1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
4 INTEROPERATE with 1.4.x versions automatically. In many cases a
5 special upgrade step is needed. Please read the user documentation
6 before upgrading any part of a 1.4.x system.
7 * WARNING: Lustre configuration and startup changes are required with
8 1.6.x releases. See https://mail.clusterfs.com/wikis/lustre/MountConf
10 * Support for kernels:
11 2.6.9-42.0.10.EL (RHEL 4)
13 2.4.21-47.0.1.EL (RHEL 3)
14 2.6.12.6 vanilla (kernel.org)
15 2.6.16.27-0.9 (SLES 10)
16 * Client support for unpatched kernels:
17 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
18 2.6.16 - 2.6.19 vanilla (kernel.org)
19 2.6.9-42.0.8EL (RHEL 4)
20 * Recommended e2fsprogs version: 1.39.cfs5
22 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
25 Severity : enhancement
27 Description: lfs setstripe use optional parameters instead of postional
30 --------------------------------------------------------------------------------
32 tbd Cluster File Systems, Inc. <info@clusterfs.com>
34 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
35 INTEROPERATE with older versions automatically. In many cases a
36 special upgrade step is needed. Please read the
37 user documentation before upgrading any part of a 1.4.x system.
38 * WARNING: Lustre configuration and startup changes are required with
39 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
41 * Support for kernels:
42 2.6.9-42.0.10.EL (RHEL 4)
44 2.4.21-47.0.1.EL (RHEL 3)
45 2.6.12.6 vanilla (kernel.org)
46 2.6.16.27-0.9 (SLES 10)
47 * Client support for unpatched kernels:
48 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
49 2.6.16 - 2.6.19 vanilla (kernel.org)
50 2.6.9-42.0.8EL (RHEL 4)
51 * Recommended e2fsprogs version: 1.39.cfs5
53 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
56 Severity : enhancement
58 Description: MountConf
59 Details : Lustre configuration is now managed via mkfs and mount
60 commands instead of lmc and lconf. New obd types (MGS, MGC)
61 are added for dynamic configuration management. See
62 https://mail.clusterfs.com/wikis/lustre/MountConf for
65 Severity : enhancement
67 Description: dynamic OST addition
68 Details : OSTs can now be added to a live filesystem
70 Severity : enhancement
72 Description: startup order invariance
73 Details : MDTs and OSTs can be started in any order. Clients only
74 require the MDT to complete startup.
76 Severity : enhancement
78 Description: parallel, asynchronous orphan cleanup
79 Details : orphan cleanup is now performed in separate threads for each
80 OST, allowing parallel non-blocking operation.
82 Severity : enhancement
84 Description: optimized stripe assignment
85 Details : stripe assignments are now made based on ost space available,
86 ost previous usage, and OSS previous usage, in order to try
87 to optimize storage space and networking resources.
89 Severity : enhancement
91 Description: Permanently set tunables
92 Details : All writable /proc/fs/lustre tunables can now be permanently
93 set on a per-server basis, at mkfs time or on a live
96 Severity : enhancement
98 Description: Lustre message v2
99 Details : Add lustre message format v2.
101 Severity : enhancement
103 Description: client OST exclusion list
104 Details : Clients can be started with a list of OSTs that should be
105 declared "inactive" for known non-responsive OSTs.
108 Frequency : SFS test only (otherwise harmless)
110 Description: SPEC SFS validation failure on NFS v2 over lustre.
111 Details : Changes the blocksize for regular files to be 2x RPC size,
112 and not depend on stripe size.
114 Severity : enhancement
116 Description: Multiple MD RPCs in flight.
117 Details : Further unserialise some read-only MDT RPCs - learn about intents.
118 To avoid overly-overloading MDT, introduce a limit on number of
119 MDT RPCs in flight for a single client and add /proc controls
120 to adjust this limit.
122 Severity : enhancement
124 Description: client read/write statistics
125 Details : Add client read/write call usage stats for performance
126 analysis of user processes.
127 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
128 file access. extents_stats shows chunk size distribution.
129 extents_stats_per_process show chunk size distribution per
132 Severity : enhancement
134 Description: per-client statistics on server
135 Details : Add ldlm and operations statistics for each client in
136 /proc/fs/lustre/mds|obdfilter/*/exports/
138 Severity : enhancement
140 Description: improved MDT statistics
141 Details : Add detailed MDT operations statistics in
142 /proc/fs/lustre/mds/*/stats
144 Severity : enhancement
146 Description: VFS operations stats
147 Details : Add client VFS call stats, trackable by pid, ppid, or gid
148 /proc/fs/lustre/llite/*/vfs_ops_stats
149 /proc/fs/lustre/llite/*/vfs_track_[pid|ppid|gid]
154 Description: Fix client-side osc byte counters
155 Details : The osc read/write byte counters in
156 /proc/fs/lustre/osc/*/stats are now working
159 Frequency : always as root on SLES
161 Description: Failure of copying files with lustre special EAs.
162 Details : Client side always return success for setxattr call for lustre
163 special xattr (currently only "trusted.lov").
168 Description: Refcount LNET uuids
169 Details : The global LNET uuid list grew linearly with every startup;
170 refcount repeated list entries instead of always adding to
173 Severity : enhancement
175 Description: Dynamic service threads
176 Details : Within a small range, start extra service threads
177 automatically when the request queue builds up.
180 Frequency : mixed-endian client/server environments
182 Description: mixed-endian crashes
183 Details : The new msg_v2 system had some failures in mixed-endian
186 Severity : enhancement
188 Description: Easy OST removal
189 Details : OSTs can be permanently deactivated with e.g. 'lctl
190 conf_param lustre-OST0001.osc.active=0'
192 Severity : enhancement
194 Description: MGS proc entries
195 Details : Added basic proc entries for the MGS showing what filesystems
198 Severity : enhancement
200 Description: provide MGS failover
201 Details : Added config lock reacquisition after MGS server failover.
203 Severity : enhancement
205 Description: add Linux 2.4 support
206 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
210 Description: a large application tries to do I/O to the same resource and dies
212 Details : Check the req->rq_arrival time after the call to
213 ost_brw_lock_get(), but before we do anything about
214 processing it & sending the BULK transfer request. This
215 should help move old stale pending locks off the queue as
216 quickly as obd_timeout.
219 Frequency : when an incorrect nid is specified during startup
221 Description: ptlrpc connect to non-existant node causes kernel crash
222 Details : LNET can't be re-entered from an event callback, which
223 happened when we expire a message after the export has been
224 cleaned up. Instead, hand the zombie cleanup off to another
227 Severity : enhancement
229 Description: plain/inodebits lock performance improvement
230 Details : Grouping plain/inodebits in granted list by their request modes
231 and bits policy, thus improving the performance of search through
235 Frequency : only for kernels with patches from Lustre below 1.4.3
237 Description: Remove old rdonly API
238 Details : Remove old rdonly API which unsed from at least lustre 1.4.3
241 Frequency : only for devices with external journals
243 Description: Set external device read-only also
244 Details : During a commanded failover stop, we set the disk device
245 read-only while the server shuts down. We now also set any
246 external journal device read-only at the same time.
249 Frequency : when upgrading from 1.4 while trying to change parameters
251 Description: The wrong (new) MDC name was used when setting parameters for
252 upgraded MDT's. Also allows changing of OSC (and MDC)
253 parameters if --writeconf is specified at tunefs upgrade time.
256 Frequency : when setting specific ost indicies
258 Description: QOS code breaks on skipped indicies
259 Details : Add checks for missing OST indicies in the QOS code, so OSTs
260 created with --index need not be sequential.
262 ------------------------------------------------------------------------------
264 TBD Cluster File Systems, Inc. <info@clusterfs.com>
266 * Support for kernels:
267 2.6.16.21-0.8 (SLES10)
268 2.6.9-42.0.8EL (RHEL 4)
270 2.4.21-47.0.1.EL (RHEL 3)
271 2.6.12.6 vanilla (kernel.org)
272 * Recommended e2fsprogs version: 1.39.cfs2-0
275 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
277 Description: System hang on startup
278 Details : This bug allowed the liblustre (e.g. catamount) client to
279 return to the app before handling all startup RPCs. This
280 could leave the node unresponsive to lustre network traffic
281 and manifested as a server ptllnd timeout.
283 Severity : enhancement
285 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
286 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
287 boolean to control whether to print peer debug info when a
288 client's RPC times out.
293 Description: make O_SYNC working on 2.6 kernels
294 Details : 2.6 kernels use different method for mark pages for write,
295 so need add a code to lustre for O_SYNC work.
300 Description: Failure to close file and release space on NFS
301 Details : Put inode details into lock acquired in ll_intent_file_open.
302 Use mdc_intent_lock in ll_intent_open to properly
303 detect all kind of errors unhandled by mdc_enqueue
308 Description: proc file read during shutdown sometimes raced obd removal,
310 Details : Add lock to prevent obd access after proc file removal.
313 Frequency : Only for files larger than 4GB on 32-bit clients.
315 Description: improperly doing page alignment of locks
316 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
317 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
322 Description: RPCs being resent when they shouldn't be
323 Details : Some RPCs that should not be resent are being resent. This
324 can cause inconsistencies in the RPC state machine. Do not
325 resend such requests.
328 Frequency : rare, only with NFS export
330 Description: Crash on NFS re-export node
331 Details : under very unusual load conditions an assertion is hit in
332 ll_intent_file_open()
335 Frequency : only if OST filesystem is corrupted
337 Description: client incorrectly hits assertion in ptlrpc_replay_req()
338 Details : for a short time RPCs with bulk IO are in the replay list,
339 but replay of bulk IOs is unimplemented. If the OST filesystem
340 is corrupted due to disk cache incoherency and then replay is
341 started it is possible to trip an assertion. Avoid putting
342 committed RPCs into the replay list at all to avoid this issue.
347 Description: large O_DIRECT requests fail under memory pressure/fragmentation
348 Details : Large single O_DIRECT read and write calls can fail to allocate
349 a sufficiently large buffer to process the request. In case of
350 allocation failure the allocation is retried with a smaller
351 buffer and broken into smaller requests.
356 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
357 > 32000 subdirectories
358 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
359 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
360 subdirectory count crosses 32000. This will aid e2fsck to
361 correctly handle more than 32000 subdirectories.
366 Description: versioning check is incomplete
367 Details : Checking the version difference of client vs. server, report
368 error if the gap is too big.
373 Description: add support PG_writeback bit
374 Details : add support for PG_writeback bit for Lustre, for more carefull
375 work with page cache in 2.6 kernel. This also fix some deadlocks
376 and remove hack for work O_SYNC with 2.6 kernel.
378 ------------------------------------------------------------------------------
380 2006-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
382 * Support for kernels:
383 2.6.16.21-0.8 (SLES10)
384 2.6.9-42.0.3EL (RHEL 4)
386 2.4.21-47.0.1.EL (RHEL 3)
387 2.6.12.6 vanilla (kernel.org)
390 * The backwards-compatible /proc/sys/portals symlink has been removed
391 in this release. Before upgrading, please ensure that you change
392 any configuration scripts or /etc/sysctl.conf files that access
393 /proc/sys/portals/* or sysctl portals.* to use the corresponding
394 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
395 in advance of the upgrade on any system running Lustre 1.4.6 or
396 newer, since /proc/sys/lnet was added in that version.
397 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
401 Frequency : MDS failover only, very rarely
403 Description: "went back in time" messages on mds failover
404 Details : The greatest transno may be lost when the current operation
405 finishes with an error (transno==0) and the client's last_rcvd
406 record is over-written. Save the greatest transno in the
407 mds_last_transno for this case.
410 Frequency : always for specific kernels and striping counts
412 Description: client may get "Matching packet too big" without ACL support
413 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
414 when trying to access files in certain configurations. The
415 clients should in fact be denied when mounting because they do
419 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
421 Description: many clients connecting with IO in progress causes connect timeouts
422 Details : Avoid synchronous journal commits to avoid delays caused by many
423 clients connecting/disconnecting when bulk IO is in progress.
424 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
425 OST_IO_PORTAL to avoid delays behind potentially many pending
429 Frequency : occasionally with multiple writers to a single file
431 Description: shared writes to file may result in wrong size reported by stat()
432 Details : Allow growing of kms when extent lock is cancelled
435 Frequency : always with random mmap IO to multi-striped file
437 Description: mmap write might be lost if we are writing to a 'hole' in stripe
438 Details : Only if the hole is at the end of OST object so that kms is too
439 small. Fix is to increase kms accordingly in ll_nopage.
442 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
444 Description: writes to a missing object would leak memory on the OST
445 Details : If there is an inconsistency between the MDS and OST filesystems,
446 such that the MDS references an object that doesn't exist, writes
447 to that object will leak memory due to incorrect cleanup in the
448 error handling path, eventually running out of memory on the OST.
453 Description: Creating too long symlink causes lustre errors
454 Details : Check symlink and name lengths before sending requests to MDS.
457 Frequency : only if flock is enabled (not on by default)
459 Description: posix locks not released on fd closure on 2.6.9+
460 Details : We failed to add posix locks to list of inode locks on 2.6.9+
461 kernels, this caused such locks not to be released on fd close and
462 then assertions on fs unmount about still used locks.
465 Frequency : MDS failover only, very rarely
467 Description: clients may get ASSERTION(granted_lock != NULL)
468 Details : When request was taking a long time, and a client was resending
469 a getattr by name lock request. The were multiple lock requests
470 with the same client lock handle and
471 mds_getattr_name->fixup_handle_for_resent_request found one of the
472 lock handles but later failed with ASSERTION(granted_lock != NULL).
477 Description: handle->h_buffer_credits > 0, assertion failure
478 Details : h_buffer_credits is zero after truncate, causing assertion
479 failure. This patch extends the transaction or creates a new
483 Frequency : NFS re-export or patchless client
484 Bugzilla : 11179, 10796
485 Description: Crash on NFS re-export node (__d_move)
486 Details : We do not want to hash the dentry if we don't have a lock.
487 But if this dentry is later used in d_move, we'd hit uninitialised
488 list head d_hash, so we just do this to init d_hash field but
489 leave dentry unhashed.
492 Frequency : NFS re-export or patchless client
494 Description: NFS exports has problem with symbolic link
495 Details : lustre client didn't properly install dentry when re-exported
496 to NFS or running patchless client.
499 Frequency : NFS re-export or patchless client
501 Description: Various nfs/patchless fixes.
502 Details : fixes reuse disconected alias for lookup process - this fixes
503 warning "find_exported_dentry: npd != pd",
504 fix permission error with open files at nfs.
505 fix apply umaks when do revalidate.
508 Frequency : occasional
510 Description: Crash on NFS re-export node
511 Details : calling clear_page() on the wrong pointer triggered oops in
512 generic_mapping_read().
515 Frequency : rarely, using O_DIRECT IO
517 Description: unaligned directio crashes client with LASSERT
518 Details : check for unaligned buffers before trying any requests.
521 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
523 Description: stale data returned from RAID cache
524 Details : If only a small amount of IO is done to the RAID device before
525 reading it again it is possible to get stale data from the RAID
526 cache instead of reading it from disk.
529 Frequency : depends on arch, kernel and compiler version, always on sles10
532 Description: recursive or deep enough symlinks cause stack overflow
533 Details : getting rid of large stack-allocated variable in
537 Frequency : depends on hardware
539 Description: lustre write performance loss in the SLES10 kernel
540 Details : the performance loss is caused by using of write barriers in the
541 ext3 code. The SLES10 kernel turns barrier support on by
542 default. The fix is to undo that change for ldiskfs.
544 ------------------------------------------------------------------------------
546 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
548 * Support for kernels:
549 2.6.9-42.0.3EL (RHEL 4)
551 2.4.21-47.0.1.EL (RHEL 3)
552 2.6.12.6 vanilla (kernel.org)
556 Frequency : quota enabled and large files being deleted
558 Description: releasing more than 4GB of quota at once hangs OST
559 Details : If a user deletes more than 4GB of files on a single OST it
560 will cause the OST to spin in an infinite loop. Release
561 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
566 Description: statfs data retrieved from /proc may be stale or zero
567 Details : When reading per-device statfs data from /proc, in the
568 {kbytes,files}_{total,free,avail} files, it may appear
569 as zero or be out of date.
572 Frequency : systems with MD RAID1 external journal devices
574 Description: lconf's call to blkid is confused by RAID1 journal devices
575 Details : Use the "blkid -l" flag to locate the MD RAID device instead
576 of returning all block devices that match the journal UUID.
579 Frequency : always, for aggregate stripe size over 4GB
581 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
582 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
583 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
586 Frequency : always if "lfs find" used on a local file/directory
588 Description: "lfs find" segfaults if used on a local file/directory
589 Details : The case where a directory component was not specified wasn't
590 handled correctly. Handle this properly.
593 Frequency : always on ppc64
595 Description: the write to an ext3 filesystem mounted with mballoc got stuck
596 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
597 perform endianness conversion.
600 Frequency : rarely (truncate to non-zero file size after write under load)
601 Bugzilla : 10730, 10687
602 Description: Files padded with zeros to next 4K multiple
603 Details : With filesystems mounted using the "extents" option (2.6 kernels)
604 it is possible that files that are truncated to a non-zero size
605 immediately after being written are filled with zero bytes beyond
606 the truncated size. No file data is lost.
608 Severity : enhancement
610 Description: Allow recovery/failover for liblustre clients.
611 Details : liblustre clients were unaware of failover configurations until
614 Severity : enhancement
616 Description: user file locks should fail when not mounting with flock option
617 Details : Set up an error-returning stub in ll_file_operations.lock field
618 to prevent incorrect behaviour when client is mounted without
619 flock option. Also, set up properly f_op->flock field for
623 Frequency : always on ia64
625 Description: "lfs df" loops on printing out MDS statfs information
626 Details : The obd_ioctl_data was not initialized and in some systems
627 this caused a failure during the ioctl that did not return
628 an error. Initialize the struct and return an error on failure.
631 Frequency : SLES 9 only
633 Description: Error of copying files with lustre special EAs as root
634 Details : Client side always return success for setxattr call for lustre
635 special xattr (currently only "trusted.lov").
638 Frequency : rarely on clusters with both ia64+i386 clients
640 Description: ia64+i686 clients doing shared IO on the same file may LBUG
641 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
642 clients are doing concurrent writes to the same file it is
643 possible that the ia64 clients may LASSERT because the OST
644 extent locks are not PAGE_SIZE aligned. Ensure that grown
645 locks are always aligned on the request boundary.
648 Frequency : specific use, occasional
650 Description: Overwriting in use executable truncates on-disk binary image
651 Details : If one node attempts to overwrite an executable in use by
652 another node, we now correctly return ETXTBSY instead of
658 Description: chmod on Lustre root is propagated to other clients
659 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
660 invalid by the follow_mount time.
663 Frequency : liblustre clients only
665 Description: Race in 'instant cancel' lock handling could lead to such locks
666 never to be granted in case of SMP MDS
667 Details : Do not destroy not yet granted but cbpending locks in
671 Frequency : replay/resend of open
673 Description: non null lock assetion failure in mds_intent_policy
674 Details : Trying to replay/resend lockless open requests resulted in
675 mds_open() returning 0 with no lock. Now it sets a flag if
676 a lock is going to be returned.
678 Severity : enhancement
680 Description: Checksum enhancements
681 Details : New checksum enhancements allow for resending RPCs that failed
684 Severity : enhancement
686 Description: Tunables on number of dirty pages in cacche
687 Details : Allow to set limit on number of dirty pages cached.
692 Description: client crash on unmount - lock still has references
693 Details : In some error handling cases it was possible to leak a lock
694 reference on a client while accessing a file. This was not
695 harmful to the client during operation, but would cause the
696 client to crash when the filesystem is unmounted.
699 Frequency : specific case, rare
701 Description: ETXTBSY on mds though file not in use by client
702 Details : ETXTBSY is no longer incorrectly returned when attempting to
703 chmod or chown a directory that the user previously tried to
704 execute or a currently-executing binary.
707 Frequency : extremely rare except on liblustre-based clients
709 Description: Lustre space not freed when files are deleted
710 Details : Clean up open-unlinked files after client eviction. Previously
711 the unlink was skipped and the files remained as orphans.
716 Description: OST failure "would be an LBUG" in waiting_locks_callback()
717 Details : In some cases it was possible to send a blocking callback to
718 a client doing a glimpse, even though that client didn't get
719 a lock granted. When the glimpse lock is cancelled on the OST
720 the freed lock is left on the waiting list and corrupted the list.
723 Frequency : all core dumps
725 Description: Broke core dumps to lustre
726 Details : Negative dentry may be unhashed if parent does not have UPDATE
727 lock, but some callers, e.g. do_coredump, expect dentry to be
728 hashed after successful create, hash it in ll_create_it.
730 ------------------------------------------------------------------------------
732 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
734 * Support for kernels:
735 2.6.9-42.0.2.EL (RHEL 4)
737 2.4.21-40.EL (RHEL 3)
738 2.6.12.6 vanilla (kernel.org)
742 Frequency : always on RHEL 3
744 Description: Number of open files grows over time
745 Details : The number of open files grows over time, whether or not
746 Lustre is started. This was due to a filp leak introduced
747 by one of our kernel patches.
749 ------------------------------------------------------------------------------
751 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
753 * Support for kernels:
756 2.4.21-40.EL (RHEL 3)
757 2.6.12.6 vanilla (kernel.org)
762 Bugzilla : 5719, 9635, 9792, 9684
763 Description: OST (or MDS) trips assertions in (re)connection under heavy load
764 Details : If a server is under heavy load and cannot reply to new
765 connection requests before the client resends the (re)connect,
766 the connection handling code can behave badly if two service
767 threads are concurrently handing separate (re)connections from
768 the same client. Add better locking to the connection handling
769 code, and ensure that only a single connection will be processed
770 for a given client UUID, even if the lock is dropped.
772 Severity : enhancement
774 Description: add TCP zero-copy support to kernel
775 Details : Add support to the kernel TCP stack to allow zero-copy bulk
776 sends if the hardware supports scatter-gather and checksumming.
777 This allows socklnd to do client-write and server-read more
778 efficiently and reduce CPU utilization from skbuf copying.
781 Frequency : only if NFS exporting from client
783 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
784 Details : If setting attributes on a file created under NFS that had
785 never been opened it would be possible to oops the client
786 if the file had no objects.
790 Bugzilla : 9326, 10402, 10897
791 Description: client crash in ptlrpcd_wake() thread when sending async RPC
792 Details : It is possible that ptlrpcd_wake() dereferences a freed async
793 RPC. In rare cases the ptlrpcd thread alread processed the RPC
794 before ptlrpcd_wake() was called and the request was freed.
797 Frequency : always for liblustre
799 Description: liblustre client does MDS+OSTs setattr RPC for each write
800 Details : When doing a write from a liblustre client, the client
801 incorrectly issued an RPC to the MDS and each OST the file was
802 striped over in order to update the timestamps. When writing
803 with small chunks and many clients this could overwhelm the MDS
804 with RPCs. In all cases it would slow down the write because
805 these RPCs are unnecessary.
807 Severity : enhancement
809 Description: allow number of MDS service threads to be changed at module load
810 Details : It is now possible to change the number of MDS service threads
811 running. Adding "options mds mds_num_threads={N}" to the MDS's
812 /etc/modprobe.conf will set the number of threads for the next
813 time Lustre is restarted (assuming the "mds" module is also
814 reloaded at that time). The default number of threads will
815 stay the same, 32 for most systems.
820 Description: OST crash if filesystem is unformatted or corrupt
821 Details : If an OST is started on a device that has never been formatted
822 or if the filesystem is corrupt and cannot even mount then the
823 error handling cleanup routines would dereference a NULL pointer.
828 Description: NULL pointer deref in llap_from_page.
829 Details : get_cache_page_nowait can return a page with NULL (or otherwise
830 incorrect) mapping if the page was truncated/reclaimed while it was
831 searched for. Check for this condition and skip such pages when
832 doing readahead. Introduce extra check to llap_from_page() to
833 verify page->mapping->host is non-NULL (so page is not anonymous).
836 Frequency : Sometimes when using sys_sendfile
838 Description: "page not covered by a lock" warnings from ll_readpage
839 Details : sendfile called ll_readpage without right page locks present.
840 Now we introduced ll_file_sendfile that does necessary locking
841 around call to generic_file_sendfile() much like we do in
845 Frequency : with certain MDS communication failures at client mount time
847 Description: NULL pointer deref after failed client mount
848 Details : a client connection request may delayed by the network layer
849 and not be sent until after the PTLRPC layer has timed out the
850 request. If the client fails the mount immediately it will try
851 to clean up before the network times out the request. Add a
852 reference from the request import to the obd device and delay
853 the cleanup until the network drops the request.
856 Frequency : occasionally during client (re)connect
858 Description: assertion failure during client (re)connect
859 Details : processing a client connection request may be delayed by the
860 client or server longer than the client connect timeout. This
861 causes the client to resend the connection request. If the
862 original connection request is replied in this interval, the
863 client may trip an assertion failure in ptlrpc_connect_interpret()
864 which thought it would be the only running connect process.
867 Frequency : only with obd_echo servers and clients that are rebooted
869 Description: kernel BUG accessing uninitialized data structure
870 Details : When running an obd_echo server it did not start the ping_evictor
871 thread, and when a client was evicted an uninitialized data
872 structure was accessed. Start the ping_evictor in the RPC
873 service startup instead of the OBD startup.
875 Severity : enhancement
876 Bugzilla : 10193 (patchless)
877 Description: Remove dependency on various unexported kernel interfaces.
878 Details : No longer need reparent_to_init, exit_mm, exit_files,
879 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
882 Frequency : rare (only users of deprecated and unsupported LDAP config)
884 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
885 Details : LDAP apparently contains 'lustreName' attributes instead of
886 'name'. A simple remapping of the name is sufficient.
889 Frequency : rare (only with non-default dump_on_timeout debug enabled)
891 Description: waiting_locks_callback trips kernel BUG if client is evicted
892 Details : Running with the dump_on_timeout debug flag turned on makes
893 it possible that the waiting_locks_callback() can try to dump
894 the Lustre kernel debug logs from an interrupt handler. Defer
895 this log dumping to the expired_lock_main() thread.
897 Severity : enhancement
899 Description: Support NFS exporting on 2.6 kernels.
900 Details : Implement non-rawops metadata methods for NFS server to use without
901 changing NFS server code.
904 Frequency : very rare (synthetic metadata workload only)
906 Description: two racing renames might cause an MDS thread to deadlock
907 Details : Running the "racer" program may cause one MDS thread to rename
908 a file from being the source of a rename to being the target of
909 a rename at exactly the same time that another thread is doing
910 so, and the second thread has already enqueued these locks after
911 doing a lookup of the target and is trying to relock them in
912 order. Ensure that we don't try to re-lock the same resource.
915 Frequency : only very large systems with liblustre clients
917 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
918 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
919 all OSTs in parallel. This allows the eviction of stale liblustre
920 clients to proceed much faster than if they were done in series,
921 and also offers similar improvements for other set_info RPCs.
926 Description: excessive CPU usage during initial read phase on client
927 Details : During the initial read phase on a client, it would agressively
928 retry readahead on the file, consuming too much CPU and impacting
929 performance (since 1.4.5.8). Improve the readahead algorithm
930 to avoid this, and also improve some other common cases (read
931 of small files in particular, where "small" is files smaller than
932 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
937 Description: MDS crash when receiving packet with unknown intent.
938 Details : Do not LBUG in unknown intent case, just return -EFAULT
940 Severity : enhancement
941 Bugzilla : 9293, 9385
942 Description: MDS RPCs are serialised on client. This is unnecessary for some.
943 Details : Do not serialize getattr (non-intent version) and statfs.
946 Frequency : occasional, when OST network is overloaded/intermittent
948 Description: client evicted by OST after bulk IO timeout
949 Details : If a client sends a bulk IO request (read or write) the OST
950 may evict the client if it is unresposive to its data GET/PUT
951 request. This is incorrect if the network is overloaded (takes
952 too long to transfer the RPC data) or dropped the OST GET/PUT
953 request. There is no need to evict the client at all, since
954 the pinger and/or lock callbacks will handle this, and the
955 client can restart the bulk request.
958 Frequency : Always when mmapping file with no objects
960 Description: client crashes when mmapping file with no objects
961 Details : Check that we actually have objects in a file before doing any
962 operations on objects in ll_vm_open, ll_vm_close and
968 Description: Request leak when working with deleted CWD
969 Details : Introduce advanced request refcount tracking for requests
970 referenced from lustre intent.
972 Severity : Enhancement
974 Description: Cache open file handles on client.
975 Details : MDS now will return special lock along with openhandle, if
976 requested and client is allowed to hold openhandle, even if unused,
977 until such a lock is revoked. Helps NFS a lot, since NFS is opening
978 closing files for every read/write openration.
980 Severity : Enhancement
982 Description: Cache open negative dentries on client when possible.
983 Details : Guard negative dentries with UPDATE lock on parent dir, drop
984 negative dentries on lock revocation.
989 Description: Remounting a client read-only wasn't possible with a zconf mount
990 Details : It wasn't possible to remount a client read-only with llmount.
992 Severity : enhancement
993 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
994 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
995 MPICH is included to add Lustre-specific ADIO interfaces.
996 This is based closely on the UFS ADIO layer and only differs
997 in file creation, in order to allow the OST striping to be set.
998 This is user-contributed code and not supported by CFS.
1003 Description: extended inode attributes (immutable, append-only) work improperly
1004 when 2.4 and 2.6 kernels are used on client/server or vice versa
1005 Details : Introduce kernel-independent values for these flags.
1007 Severity : enhancement
1010 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1011 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1012 granularity. Now, Fractional MB granularity is be supported,
1013 this is very useful for low memory system.
1015 Severity : enhancement
1017 Description: Getattr by fid
1018 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1019 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1020 have locks and avoids getattr rpc requests for it on every stat.
1023 Frequency : Always, for filesystems larger than 2TB
1025 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1026 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1027 filesystems larger than 2TB to be mounted with mballoc enabled.
1030 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1032 Description: filesystem corruption for non-standard kernels and very large OSTs
1033 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1034 filesystems larger than 2TB is mounted then the kernel will
1035 silently corrupt the start of the filesystem. CONFIG_LBD is
1036 enabled for all CFS-supported kernels, but the possibility of
1037 this happening with a modified kernel config exists.
1039 Severity : enhancement
1041 Description: add client O_DIRECT support for 2.6 kernels
1042 Details : It is now possible to do O_DIRECT reads and writes to files
1043 in the Lustre client mountpoint on 2.6 kernel clients.
1045 Severity : enhancement
1047 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1048 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1049 parallel, not waiting for response from every OST before sending
1050 a rpc to the next OST.
1055 Description: setattr vs write race when updating file timestamps
1056 Details : Client processes that update a file timestamp into the past
1057 right after writing to the file (e.g. tar) it is possible that
1058 the updated file modification time can be reset to the current
1059 time due to a race between processing the setattr and write RPC.
1061 Severity : enhancement
1063 Description: Bring 'lfs find' closer in line with regular Linux find.
1064 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1065 -print0 options and obtains all the needed info through the lustre
1068 Severity : enhancement
1070 Description: support up to 1024 configured devices on one node
1071 Details : change obd_dev array from statically allocated to dynamically
1072 allocated structs as they are first used to reduce memory usage
1077 Description: Flush dirty partially truncated pages during truncate
1078 Details : Immediatelly flush partially truncated pages in filter_setattr,
1079 this way we completely avoid having any pages in page cache on OST
1080 and can retire ugly workarounds during writes to flush such pages.
1085 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1086 Details : protect inode from truncation within vfs_unlink() context
1087 just take a reference before calling vfs_unlink() and release it
1088 when parent's i_sem is free.
1091 Frequency : always, if extents are used on OSTs
1093 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1094 Details : OSTs using the extents format would not zero the high 16 bits of
1095 the index physical block number. This is not a problem for any
1096 OST filesystems smaller than 16TB, and no kernels support ext3
1097 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1098 new/modified files) and can be fixed for existing filesystems
1099 with e2fsprogs-1.39-cfs1.
1104 Description: import connection selection may be incorrect if timer wraps
1105 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1106 import connections to be ignored if the 32-bit jiffies counter
1107 wraps. Use a 64-bit jiffies counter.
1110 Frequency : during server recovery
1112 Description: crash after server is denying duplicate export
1113 Details : If clients are resending connect requests to the server, the
1114 server refuses to allow a client to connect multiple times.
1115 Fixed a bug in the handling of this case.
1118 Frequency : very large clusters immediately after boot
1120 Description: LNET request buffers exhausted under heavy short-term load
1121 Details : If a large number of client requests are generated on a service
1122 that has previously never seen so many requests it is possible
1123 that the request buffer growth cannot keep up with the spike in
1124 demand. Instead of dropping incoming requests, they are held in
1125 the LND until the RPC service can accept more requests.
1128 Frequency : Sometimes during replay
1130 Description: Assertion failure in ll_local_open after replay.
1131 Details : If replay happened on an open request reply before we were able
1132 to set replay handler, reply will become not swabbed tripping the
1133 assertion in ll_local_open. Now we set the handler right after
1134 recognising of open request
1137 Frequency : very rare
1139 Description: kernel reports "badness in vsnprintf"
1140 Details : Reading from the "recovery_status" /proc file in small chunks
1141 may cause a negative length in lprocfs_obd_rd_recovery_status()
1142 call to vsnprintf() (which is otherwise harmless). Exit early
1143 if there is no more space in the output buffer.
1145 Severity : enhancement
1147 Description: clear OBD RPC statistics by writing to them
1148 Details : It is now possible to clear the OBD RPC statistics by writing
1149 to the "stats" file.
1154 Description: Client mtime is not the same on different clients after utimes
1155 Details : In some cases, the client was using the utimes() syscall on
1156 a file cached on another node. The clients now validate the
1157 ctime from the MDS + OSTs to determine which one is right.
1162 Description: Inability to activate failout mode
1163 Details : lconf script incorrectly assumed that in python string's numeric
1164 value is used in comparisons.
1167 Frequency : always with multiple stripes per file
1169 Description: Inefficient object allocation for mutli-stripe files
1170 Details : When selecting which OSTs to stripe files over, for files with
1171 a stripe count that divides evenly into the number of OSTs,
1172 the MDS is always picking the same starting OST for each file.
1173 Return the OST selection heuristic to the original design.
1178 Description: mount failures may take full timeout to return an error
1179 Details : Under some heavy load conditions it is possible that a
1180 failed mount can wait for the full obd_timeout interval,
1181 possibly several minutes, before reporting an error.
1182 Instead return an error as soon as the status is known.
1184 ------------------------------------------------------------------------------
1186 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
1188 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1189 INTEROPERATE with older versions automatically. Please read the
1190 user documentation before upgrading any part of a live system.
1191 * WARNING: Lustre networking configuration changes are required with
1192 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1195 * Support for kernels:
1196 2.6.9-22.0.2.EL (RHEL 4)
1197 2.6.5-7.244 (SLES 9)
1198 2.6.12.6 vanilla (kernel.org)
1201 Severity : enhancement
1202 Bugzilla : 7981/8208
1203 Description: Introduced Lustre Networking (LNET)
1204 Details : LNET is new networking infrastructure for Lustre, it includes
1205 a reorganized network configuration mode (see the user
1206 documentation for full details) as well as support for routing
1207 between different network fabrics. Lustre Networking Devices
1208 (LNDs) for the supported network fabrics have also been
1209 created for this new infrastructure.
1211 Severity : enhancement
1212 Description: Introduced Access control lists
1213 Details : clients can set ACLs on files and directories in order to have
1214 more fine-grained permissions than the standard Unix UGO+RWX.
1215 The MDS must be started with the "-o acl" mount option.
1217 Severity : enhancement
1218 Description: Introduced filesystem quotas
1219 Details : Administrators may now establish per-user quotas on the
1222 Severity : enhancement
1224 Description: Configuration change for the XT3
1225 The PTLLND is now used to run Lustre over Portals on the XT3
1226 The configure option(s) --with-cray-portals are no longer used.
1227 Rather --with-portals=<path-to-portals-includes> is used to
1228 enable building on the XT3. In addition to enable XT3 specific
1229 features the option --enable-cray-xt3 must be used.
1234 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1235 Details : A race between allocating a new llcd and re-getting the llcd_lock
1236 allowed another thread to grab newly-allocated llcd.
1238 Severity : enhancement
1240 Description: 2.6 OST async journal commit and locking fix to improve performance
1241 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1242 2.6 kernels are now async as they already were in 2.4 kernels so
1243 that they can commit concurrently with the network bulk transfer.
1244 For block-allocated files the filter allocation semaphore is held
1245 to avoid filesystem fragmentation during allocation. BKL lock
1246 removed for 2.6 xattr operations where it is no longer needed.
1251 Description: lconf incorrectly determined whether two IP networks could talk
1252 Details : In some more complicated routing and multiple-network
1253 configurations, lconf will avoid trying to make a network
1254 connection to a disjoint part of the IP space. It was doing the
1255 math incorrectly for one set of cases.
1260 Description: Fix for potential infinite loop processing records in an llog.
1261 Details : If an llog record is corrupted/zeroed, it is possible to loop
1262 forever in llog_process(). Validate the llog record length
1263 and skip the remainder of the block on error.
1266 Frequency : occasional (liblustre only)
1268 Description: liblustre could not open files whose last component is a symlink
1269 Details : sysio_path_walk() would incorrectly pass the open intent to
1270 intermediate path components.
1273 Frequency : rare (liblustre only with non-standard tuning)
1274 Bugzilla : 7201 (7350)
1275 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1276 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1277 after it has been released, only do this for referenced locks
1279 Severity : enhancement
1281 Description: specifying an (invalid) directory default stripe_size of -1
1282 would reset the directory default striping
1283 Details : stripe_size -1 was used internally to signal directory stripe
1284 removal, now use "all default" to signal dir stripe removal
1285 as a directory striping of "all default" is not useful
1288 Frequency : common for large clusters running liblustre clients
1290 Description: doing an ls when liblustre clients are running is slow
1291 Details : sending a glimpse AST to a liblustre client waits for every AST
1292 to time out, as liblustre clients will not respond. Since they
1293 cannot cache data we refresh the OST lock LVB from disk instead.
1295 Severity : enhancement
1297 Description: doing an ls at the same time as file IO can be slow
1298 Details : enqueue and other "small" requests can be blocked behind many
1299 large IO requests. Create a new OST IO portal for non-IO
1300 requests so they can be processed faster.
1303 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1305 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1306 created file with mode 000
1307 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1308 allow an ftruncate() as a non-root user to a file with mode 000.
1309 Lustre now respects this flag to disable mode checks when
1310 truncating a file owned by the user
1313 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1316 Description: Revoking locks from clients that went dead or catatonic might take
1318 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1319 cancellation of such locks instant on servers without waiting for
1320 any reply from clients. Clients drops these locks when cancel
1321 notification from server is received without replying.
1324 Frequency : liblustre-only, when liblustre client dies or becomes busy
1326 Description: Doing ls on Linux clients can take a long time with active
1328 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1329 granting such locks to it in the first place if possible. Locks
1330 are taken by proxy on the OST during the read or write and
1331 dropped immediately afterward. Add connect flags handling, do
1332 not grant locks to liblustre clients for glimpse ASTs.
1334 Severity : enhancement
1336 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1337 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1338 sequential reads and behaves badly with applications doing random
1339 reads. Improve it by reading ahead at least read region, and
1340 avoiding excessive large RPC for small reads.
1342 Severity : enhancement
1344 Description: Creating more than 1000 files for a single job may cause a load
1345 imbalance on the OSTs if there are also a large number of OSTs.
1346 Details : qos_prep_create() uses an OST index reseed value that is an
1347 even multiple of the number of available OSTs so that if the
1348 reseed happens in the middle of the object allocation it will
1349 still utilize the OSTs as uniformly as possible.
1354 Description: OST or MDS may oops in ping_evictor_main()
1355 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1356 but doesn't restart at beginning of obd_exports_timed list
1359 Severity : enhancement
1361 Description: improve by-nid export eviction on the MDS and OST
1362 Details : allow multiple exports with the same NID to be evicted at one
1363 time without re-searching the exports list.
1366 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1368 Description: MDS may oops in groups_free()
1369 Details : in rare race conditions a newly allocated group_info struct is
1370 freed again, and this can be NULL. The 2.4 compatibility code
1371 for groups_free() checked for a NULL pointer, but 2.6 did not.
1374 Frequency : common for liblustre clients doing little filesystem IO
1375 Bugzilla : 9352, 7313
1376 Description: server may evict liblustre clients accessing contended locks
1377 Details : if a client is granted a lock or receives a completion AST
1378 with a blocking AST already set it would not reply to the AST
1379 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1382 Frequency : lfs setstripe, only systems with more than 160 OSTs
1384 Description: unable to set striping with a starting offset beyond OST 160
1385 Details : llapi_create_file() incorrectly limited the starting stripe
1386 index to the maximum single-file stripe count.
1389 Frequency : LDAP users only
1391 Description: lconf did not handle in-kernel recovery with LDAP properly
1392 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1394 Severity : enhancement
1396 Description: bind OST threads to NUMA nodes to improve performance
1397 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1398 node and do their allocations there to localize memory access
1400 Severity : enhancement
1402 Description: llmount can determine client NID directly from Myrinet (GM)
1403 Details : the client NID code from gmnalnid was moved directly into
1404 llmount, removing the need to use this or specifying the
1405 client NID explicitly when mounting GM clients with zeroconf
1408 Frequency : if client is started with down MDS
1410 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1411 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1412 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1413 ptlrpc_import_delayed_req() because we didn't check if the
1414 request was interrupted, and we also didn't break out of the
1415 event loop if there was no timeout
1420 Description: data loss during non-page-aligned writes to a single file from
1421 both multiple nodes and multiple threads on one node at same time
1422 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1423 inconsistency led to false short-reads, that were cached and later
1424 used by ->prepare_write() to fill in partially written page,
1425 leading to data loss.
1428 Frequency : always, if lconf --abort_recovery used
1430 Description: lconf --abort_recovery fails with 'Operation not supported'
1431 Details : lconf was attempting to abort recovery on the MDT device and not
1434 Severity : enhancement
1436 Description: remove cleanup logs
1437 Details : replace lconf-generated cleanup logs with lustre internal
1438 cleanup routines. Eliminates the need for client-cleanup and
1441 Severity : enhancement
1443 Description: add support for EAs (user and system) on lustre filesystems
1444 Details : it is now possible to store extended attributes in the Lustre
1445 client filesystem, and with the user_xattr mount option it
1446 is possible to allow users to store EAs on their files also
1448 Severity : enhancement
1450 Description: Add possibility (config option) to show minimal available OST free
1452 Details : When compiled with --enable-mindf configure option, statfs(2)
1453 (and so, df) will return least minimal free space available from
1454 all OSTs as amount of free space on FS, instead of summary of
1455 free spaces of all OSTs.
1457 Severity : enhancement
1459 Description: do not expand extent locks acquired on OST-side
1460 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1461 by server: they are not cached anyway.
1464 Frequency : when mmap is used/binaries executed from Lustre
1466 Description: Unmmap pages before throwing them away from read cache.
1467 Details : llap_shrink cache now attempts to unmap pages before discarding
1468 them (if unmapping failed - do not discard). SLES9 kernel has
1469 extra checks that trigger if this unmapping is not done first.
1474 Description: lconf didn't resolve symlinks before checking to see whether a
1475 given mountpoint was already in use
1478 Frequency : when migrating failover services
1479 Bugzilla : 6395, 9514
1480 Description: When migrating a subset of services from a node (e.g. failback
1481 from a failover service node) the remaining services would
1482 time out and evict clients.
1483 Details : lconf --force (implied by --failover) sets the global obd_timeout
1484 to 5 seconds in order to quickly disconnect, but this caused
1485 other RPCs to time out too quickly. Do not change the global
1486 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1488 Severity : enhancement
1489 Frequency : if MDS is started with down OST
1490 Bugzilla : 9439,5706
1491 Description: Allow startup/shutdown of an MDS without depending on the
1492 availability of the OSTs.
1493 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1494 Add appropriate locking and lov-osc refcounts for safe
1495 cleaning. Add osc abort_inflight calls in case the
1496 synchronize never started.
1499 Frequency : occasional (Cray XT3 only)
1501 Description: root not authorized to access files in CRAY_PORTALS environment
1502 Details : The client process capabilities were not honoured on the MDS in
1503 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1504 been accessed by an authorized user then root was able to access
1505 the file on the local client also. The root user capabilities
1506 are now allowed on the MDS, as this environment has secure UID.
1509 Frequency : occasional
1511 Description: ldiskfs "too long searching" message happens too often
1512 Details : A debugging message (otherwise harmless) prints too often on
1513 the OST console. This has been reduced to only happen when
1514 there are fragmentation problems on the filesystem.
1519 Description: Division by zero in statfs when all OSCs are inactive
1520 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
1521 lov_statfs divides by value returned by lov_get_stripecnt().
1525 Bugzilla : 9489, 3273
1526 Description: First write from each client to each OST was only 4kB in size,
1527 to initialize client writeback cache, which caused sub-optimal
1528 RPCs and poor layout on disk for the first writen file.
1529 Details : Clients now request an initial cache grant at (re)connect time
1530 and so that they can start streaming writes to the cache right
1531 away and always do full-sized RPCs if there is enough data.
1532 If the OST is rebooted the client also re-establishes its grant
1533 so that client cached writes will be honoured under the grant.
1538 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
1539 Details : Now I/O RPCs go to different portal number and (presumably) fast
1540 lock requests (and glimses) and other RPCs get their own service
1541 threads pool that should be able to service those RPCs
1544 Severity : enhancement
1546 Description: Ability to exchange lustre version between client and servers and
1547 issue warnings at client side if client is too old. Also for
1548 liblustre clients there is ability to refuse connection of too old
1550 Details : New 'version' field is added to connect data structure that is
1551 filled with version info. That info is later checked by server and
1555 Frequency : rare, liblustre only.
1556 Bugzilla : 9296, 9581
1557 Description: Two simultaneous writes from liblustre at offset within same page
1558 might proceed at the same time overwriting eachother with stale
1560 Details : I/O lock withing llu_file_prwv was released too early, before data
1561 actually was hitting the wire. Extended lock-holding time until
1562 server acknowledges receiving data.
1565 Frequency : extremely rare. Never observed in practice.
1567 Description: avoid generating lustre_handle cookie of 0.
1568 Details : class_handle_hash() generates handle cookies by incrementing
1569 global counter, and can hit 0 occasionaly (this is unlikely, but
1570 not impossible, because initial value of cookie counter is
1571 selected randonly). Value of 0 is used as a sentinel meaning
1572 "unassigned handle" --- avoid it. Also coalesce two critical
1573 sections in this function into one.
1575 Severity : enhancement
1577 Description: allow liblustre clients to delegate truncate locking to OST
1578 Details : To avoid overhead of locking, liblustre client instructs OST to
1579 take extent lock in ost_punch() on client's behalf. New connection
1580 flag is added to handle backward compatibility.
1582 Severity : enhancement
1583 Bugzilla : 4928, 7341, 9758
1584 Description: allow number of OST service threads to be specified
1585 Details : a module parameter allows the number of OST service threads
1586 to be specified via "options ost ost_num_threads={N}" in the
1587 OSS's /etc/modules.conf or /etc/modprobe.conf.
1591 Bugzilla : 6146, 9635, 9895
1592 Description: servers crash with bad pointer in target_handle_connect()
1593 Details : In rare cases when a client is reconnecting it was possible that
1594 the connection request was the last reference for that export.
1595 We would temporarily drop the export reference and get a new
1596 one, but this may have been the last reference and the export
1597 was just destroyed. Get new reference before dropping old one.
1599 Severity : enhancement
1600 Frequency : if client is started with failover MDS
1602 Description: Allow multiple MDS hostnames in the mount command
1603 Details : Try to read the configuration from all specified MDS
1604 hostnames during a client mount in case the "primary"
1607 Severity : enhancement
1609 Description: Stop sending data to evicted clients as soon as possible.
1610 Details : Check if the client we are about to send or are sending data to
1611 was evicted already. (Check is done every second of waiting,
1612 for which l_wait_event interface was extended to allow checking
1613 of exit condition at specified intervals).
1616 Frequency : rare, normally only when NFS exporting is done from client
1618 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
1620 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
1623 Frequency : always, if srand() is called before liblustre initialization
1625 Description: Liblustre uses system PRNG disturbing its usage by user application
1626 Details : Introduce internal to lustre fast and high-quality PRNG for
1627 lustre usage and make liblustre and some other places in generic
1628 lustre code to use it.
1630 Severity : enhancement
1631 Bugzilla : 9477, 9557, 9870
1632 Description: Verify that the MDS configuration logs are updated when xml is
1633 Details : Check if the .xml configuration logs are newer than the config
1634 logs stored on the MDS and report an error if this is the case.
1635 Request --write-conf, or allow starting with --old_conf.
1637 Severity : enhancement
1639 Description: Handle symlinks in the path when checking if Lustre is mounted.
1640 Details : Resolve intermediate symlinks when checking if a client has
1641 mounted a filesystem to avoid duplicate client mounts.
1646 Description: lconf can hit an error exception but still return success.
1647 Details : The lconf command catches the Command error exception at the top
1648 level script context and will exit with the associated exit
1649 status, but doesn't ensure that this exit status is non-zero.
1654 Description: failure of ptlrpc thread startup can cause oops
1655 Details : Starting a ptlrpc service thread can fail if there are a large
1656 number of threads or the server memory is very fragmented.
1657 Handle this without oopsing.
1660 Frequency : always, only if liblustre and non-default acceptor port was used
1662 Description: liblustre cannot connect to servers with non-default acceptor port
1663 Details : tcpnal_set_default_params() was not called and was therefore
1664 ignoring the environment varaible TCPNAL_PORT, as well as other
1665 TCPNAL_ environment variables
1670 Description: two objects could be created on the same OST for a single file
1671 Details : If an OST is down, in some cases it was possible to create two
1672 objects on a single OST for a single file. No problems other
1673 than potential performance impact and spurious error messages.
1677 Bugzilla : 5681, 9562
1678 Description: Client may oops in ll_unhash_aliases
1679 Details : Client dcache may become inconsistent in race condition.
1680 In some cases "getcwd" can fail if the current directory is
1686 Description: Inode refcounting problems in NFS export code
1687 Details : link_raw functions used to call d_instantiate without obtaining
1688 extra inode reference first.
1692 Bugzilla : 9942, 9903
1693 Description: Referencing freed requests leading to crash, memleaks with NFS.
1694 Details : We used to require that call to ll_revalidate_it was always
1695 followed by ll_lookup_it. Also with revalidate_special() it is
1696 possible to call ll_revalidate_it() twice for the same dentry
1697 even if first occurence returned success. This fix changes semantic
1698 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
1699 reference on a request referred from the intent.
1700 ll_intent_release() then releases such a request.
1703 Frequency : rare, normally benchmark loads only
1705 Description: unlinked inodes were kept in memory on the client
1706 Details : If a client is repeatedly creating and unlinking files it
1707 can accumulate a lot of stale inodes in the inode slab cache.
1708 If there is no other client load running this can cause the
1709 client node to run out of memory. Instead flush old inodes
1710 from client cache that have the same inode number as a new inode.
1713 Frequency : SLES9 2.6.5 kernel and long filenames only
1714 Bugzilla : 9969, 10379
1715 Description: utime reports stale NFS file handle
1716 Details : SLES9 uses out-of-dentry names in some cases, which confused
1717 the lustre dentry revalidation. Change it to always use the
1721 Frequency : rare, unless heavy write-truncate concurrency is continuous
1722 Bugzilla : 4180, 6984, 7171, 9963, 9331
1723 Description: OST becomes very slow and/or deadlocked during object unlink
1724 Details : filter_destroy() was holding onto the parent directory lock
1725 while truncating+unlinking objects. For very large objects this
1726 may block other threads for a long time and slow overall OST
1727 responsiveness. It may also be possible to get a lock ordering
1728 deadlock in this case, or run out of journal credits because of
1729 the combined truncate+unlink. Solution is to do object truncate
1730 first in one transaction without parent lock, and then do the
1731 final unlink in a new transaction with the parent lock. This
1732 reduces the lock hold time dramatically.
1735 Frequency : rare, 2.4 kernels only
1737 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
1738 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
1739 before unmounting their filesystem in order to avoid deadlock.
1740 The kernel_locked() function in 2.4 kernels only checks whether
1741 the kernel lock is held, not whether it is this process that is
1742 holding it as 2.6 kernels do.
1747 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
1748 Details : The client ptlrpc code may be trying to reconnect to a down
1749 server before a previous connection attempt has timed out.
1750 Increase the reconnect interval to be longer than the connection
1751 timeout interval to avoid sending duplicate connections to
1755 Frequency : echo_client brw_test command
1757 Description: fix echo_client to work with OST preallocated code
1758 Details : OST preallocation code (5137) didn't take echo_client IO path
1759 into account: echo_client calls filter methods outside of any
1760 OST thread and, hence, there is no per-thread preallocated
1761 pages and buffers to use. Solution: hijack pga pages for IO. As
1762 a byproduct, this avoids unnecessary data copying.
1766 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
1767 Description: Client can oops in mdc_commit_close() after open replay
1768 Details : It was possible for the MDS to return an open request with no
1769 transaction number in mds_finish_transno() if the client was
1770 evicted, but without actually returning an error. Clients
1771 would later try to replay that open and may trip an assertion
1772 Simplify the client close codepath, and always return an error
1773 from the MDS in case the open is not successful.
1776 Frequency : rare, 2.6 OSTs only
1778 Description: OST may deadlock under high load on fragmented files
1779 Details : If there was a heavy load and highly-fragmented OST filesystems
1780 it was possible to have all the OST threads deadlock waiting on
1781 allocation of biovecs, because the biovecs were not released
1782 until the entire RPC IO was completed. Instead, release biovecs
1783 as soon as they are complete to ensure forward IO progress.
1785 Severity : enhancement
1787 Description: Support for specifying external journal device at mount
1788 Details : If an OST or MDS device is formatted with an external journal
1789 device, this device major/minor is stored in the ext3 superblock
1790 and may not be valid for failover. Allow detecting and
1791 specifying the external journal at mount time.
1796 Description: Mounting an MDS with pending unlinked files may cause oops
1797 Details : target_finish_recovery() calls mds_postrecov() which returned
1798 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
1799 considers this an error and immediately begins cleaning up the
1800 lov, just after starting the mds_lov process
1802 Severity : enhancement
1804 Description: Implement 'lfs df' to report actual free space on per-OST basis
1805 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
1806 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
1807 usage of objects; '-h' to report in human readable format.
1809 ------------------------------------------------------------------------------
1811 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
1818 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
1819 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
1820 at startup during orphan unlinking. Instead, skip buddy update
1821 before setup, as it will be regenerated after recovery is complete.
1826 Description: If an OST is inactive, its locks might reference stale inodes.
1827 Details : lov_change_cbdata() must iterate over all namespaces, even if
1828 they are inactive to clear inode references from the lock.
1830 Severity : enhancement
1831 Frequency : occasional, if non-standard max_dirty_mb used
1833 Description: Client will block write RPCs if not enough grant
1834 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
1835 then the client will block writes while waiting for another RPC
1836 to complete instead of consuming its dirty limit. With change
1837 we get improved performance when max_dirty_mb is small.
1839 Severity : enhancement
1840 Bugzilla : 3389, 6253
1841 Description: Add support for supplementary groups on the MDS.
1842 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
1843 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
1844 lookups for user supplementary groups into a cache.
1848 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
1849 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
1850 for liblustre O_EXCL is enforced by MDS.
1854 Description: readdir never returns NULL in liblustre.
1855 Details : Corrected llu_iop_getdirentries logic, to return offset of next
1856 dentry in struct dirent.
1860 Frequency : liblustre only, depends on application IO pattern
1861 Description: liblustre clients evicted if not contacting servers
1862 Details : Don't put liblustre clients into the ping_evictor list, so
1863 they will not be evicted by the pinger ever.
1865 Severity : enhancement
1867 Description: Add ability to evict clients by NID from MDS.
1868 Details : By echoing "nid:$NID" string into
1869 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
1870 $NID would be instantly evicted from this MDS and from all active
1871 OSTs connected to it.
1875 Description: Do not query file size twice, somewhat slowing stat(2) calls.
1876 Details : lookup_it_finish() used to query file size from OSTs that was not
1881 Description: service threads change working directory to that of init
1882 Details : Starting lustre service threads may pin the working directory
1883 of the parent thread, making that filesystem busy. Threads
1884 now change to the working directory of init to avoid this.
1888 Frequency : during shutdown only
1889 Description: shutdown with a failed MDS or OST can cause unmount to hang
1890 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
1895 Frequency : 2.6 only
1896 Description: chmod/chown may include an extra supplementary group
1897 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
1898 supplementary group and if none is specified this is used.
1901 Bugzilla : 5479 (6816)
1903 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
1904 Details : If lookup is in progress on a file that is unlinked we might try
1905 to revalidate the inode and fail in revalidate after lookup is
1906 complete and ll_file_open() enqueues the open again but
1907 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
1910 Frequency : always, if lconf --abort_recovery used
1912 Description: lconf --abort_recovery fails with 'Operation not supported'
1913 Details : lconf was attempting to abort recovery on the MDT device and not
1916 ------------------------------------------------------------------------------
1918 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
1923 Frequency : rare (only unsupported configurations with a node running as an
1925 Bugzilla : 6514, 5137
1926 Description: Mounting a Lustre file system on a node running as an OST could
1928 Details : OSTs now preallocates memory needed to write out data at
1929 startup, instead of when needed, to avoid having to
1930 allocate memory in possibly low memory situations.
1931 Specifically, if the file system is mounted on on OST,
1932 memory pressure could force it to try to write out data,
1933 which it needed to allocate memory to do. Due to the low
1934 memory, it would be unable to do so and the node would
1935 become unresponsive.
1937 Severity : enhancement
1939 Description: Addition of lconf --service command line option
1940 Details : lconf now accepts a '--service <arg>' option, which is
1941 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
1943 Severity : enhancement
1945 Description: Failover mode is now the default for OSTs.
1946 Details : By default, OSTs will now run in failover mode. To return to
1947 the old behaviour, add '--failout' to the lmc line for OSTs.
1949 Severity : enhancement
1951 Description: Health checks are now provided for MDS and OSTs
1952 Details : Additional detailed health check information on MSD and OSTs
1953 is now provided through the procfs health_check value.
1956 Frequency : occasional, depends on IO load
1958 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
1959 after numerous create/delete cycles
1960 Details : The ext3 inode allocation policy would not allocate new inodes
1961 very well on the OSTs because there are no new directories
1962 being created. Instead we look for groups with free space if
1963 the parent directories are nearly full.
1968 Description: Network or server problems during mount may cause partially
1969 mounted clients instead of returning an error.
1970 Details : The config llog parsing code may overwrite the error return
1971 code during mount error handling, returning success instead
1977 Description: MDS can fail to allocate large reply buffers
1978 Details : After long uptimes the MDS can fail to allocate large reply
1979 buffers (e.g. zconf client mount config records) due to memory
1980 fragmentation or consumption by the buffer cache. Preallocate
1981 some large reply buffers so that these replies can be sent even
1982 under memory pressure.
1986 Frequency : rare (liblustre)
1987 Description: fsx running with liblustre complained that using truncate() to
1988 extend the file doesn't work. This patch corrects that issue.
1989 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
1990 ATTR_SIZE and lsm use in llu_setattr_raw.
1994 Frequency : rare, only 2.6 kernels
1995 Description: Unusual file access patterns on the MDS may result in inode
1996 data being lost in very rare circumstances.
1997 Details : Bad interaction between the ea-in-inode patch and the "no-read"
1998 code in the 2.6 kernel caused the inode and/or EA data not to
1999 be read from disk, causing single-file corruption.
2003 Frequency : rare, only 2.6 filesystems using extents
2004 Description: Heavy concurrent write and delete load may cause data corruption.
2005 Details : It was possible under high-load situations to have an extent
2006 metadata block in the block device cache from a just-unlinked
2007 file overwrite a newly-allocated data block. We now unmap any
2008 metadata buffers that alias just-allocated data blocks.
2012 Frequency : filesystems with default stripe_count larger than 77
2013 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2014 Details : lconf specifies an inode size of 4096 bytes when the default
2015 stripe_count is larger than 77. This conflicts with the default
2016 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2017 this case to avoid pinning too much memory for large EAs.
2019 ------------------------------------------------------------------------------
2021 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2026 Frequency : rare (extremely heavy IO load with hundreds of clients)
2028 Description: Client is evicted, gets IO error writing to file
2029 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2030 caused clients to be evicted for AST timeouts. The fixes in
2031 bug 5192 mean we no longer need to have such short AST timeouts
2032 so ldlm_timeout has been increased.
2035 Frequency : occasional during --force or --failover shutdown under load
2036 Bugzilla : 5949, 4834
2037 Description: Server oops/LBUG if stopped with --force or --failover under load
2038 Details : a collection of import/export refcount and cleanup ordering
2039 issues fixed for safer force cleanup
2042 Frequency : only filesystems larger than 120 OSTs
2043 Bugzilla : 5990, 6223
2044 Description: lfs getstripe would oops on a very large filesystem
2045 Details : lov_getconfig used kfree on vmalloc'd memory
2048 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2049 Bugzilla : 6242, 6243
2050 Description: reading from files that had been truncated to a non-zero size
2051 but never opened returned no data
2052 Details : ll_file_read() reads zeros from no-object files to EOF
2057 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2058 Details : There's little harm in aborting MDS/OSS recovery and letting it
2059 try again, so I removed the LASSERT and return an error instead.
2061 Severity : enhancement
2063 Description: New debugging infrastructure for tracking down data corruption
2064 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2065 (b) cover more of the client-side code path, and (c) try to narrow
2066 down where problems occurred
2070 Bugzilla : 3819, 4364, 4397, 6313
2071 Description: Racing close and eviction MDS could cause assertion in mds_close
2072 Details : It was possible to get multiple mfd references during close and
2073 client eviction, leading to one thread referencing a freed mfd.
2075 Severity: : enhancement
2076 Bugzilla : 3262, 6359
2077 Description: Attempts to reconnect to servers are now more aggressive.
2078 Details : This builds on the enhanced upcall-less recovery that was added
2079 in 1.4.2. When trying to reconnect to servers, clients will
2080 now try each server in the failover group every 10 seconds. By
2081 default, clients would previously try one server every 25 seconds.
2086 Description: After recovery, certain operations trigger a failed
2087 assertion on a client.
2088 Details : Failing over an mds, using lconf -d --failover, while a
2089 client was doing a readdir() call would cause the client to
2090 LBUG after recovery completed and the readdir() was resent.
2092 Severity : enhancement
2094 Description: Default groups are now added by lconf
2095 Details : You can now run lconf --group <servicename> without having to
2096 manually add groups with lmc.
2099 Frequency : occasional
2101 Description: Nodes with an elan id of 0 trigger a failed assertion
2104 Frequency : always when accessing e.g. tty/console device nodes
2106 Description: tty and some other devices nodes cannot be used on lustre
2107 Details : file's private_data field is used by device data and lustre
2108 values in there got lost. New field was added to struct file to
2109 store fs-specific private data.
2112 Frequency : when exporting Lustre via NFS
2114 Description: NFSD failed occasionally when looking up a path component
2115 Details : NFSD is looking up ".." which was broken in ext3 directories
2116 that had grown large enough to become hashed.
2119 Frequency : Clusters with multiple interfaces not on the same subnet
2121 Description: Nodes will repeatedly try to reconnect to an interface which it
2122 cannot reach and report an error to the log.
2123 Details : Extra peer list entries will be created by lconf with some peers
2124 unreachable. lconf now validates the peer before adding it.
2127 Frequency : Only if a default stripe is set on the filesystem root.
2129 Description: Setting a default stripe on the filesystem root prevented the
2130 filesystem from being remounted.
2131 Details : The client was sending extra request flags in the root getattr
2132 request and did not allocate a reply buffer for the dir EA.
2135 Frequency : occasional, higher if lots of files are accessed by one client
2136 Bugzilla : 6159, 6097
2137 Description: Client trips assertion regarding lsm mismatch/magic
2138 Details : While revalidating inodes the VFS looks up inodes with ifind()
2139 and in rare cases can find an inode that is being freed.
2140 The ll_test_inode() code will free the lsm during ifind()
2141 when it finds an existing inode and then the VFS later attaches
2142 this free lsm to a new inode.
2146 Bugzilla : 6422, 7030
2147 Description: MDS deadlock between mkdir and client eviction
2148 Details : Creating a new file via mkdir or mknod (starting a transaction
2149 and getting the ns lock) can deadlock with client eviction
2150 (gets ns lock and trying to finish a synchronous transaction).
2153 Frequency : occasional
2154 Description: While starting a server, the fsfilt_ext3 module could not be
2156 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2157 kernels. Previously, lconf would still use the ext3 name
2158 when trying to load modules. Now, it will correctly use
2159 ext3 on 2.4 and ldiskfs on 2.6.
2161 Severity : enhancement
2162 Description: The default stripe count has been changed to 1
2163 Details : The interpretation of the default stripe count (0, to lfs
2164 or lmc) has been changed to mean striping across a single
2165 OST, rather than all available. For general usage we have
2166 found a stripe count of 1 or 2 works best.
2168 Severity : enhancement
2169 Description: Add support for compiling against Cray portals.
2170 Details : Conditional compiling for some areas that are different
2174 Frequency : occasional
2175 Bugzilla : 6409, 6834
2176 Description: Creating files with an explicit stripe count may lead to
2177 a failed assertion on the MDS
2178 Details : If some OSTs are full or unavailable, creating files may
2179 trigger a failed assertion on the MDS. Now, Lustre will
2180 try to use other servers or return an error to the
2184 Frequency : occasional
2186 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2187 caused lock timeout during readahead (since 1.4.2).
2188 Details : Processes doing readahead might match a lock that hasn't been
2189 granted yet if there are overlapping and conflicting lock
2190 requests. The readahead process waits on ungranted lock
2191 (original lock is CBPENDING), while OST waits for that process
2192 to cancel CBPENDING read lock and eventually evicts client.
2194 Severity : enhancement
2196 Description: Initial enabling of flock support for clients
2197 Details : Implements fcntl advisory locking and file status functions.
2198 This feature is provided as an optional mount flag (default
2199 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2200 locking are implemented yet, and those that are are not guaranteed
2201 to be completely correct in production environments.
2202 mount -t lustre -o [flock|noflock] ...
2205 Frequency : occasional
2207 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2208 assertion in the ext3 JBD (journaling) layer.
2209 Details : The b_committed_data struct is protected by the big kernel lock
2210 in 2.4 kernels, serializing journal_commit_transaction() and
2211 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2212 access to this struct. In 2.6 kernels there is finer grained
2213 locking to improve SMP performance of the JBD layer.
2217 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2219 -----------------------------------------------------------------------------
2221 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2223 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2224 versions of Lustre. Please update all servers and clients to
2225 version 1.4.2 or later at the same time. You must also run
2226 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2227 to update the configuration logs.
2229 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2230 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2231 - swab llog records of type '0' so we get proper header size/idx (5861)
2232 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2233 - fix rename of one directory over another leaking an inode (5953)
2234 - avoid SetPageDirty on 2.6 (5981)
2235 - don't re-add just-being-destroyed locks to the waiting list (5653)
2236 - when creating new directories, inherit the parent's custom
2237 striping settings if present parent (3048)
2238 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2239 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2240 - don't hold client locks on temporary worklist from l_lru (5666)
2241 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2242 - automatically evict dead clients (5921)
2243 - Update file size properly in create+truncate+fstat case (6196)
2244 - Do not unhash mountpoint dentries, do not allow removal of
2246 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2247 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2248 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2249 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2250 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2251 - Correctly handle reads of files with no objects (6243)
2252 - lctl recover will also mark a device active if deactivate used (5933)
2254 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2255 - allow --write-conf on an MDS with different nettype than client (5619)
2256 - don't write config llogs to MDS for mounts not from that MDS (5617)
2257 - lconf should create multiple TCP connections from a client (5201)
2258 - init scripts are now turned off by default; run chkconfig --on
2259 lustre and chkconfig --on lustrefs to use them
2260 - upcalls are no longer needed for clients to recover to failover
2262 - add --abort-recovery option to lconf to abort recovery on device
2264 - add support for an arbitrary number of OSTs (3026)
2265 - Quota support protocol changes.
2266 - forward compatibility changes to wire structs (6007)
2267 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2268 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2269 - improved "lustre status" script
2270 - initialize blocksize for non-regular files (6062)
2271 - added --disable-server and --disable-client configure options (5782)
2272 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2273 - Vanilla 2.4.29 support
2274 - increase maximum number of obd devices to 520 (6242)
2275 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2276 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2277 SLES 9 2.6.5 kernels
2278 - limit stripes per file to 160 (the maximum EA size) (6093)
2280 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2283 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2284 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2285 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2286 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2287 - small ext3 extents cleanups and fixes (5733)
2288 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2289 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2290 - avoid needless client->OST connect, fix handle mismatch (5317)
2291 - fix DLM error path that led to out-of-sync client, long delays (5779)
2292 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2293 - fix several locking issues related to i_size (5492,5624,5654,5672)
2294 - don't move pending lock onto export if it is already evicted (5683)
2295 - fix kernel oops when creating .foo in unlinked directory (5548)
2296 - fix deadlock in obdfilter statistics vs. object create (5811)
2297 - use time_{before,after} to avoid timer jiffies wrap (5882)
2298 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2299 - Do not leak request if server was not able to process it (5154)
2300 - If mds_open unable to find parent dir, make that negative lookup(5154)
2301 - don't create new directories with extent-mapping (5909, 5936)
2303 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2304 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2305 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2306 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2307 - support for RHEL4 kernel on i686 (5773)
2308 - provide error messages when incompatible logs are encountered (5898)
2310 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2311 * version 1.4.0.10 (1.4.1 release candidate 1)
2313 - don't keep a lock reference when lock is not granted (4238)
2314 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2315 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2316 - drop import inflight refcount on signal_completed_replay error (5255)
2317 - unlock page after async write error during send (3677)
2318 - handle missing objects in filter_preprw_read properly (5265)
2319 - no transno return for symlink open, don't save no-trasno open (3440)
2320 - don't try to complete elan receive that already failed (4012)
2321 - free RPC server reply state on error (5406)
2322 - clean up thread from ptlrpc_start_thread() on error (5160)
2323 - readahead could read extra page into cache that wasn't ejected (5388)
2324 - prevent races in class_attach/setup/cleanup/detach (5260)
2325 - don't dereference de->d_inode after l_dput of de (5458)
2326 - use "int" for stripe value returned from lock_to_stripe (5544)
2327 - mballoc allocation and error-checking fixes in 2.6 (5504)
2328 - block device patches to fix I/O request sizes in 2.6 (5482)
2329 - look up hostnames for IB nals (5602)
2330 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2331 - don't start multiple acceptors for the same port (5277)
2332 - fix incorrect LASSERT in mds_getattr_name (5635)
2333 - export a proc file for general "ping" checking (5628)
2334 - fix "lfs check" to not block when the MDS is down (5628)
2336 - service request history (4965)
2337 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2338 - create an "evict_client" /proc entry on OSTs, like the MDS has
2339 - fix mount usage message, return errors per mount(8) (5168)
2340 - change grep [] to grep "[]" in tests so they work in more UMLs
2341 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2342 - remove ancient LOV_MAGIC_V0 EA support (5047)
2343 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2344 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2345 - allow readahead allocations to fail when low on memory (5383)
2346 - mmap locking landed again, after considerable improvement (2828)
2347 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2349 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2352 - send OST transaction number in read/write reply to free req (4966)
2353 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2354 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2355 - the watchdog thread now runs as interruptible (5246)
2356 - flock/lockf fixes (but it's still disabled, pending 5135)
2357 - don't use EXT3 constants in llite code (5094)
2358 - memory shortage at startup could cause assertion (5176)
2360 - reorganization of lov code
2361 - single portals codebase
2363 - add extents/mballoc support (5025)
2364 - direct I/O reads in the obdfilter (4048)
2365 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2367 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2370 - send OST transaction number in read/write reply to free req (4966)
2371 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2372 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2373 - the watchdog thread now runs as interruptible (5246)
2374 - handle missing objects in filter_preprw_read properly (5265)
2375 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2376 - drop import inflight refcount on signal_completed_replay error (5255)
2377 - unlock page after async write error during send (3677)
2378 - return original error code on reconstructed replies (3761)
2379 - no transno return for symlink open, don't save no-trasno open (3440)
2381 - add pid to ldlm debugging output (4922)
2382 - bump the watchdog timeouts -- we can't handle 30sec yet
2383 - extra debugging for orphan dentry/inode bug (5259)
2385 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2388 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2389 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2390 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2391 - fix NULL dereference in /proc/sys/portals/routes (4827)
2392 - allow failed mdc_close() operations to be interrupted (4561)
2393 - stop precreate on OST before MDS would time out on it (4778)
2394 - don't send partial-page writes before EOF from client (4410)
2395 - discard client grant for sub-page writes on large-page clients (4520)
2396 - don't free dentries not owned by NFS code, check generation (4806)
2397 - fix lsm leak if mds_create_objects() fails (4801)
2398 - limit debug_daemon file size, always print CERROR messages (4789)
2399 - use transno after validating reply (3892)
2400 - process timed out requests if import state changes (3754)
2401 - update mtime on OST during writes, return in glimpse (4829)
2402 - add mkfsoptions to LDAP (4679)
2403 - use ->max_readahead method instead of zapping global ra (5039)
2404 - don't interrupt __l_wait_event() during strace
2406 - add software watchdogs to catch hung threads quickly (4941)
2407 - make lustrefs init script start after nfs is mounted
2408 - fix CWARN/ERROR duplication (4930)
2409 - return async write errors to application if possible (2248)
2410 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2411 - print NAL number in %x format (4645)
2412 - update barely-supported suse-2.4.21-171 series (4842)
2413 - support for sles 9 %post scripts
2414 - support for building 2.6 kernel-source packages
2415 - support for sles km_* packages
2417 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2420 - ignore -ENOENT errors in osc_destroy (3639)
2421 - notify osc create thread that OSC is being cleaned up (4600)
2422 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2423 - reconstruct ost_handle() like mds_handle() (4657)
2424 - create a new thread to do import eviction to avoid deadlock (3969)
2425 - let lconf resolve symlinked-to devices (4629)
2426 - don't unlink "objects" from directory with default EA (4554)
2427 - hold socknal file ref over connect in case target is down (4394)
2428 - allow more than 32000 subdirectories in a single directory (3244)
2429 - fix blocks count for O_DIRECT writes (3751)
2430 - OST returns ENOSPC from object create when no space left (4539)
2431 - don't send truncate RPC if file size isn't changing (4410)
2432 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2433 - bind to privileged port in socknal and tcpnal (3689)
2435 - rate limit CERROR/CWARN console message to avoid overload (4519)
2436 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2437 - basic mmap support (3918)
2438 - kernel patch series update from b1_4 (4711)
2440 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2443 - avoid crash during MDS cleanup with OST shut down (2775)
2444 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2445 - don't use bad inodes on the MDS (3744)
2446 - dynamic object preallocation to improve recovery speed (4236)
2447 - don't hold spinlock over lock dumping or change debug flags (4401)
2448 - don't zero obd_dev when it is force cleaned (3651)
2449 - print grants to console if they go negative (4431)
2450 - "lctl deactivate" will stop automatic recovery attempts (3406)
2451 - look for existing locks in ldlm_handle_enqueue() (3764)
2452 - don't resolve lock handle twice in recovery avoiding race (4401)
2453 - revalidate should check working dir is a directory (4134)
2455 - don't always mark "slow" obdfilter messages as errors (4418)
2457 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2460 - don't close LustreDB during write_conf until it is done (3860)
2461 - fix typo in lconf for_each_profile (3821)
2462 - allow dumping logs from multiple threads at one time (3820)
2463 - don't allow multiple threads in OSC recovery (3812)
2464 - fix debug_size parameters (3864)
2465 - fix mds_postrecov to initialize import for llog ctxt (3121)
2466 - replace config semaphore with spinlock (3306)
2467 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2468 - don't allow enqueue to complete on a destroyed export (3822)
2469 - down write_lock before checking llog header bitmap (3825)
2470 - recover from lock replay timeout (3764)
2471 - up llog sem before sending rpc (3652)
2472 - reduce ns lock hold times when setting kms (3267)
2473 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2474 - fix NULL deref and obd_dev leak on setup error (3312)
2475 - replace some LBUG about llog ops with error handling (3841)
2476 - don't match INVALID dentries from d_lookup and spin (3784)
2477 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2478 - fix invalid assertion in ptlrpc_set_wait (3880)
2480 - add libwrap support for the TCP acceptor (3996)
2481 - add /proc/sys/portals/routes for non-root route listing (3994)
2482 - allow setting MDS UUID in .xml (2580)
2483 - print the stack of a process that LBUGs (4228)
2485 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2488 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2489 - make sure to unset replay flag from failed open requests (3440)
2490 - if default stripe count is 0, use OST count for inode size (3636)
2491 - update parent mtime/ctime on client for create/unlink (2611)
2492 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2493 - free recovery state on server during a forced cleanup (3571)
2494 - unregister_reply for resent reqs (3063)
2495 - loop back devices mounting and status check on 2.6 (3563)
2496 - fix resource-creation race that can provoke i_size == 0 (3513)
2497 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2498 - more debugging for page-accounting assertion (3746)
2499 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2500 - avoid deadlock after precreation failure (3758)
2501 - fix race and lock order deadlock in orphan handling (3450, 3750)
2502 - add validity checks when grabbing inodes from l_ast_data (3599)
2504 - add /proc/.../recovery_status to obdfilter (3428)
2505 - lightweight CDEBUG infrastructure, debug daemon (3668)
2506 - change default OSC RPC parameters to be better on small clusters
2507 - turn off OST read cache for files smaller than 32MB
2508 - install man pages and include them in rpms (3100)
2509 - add new init script for (un)mounting lustre filesystems (2593)
2510 - run chkconfig in %post for init scripts (3701)
2511 - drop scimac NAL (unmaintained)
2513 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
2516 - clean kiobufs before and after use (3485)
2517 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
2518 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
2519 - change init script to fail more clearly if not run as root (1528)
2520 - allow clients to reconnect during replay (1742)
2521 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
2522 - don't do DNS lookups on NIDs too small for IP addresses (3442)
2523 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
2524 - fix cond_resched (3554)
2525 - only evict unfinished clients after recovery (3515)
2526 - allow bulk resend, prevent data loss (3570)
2527 - dynamic ptlrpc request buffer allocation (2102)
2528 - don't allow unlinking open directory if it isn't empty (2904)
2529 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
2530 - remove extraneous obd dereference causing LASSERT failure (3334)
2531 - don't use get_cycles() when creating temp. files on the mds (3156)
2532 - hold i_sem when setting i_size in ll_extent_lock() (3564)
2533 - handle EEXIST for set-stripe, set proper directory name (3336)
2535 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
2536 - fix ksocknal_fmb_callback() error messages (2918)
2538 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
2541 - don't copy lvb into (possibly NULL) reply on error (2983)
2542 - don't deref dentry after dput, don't free lvb on error (2922)
2543 - use the kms to determine writeback rpc length (2947)
2544 - increment oti_logcookies when osc is inactive (2948)
2545 - update client's i_blocks count via lvb messages (2543)
2546 - handle intent open/close of special files properly (1557)
2547 - mount MDS with errors=remount-ro, like obdfilter (2009)
2548 - initialize lock handle to avoid ASSERT on error cleanup (3057)
2549 - don't use cancelling-locks' kms values (2947)
2550 - use highest lock extent for kms, not last one (2925)
2551 - don't dereference ERR_PTR() dentry in error handling path (3107)
2552 - fix thread race in portals_debug_dumplog() (3122)
2553 - create lprocfs device entries at setup instead of at attach (1519)
2554 - common AST error handler, don't evict client on completion race (3145)
2555 - zero nameidata in detach_mnt in 2.6 (3118)
2556 - verify d_inode after revalidate_special is valid in 2.6 (3116)
2557 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
2558 - initialize RPC timeout timer earlier for 2.6 (3219)
2559 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
2560 - print nal/nid for unknown nid (3258)
2561 - additional checks for oscc recovery before doing precreate (3284)
2562 - fix ll_extent_lock() error return code for 64-bit systems (3043)
2563 - don't crash in mdc_close for bad permissions on open (3285)
2564 - zero i_rdev for non-device files (3147)
2565 - clear page->private before handing to FS, better assertion (3119)
2566 - tune the read pipeline (3236)
2567 - fix incorrect decref of invalidated dentry (2350)
2568 - provide read-ahead stats and refine rpc in flight stats (3328)
2569 - don't hold journal transaction open across create RPC (3313)
2570 - update atime on MDS at close time (3265)
2571 - close LDAP connection when recovering to avoid server load (3315)
2572 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
2573 - don't leak open file on MDS after open resend (3325)
2574 - serialize filter_precreate and filter_destroy_precreated (3329)
2575 - loop device shouldn't call sync_dev() for nul device (3092)
2576 - clear page cache after eviction (2766)
2577 - resynchronize MDS->OST in background (2824)
2578 - refuse to mount the same filesystem twice on same mountpoint (3394)
2579 - allow llmount to create routes for mounting behind routers (3320)
2580 - push lock cancellation to blocking thread for glimpse ASTs (3409)
2581 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
2582 - fix rare problem with rename on htree directories (3417)
2584 - allow default OST striping configuration per directory (1414)
2585 - fix compilation for qswnal for 2.6 kernels (3125)
2586 - increase maximum number of MDS request buffers for large systems
2587 - change liblustreapi to be useful for external progs like lfsck (3098)
2588 - increase local configuration timeout for slow disks (3353)
2589 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
2591 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2594 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
2595 - don't overwrite extent policy data in reply if lock was blocked (2901)
2596 - drop filter export grants atomically with removal from device (2663)
2597 - del obd_self_export from work_list in class_disconnect_exports (2908)
2598 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
2599 - swab reply message in mdc_close, other PPC fixes (2464)
2600 - fix destroying of named logs (2325)
2601 - overwrite old logs when running lconf --write_conf (2264)
2602 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
2603 - fix race in target_handle_connect (2898)
2604 - mds_reint_create() should take same inode create lock (2926)
2605 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
2606 - don't close files for self_export to avoid uninitialized obd (2936)
2607 - allow MDS with the same name as client node (2939)
2608 - hold dentry reference for closed log files for unlink (2325)
2609 - reserve space for all logs during transactions (2059)
2610 - don't evict page beyond end of stripe extent (2925)
2611 - don't oops on a deleted current working directory (2399)
2612 - handle hard links to targets without a parent properly (2517)
2613 - don't dereference NULL lock when racing during eviction (2867)
2614 - don't grow lock extents when lots of conflicting locks (2919)
2616 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
2619 - account for cache space usage on clients to avoid data loss (974)
2620 - lfsck support in lustre kernel code (2349)
2621 - reduce journal credits needed for BRW writes (2370)
2622 - orphan handling to avoid losing space on client/server crashes
2623 - ptlrpcd can be blocked, stopping ALL progress (2477)
2624 - use lock value blocks to assist in proper KMS, faster stat (1021)
2625 - takes i_sem instead of DLM locks internally on obdfilter (2720)
2626 - recovery for initial connections (2355)
2627 - fixes for mds_cleanup_orphans (1934)
2628 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
2629 - block all file creations until orphan recovery completes (1901)
2630 - client remove rq_connection from request struct (2423)
2631 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
2632 - recovery timer race (2670)
2633 - mdc_close recovey bug (2532)
2634 - ptlrpc cleanup bug (2710)
2635 - mds timeout on local locks (2588)
2636 - namespace lock held during RPCs (2431)
2637 - handle interrupted sync write properly (2503)
2638 - don't try to handle a message that hasn't been replied to (2699)
2639 - client assert failure during cleanup after abort recovery (2701)
2640 - leak mdc device after failed mount (2712)
2641 - ptlrpc_check_set allows timedout requests to complete (2714)
2642 - wait for inflight reqs when ptlrpcd finishes (2710)
2643 - make sure unregistered services are removed from the srv_list
2644 - reset bulk XID's when resending them (caught by 1138 test)
2645 - unregister_bulk after timeout
2646 - fix lconf error (2694)
2647 - handle write after unfinished setstripe, stripe-only getstripe (2388)
2648 - readahead locks pages, leaves pending causing memory pressure (2673)
2649 - increase OST request buffers to 4096 on large machines (2729)
2650 - fix up permission of existing directories in simple_mkdir (2661)
2651 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
2652 - don't assign transno to errored transactions (2742)
2653 - don't delete objects on OST if given a bogus objid from MDS (2751)
2654 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
2655 - if rq_no_resend, then timeout request after recovery (2432)
2656 - fix MDS llog_logid record size, 64-bit array alignment (2733)
2657 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
2658 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
2659 - MDS recovery shouldn't delete objects that it has given out (2730)
2660 - if enqueue arrives after completion, don't clobber LVB (2819)
2661 - don't unlock pages twice when trigger_group_io returns error (2814)
2662 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
2663 - don't write pages to disk if there was an error (1450)
2664 - don't ping imports that have recovery disabled (2676)
2665 - take buffered bytes into account when balancing socknal conn (2817)
2666 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
2667 - reconnect unlink llog connection after MDS reconnects to OST (2816)
2668 - remove little-endian swabbing of llog records (1987)
2669 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
2670 - retry reposting request buffers if they fail (1191)
2671 - grow extent at grant time to avoid granting a revoked lock (2809)
2672 - lock revoke doesn't evict page if covered by a second lock (2765)
2673 - disable VM readahead to avoid reading outside lock extents (2805)
2675 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
2676 - updated kernel patches for hp-2.4.20 kernel (2681)
2678 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
2681 - fix truncated write corruption (2366)
2682 - fix for failed assertion in iopen_connect_dentry (1792,2517)
2684 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
2685 - ensure len(uuid) < 37 in lmc (1171)
2686 - fix ia64 OOPS in llog_test (2255)
2687 - zero end of page at obdfilter for partial page writes (2648)
2688 - don't leave stale dentries around after renames (bug 2428)
2689 - fix timeouts when evicting a client with a single lock held (2642)
2690 - set deadline for the initial HELLO message to drain (2634)
2691 - print out dotted-quad IP addresses in the socknal (2302)
2693 - additional debugging for MDS client eviction problem (2443)
2694 - fix mkfsoptions support for osts (2603, 2604)
2696 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
2699 - add series for the vanilla 2.6.0 kernel
2700 - add series for the vanilla 2.4.24 kernel
2701 - add series for a cray x86/64 UL kernel drop
2702 - fix xattr patches for the vanilla 2.4.19 series
2704 - generate true UUIDs in lmc (1171)
2705 - have portals stack dumping break in UML (2466)
2706 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
2707 - allocate with _NOFS in ldlm to avoid deadlock (1933)
2708 - wake callback waiting threads on client eviction (2460)
2709 - Add --ptldebug and --subsystem to lmc (1719)
2710 - update assertion to allow safe interrupt allocation
2711 - set rq_no_resend for cancel requests (2432)
2712 - recalculate ptlrpcd timeout after resend (2494)
2713 - call vfs_rmdir when removing pending directories (2368)
2714 - fix renaming a file to itself (2429)
2715 - lmc creates a default one-stripe lov (2454)
2716 - expand procfs space to handle large clusters (2326)
2717 - increase UML stack to avoid overflow
2718 - update lconf's list of debug and subsystem masks
2719 - fix lfs find --obd (2510)
2720 - /proc tunable for disabling filter read caching (2591)
2721 - stop rpm packages from altering slapd.conf (2301)
2722 - disable nagle in the socknal under 0conf (2578)
2723 - choose mds inode size based on stripe count (2572)
2724 - fix kernel-source rpm problems (2516)
2726 - add --disable-doc to avoid pdf generation (2421)
2727 - update documentation, tests, type-os, comments
2728 - avoid format warnings on ia64
2729 - remove the TOE NAL
2730 - tiny code cleanups by removing unused fields
2732 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
2735 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
2736 - remove the most blatant lies from BUILDING file (2371)
2737 - change default debug level to reasonable production setting
2738 - reduce client side cache size to reduce cache flush time
2739 - reduce max RPCs in flight to avoid unnecessary file fragmentation
2740 - make TCP zerocopy and pinger support enabled by default (2476)
2741 - sync writes completed after process exits caused crashes (2319)
2742 - maintain correct mount count on the MDS (2356)
2743 - backout 1557, because 2316 wasn't really fixed
2744 - better file I/O statistics gathering in /proc
2745 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
2746 - another kernel patch to fix zero-copy TCP function export
2747 - don't take duplicate lock when processing re-sent getattr (2420)
2748 - lctl uses obd_self_export instead of creating new conn (2353)
2749 - MDS/OST recovery case which requires object creation asserted (2425)
2750 - move lfs from /usr/sbin to /usr/bin in packages
2751 - fix race between mds_client_add and mds_client_free (2417)
2752 - use kmalloc instead of slabs in portals (2430)
2753 - don't create duplicate records when a failover MDS is present (2442)
2754 - remove unnecessary mount age check (2332)
2755 - don't remove directory inodes from locks prematurely (2451)
2756 - don't break if MDS service name is the same as hostname (2103)
2757 - fix races in client write RPC generation when cache full (2482)
2759 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
2762 - remove now-unused request->rq_obd (278)
2763 - if an allocation fails, print out how much memory we've used (1933)
2764 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
2765 - add the "configurable stack size" patch to most series files (1256)
2766 - ability to write large log records, for 100+ OST configs (2306)
2767 - fix NULL deref when filter_prep fails (2314)
2768 - fix operator precedence error in filter_sync
2769 - dynamic allocation of socknal TX descriptors (2315)
2770 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
2771 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
2772 - ensure that CWARN messages in Portals always get to the syslog
2773 - __init/__exit are not for prototype decls (ldlm_init/exit)
2774 - x86-64 compile warning fixes
2775 - fix gateway LMC keyword conflict (2318)
2776 - fix MDS lock inversions in getattr/reint paths (1844)
2777 - fix a rare lock re-ordering bug, which caused deadlock (2322)
2778 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
2779 - DLM race condition prevented some lock evictions (2328)
2780 - ENOMEM detection and retry on socknal sends (2230)
2781 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
2782 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
2783 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
2784 - better error messages when a client is rejected during recovery (1505)
2785 - avoid cancelling locks which were never granted, after failure (2330)
2786 - fix i_sem/journal inversion in mds_client_add (2333)
2787 - fix truncate/getattr lock cycle deadlock (2334)
2788 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
2789 - fix two rare exit paths which could leak an l_lock() ref (2321)
2790 - fencepost error in MDS/OST orphan recovery (2226)
2791 - make log record alignment 8 bytes (1988)
2792 - lstripe now fails when requested offset > ost_count (2237)
2793 - ensure that all kernel series have a complete list.h (1607)
2794 - fix crashes in special-file operations (2316)
2795 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
2797 - allow configurable automake binary, for testing new versions
2798 - small update to the lfs documentation
2800 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
2802 * fix negative export reference count in fsfilt_sync (2312)
2804 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
2805 * release candidate 0.9.1
2807 - orphans are moved into the PENDING directory for possible recovery
2808 - replayed opens now open by fid for orphan/rename safety (1042)
2809 - last close of an orphan inode generates a transno (683)
2810 - chdir() and mount() now pin the directory entry (1020)
2811 - avoid CERROR in normal ll_setattr_raw() error case (1500)
2812 - discard very old requests without processing them (1502)
2813 - remove some common, well-understood CERRORs (1505)
2814 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
2815 - clear "grant" flags in OST replies until OST grant code lands (1644)
2816 - fix read performance by not clobbering i_blksize on client (1598)
2817 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
2818 - make LRU size a /proc tunable, clears locks when reduced (707)
2819 - fix some lprocfs rot that prevented ptlbd from loading (1732)
2820 - server locks take references on exports now (1558)
2821 - build fixes for 2.4.20-rh trees (1663)
2822 - return an error from lov_create if all OSCs are inactive (1751)
2823 - fix import levels when a reconnect happens without a timeout (1597)
2824 - exit early from mds_open if we get a lookup error (1749)
2825 - partial page read at EOF wouldn't wait for disk before sending (1642)
2826 - avoid NULL deref in obdfilter when reading page past EOF (1592)
2827 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
2828 - fix LBUG in ll_it_open_error with rc = -2 (1861)
2829 - write/truncate lock inversion (1639)
2830 - Don't auto-load obdclass, portals modules during cleanup (1495)
2831 - fix timestamps from jumping to "now" (1763)
2832 - extra journal assertions (1648)
2833 - add an extra multiunlink test (1771)
2834 - fix read_record/write_record API (1776)
2835 - fix leak of offset_extent, possible incorrect i_size later (1772)
2836 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
2837 - Debugging for the kqswnal_get_idle_tx problems (1820)
2838 - Allow recovery to be attempted multiple times (1536)
2839 - Write out MDS last_rcvd file after it is first created (1600)
2840 - Fix tx_descriptor leak in failed transmit situations (1827)
2841 - ext3 journaling fixes for assertion failure after IO error (1871)
2842 - class_export_put() on freed export after completion AST error (1896)
2843 - Fix revalidate looping in VFS (1322)
2844 - Don't access a freed export during MDS_REINT timeout (1521)
2845 - Add open-unlink recovery support on the MDS (1673,1764)
2846 - Return an error if no MDS data was read from last_rcvd (1946)
2847 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
2848 - Don't setup a disk device that doesn't match exported UUID (317)
2849 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
2850 - avoid committing NULL handle in force close
2851 - local.sh is now a one-stripe LOV configuration
2852 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
2853 - fix ext3 htree duplicate directory entry corruption (1516)
2854 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
2855 - update to new LOV EA format (2097)
2856 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
2858 2003-06-15 Phil Schwan <phil@clusterfs.com>
2861 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
2862 - per-import/export recovery handling (958, 931, 959)
2863 - multiple last-rcvd slots, for serving multiple FSes (949)
2864 - connections are again shared between multiple imp/exports (963, 964)
2865 - "umount -f" would hang if any requests needed to be sent (393, 978)
2866 - avoid pinning large req buffer by copying for queued messages (989)
2867 - add "uuid" to "lctl device" command to help upcalls (991)
2868 - "open" RPCs with transnos would confuse recovery counters (1037)
2869 - do proper endian conversion of all wire messages (288, 340, 891)
2870 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
2871 - call waiting locks callback from LDLM recovery thread (1127, 1151)
2872 - fix ptlrpc_connection leak in target_handle_connect (1174)
2873 - fix import refcounting bug in OST and MDS cleanup (1134)
2874 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
2875 - fix very unlikely obd_types race condition (501)
2876 - remove osc_open hack for echo_client (1187)
2877 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
2878 - a failure in read_inode2 leads to deadlock (1139)
2879 - cancel ack-locks as soon as transaction is committed (1072)
2880 - fix major leaks and crashes in the bulk I/O path (937, 1057)
2881 - make sure to commitrw after any preprw to avoid deadlock (1162)
2882 - failing to execute a file in a lustre FS would lock inode (1203)
2883 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
2884 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
2885 - obd_uuid2tgt would walk off the end of the list (1255)
2886 - on IA64 the osc would give portals incorrect bulk size (1258)
2887 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
2888 - fix lock inversion caused by new llite matching code (1282)
2889 - limit the number of dirty pages on a client to 10MB (1286)
2890 - timed out locks were not being corrected cancelled (1289)
2891 - fix O_DIRECT above 4GB on IA-32 (1292)
2892 * major user-visible changes
2893 - fail out/fail over policy now controlled by the upcall (993)
2895 - add OBD_PING to check server availability and failure (954)
2896 - lustre messages are now sent in sending host order (288, 340, 891)
2897 - add eadatalen to MDS getattr reply (340)
2898 - OST read replies may contain second buffer, with per-page status (593)
2900 2003-03-11 Phil Schwan <phil@clusterfs.com>
2903 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
2904 - failed open()s could cause deadlock; fixed (867, 869)
2905 - stop cancelling OST locks when files are closed (481)
2906 - overlapping XID spaces caused network corruption (851, 853)
2907 - fix unsafe fsfilt counter arithmetic; change to atomic_t
2908 - setattr_raw added, to do single-RPC, server-side setattrs
2909 - lmc/lconf syntax change for OST UUIDs
2910 - fix crashy race condition between ptlrpc_free_req and osc_close
2911 - don't use request in mdc_enqueue if we hit a timeout (889)
2912 - don't set the inode i_size for regular files from the MDS (896)
2913 - handle out of order completion AST (842)
2914 - don't LBUG if a lock request times out after receiving AST (913)
2915 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
2916 - if a bad lock AST arrives, send an error instead of dropping entirely
2917 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
2918 - fix leak in bulk IO when only partially completed (899, 900, 926)
2919 - fix O_DIRECT for ia64 (55)
2920 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
2921 - C-z after timeout could hang a process forever; fixed (977)
2923 - client-side I/O cache (678, 924, 929, 941, 970)
2925 - READPAGE and SETATTRs which don't take server-side locks get
2928 2003-02-11 Phil Schwan <phil@clusterfs.com>
2931 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
2932 - Fix fsfilt_extN_readpage() to read a full page of directory
2933 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
2934 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
2935 - Fix imbalanced LOV object allocation and out-of-bound access (469)
2936 - Most intent operations were removed, in favour of a new RPC mode
2937 that does a single RPC to the server and bypasses most of the VFS
2938 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
2939 - Aggressively cancel local locks on DLM servers
2940 - mds_reint_unlink sends EA to the client if it's the last nlink.
2941 client uses that EA to unlink OST objects.
2942 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
2943 - recursive symlinks were fixed (439)
2944 - fixed NULL deref in DEBUG_REQ
2945 - filter_update_lastobjid no longer calls sync, which annoyed extN
2946 - fixed multi-client small-writes to a single file problem (445)
2947 - fixed mtime updates during file writes (607)
2948 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
2949 - fixed bug in obd_brw_read/write() (under guise of testing 367)
2950 - fixed Linux OST size reporting problem (444, 656)
2951 - OST now updates object mtime with writes or setattr (607, 619)
2952 - client verifies file size before zeroing page past EOF (445)
2953 - OST now writes last allocated objid to disk with allocation (108)
2954 - LOV on echo now works (409)
2956 - mds_reint_unlink sends a new buffer, with the EA included. this
2957 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
2958 set if a regular file was being unlinked, and it was the last link
2959 - use PtlGet from the target for bulk writes (315)
2960 - OST now updates object mtime with writes or setattr (607, 619)
2961 - LDLM now has a grant-time callback to revalidate locked items, if
2963 - Many MDS operations were reorganized to combat race conditions
2965 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
2966 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
2968 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
2971 - Fully reactivate OST imports after reconnection (512, others)
2972 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
2973 - More graceful error handling for truncating on dead OST (515)
2974 - Don't error out unless we're actually accessing dead stripes (474)
2975 - Fix garbage sizes when stripes are missing (410)
2976 - LRU counters were broken, causing constant lock purge (433, 432)
2977 - garbage on read from stripes with failed OSTs (441)
2978 - mark OSCs as active before reconnecting during recovery (438)
2979 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
2980 - lfind did not preserve OST order in output (443)
2981 - symlinks cause hung clients, incorrect data (439)
2982 - stop dereferencing request after dropping refcount (457)
2983 - don't LASSERT(spin_is_locked) on non-SMP (455)
2984 - fixes for many rename() bugs
2985 - fstat didn't correctly synchronize attributes (399)
2986 - server must handle lock cancellation during blocking AST prep (487)
2987 - bulk descriptors were free()d too soon (511)
2988 - fix paths in lconf, which would load incorrect modules (451, 507)
2989 - fix confusing lconf 'host not found' error message (386)
2990 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
2991 - fix race condition in mdc_blocking_ast() for inode access (526)
2992 - fix lov_unpackmd() unpacking wrong number of stripes (537)
2993 - fix lov_set_osc_active() marking wrong OSC inactive (440)
2994 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
2995 - fix multiple writes of stripe MD to MDS (358, maybe 519)
2996 - fix lstripe in several ways (kernel side) (527)
2997 - fix request leak in ldlm_cli_enqueue (262)
2998 - incorrect OSC was marked inactive after OST failure
2999 - call mds_fs_cleanup before unmounting filesystem (524)
3000 - fix races between taking ns_lock and ldlm_lock_change_resource
3001 - fix races updating LOV export open file list
3002 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3003 - fix recovery NULL deref in ldlm_cli_cancel_unused
3004 - fix some DLM races by using new hash table for lock handles (419)
3005 - permit the client to specify desired inodes, at replay
3006 - duplicate requests when we queue them for replay reintegration
3007 - fix last_rcvd offset calculation
3008 - sync after each recovered transaction, so we always make progress
3009 - never, not always, ERESTART requests without transnos
3010 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3012 - skip replay if the MDS says that the client is already connected
3013 - don't check for a recovery-enabled export to match lctl's UUID
3014 - don't INC_USE_COUNT for phantom exports
3015 - don't crash when cleaning up phantom exports (567)
3016 - don't double-finish or set replay data for errored mdc_open requests
3017 - abort requests when they time out, so we don't get old replies
3018 - send/receive replies for AST messages again
3019 - if the client says that it doesn't have the lock, cancel it on the
3021 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3022 mark it as destroyed, it will all work out when decref is called
3023 - fix module use counts (22, 581)
3025 - ASTs now expect a reply (server cancels lock on error reply)
3027 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3030 - fix many simultaneous client startup (392)
3031 - fix dentry->d_it clobbering
3032 - credentials weren't being shipped for readdir/getattr operations
3033 - remove invalid assertions triggered during some concurrent MD
3035 - proper Lustre versions added (336, 389)
3036 - fix memory leak for create error case (398)
3037 - fix LOV locking bug that would get cli/srv out of sync
3038 - fix echo client over LOV (409)
3039 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3040 - fix double-O_EXCL intent crash (424)
3041 - avoid sending multiple lock CANCELs (352)
3043 - MDS can do multi-client recovery (modulo bugs in new code)
3045 - many updates, edits, cleanups
3047 2002-11-18 Phil Schwan <phil@clusterfs.com>
3050 - fix null d_it dereference (346)
3051 - fix full OST/dbench hang (333)
3052 - fix permission problem with file removal (286)
3053 - fix removal of OSCs from LOV when they fail
3054 - fix NULL deref during bulk timeout (214)
3055 - fix problems related to multiple filesystems on one MDS (241)
3056 - fixed serious subtle metadata locking bugs
3057 - free locks on clients when inodes are removed due to memory
3059 - fix inode pointer in lock data (285)
3060 - partial support for multiple MDS on a single host (241)
3061 - data locks weren't cancelled at clear_inode time (290, 311)
3062 - intent locks could lead to unbounded lock growth (205)
3063 - added a maximum lock count, an LRU list, and a flusher
3064 - fix multiple rename (365)
3065 - properly abstracted the echo client
3066 - OSC locked 1 byte too many; fixed
3067 - rewrote brw callback code:
3068 - fixed recovery bugs related to LOVs (306)
3069 - fixed too-many-pages-in-one-write crash (191)
3070 - fixed (again) crash in sync_io_timeout (214)
3071 - probably fixed callback-related race (385)
3073 - Add capability to MDS protocol
3074 - LDLM cancellations and callbacks on different portals
3076 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3079 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3080 - defer open object destruction to close time (601981, 138)
3081 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3082 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3083 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3084 - don't call obd_setattr in ll_file_release for destroyed objects
3085 * protocol change to lustre_msg: move |version| and add |flags|
3086 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3087 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3088 * added replay of create, unlink, link and rename operations during
3089 MDS failover; recovery should be much more robust now
3090 * remove failed OSCs from LOVs (only lov_create uses this so far)
3091 * the lustre-HOWTO was brought (more) up to date (582544)
3093 2002-10-23 Phil Schwan <phil@clusterfs.com>
3096 - in-use dentries weren't being reused properly (617851)
3097 - prevent multiple LDLM setup (599178)
3098 - fix LOV size calculations for truncate (617853)
3099 - fix client handling of MDS intent errors (POSIX)
3100 - fix permission bug in lovstripe.c test (624321)
3101 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3102 - truncate past end of file could corrupt data
3103 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3104 - a race in recovery could return ETIMEDOUT to apps (623947)
3105 - building outside the source directory was fixed
3106 * the lustre-HOWTO was brought (more) up to date (582544)
3107 * major progress was made on recovery functionality
3109 2002-10-10 Phil Schwan <phil@clusterfs.com>
3112 - recovery deadlock fix
3113 - rm -rf causes LBUG fix (617817)
3114 - file open by multiple tasks fix (618962)
3115 - directory permissions bugs (602707 and 620007)
3116 - journal_stop fixed with locking (611313)
3117 - O_APPEND failures resolved (618273, perhaps 614459)
3118 - lconf PATH fix (619770)
3119 - IA64 build fix (621450)
3120 - RPC buffer sizes scale with amount of memory
3122 2002-10-01 Phil Schwan <phil@clusterfs.com>
3125 - locks would be cancelled without throwing away data pages,
3126 resulting in inconsistent data (605627)
3127 - inode attributes were not always being refreshed (605627, 612449)
3128 - lconf now continues to cleanup after lctl reports an error
3129 - MDS now enforces user permissions (602707)
3130 - lprocfs cleanup fixed, but not yet enabled (614157)
3131 - fixed infinite server hang, should a client not respond to an AST
3132 - avoid going into recovery if user calls readlink() with a buffer
3133 that's too small (613941)
3134 - AST RPCs no longer require replies (614867) -- this may be changed
3135 - don't crash server if client sends an IOV that's too big (611336)
3136 - fixed lock conversion deadlock (611892)
3137 - fixed the following of symlinks (614622)
3138 * recovery: the server can remove locks from a client that dies, other
3139 clients can make progress
3140 * more extN patch fixes
3141 * compile-time configurable ptlrpc buffer allocations
3143 - collaborative read cache document
3144 - Lustre Lite Performance CDR document-in-progress
3146 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3149 - fix typo in patch-2.4.18
3151 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3154 - clear ptlrpc request each time in handle_incoming_request()
3155 - unlink of files now destroys the object on the OST
3157 2002-09-19 Peter Braam <braam@clusterfs.com>
3159 * add hard link support
3160 * change obdfile creation method
3161 * kernel patch changed
3163 2002-09-19 Peter Braam <braam@clusterfs.com>
3166 - stack overflow bug in extN fixed
3168 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3170 * documentation updates
3171 - add man pages for config tools
3172 - update tests/README to describe testing with new config tools
3173 - finish metadata API descriptions
3174 * bug fixes and cleanups
3175 - statfs workaround for 16TB limit
3176 - LOV stripe allocation improved, can stripe on subset of OSTs
3177 - LOV file size/IO offset was wrong for files > 4GB in size
3178 - object EA data was being dropped, caused files to be unreadable
3179 - memory overflow with non-LOV OST caused memory corruption
3180 - fixed regression tests to work with new config tools, obdfilter
3181 - fixed bug when directory size became larger than 1 block
3182 - fixed bug (for single client case) when PWD was deleted
3183 - invalidate local directory pages when doing intent-based ops
3184 - avoid LDLM oops when lock callback contained bad data
3186 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3188 * documentation updates
3189 * bug fixes and cleanups
3190 - configuration tools
3193 - 64-bit compile warnings
3194 - 64-bit internal statfs data
3196 * test_brw on persistent OST devices
3198 * lprocfs (disabled)
3200 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3202 * documentation updates
3203 * bug fixes and cleanups
3204 * configuration tools
3206 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3209 * many small fixes to 0_5_4
3210 * io/network handling
3211 * thinkos in MDS operations
3213 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3216 * crucial basic fixes to 0.5.3
3217 * IOR, Iozone work over Elan
3220 2002-08-07 Phil Schwan <phil@clusterfs.com>
3221 * version 0_5_3, our first alpha
3222 * we use the new Portals iovs
3223 * documentation updates
3224 * bug fixes and cleanups
3225 * small changes in the DLM wire protocol
3227 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3228 * version 0_5_1 with some initial stability,
3229 * locking on MD and file I/O.
3230 * documentation updates
3231 * several bug fixes since 0.5.0
3232 * small changes in wire protocol
3234 2002-07-18 Phil Schwan <phil@clusterfs.com>
3236 * delivered as Lustre Light Alpha
3237 * fixed a crash after handling invalid MDS requests
3238 * fixed directory pages for architectures with non-4k pages sizes
3240 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3241 * release version v0_4_4
3242 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3244 * Fixes a number of interruption problems with OST operations.
3245 * Update documentation for portals header changes
3246 * Move all wire protocol structs/defines to lustre_idl.h
3247 * Fixes symlink length bug.
3248 * Add tcpdump to repository.
3250 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3251 * release version v0_4_3
3252 * Fixes statfs for inodes on extN.
3253 * Fixes bug in runtests which would delete /etc/hosts.
3254 * Use 64-bit object IDs wherever possible (not into VFS though)
3255 Remove ost_get_info, which is unused by lustre, and out of date.
3257 2002-07-03 Peter Braam <braam@clusterfs.com>
3258 * release version v0_4_2 Fixes a lookup error (type not passed)
3259 * move forward to head of Portals
3260 * move forward to latest Lustre kernel
3262 2002-06-25 Peter Braam <braam@clusterfs.com>
3263 * release version v0_4_1. Hopefully stable on single node use.