (5265)
2483 - no transno return for symlink open, don't save no-trasno open (3440)
2484 - don't try to complete elan receive that already failed (4012)
2485 - free RPC server reply state on error (5406)
2486 - clean up thread from ptlrpc_start_thread() on error (5160)
2487 - readahead could read extra page into cache that wasn't ejected (5388)
2488 - prevent races in class_attach/setup/cleanup/detach (5260)
2489 - don't dereference de->d_inode after l_dput of de (5458)
2490 - use "int" for stripe value returned from lock_to_stripe (5544)
2491 - mballoc allocation and error-checking fixes in 2.6 (5504)
2492 - block device patches to fix I/O request sizes in 2.6 (5482)
2493 - look up hostnames for IB nals (5602)
2494 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2495 - don't start multiple acceptors for the same port (5277)
2496 - fix incorrect LASSERT in mds_getattr_name (5635)
2497 - export a proc file for general "ping" checking (5628)
2498 - fix "lfs check" to not block when the MDS is down (5628)
2500 - service request history (4965)
2501 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2502 - create an "evict_client" /proc entry on OSTs, like the MDS has
2503 - fix mount usage message, return errors per mount(8) (5168)
2504 - change grep [] to grep "[]" in tests so they work in more UMLs
2505 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2506 - remove ancient LOV_MAGIC_V0 EA support (5047)
2507 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2508 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2509 - allow readahead allocations to fail when low on memory (5383)
2510 - mmap locking landed again, after considerable improvement (2828)
2511 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2513 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2516 - send OST transaction number in read/write reply to free req (4966)
2517 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2518 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2519 - the watchdog thread now runs as interruptible (5246)
2520 - flock/lockf fixes (but it's still disabled, pending 5135)
2521 - don't use EXT3 constants in llite code (5094)
2522 - memory shortage at startup could cause assertion (5176)
2524 - reorganization of lov code
2525 - single portals codebase
2527 - add extents/mballoc support (5025)
2528 - direct I/O reads in the obdfilter (4048)
2529 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2531 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2534 - send OST transaction number in read/write reply to free req (4966)
2535 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2536 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2537 - the watchdog thread now runs as interruptible (5246)
2538 - handle missing objects in filter_preprw_read properly (5265)
2539 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2540 - drop import inflight refcount on signal_completed_replay error (5255)
2541 - unlock page after async write error during send (3677)
2542 - return original error code on reconstructed replies (3761)
2543 - no transno return for symlink open, don't save no-trasno open (3440)
2545 - add pid to ldlm debugging output (4922)
2546 - bump the watchdog timeouts -- we can't handle 30sec yet
2547 - extra debugging for orphan dentry/inode bug (5259)
2549 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2552 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2553 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2554 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2555 - fix NULL dereference in /proc/sys/portals/routes (4827)
2556 - allow failed mdc_close() operations to be interrupted (4561)
2557 - stop precreate on OST before MDS would time out on it (4778)
2558 - don't send partial-page writes before EOF from client (4410)
2559 - discard client grant for sub-page writes on large-page clients (4520)
2560 - don't free dentries not owned by NFS code, check generation (4806)
2561 - fix lsm leak if mds_create_objects() fails (4801)
2562 - limit debug_daemon file size, always print CERROR messages (4789)
2563 - use transno after validating reply (3892)
2564 - process timed out requests if import state changes (3754)
2565 - update mtime on OST during writes, return in glimpse (4829)
2566 - add mkfsoptions to LDAP (4679)
2567 - use ->max_readahead method instead of zapping global ra (5039)
2568 - don't interrupt __l_wait_event() during strace
2570 - add software watchdogs to catch hung threads quickly (4941)
2571 - make lustrefs init script start after nfs is mounted
2572 - fix CWARN/ERROR duplication (4930)
2573 - return async write errors to application if possible (2248)
2574 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2575 - print NAL number in %x format (4645)
2576 - update barely-supported suse-2.4.21-171 series (4842)
2577 - support for sles 9 %post scripts
2578 - support for building 2.6 kernel-source packages
2579 - support for sles km_* packages
2581 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2584 - ignore -ENOENT errors in osc_destroy (3639)
2585 - notify osc create thread that OSC is being cleaned up (4600)
2586 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2587 - reconstruct ost_handle() like mds_handle() (4657)
2588 - create a new thread to do import eviction to avoid deadlock (3969)
2589 - let lconf resolve symlinked-to devices (4629)
2590 - don't unlink "objects" from directory with default EA (4554)
2591 - hold socknal file ref over connect in case target is down (4394)
2592 - allow more than 32000 subdirectories in a single directory (3244)
2593 - fix blocks count for O_DIRECT writes (3751)
2594 - OST returns ENOSPC from object create when no space left (4539)
2595 - don't send truncate RPC if file size isn't changing (4410)
2596 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2597 - bind to privileged port in socknal and tcpnal (3689)
2599 - rate limit CERROR/CWARN console message to avoid overload (4519)
2600 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2601 - basic mmap support (3918)
2602 - kernel patch series update from b1_4 (4711)
2604 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2607 - avoid crash during MDS cleanup with OST shut down (2775)
2608 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2609 - don't use bad inodes on the MDS (3744)
2610 - dynamic object preallocation to improve recovery speed (4236)
2611 - don't hold spinlock over lock dumping or change debug flags (4401)
2612 - don't zero obd_dev when it is force cleaned (3651)
2613 - print grants to console if they go negative (4431)
2614 - "lctl deactivate" will stop automatic recovery attempts (3406)
2615 - look for existing locks in ldlm_handle_enqueue() (3764)
2616 - don't resolve lock handle twice in recovery avoiding race (4401)
2617 - revalidate should check working dir is a directory (4134)
2619 - don't always mark "slow" obdfilter messages as errors (4418)
2621 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2624 - don't close LustreDB during write_conf until it is done (3860)
2625 - fix typo in lconf for_each_profile (3821)
2626 - allow dumping logs from multiple threads at one time (3820)
2627 - don't allow multiple threads in OSC recovery (3812)
2628 - fix debug_size parameters (3864)
2629 - fix mds_postrecov to initialize import for llog ctxt (3121)
2630 - replace config semaphore with spinlock (3306)
2631 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2632 - don't allow enqueue to complete on a destroyed export (3822)
2633 - down write_lock before checking llog header bitmap (3825)
2634 - recover from lock replay timeout (3764)
2635 - up llog sem before sending rpc (3652)
2636 - reduce ns lock hold times when setting kms (3267)
2637 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2638 - fix NULL deref and obd_dev leak on setup error (3312)
2639 - replace some LBUG about llog ops with error handling (3841)
2640 - don't match INVALID dentries from d_lookup and spin (3784)
2641 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2642 - fix invalid assertion in ptlrpc_set_wait (3880)
2644 - add libwrap support for the TCP acceptor (3996)
2645 - add /proc/sys/portals/routes for non-root route listing (3994)
2646 - allow setting MDS UUID in .xml (2580)
2647 - print the stack of a process that LBUGs (4228)
2649 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2652 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2653 - make sure to unset replay flag from failed open requests (3440)
2654 - if default stripe count is 0, use OST count for inode size (3636)
2655 - update parent mtime/ctime on client for create/unlink (2611)
2656 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2657 - free recovery state on server during a forced cleanup (3571)
2658 - unregister_reply for resent reqs (3063)
2659 - loop back devices mounting and status check on 2.6 (3563)
2660 - fix resource-creation race that can provoke i_size == 0 (3513)
2661 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2662 - more debugging for page-accounting assertion (3746)
2663 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2664 - avoid deadlock after precreation failure (3758)
2665 - fix race and lock order deadlock in orphan handling (3450, 3750)
2666 - add validity checks when grabbing inodes from l_ast_data (3599)
2668 - add /proc/.../recovery_status to obdfilter (3428)
2669 - lightweight CDEBUG infrastructure, debug daemon (3668)
2670 - change default OSC RPC parameters to be better on small clusters
2671 - turn off OST read cache for files smaller than 32MB
2672 - install man pages and include them in rpms (3100)
2673 - add new init script for (un)mounting lustre filesystems (2593)
2674 - run chkconfig in %post for init scripts (3701)
2675 - drop scimac NAL (unmaintained)
2677 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
2680 - clean kiobufs before and after use (3485)
2681 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
2682 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
2683 - change init script to fail more clearly if not run as root (1528)
2684 - allow clients to reconnect during replay (1742)
2685 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
2686 - don't do DNS lookups on NIDs too small for IP addresses (3442)
2687 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
2688 - fix cond_resched (3554)
2689 - only evict unfinished clients after recovery (3515)
2690 - allow bulk resend, prevent data loss (3570)
2691 - dynamic ptlrpc request buffer allocation (2102)
2692 - don't allow unlinking open directory if it isn't empty (2904)
2693 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
2694 - remove extraneous obd dereference causing LASSERT failure (3334)
2695 - don't use get_cycles() when creating temp. files on the mds (3156)
2696 - hold i_sem when setting i_size in ll_extent_lock() (3564)
2697 - handle EEXIST for set-stripe, set proper directory name (3336)
2699 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
2700 - fix ksocknal_fmb_callback() error messages (2918)
2702 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
2705 - don't copy lvb into (possibly NULL) reply on error (2983)
2706 - don't deref dentry after dput, don't free lvb on error (2922)
2707 - use the kms to determine writeback rpc length (2947)
2708 - increment oti_logcookies when osc is inactive (2948)
2709 - update client's i_blocks count via lvb messages (2543)
2710 - handle intent open/close of special files properly (1557)
2711 - mount MDS with errors=remount-ro, like obdfilter (2009)
2712 - initialize lock handle to avoid ASSERT on error cleanup (3057)
2713 - don't use cancelling-locks' kms values (2947)
2714 - use highest lock extent for kms, not last one (2925)
2715 - don't dereference ERR_PTR() dentry in error handling path (3107)
2716 - fix thread race in portals_debug_dumplog() (3122)
2717 - create lprocfs device entries at setup instead of at attach (1519)
2718 - common AST error handler, don't evict client on completion race (3145)
2719 - zero nameidata in detach_mnt in 2.6 (3118)
2720 - verify d_inode after revalidate_special is valid in 2.6 (3116)
2721 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
2722 - initialize RPC timeout timer earlier for 2.6 (3219)
2723 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
2724 - print nal/nid for unknown nid (3258)
2725 - additional checks for oscc recovery before doing precreate (3284)
2726 - fix ll_extent_lock() error return code for 64-bit systems (3043)
2727 - don't crash in mdc_close for bad permissions on open (3285)
2728 - zero i_rdev for non-device files (3147)
2729 - clear page->private before handing to FS, better assertion (3119)
2730 - tune the read pipeline (3236)
2731 - fix incorrect decref of invalidated dentry (2350)
2732 - provide read-ahead stats and refine rpc in flight stats (3328)
2733 - don't hold journal transaction open across create RPC (3313)
2734 - update atime on MDS at close time (3265)
2735 - close LDAP connection when recovering to avoid server load (3315)
2736 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
2737 - don't leak open file on MDS after open resend (3325)
2738 - serialize filter_precreate and filter_destroy_precreated (3329)
2739 - loop device shouldn't call sync_dev() for nul device (3092)
2740 - clear page cache after eviction (2766)
2741 - resynchronize MDS->OST in background (2824)
2742 - refuse to mount the same filesystem twice on same mountpoint (3394)
2743 - allow llmount to create routes for mounting behind routers (3320)
2744 - push lock cancellation to blocking thread for glimpse ASTs (3409)
2745 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
2746 - fix rare problem with rename on htree directories (3417)
2748 - allow default OST striping configuration per directory (1414)
2749 - fix compilation for qswnal for 2.6 kernels (3125)
2750 - increase maximum number of MDS request buffers for large systems
2751 - change liblustreapi to be useful for external progs like lfsck (3098)
2752 - increase local configuration timeout for slow disks (3353)
2753 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
2755 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2758 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
2759 - don't overwrite extent policy data in reply if lock was blocked (2901)
2760 - drop filter export grants atomically with removal from device (2663)
2761 - del obd_self_export from work_list in class_disconnect_exports (2908)
2762 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
2763 - swab reply message in mdc_close, other PPC fixes (2464)
2764 - fix destroying of named logs (2325)
2765 - overwrite old logs when running lconf --write_conf (2264)
2766 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
2767 - fix race in target_handle_connect (2898)
2768 - mds_reint_create() should take same inode create lock (2926)
2769 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
2770 - don't close files for self_export to avoid uninitialized obd (2936)
2771 - allow MDS with the same name as client node (2939)
2772 - hold dentry reference for closed log files for unlink (2325)
2773 - reserve space for all logs during transactions (2059)
2774 - don't evict page beyond end of stripe extent (2925)
2775 - don't oops on a deleted current working directory (2399)
2776 - handle hard links to targets without a parent properly (2517)
2777 - don't dereference NULL lock when racing during eviction (2867)
2778 - don't grow lock extents when lots of conflicting locks (2919)
2780 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
2783 - account for cache space usage on clients to avoid data loss (974)
2784 - lfsck support in lustre kernel code (2349)
2785 - reduce journal credits needed for BRW writes (2370)
2786 - orphan handling to avoid losing space on client/server crashes
2787 - ptlrpcd can be blocked, stopping ALL progress (2477)
2788 - use lock value blocks to assist in proper KMS, faster stat (1021)
2789 - takes i_sem instead of DLM locks internally on obdfilter (2720)
2790 - recovery for initial connections (2355)
2791 - fixes for mds_cleanup_orphans (1934)
2792 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
2793 - block all file creations until orphan recovery completes (1901)
2794 - client remove rq_connection from request struct (2423)
2795 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
2796 - recovery timer race (2670)
2797 - mdc_close recovey bug (2532)
2798 - ptlrpc cleanup bug (2710)
2799 - mds timeout on local locks (2588)
2800 - namespace lock held during RPCs (2431)
2801 - handle interrupted sync write properly (2503)
2802 - don't try to handle a message that hasn't been replied to (2699)
2803 - client assert failure during cleanup after abort recovery (2701)
2804 - leak mdc device after failed mount (2712)
2805 - ptlrpc_check_set allows timedout requests to complete (2714)
2806 - wait for inflight reqs when ptlrpcd finishes (2710)
2807 - make sure unregistered services are removed from the srv_list
2808 - reset bulk XID's when resending them (caught by 1138 test)
2809 - unregister_bulk after timeout
2810 - fix lconf error (2694)
2811 - handle write after unfinished setstripe, stripe-only getstripe (2388)
2812 - readahead locks pages, leaves pending causing memory pressure (2673)
2813 - increase OST request buffers to 4096 on large machines (2729)
2814 - fix up permission of existing directories in simple_mkdir (2661)
2815 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
2816 - don't assign transno to errored transactions (2742)
2817 - don't delete objects on OST if given a bogus objid from MDS (2751)
2818 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
2819 - if rq_no_resend, then timeout request after recovery (2432)
2820 - fix MDS llog_logid record size, 64-bit array alignment (2733)
2821 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
2822 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
2823 - MDS recovery shouldn't delete objects that it has given out (2730)
2824 - if enqueue arrives after completion, don't clobber LVB (2819)
2825 - don't unlock pages twice when trigger_group_io returns error (2814)
2826 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
2827 - don't write pages to disk if there was an error (1450)
2828 - don't ping imports that have recovery disabled (2676)
2829 - take buffered bytes into account when balancing socknal conn (2817)
2830 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
2831 - reconnect unlink llog connection after MDS reconnects to OST (2816)
2832 - remove little-endian swabbing of llog records (1987)
2833 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
2834 - retry reposting request buffers if they fail (1191)
2835 - grow extent at grant time to avoid granting a revoked lock (2809)
2836 - lock revoke doesn't evict page if covered by a second lock (2765)
2837 - disable VM readahead to avoid reading outside lock extents (2805)
2839 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
2840 - updated kernel patches for hp-2.4.20 kernel (2681)
2842 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
2845 - fix truncated write corruption (2366)
2846 - fix for failed assertion in iopen_connect_dentry (1792,2517)
2848 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
2849 - ensure len(uuid) < 37 in lmc (1171)
2850 - fix ia64 OOPS in llog_test (2255)
2851 - zero end of page at obdfilter for partial page writes (2648)
2852 - don't leave stale dentries around after renames (bug 2428)
2853 - fix timeouts when evicting a client with a single lock held (2642)
2854 - set deadline for the initial HELLO message to drain (2634)
2855 - print out dotted-quad IP addresses in the socknal (2302)
2857 - additional debugging for MDS client eviction problem (2443)
2858 - fix mkfsoptions support for osts (2603, 2604)
2860 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
2863 - add series for the vanilla 2.6.0 kernel
2864 - add series for the vanilla 2.4.24 kernel
2865 - add series for a cray x86/64 UL kernel drop
2866 - fix xattr patches for the vanilla 2.4.19 series
2868 - generate true UUIDs in lmc (1171)
2869 - have portals stack dumping break in UML (2466)
2870 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
2871 - allocate with _NOFS in ldlm to avoid deadlock (1933)
2872 - wake callback waiting threads on client eviction (2460)
2873 - Add --ptldebug and --subsystem to lmc (1719)
2874 - update assertion to allow safe interrupt allocation
2875 - set rq_no_resend for cancel requests (2432)
2876 - recalculate ptlrpcd timeout after resend (2494)
2877 - call vfs_rmdir when removing pending directories (2368)
2878 - fix renaming a file to itself (2429)
2879 - lmc creates a default one-stripe lov (2454)
2880 - expand procfs space to handle large clusters (2326)
2881 - increase UML stack to avoid overflow
2882 - update lconf's list of debug and subsystem masks
2883 - fix lfs find --obd (2510)
2884 - /proc tunable for disabling filter read caching (2591)
2885 - stop rpm packages from altering slapd.conf (2301)
2886 - disable nagle in the socknal under 0conf (2578)
2887 - choose mds inode size based on stripe count (2572)
2888 - fix kernel-source rpm problems (2516)
2890 - add --disable-doc to avoid pdf generation (2421)
2891 - update documentation, tests, type-os, comments
2892 - avoid format warnings on ia64
2893 - remove the TOE NAL
2894 - tiny code cleanups by removing unused fields
2896 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
2899 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
2900 - remove the most blatant lies from BUILDING file (2371)
2901 - change default debug level to reasonable production setting
2902 - reduce client side cache size to reduce cache flush time
2903 - reduce max RPCs in flight to avoid unnecessary file fragmentation
2904 - make TCP zerocopy and pinger support enabled by default (2476)
2905 - sync writes completed after process exits caused crashes (2319)
2906 - maintain correct mount count on the MDS (2356)
2907 - backout 1557, because 2316 wasn't really fixed
2908 - better file I/O statistics gathering in /proc
2909 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
2910 - another kernel patch to fix zero-copy TCP function export
2911 - don't take duplicate lock when processing re-sent getattr (2420)
2912 - lctl uses obd_self_export instead of creating new conn (2353)
2913 - MDS/OST recovery case which requires object creation asserted (2425)
2914 - move lfs from /usr/sbin to /usr/bin in packages
2915 - fix race between mds_client_add and mds_client_free (2417)
2916 - use kmalloc instead of slabs in portals (2430)
2917 - don't create duplicate records when a failover MDS is present (2442)
2918 - remove unnecessary mount age check (2332)
2919 - don't remove directory inodes from locks prematurely (2451)
2920 - don't break if MDS service name is the same as hostname (2103)
2921 - fix races in client write RPC generation when cache full (2482)
2923 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
2926 - remove now-unused request->rq_obd (278)
2927 - if an allocation fails, print out how much memory we've used (1933)
2928 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
2929 - add the "configurable stack size" patch to most series files (1256)
2930 - ability to write large log records, for 100+ OST configs (2306)
2931 - fix NULL deref when filter_prep fails (2314)
2932 - fix operator precedence error in filter_sync
2933 - dynamic allocation of socknal TX descriptors (2315)
2934 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
2935 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
2936 - ensure that CWARN messages in Portals always get to the syslog
2937 - __init/__exit are not for prototype decls (ldlm_init/exit)
2938 - x86-64 compile warning fixes
2939 - fix gateway LMC keyword conflict (2318)
2940 - fix MDS lock inversions in getattr/reint paths (1844)
2941 - fix a rare lock re-ordering bug, which caused deadlock (2322)
2942 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
2943 - DLM race condition prevented some lock evictions (2328)
2944 - ENOMEM detection and retry on socknal sends (2230)
2945 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
2946 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
2947 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
2948 - better error messages when a client is rejected during recovery (1505)
2949 - avoid cancelling locks which were never granted, after failure (2330)
2950 - fix i_sem/journal inversion in mds_client_add (2333)
2951 - fix truncate/getattr lock cycle deadlock (2334)
2952 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
2953 - fix two rare exit paths which could leak an l_lock() ref (2321)
2954 - fencepost error in MDS/OST orphan recovery (2226)
2955 - make log record alignment 8 bytes (1988)
2956 - lstripe now fails when requested offset > ost_count (2237)
2957 - ensure that all kernel series have a complete list.h (1607)
2958 - fix crashes in special-file operations (2316)
2959 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
2961 - allow configurable automake binary, for testing new versions
2962 - small update to the lfs documentation
2964 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
2966 * fix negative export reference count in fsfilt_sync (2312)
2968 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
2969 * release candidate 0.9.1
2971 - orphans are moved into the PENDING directory for possible recovery
2972 - replayed opens now open by fid for orphan/rename safety (1042)
2973 - last close of an orphan inode generates a transno (683)
2974 - chdir() and mount() now pin the directory entry (1020)
2975 - avoid CERROR in normal ll_setattr_raw() error case (1500)
2976 - discard very old requests without processing them (1502)
2977 - remove some common, well-understood CERRORs (1505)
2978 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
2979 - clear "grant" flags in OST replies until OST grant code lands (1644)
2980 - fix read performance by not clobbering i_blksize on client (1598)
2981 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
2982 - make LRU size a /proc tunable, clears locks when reduced (707)
2983 - fix some lprocfs rot that prevented ptlbd from loading (1732)
2984 - server locks take references on exports now (1558)
2985 - build fixes for 2.4.20-rh trees (1663)
2986 - return an error from lov_create if all OSCs are inactive (1751)
2987 - fix import levels when a reconnect happens without a timeout (1597)
2988 - exit early from mds_open if we get a lookup error (1749)
2989 - partial page read at EOF wouldn't wait for disk before sending (1642)
2990 - avoid NULL deref in obdfilter when reading page past EOF (1592)
2991 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
2992 - fix LBUG in ll_it_open_error with rc = -2 (1861)
2993 - write/truncate lock inversion (1639)
2994 - Don't auto-load obdclass, portals modules during cleanup (1495)
2995 - fix timestamps from jumping to "now" (1763)
2996 - extra journal assertions (1648)
2997 - add an extra multiunlink test (1771)
2998 - fix read_record/write_record API (1776)
2999 - fix leak of offset_extent, possible incorrect i_size later (1772)
3000 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3001 - Debugging for the kqswnal_get_idle_tx problems (1820)
3002 - Allow recovery to be attempted multiple times (1536)
3003 - Write out MDS last_rcvd file after it is first created (1600)
3004 - Fix tx_descriptor leak in failed transmit situations (1827)
3005 - ext3 journaling fixes for assertion failure after IO error (1871)
3006 - class_export_put() on freed export after completion AST error (1896)
3007 - Fix revalidate looping in VFS (1322)
3008 - Don't access a freed export during MDS_REINT timeout (1521)
3009 - Add open-unlink recovery support on the MDS (1673,1764)
3010 - Return an error if no MDS data was read from last_rcvd (1946)
3011 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3012 - Don't setup a disk device that doesn't match exported UUID (317)
3013 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3014 - avoid committing NULL handle in force close
3015 - local.sh is now a one-stripe LOV configuration
3016 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3017 - fix ext3 htree duplicate directory entry corruption (1516)
3018 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3019 - update to new LOV EA format (2097)
3020 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3022 2003-06-15 Phil Schwan <phil@clusterfs.com>
3025 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3026 - per-import/export recovery handling (958, 931, 959)
3027 - multiple last-rcvd slots, for serving multiple FSes (949)
3028 - connections are again shared between multiple imp/exports (963, 964)
3029 - "umount -f" would hang if any requests needed to be sent (393, 978)
3030 - avoid pinning large req buffer by copying for queued messages (989)
3031 - add "uuid" to "lctl device" command to help upcalls (991)
3032 - "open" RPCs with transnos would confuse recovery counters (1037)
3033 - do proper endian conversion of all wire messages (288, 340, 891)
3034 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3035 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3036 - fix ptlrpc_connection leak in target_handle_connect (1174)
3037 - fix import refcounting bug in OST and MDS cleanup (1134)
3038 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3039 - fix very unlikely obd_types race condition (501)
3040 - remove osc_open hack for echo_client (1187)
3041 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3042 - a failure in read_inode2 leads to deadlock (1139)
3043 - cancel ack-locks as soon as transaction is committed (1072)
3044 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3045 - make sure to commitrw after any preprw to avoid deadlock (1162)
3046 - failing to execute a file in a lustre FS would lock inode (1203)
3047 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3048 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3049 - obd_uuid2tgt would walk off the end of the list (1255)
3050 - on IA64 the osc would give portals incorrect bulk size (1258)
3051 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3052 - fix lock inversion caused by new llite matching code (1282)
3053 - limit the number of dirty pages on a client to 10MB (1286)
3054 - timed out locks were not being corrected cancelled (1289)
3055 - fix O_DIRECT above 4GB on IA-32 (1292)
3056 * major user-visible changes
3057 - fail out/fail over policy now controlled by the upcall (993)
3059 - add OBD_PING to check server availability and failure (954)
3060 - lustre messages are now sent in sending host order (288, 340, 891)
3061 - add eadatalen to MDS getattr reply (340)
3062 - OST read replies may contain second buffer, with per-page status (593)
3064 2003-03-11 Phil Schwan <phil@clusterfs.com>
3067 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3068 - failed open()s could cause deadlock; fixed (867, 869)
3069 - stop cancelling OST locks when files are closed (481)
3070 - overlapping XID spaces caused network corruption (851, 853)
3071 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3072 - setattr_raw added, to do single-RPC, server-side setattrs
3073 - lmc/lconf syntax change for OST UUIDs
3074 - fix crashy race condition between ptlrpc_free_req and osc_close
3075 - don't use request in mdc_enqueue if we hit a timeout (889)
3076 - don't set the inode i_size for regular files from the MDS (896)
3077 - handle out of order completion AST (842)
3078 - don't LBUG if a lock request times out after receiving AST (913)
3079 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3080 - if a bad lock AST arrives, send an error instead of dropping entirely
3081 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3082 - fix leak in bulk IO when only partially completed (899, 900, 926)
3083 - fix O_DIRECT for ia64 (55)
3084 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3085 - C-z after timeout could hang a process forever; fixed (977)
3087 - client-side I/O cache (678, 924, 929, 941, 970)
3089 - READPAGE and SETATTRs which don't take server-side locks get
3092 2003-02-11 Phil Schwan <phil@clusterfs.com>
3095 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3096 - Fix fsfilt_extN_readpage() to read a full page of directory
3097 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3098 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3099 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3100 - Most intent operations were removed, in favour of a new RPC mode
3101 that does a single RPC to the server and bypasses most of the VFS
3102 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3103 - Aggressively cancel local locks on DLM servers
3104 - mds_reint_unlink sends EA to the client if it's the last nlink.
3105 client uses that EA to unlink OST objects.
3106 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3107 - recursive symlinks were fixed (439)
3108 - fixed NULL deref in DEBUG_REQ
3109 - filter_update_lastobjid no longer calls sync, which annoyed extN
3110 - fixed multi-client small-writes to a single file problem (445)
3111 - fixed mtime updates during file writes (607)
3112 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3113 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3114 - fixed Linux OST size reporting problem (444, 656)
3115 - OST now updates object mtime with writes or setattr (607, 619)
3116 - client verifies file size before zeroing page past EOF (445)
3117 - OST now writes last allocated objid to disk with allocation (108)
3118 - LOV on echo now works (409)
3120 - mds_reint_unlink sends a new buffer, with the EA included. this
3121 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3122 set if a regular file was being unlinked, and it was the last link
3123 - use PtlGet from the target for bulk writes (315)
3124 - OST now updates object mtime with writes or setattr (607, 619)
3125 - LDLM now has a grant-time callback to revalidate locked items, if
3127 - Many MDS operations were reorganized to combat race conditions
3129 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3130 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3132 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3135 - Fully reactivate OST imports after reconnection (512, others)
3136 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3137 - More graceful error handling for truncating on dead OST (515)
3138 - Don't error out unless we're actually accessing dead stripes (474)
3139 - Fix garbage sizes when stripes are missing (410)
3140 - LRU counters were broken, causing constant lock purge (433, 432)
3141 - garbage on read from stripes with failed OSTs (441)
3142 - mark OSCs as active before reconnecting during recovery (438)
3143 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3144 - lfind did not preserve OST order in output (443)
3145 - symlinks cause hung clients, incorrect data (439)
3146 - stop dereferencing request after dropping refcount (457)
3147 - don't LASSERT(spin_is_locked) on non-SMP (455)
3148 - fixes for many rename() bugs
3149 - fstat didn't correctly synchronize attributes (399)
3150 - server must handle lock cancellation during blocking AST prep (487)
3151 - bulk descriptors were free()d too soon (511)
3152 - fix paths in lconf, which would load incorrect modules (451, 507)
3153 - fix confusing lconf 'host not found' error message (386)
3154 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3155 - fix race condition in mdc_blocking_ast() for inode access (526)
3156 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3157 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3158 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3159 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3160 - fix lstripe in several ways (kernel side) (527)
3161 - fix request leak in ldlm_cli_enqueue (262)
3162 - incorrect OSC was marked inactive after OST failure
3163 - call mds_fs_cleanup before unmounting filesystem (524)
3164 - fix races between taking ns_lock and ldlm_lock_change_resource
3165 - fix races updating LOV export open file list
3166 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3167 - fix recovery NULL deref in ldlm_cli_cancel_unused
3168 - fix some DLM races by using new hash table for lock handles (419)
3169 - permit the client to specify desired inodes, at replay
3170 - duplicate requests when we queue them for replay reintegration
3171 - fix last_rcvd offset calculation
3172 - sync after each recovered transaction, so we always make progress
3173 - never, not always, ERESTART requests without transnos
3174 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3176 - skip replay if the MDS says that the client is already connected
3177 - don't check for a recovery-enabled export to match lctl's UUID
3178 - don't INC_USE_COUNT for phantom exports
3179 - don't crash when cleaning up phantom exports (567)
3180 - don't double-finish or set replay data for errored mdc_open requests
3181 - abort requests when they time out, so we don't get old replies
3182 - send/receive replies for AST messages again
3183 - if the client says that it doesn't have the lock, cancel it on the
3185 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3186 mark it as destroyed, it will all work out when decref is called
3187 - fix module use counts (22, 581)
3189 - ASTs now expect a reply (server cancels lock on error reply)
3191 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3194 - fix many simultaneous client startup (392)
3195 - fix dentry->d_it clobbering
3196 - credentials weren't being shipped for readdir/getattr operations
3197 - remove invalid assertions triggered during some concurrent MD
3199 - proper Lustre versions added (336, 389)
3200 - fix memory leak for create error case (398)
3201 - fix LOV locking bug that would get cli/srv out of sync
3202 - fix echo client over LOV (409)
3203 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3204 - fix double-O_EXCL intent crash (424)
3205 - avoid sending multiple lock CANCELs (352)
3207 - MDS can do multi-client recovery (modulo bugs in new code)
3209 - many updates, edits, cleanups
3211 2002-11-18 Phil Schwan <phil@clusterfs.com>
3214 - fix null d_it dereference (346)
3215 - fix full OST/dbench hang (333)
3216 - fix permission problem with file removal (286)
3217 - fix removal of OSCs from LOV when they fail
3218 - fix NULL deref during bulk timeout (214)
3219 - fix problems related to multiple filesystems on one MDS (241)
3220 - fixed serious subtle metadata locking bugs
3221 - free locks on clients when inodes are removed due to memory
3223 - fix inode pointer in lock data (285)
3224 - partial support for multiple MDS on a single host (241)
3225 - data locks weren't cancelled at clear_inode time (290, 311)
3226 - intent locks could lead to unbounded lock growth (205)
3227 - added a maximum lock count, an LRU list, and a flusher
3228 - fix multiple rename (365)
3229 - properly abstracted the echo client
3230 - OSC locked 1 byte too many; fixed
3231 - rewrote brw callback code:
3232 - fixed recovery bugs related to LOVs (306)
3233 - fixed too-many-pages-in-one-write crash (191)
3234 - fixed (again) crash in sync_io_timeout (214)
3235 - probably fixed callback-related race (385)
3237 - Add capability to MDS protocol
3238 - LDLM cancellations and callbacks on different portals
3240 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3243 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3244 - defer open object destruction to close time (601981, 138)
3245 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3246 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3247 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3248 - don't call obd_setattr in ll_file_release for destroyed objects
3249 * protocol change to lustre_msg: move |version| and add |flags|
3250 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3251 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3252 * added replay of create, unlink, link and rename operations during
3253 MDS failover; recovery should be much more robust now
3254 * remove failed OSCs from LOVs (only lov_create uses this so far)
3255 * the lustre-HOWTO was brought (more) up to date (582544)
3257 2002-10-23 Phil Schwan <phil@clusterfs.com>
3260 - in-use dentries weren't being reused properly (617851)
3261 - prevent multiple LDLM setup (599178)
3262 - fix LOV size calculations for truncate (617853)
3263 - fix client handling of MDS intent errors (POSIX)
3264 - fix permission bug in lovstripe.c test (624321)
3265 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3266 - truncate past end of file could corrupt data
3267 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3268 - a race in recovery could return ETIMEDOUT to apps (623947)
3269 - building outside the source directory was fixed
3270 * the lustre-HOWTO was brought (more) up to date (582544)
3271 * major progress was made on recovery functionality
3273 2002-10-10 Phil Schwan <phil@clusterfs.com>
3276 - recovery deadlock fix
3277 - rm -rf causes LBUG fix (617817)
3278 - file open by multiple tasks fix (618962)
3279 - directory permissions bugs (602707 and 620007)
3280 - journal_stop fixed with locking (611313)
3281 - O_APPEND failures resolved (618273, perhaps 614459)
3282 - lconf PATH fix (619770)
3283 - IA64 build fix (621450)
3284 - RPC buffer sizes scale with amount of memory
3286 2002-10-01 Phil Schwan <phil@clusterfs.com>
3289 - locks would be cancelled without throwing away data pages,
3290 resulting in inconsistent data (605627)
3291 - inode attributes were not always being refreshed (605627, 612449)
3292 - lconf now continues to cleanup after lctl reports an error
3293 - MDS now enforces user permissions (602707)
3294 - lprocfs cleanup fixed, but not yet enabled (614157)
3295 - fixed infinite server hang, should a client not respond to an AST
3296 - avoid going into recovery if user calls readlink() with a buffer
3297 that's too small (613941)
3298 - AST RPCs no longer require replies (614867) -- this may be changed
3299 - don't crash server if client sends an IOV that's too big (611336)
3300 - fixed lock conversion deadlock (611892)
3301 - fixed the following of symlinks (614622)
3302 * recovery: the server can remove locks from a client that dies, other
3303 clients can make progress
3304 * more extN patch fixes
3305 * compile-time configurable ptlrpc buffer allocations
3307 - collaborative read cache document
3308 - Lustre Lite Performance CDR document-in-progress
3310 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3313 - fix typo in patch-2.4.18
3315 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3318 - clear ptlrpc request each time in handle_incoming_request()
3319 - unlink of files now destroys the object on the OST
3321 2002-09-19 Peter Braam <braam@clusterfs.com>
3323 * add hard link support
3324 * change obdfile creation method
3325 * kernel patch changed
3327 2002-09-19 Peter Braam <braam@clusterfs.com>
3330 - stack overflow bug in extN fixed
3332 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3334 * documentation updates
3335 - add man pages for config tools
3336 - update tests/README to describe testing with new config tools
3337 - finish metadata API descriptions
3338 * bug fixes and cleanups
3339 - statfs workaround for 16TB limit
3340 - LOV stripe allocation improved, can stripe on subset of OSTs
3341 - LOV file size/IO offset was wrong for files > 4GB in size
3342 - object EA data was being dropped, caused files to be unreadable
3343 - memory overflow with non-LOV OST caused memory corruption
3344 - fixed regression tests to work with new config tools, obdfilter
3345 - fixed bug when directory size became larger than 1 block
3346 - fixed bug (for single client case) when PWD was deleted
3347 - invalidate local directory pages when doing intent-based ops
3348 - avoid LDLM oops when lock callback contained bad data
3350 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3352 * documentation updates
3353 * bug fixes and cleanups
3354 - configuration tools
3357 - 64-bit compile warnings
3358 - 64-bit internal statfs data
3360 * test_brw on persistent OST devices
3362 * lprocfs (disabled)
3364 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3366 * documentation updates
3367 * bug fixes and cleanups
3368 * configuration tools
3370 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3373 * many small fixes to 0_5_4
3374 * io/network handling
3375 * thinkos in MDS operations
3377 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3380 * crucial basic fixes to 0.5.3
3381 * IOR, Iozone work over Elan
3384 2002-08-07 Phil Schwan <phil@clusterfs.com>
3385 * version 0_5_3, our first alpha
3386 * we use the new Portals iovs
3387 * documentation updates
3388 * bug fixes and cleanups
3389 * small changes in the DLM wire protocol
3391 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3392 * version 0_5_1 with some initial stability,
3393 * locking on MD and file I/O.
3394 * documentation updates
3395 * several bug fixes since 0.5.0
3396 * small changes in wire protocol
3398 2002-07-18 Phil Schwan <phil@clusterfs.com>
3400 * delivered as Lustre Light Alpha
3401 * fixed a crash after handling invalid MDS requests
3402 * fixed directory pages for architectures with non-4k pages sizes
3404 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3405 * release version v0_4_4
3406 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3408 * Fixes a number of interruption problems with OST operations.
3409 * Update documentation for portals header changes
3410 * Move all wire protocol structs/defines to lustre_idl.h
3411 * Fixes symlink length bug.
3412 * Add tcpdump to repository.
3414 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3415 * release version v0_4_3
3416 * Fixes statfs for inodes on extN.
3417 * Fixes bug in runtests which would delete /etc/hosts.
3418 * Use 64-bit object IDs wherever possible (not into VFS though)
3419 Remove ost_get_info, which is unused by lustre, and out of date.
3421 2002-07-03 Peter Braam <braam@clusterfs.com>
3422 * release version v0_4_2 Fixes a lookup error (type not passed)
3423 * move forward to head of Portals
3424 * move forward to latest Lustre kernel
3426 2002-06-25 Peter Braam <braam@clusterfs.com>
3427 * release version v0_4_1. Hopefully stable on single node use.