1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
4 INTEROPERATE with 1.4.x versions automatically. In many cases a
5 special upgrade step is needed. Please read the user documentation
6 before upgrading any part of a 1.4.x system.
7 * WARNING: Lustre configuration and startup changes are required with
8 1.6.x releases. See https://mail.clusterfs.com/wikis/lustre/MountConf
10 * Support for kernels:
11 2.6.9-42.0.10.EL (RHEL 4)
13 2.4.21-47.0.1.EL (RHEL 3)
14 2.6.12.6 vanilla (kernel.org)
15 2.6.16.27-0.9 (SLES 10)
16 * Client support for unpatched kernels:
17 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
18 2.6.16 - 2.6.19 vanilla (kernel.org)
19 2.6.9-42.0.8EL (RHEL 4)
20 * Recommended e2fsprogs version: 1.39.cfs6
22 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
28 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
29 Details : Due to a race condition, liblustre clients were occasionally
32 Severity : enhancement
34 Description: lfs setstripe use optional parameters instead of postional
37 Severity : enhancement
39 Description: Nanosecond timestamp support for ldiskfs
40 Details : The on-disk ldiskfs filesystem has added support for nanosecond
41 resolution timestamps. There is not yet support for this at
42 the Lustre filesystem level.
45 Frequency : during server recovery
47 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
48 Details : request with rq_no_resend flag not awake l_wait_event if they get a
52 Frequency : nfs export on patchless client
54 Description: connectathon hang when test nfs export over patchless client
55 Details : Disconnected dentry cannot be found with lookup, so we do not need
56 to unhash it or make it invalid
59 Description: ENOENT returned for valid filehandle during dbench.
60 Details: : Check if a directory has children when invalidating dentries
61 associated with an inode during lock cancellation. This fixes
62 an incorrect ENOENT sometimes seen for valid filehandles during
65 --------------------------------------------------------------------------------
67 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
69 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
70 INTEROPERATE with older versions automatically. In many cases a
71 special upgrade step is needed. Please read the
72 user documentation before upgrading any part of a 1.4.x system.
73 * WARNING: Lustre configuration and startup changes are required with
74 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
76 * Support for kernels:
77 2.4.21-47.0.1.EL (RHEL 3)
79 2.6.9-42.0.10.EL (RHEL 4)
80 2.6.12.6 vanilla (kernel.org)
81 2.6.16.27-0.9 (SLES10)
82 * Client support for unpatched kernels:
83 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
84 2.6.16 - 2.6.19 vanilla (kernel.org)
85 2.6.9-42.0.8EL (RHEL 4)
86 * Recommended e2fsprogs version: 1.39.cfs6
87 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
90 Severity : enhancement
92 Description: MountConf
93 Details : Lustre configuration is now managed via mkfs and mount
94 commands instead of lmc and lconf. New obd types (MGS, MGC)
95 are added for dynamic configuration management. See
96 https://mail.clusterfs.com/wikis/lustre/MountConf for
99 Severity : enhancement
101 Description: dynamic OST addition
102 Details : OSTs can now be added to a live filesystem
104 Severity : enhancement
106 Description: startup order invariance
107 Details : MDTs and OSTs can be started in any order. Clients only
108 require the MDT to complete startup.
110 Severity : enhancement
112 Description: parallel, asynchronous orphan cleanup
113 Details : orphan cleanup is now performed in separate threads for each
114 OST, allowing parallel non-blocking operation.
116 Severity : enhancement
118 Description: optimized stripe assignment
119 Details : stripe assignments are now made based on ost space available,
120 ost previous usage, and OSS previous usage, in order to try
121 to optimize storage space and networking resources.
123 Severity : enhancement
125 Description: Permanently set tunables
126 Details : All writable /proc/fs/lustre tunables can now be permanently
127 set on a per-server basis, at mkfs time or on a live
130 Severity : enhancement
132 Description: Lustre message v2
133 Details : Add lustre message format v2.
135 Severity : enhancement
137 Description: client OST exclusion list
138 Details : Clients can be started with a list of OSTs that should be
139 declared "inactive" for known non-responsive OSTs.
143 Description: ENOENT returned for valid filehandle during dbench.
144 Details : Check if a directory has children when invalidating dentries
145 associated with an inode during lock cancellation. This fixes
146 an incorrect ENOENT sometimes seen for valid filehandles during
150 Frequency : SFS test only (otherwise harmless)
152 Description: SPEC SFS validation failure on NFS v2 over lustre.
153 Details : Changes the blocksize for regular files to be 2x RPC size,
154 and not depend on stripe size.
156 Severity : enhancement
158 Description: fine-grained SMP locking inside DLM
159 Details : Improve DLM performance on SMP systems by removing the single
160 per-namespace lock and replace it with per-resource locks.
162 Severity : enhancement
164 Description: don't hold multiple extent locks at one time
165 Details : To avoid client eviction during large writes, locks are not
166 held on multiple stripes at one time or for very large writes.
167 Otherwise, clients can block waiting for a lock on a failed OST
168 while holding locks on other OSTs and be evicted.
170 Severity : enhancement
172 Description: Multiple MD RPCs in flight.
173 Details : Further unserialise some read-only MDT RPCs - learn about intents.
174 To avoid overly-overloading MDT, introduce a limit on number of
175 MDT RPCs in flight for a single client and add /proc controls
176 to adjust this limit.
178 Severity : enhancement
180 Description: client read/write statistics
181 Details : Add client read/write call usage stats for performance
182 analysis of user processes.
183 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
184 file access. extents_stats shows chunk size distribution.
185 extents_stats_per_process show chunk size distribution per
188 Severity : enhancement
190 Description: per-client statistics on server
191 Details : Add ldlm and operations statistics for each client in
192 /proc/fs/lustre/mds|obdfilter/*/exports/
194 Severity : enhancement
196 Description: improved MDT statistics
197 Details : Add detailed MDT operations statistics in
198 /proc/fs/lustre/mds/*/stats
200 Severity : enhancement
202 Description: VFS operations stats
203 Details : Add client VFS call stats, trackable by pid, ppid, or gid
204 /proc/fs/lustre/llite/*/vfs_ops_stats
205 /proc/fs/lustre/llite/*/vfs_track_[pid|ppid|gid]
210 Description: Fix client-side osc byte counters
211 Details : The osc read/write byte counters in
212 /proc/fs/lustre/osc/*/stats are now working
215 Frequency : always as root on SLES
217 Description: Failure of copying files with lustre special EAs.
218 Details : Client side always return success for setxattr call for lustre
219 special xattr (currently only "trusted.lov").
224 Description: Refcount LNET uuids
225 Details : The global LNET uuid list grew linearly with every startup;
226 refcount repeated list entries instead of always adding to
229 Severity : enhancement
231 Description: Dynamic service threads
232 Details : Within a small range, start extra service threads
233 automatically when the request queue builds up.
236 Frequency : mixed-endian client/server environments
238 Description: mixed-endian crashes
239 Details : The new msg_v2 system had some failures in mixed-endian
242 Severity : enhancement
244 Description: Easy OST removal
245 Details : OSTs can be permanently deactivated with e.g. 'lctl
246 conf_param lustre-OST0001.osc.active=0'
248 Severity : enhancement
250 Description: MGS proc entries
251 Details : Added basic proc entries for the MGS showing what filesystems
254 Severity : enhancement
256 Description: provide MGS failover
257 Details : Added config lock reacquisition after MGS server failover.
259 Severity : enhancement
261 Description: add Linux 2.4 support
262 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
266 Description: a large application tries to do I/O to the same resource and dies
268 Details : Check the req->rq_arrival time after the call to
269 ost_brw_lock_get(), but before we do anything about
270 processing it & sending the BULK transfer request. This
271 should help move old stale pending locks off the queue as
272 quickly as obd_timeout.
275 Frequency : when an incorrect nid is specified during startup
277 Description: ptlrpc connect to non-existant node causes kernel crash
278 Details : LNET can't be re-entered from an event callback, which
279 happened when we expire a message after the export has been
280 cleaned up. Instead, hand the zombie cleanup off to another
283 Severity : enhancement
285 Description: plain/inodebits lock performance improvement
286 Details : Grouping plain/inodebits in granted list by their request modes
287 and bits policy, thus improving the performance of search through
291 Frequency : only if OST filesystem is corrupted
293 Description: client incorrectly hits assertion in ptlrpc_replay_req()
294 Details : for a short time RPCs with bulk IO are in the replay list,
295 but replay of bulk IOs is unimplemented. If the OST filesystem
296 is corrupted due to disk cache incoherency and then replay is
297 started it is possible to trip an assertion. Avoid putting
298 committed RPCs into the replay list at all to avoid this issue.
301 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
303 Description: System hang on startup
304 Details : This bug allowed the liblustre (e.g. catamount) client to
305 return to the app before handling all startup RPCs. This
306 could leave the node unresponsive to lustre network traffic
307 and manifested as a server ptllnd timeout.
309 Severity : enhancement
311 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
312 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
313 boolean to control whether to print peer debug info when a
314 client's RPC times out.
317 Frequency : only for kernels with patches from Lustre below 1.4.3
319 Description: Remove old rdonly API
320 Details : Remove old rdonly API which unused from at least lustre 1.4.3
323 Frequency : only for devices with external journals
325 Description: Set external device read-only also
326 Details : During a commanded failover stop, we set the disk device
327 read-only while the server shuts down. We now also set any
328 external journal device read-only at the same time.
331 Frequency : when upgrading from 1.4 while trying to change parameters
333 Description: The wrong (new) MDC name was used when setting parameters for
334 upgraded MDT's. Also allows changing of OSC (and MDC)
335 parameters if --writeconf is specified at tunefs upgrade time.
338 Frequency : when setting specific ost indicies
340 Description: QOS code breaks on skipped indicies
341 Details : Add checks for missing OST indicies in the QOS code, so OSTs
342 created with --index need not be sequential.
344 Severity : enhancement
346 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
347 Details : The uninit_groups feature works in conjunction with the kernel
348 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
349 up the pass1 processing of e2fsck. This is a read-only feature
350 in ldiskfs2 only, so older kernels and current ldiskfs cannot
351 mount filesystems that have had this feature enabled.
353 Severity : enhancement
355 Description: Improve multi-block allocation algorithm to avoid fragmentation
356 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
357 allocation locality and avoid filesystem fragmentation.
359 ------------------------------------------------------------------------------
361 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
363 * Support for kernels:
364 2.4.21-47.0.1.EL (RHEL 3)
366 2.6.9-42.0.10.EL (RHEL 4)
367 2.6.12.6 vanilla (kernel.org)
368 2.6.16.27-0.9 (SLES 10)
369 * Recommended e2fsprogs version: 1.39.cfs5
371 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
375 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
377 Description: System hang on startup
378 Details : This bug allowed the liblustre (e.g. catamount) client to
379 return to the app before handling all startup RPCs. This
380 could leave the node unresponsive to lustre network traffic
381 and manifested as a server ptllnd timeout.
383 Severity : enhancement
385 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
386 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
387 boolean to control whether to print peer debug info when a
388 client's RPC times out.
393 Description: make O_SYNC working on 2.6 kernels
394 Details : 2.6 kernels use different method for mark pages for write,
395 so need add a code to lustre for O_SYNC work.
400 Description: Failure to close file and release space on NFS
401 Details : Put inode details into lock acquired in ll_intent_file_open.
402 Use mdc_intent_lock in ll_intent_open to properly
403 detect all kind of errors unhandled by mdc_enqueue.
408 Description: proc file read during shutdown sometimes raced obd removal,
410 Details : Add lock to prevent obd access after proc file removal.
413 Frequency : Only for files larger than 4GB on 32-bit clients.
415 Description: improperly doing page alignment of locks
416 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
417 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
422 Description: RPCs being resent when they shouldn't be
423 Details : Some RPCs that should not be resent are being resent. This
424 can cause inconsistencies in the RPC state machine. Do not
425 resend such requests.
428 Frequency : rare, only with NFS export
430 Description: Crash on NFS re-export node
431 Details : under very unusual load conditions an assertion is hit in
432 ll_intent_file_open()
435 Frequency : only if OST filesystem is corrupted
437 Description: client incorrectly hits assertion in ptlrpc_replay_req()
438 Details : for a short time RPCs with bulk IO are in the replay list,
439 but replay of bulk IOs is unimplemented. If the OST filesystem
440 is corrupted due to disk cache incoherency and then replay is
441 started it is possible to trip an assertion. Avoid putting
442 committed RPCs into the replay list at all to avoid this issue.
447 Description: large O_DIRECT requests fail under memory pressure/fragmentation
448 Details : Large single O_DIRECT read and write calls can fail to allocate
449 a sufficiently large buffer to process the request. In case of
450 allocation failure the allocation is retried with a smaller
451 buffer and broken into smaller requests.
453 Severity : enhancement
455 Description: Add -o localflock option to simulate old noflock behaviour.
456 Details : This will achieve local-only flock/fcntl locks coherentness.
461 Description: versioning check is incomplete
462 Details : Checking the version difference of client vs. server, report
463 error if the gap is too big.
468 Description: add support PG_writeback bit
469 Details : add support for PG_writeback bit for Lustre, for more carefull
470 work with page cache in 2.6 kernel. This also fix some deadlocks
471 and remove hack for work O_SYNC with 2.6 kernel.
473 Severity : enhancement
475 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
476 Details : The uninit_groups feature works in conjunction with the kernel
477 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
478 up the pass1 processing of e2fsck. This is a read-only feature
479 in ldiskfs2 only, so older kernels and current ldiskfs cannot
480 mount filesystems that have had this feature enabled.
482 Severity : enhancement
484 Description: Improve multi-block allocation algorithm to avoid fragmentation
485 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
486 allocation locality and avoid filesystem fragmentation.
488 ------------------------------------------------------------------------------
490 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
492 * Support for kernels:
493 2.6.9-42.0.3EL (RHEL 4)
495 2.4.21-47.0.1.EL (RHEL 3)
496 2.6.12.6 vanilla (kernel.org)
497 * Recommended e2fsprogs version: 1.39.cfs2-0
499 * The backwards-compatible /proc/sys/portals symlink has been removed
500 in this release. Before upgrading, please ensure that you change
501 any configuration scripts or /etc/sysctl.conf files that access
502 /proc/sys/portals/* or sysctl portals.* to use the corresponding
503 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
504 in advance of the upgrade on any system running Lustre 1.4.6 or
505 newer, since /proc/sys/lnet was added in that version.
506 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
510 Frequency : only when quota is used
512 Description: avoid scanning export list for quota master
513 Details : Change the algorithms to avoid scanning export list in order
514 to improve the efficiency.
517 Frequency : MDS failover only, very rarely
519 Description: "went back in time" messages on mds failover
520 Details : The greatest transno may be lost when the current operation
521 finishes with an error (transno==0) and the client's last_rcvd
522 record is over-written. Save the greatest transno in the
523 mds_last_transno for this case.
526 Frequency : always for specific kernels and striping counts
528 Description: client may get "Matching packet too big" without ACL support
529 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
530 when trying to access files in certain configurations. The
531 clients should in fact be denied when mounting because they do
535 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
537 Description: many clients connecting with IO in progress causes connect timeouts
538 Details : Avoid synchronous journal commits to avoid delays caused by many
539 clients connecting/disconnecting when bulk IO is in progress.
540 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
541 OST_IO_PORTAL to avoid delays behind potentially many pending
545 Frequency : occasionally with multiple writers to a single file
547 Description: shared writes to file may result in wrong size reported by stat()
548 Details : Allow growing of kms when extent lock is cancelled
551 Frequency : always with random mmap IO to multi-striped file
553 Description: mmap write might be lost if we are writing to a 'hole' in stripe
554 Details : Only if the hole is at the end of OST object so that kms is too
555 small. Fix is to increase kms accordingly in ll_nopage.
558 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
560 Description: writes to a missing object would leak memory on the OST
561 Details : If there is an inconsistency between the MDS and OST filesystems,
562 such that the MDS references an object that doesn't exist, writes
563 to that object will leak memory due to incorrect cleanup in the
564 error handling path, eventually running out of memory on the OST.
569 Description: Creating too long symlink causes lustre errors
570 Details : Check symlink and name lengths before sending requests to MDS.
573 Frequency : only if flock is enabled (not on by default)
575 Description: posix locks not released on fd closure on 2.6.9+
576 Details : We failed to add posix locks to list of inode locks on 2.6.9+
577 kernels, this caused such locks not to be released on fd close and
578 then assertions on fs unmount about still used locks.
581 Frequency : MDS failover only, very rarely
583 Description: clients may get ASSERTION(granted_lock != NULL)
584 Details : When request was taking a long time, and a client was resending
585 a getattr by name lock request. The were multiple lock requests
586 with the same client lock handle and
587 mds_getattr_name->fixup_handle_for_resent_request found one of the
588 lock handles but later failed with ASSERTION(granted_lock != NULL).
593 Description: handle->h_buffer_credits > 0, assertion failure
594 Details : h_buffer_credits is zero after truncate, causing assertion
595 failure. This patch extends the transaction or creates a new
599 Frequency : NFS re-export or patchless client
600 Bugzilla : 11179, 10796
601 Description: Crash on NFS re-export node (__d_move)
602 Details : We do not want to hash the dentry if we don't have a lock.
603 But if this dentry is later used in d_move, we'd hit uninitialised
604 list head d_hash, so we just do this to init d_hash field but
605 leave dentry unhashed.
608 Frequency : NFS re-export or patchless client
610 Description: NFS exports has problem with symbolic link
611 Details : lustre client didn't properly install dentry when re-exported
612 to NFS or running patchless client.
615 Frequency : NFS re-export or patchless client
617 Description: Various nfs/patchless fixes.
618 Details : fixes reuse disconected alias for lookup process - this fixes
619 warning "find_exported_dentry: npd != pd",
620 fix permission error with open files at nfs.
621 fix apply umask when do revalidate.
624 Frequency : occasional
626 Description: Crash on NFS re-export node
627 Details : calling clear_page() on the wrong pointer triggered oops in
628 generic_mapping_read().
631 Frequency : rarely, using O_DIRECT IO
633 Description: unaligned directio crashes client with LASSERT
634 Details : check for unaligned buffers before trying any requests.
637 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
639 Description: stale data returned from RAID cache
640 Details : If only a small amount of IO is done to the RAID device before
641 reading it again it is possible to get stale data from the RAID
642 cache instead of reading it from disk.
645 Frequency : always for sles10 kernel
647 Description: sles10 support
648 Details : ll_follow_link: compile fixes and using of nd_set_link
652 Frequency : depends on arch, kernel and compiler version, always on sles10
655 Description: recursive or deep enough symlinks cause stack overflow
656 Details : getting rid of large stack-allocated variable in
660 Frequency : depends on hardware
662 Description: lustre write performance loss in the SLES10 kernel
663 Details : the performance loss is caused by using of write barriers in the
664 ext3 code. The SLES10 kernel turns barrier support on by
665 default. The fix is to undo that change for ldiskfs.
668 ------------------------------------------------------------------------------
670 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
672 * Support for kernels:
673 2.6.9-42.0.3EL (RHEL 4)
675 2.4.21-47.0.1.EL (RHEL 3)
676 2.6.12.6 vanilla (kernel.org)
680 Frequency : quota enabled and large files being deleted
682 Description: releasing more than 4GB of quota at once hangs OST
683 Details : If a user deletes more than 4GB of files on a single OST it
684 will cause the OST to spin in an infinite loop. Release
685 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
690 Description: statfs data retrieved from /proc may be stale or zero
691 Details : When reading per-device statfs data from /proc, in the
692 {kbytes,files}_{total,free,avail} files, it may appear
693 as zero or be out of date.
696 Frequency : systems with MD RAID1 external journal devices
698 Description: lconf's call to blkid is confused by RAID1 journal devices
699 Details : Use the "blkid -l" flag to locate the MD RAID device instead
700 of returning all block devices that match the journal UUID.
703 Frequency : always, for aggregate stripe size over 4GB
705 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
706 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
707 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
710 Frequency : always if "lfs find" used on a local file/directory
712 Description: "lfs find" segfaults if used on a local file/directory
713 Details : The case where a directory component was not specified wasn't
714 handled correctly. Handle this properly.
717 Frequency : always on ppc64
719 Description: the write to an ext3 filesystem mounted with mballoc got stuck
720 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
721 perform endianness conversion.
724 Frequency : rarely (truncate to non-zero file size after write under load)
725 Bugzilla : 10730, 10687
726 Description: Files padded with zeros to next 4K multiple
727 Details : With filesystems mounted using the "extents" option (2.6 kernels)
728 it is possible that files that are truncated to a non-zero size
729 immediately after being written are filled with zero bytes beyond
730 the truncated size. No file data is lost.
732 Severity : enhancement
734 Description: Allow recovery/failover for liblustre clients.
735 Details : liblustre clients were unaware of failover configurations until
738 Severity : enhancement
740 Description: user file locks should fail when not mounting with flock option
741 Details : Set up an error-returning stub in ll_file_operations.lock field
742 to prevent incorrect behaviour when client is mounted without
743 flock option. Also, set up properly f_op->flock field for
747 Frequency : always on ia64
749 Description: "lfs df" loops on printing out MDS statfs information
750 Details : The obd_ioctl_data was not initialized and in some systems
751 this caused a failure during the ioctl that did not return
752 an error. Initialize the struct and return an error on failure.
755 Frequency : SLES 9 only
757 Description: Error of copying files with lustre special EAs as root
758 Details : Client side always return success for setxattr call for lustre
759 special xattr (currently only "trusted.lov").
762 Frequency : rarely on clusters with both ia64+i386 clients
764 Description: ia64+i686 clients doing shared IO on the same file may LBUG
765 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
766 clients are doing concurrent writes to the same file it is
767 possible that the ia64 clients may LASSERT because the OST
768 extent locks are not PAGE_SIZE aligned. Ensure that grown
769 locks are always aligned on the request boundary.
772 Frequency : specific use, occasional
774 Description: Overwriting in use executable truncates on-disk binary image
775 Details : If one node attempts to overwrite an executable in use by
776 another node, we now correctly return ETXTBSY instead of
779 Severity : enhancement
781 Description: Async OSC create to avoid the blocking unnecessarily.
782 Details : If a OST has no remain object, system will block on the creating
783 when need to create a new object on this OST. Now, ways use
784 pre-created objects when available, instead of blocking on an
785 empty osc while others are not empty. If we must block, we block
786 for the shortest possible period of time.
791 Description: chmod on Lustre root is propagated to other clients
792 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
793 invalid by the follow_mount time.
796 Frequency : liblustre clients only
798 Description: Race in 'instant cancel' lock handling could lead to such locks
799 never to be granted in case of SMP MDS
800 Details : Do not destroy not yet granted but cbpending locks in
804 Frequency : replay/resend of open
806 Description: non null lock assetion failure in mds_intent_policy
807 Details : Trying to replay/resend lockless open requests resulted in
808 mds_open() returning 0 with no lock. Now it sets a flag if
809 a lock is going to be returned.
811 Severity : enhancement
813 Description: Checksum enhancements
814 Details : New checksum enhancements allow for resending RPCs that failed
817 Severity : enhancement
819 Description: Tunables on number of dirty pages in cacche
820 Details : Allow to set limit on number of dirty pages cached.
825 Description: client crash on unmount - lock still has references
826 Details : In some error handling cases it was possible to leak a lock
827 reference on a client while accessing a file. This was not
828 harmful to the client during operation, but would cause the
829 client to crash when the filesystem is unmounted.
832 Frequency : specific case, rare
834 Description: ETXTBSY on mds though file not in use by client
835 Details : ETXTBSY is no longer incorrectly returned when attempting to
836 chmod or chown a directory that the user previously tried to
837 execute or a currently-executing binary.
840 Frequency : extremely rare except on liblustre-based clients
842 Description: Lustre space not freed when files are deleted
843 Details : Clean up open-unlinked files after client eviction. Previously
844 the unlink was skipped and the files remained as orphans.
849 Description: OST failure "would be an LBUG" in waiting_locks_callback()
850 Details : In some cases it was possible to send a blocking callback to
851 a client doing a glimpse, even though that client didn't get
852 a lock granted. When the glimpse lock is cancelled on the OST
853 the freed lock is left on the waiting list and corrupted the list.
856 Frequency : all core dumps
858 Description: Broke core dumps to lustre
859 Details : Negative dentry may be unhashed if parent does not have UPDATE
860 lock, but some callers, e.g. do_coredump, expect dentry to be
861 hashed after successful create, hash it in ll_create_it.
863 ------------------------------------------------------------------------------
865 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
867 * Support for kernels:
868 2.6.9-42.0.2.EL (RHEL 4)
870 2.4.21-40.EL (RHEL 3)
871 2.6.12.6 vanilla (kernel.org)
875 Frequency : always on RHEL 3
877 Description: Number of open files grows over time
878 Details : The number of open files grows over time, whether or not
879 Lustre is started. This was due to a filp leak introduced
880 by one of our kernel patches.
882 ------------------------------------------------------------------------------
884 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
886 * Support for kernels:
889 2.4.21-40.EL (RHEL 3)
890 2.6.12.6 vanilla (kernel.org)
895 Bugzilla : 5719, 9635, 9792, 9684
896 Description: OST (or MDS) trips assertions in (re)connection under heavy load
897 Details : If a server is under heavy load and cannot reply to new
898 connection requests before the client resends the (re)connect,
899 the connection handling code can behave badly if two service
900 threads are concurrently handing separate (re)connections from
901 the same client. Add better locking to the connection handling
902 code, and ensure that only a single connection will be processed
903 for a given client UUID, even if the lock is dropped.
905 Severity : enhancement
907 Description: add TCP zero-copy support to kernel
908 Details : Add support to the kernel TCP stack to allow zero-copy bulk
909 sends if the hardware supports scatter-gather and checksumming.
910 This allows socklnd to do client-write and server-read more
911 efficiently and reduce CPU utilization from skbuf copying.
914 Frequency : only if NFS exporting from client
916 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
917 Details : If setting attributes on a file created under NFS that had
918 never been opened it would be possible to oops the client
919 if the file had no objects.
923 Bugzilla : 9326, 10402, 10897
924 Description: client crash in ptlrpcd_wake() thread when sending async RPC
925 Details : It is possible that ptlrpcd_wake() dereferences a freed async
926 RPC. In rare cases the ptlrpcd thread alread processed the RPC
927 before ptlrpcd_wake() was called and the request was freed.
930 Frequency : always for liblustre
932 Description: liblustre client does MDS+OSTs setattr RPC for each write
933 Details : When doing a write from a liblustre client, the client
934 incorrectly issued an RPC to the MDS and each OST the file was
935 striped over in order to update the timestamps. When writing
936 with small chunks and many clients this could overwhelm the MDS
937 with RPCs. In all cases it would slow down the write because
938 these RPCs are unnecessary.
940 Severity : enhancement
942 Description: allow number of MDS service threads to be changed at module load
943 Details : It is now possible to change the number of MDS service threads
944 running. Adding "options mds mds_num_threads={N}" to the MDS's
945 /etc/modprobe.conf will set the number of threads for the next
946 time Lustre is restarted (assuming the "mds" module is also
947 reloaded at that time). The default number of threads will
948 stay the same, 32 for most systems.
953 Description: OST crash if filesystem is unformatted or corrupt
954 Details : If an OST is started on a device that has never been formatted
955 or if the filesystem is corrupt and cannot even mount then the
956 error handling cleanup routines would dereference a NULL pointer.
961 Description: NULL pointer deref in llap_from_page.
962 Details : get_cache_page_nowait can return a page with NULL (or otherwise
963 incorrect) mapping if the page was truncated/reclaimed while it was
964 searched for. Check for this condition and skip such pages when
965 doing readahead. Introduce extra check to llap_from_page() to
966 verify page->mapping->host is non-NULL (so page is not anonymous).
969 Frequency : Sometimes when using sys_sendfile
971 Description: "page not covered by a lock" warnings from ll_readpage
972 Details : sendfile called ll_readpage without right page locks present.
973 Now we introduced ll_file_sendfile that does necessary locking
974 around call to generic_file_sendfile() much like we do in
978 Frequency : with certain MDS communication failures at client mount time
980 Description: NULL pointer deref after failed client mount
981 Details : a client connection request may delayed by the network layer
982 and not be sent until after the PTLRPC layer has timed out the
983 request. If the client fails the mount immediately it will try
984 to clean up before the network times out the request. Add a
985 reference from the request import to the obd device and delay
986 the cleanup until the network drops the request.
989 Frequency : occasionally during client (re)connect
991 Description: assertion failure during client (re)connect
992 Details : processing a client connection request may be delayed by the
993 client or server longer than the client connect timeout. This
994 causes the client to resend the connection request. If the
995 original connection request is replied in this interval, the
996 client may trip an assertion failure in ptlrpc_connect_interpret()
997 which thought it would be the only running connect process.
1000 Frequency : only with obd_echo servers and clients that are rebooted
1002 Description: kernel BUG accessing uninitialized data structure
1003 Details : When running an obd_echo server it did not start the ping_evictor
1004 thread, and when a client was evicted an uninitialized data
1005 structure was accessed. Start the ping_evictor in the RPC
1006 service startup instead of the OBD startup.
1008 Severity : enhancement
1009 Bugzilla : 10193 (patchless)
1010 Description: Remove dependency on various unexported kernel interfaces.
1011 Details : No longer need reparent_to_init, exit_mm, exit_files,
1012 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1015 Frequency : rare (only users of deprecated and unsupported LDAP config)
1017 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1018 Details : LDAP apparently contains 'lustreName' attributes instead of
1019 'name'. A simple remapping of the name is sufficient.
1022 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1024 Description: waiting_locks_callback trips kernel BUG if client is evicted
1025 Details : Running with the dump_on_timeout debug flag turned on makes
1026 it possible that the waiting_locks_callback() can try to dump
1027 the Lustre kernel debug logs from an interrupt handler. Defer
1028 this log dumping to the expired_lock_main() thread.
1030 Severity : enhancement
1032 Description: Support NFS exporting on 2.6 kernels.
1033 Details : Implement non-rawops metadata methods for NFS server to use without
1034 changing NFS server code.
1037 Frequency : very rare (synthetic metadata workload only)
1039 Description: two racing renames might cause an MDS thread to deadlock
1040 Details : Running the "racer" program may cause one MDS thread to rename
1041 a file from being the source of a rename to being the target of
1042 a rename at exactly the same time that another thread is doing
1043 so, and the second thread has already enqueued these locks after
1044 doing a lookup of the target and is trying to relock them in
1045 order. Ensure that we don't try to re-lock the same resource.
1048 Frequency : only very large systems with liblustre clients
1050 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1051 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1052 all OSTs in parallel. This allows the eviction of stale liblustre
1053 clients to proceed much faster than if they were done in series,
1054 and also offers similar improvements for other set_info RPCs.
1059 Description: excessive CPU usage during initial read phase on client
1060 Details : During the initial read phase on a client, it would agressively
1061 retry readahead on the file, consuming too much CPU and impacting
1062 performance (since 1.4.5.8). Improve the readahead algorithm
1063 to avoid this, and also improve some other common cases (read
1064 of small files in particular, where "small" is files smaller than
1065 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1070 Description: MDS crash when receiving packet with unknown intent.
1071 Details : Do not LBUG in unknown intent case, just return -EFAULT
1073 Severity : enhancement
1074 Bugzilla : 9293, 9385
1075 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1076 Details : Do not serialize getattr (non-intent version) and statfs.
1079 Frequency : occasional, when OST network is overloaded/intermittent
1081 Description: client evicted by OST after bulk IO timeout
1082 Details : If a client sends a bulk IO request (read or write) the OST
1083 may evict the client if it is unresposive to its data GET/PUT
1084 request. This is incorrect if the network is overloaded (takes
1085 too long to transfer the RPC data) or dropped the OST GET/PUT
1086 request. There is no need to evict the client at all, since
1087 the pinger and/or lock callbacks will handle this, and the
1088 client can restart the bulk request.
1091 Frequency : Always when mmapping file with no objects
1093 Description: client crashes when mmapping file with no objects
1094 Details : Check that we actually have objects in a file before doing any
1095 operations on objects in ll_vm_open, ll_vm_close and
1101 Description: Request leak when working with deleted CWD
1102 Details : Introduce advanced request refcount tracking for requests
1103 referenced from lustre intent.
1105 Severity : Enhancement
1107 Description: Cache open file handles on client.
1108 Details : MDS now will return special lock along with openhandle, if
1109 requested and client is allowed to hold openhandle, even if unused,
1110 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1111 closing files for every read/write openration.
1113 Severity : Enhancement
1115 Description: Cache open negative dentries on client when possible.
1116 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1117 negative dentries on lock revocation.
1122 Description: Remounting a client read-only wasn't possible with a zconf mount
1123 Details : It wasn't possible to remount a client read-only with llmount.
1125 Severity : enhancement
1126 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1127 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1128 MPICH is included to add Lustre-specific ADIO interfaces.
1129 This is based closely on the UFS ADIO layer and only differs
1130 in file creation, in order to allow the OST striping to be set.
1131 This is user-contributed code and not supported by CFS.
1136 Description: extended inode attributes (immutable, append-only) work improperly
1137 when 2.4 and 2.6 kernels are used on client/server or vice versa
1138 Details : Introduce kernel-independent values for these flags.
1140 Severity : enhancement
1143 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1144 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1145 granularity. Now, Fractional MB granularity is be supported,
1146 this is very useful for low memory system.
1148 Severity : enhancement
1150 Description: Getattr by fid
1151 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1152 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1153 have locks and avoids getattr rpc requests for it on every stat.
1156 Frequency : Always, for filesystems larger than 2TB
1158 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1159 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1160 filesystems larger than 2TB to be mounted with mballoc enabled.
1163 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1165 Description: filesystem corruption for non-standard kernels and very large OSTs
1166 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1167 filesystems larger than 2TB is mounted then the kernel will
1168 silently corrupt the start of the filesystem. CONFIG_LBD is
1169 enabled for all CFS-supported kernels, but the possibility of
1170 this happening with a modified kernel config exists.
1172 Severity : enhancement
1174 Description: add client O_DIRECT support for 2.6 kernels
1175 Details : It is now possible to do O_DIRECT reads and writes to files
1176 in the Lustre client mountpoint on 2.6 kernel clients.
1178 Severity : enhancement
1180 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1181 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1182 parallel, not waiting for response from every OST before sending
1183 a rpc to the next OST.
1188 Description: setattr vs write race when updating file timestamps
1189 Details : Client processes that update a file timestamp into the past
1190 right after writing to the file (e.g. tar) it is possible that
1191 the updated file modification time can be reset to the current
1192 time due to a race between processing the setattr and write RPC.
1194 Severity : enhancement
1196 Description: Bring 'lfs find' closer in line with regular Linux find.
1197 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1198 -print0 options and obtains all the needed info through the lustre
1201 Severity : enhancement
1203 Description: support up to 1024 configured devices on one node
1204 Details : change obd_dev array from statically allocated to dynamically
1205 allocated structs as they are first used to reduce memory usage
1210 Description: Flush dirty partially truncated pages during truncate
1211 Details : Immediatelly flush partially truncated pages in filter_setattr,
1212 this way we completely avoid having any pages in page cache on OST
1213 and can retire ugly workarounds during writes to flush such pages.
1218 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1219 Details : protect inode from truncation within vfs_unlink() context
1220 just take a reference before calling vfs_unlink() and release it
1221 when parent's i_sem is free.
1224 Frequency : always, if extents are used on OSTs
1226 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1227 Details : OSTs using the extents format would not zero the high 16 bits of
1228 the index physical block number. This is not a problem for any
1229 OST filesystems smaller than 16TB, and no kernels support ext3
1230 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1231 new/modified files) and can be fixed for existing filesystems
1232 with e2fsprogs-1.39-cfs1.
1237 Description: import connection selection may be incorrect if timer wraps
1238 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1239 import connections to be ignored if the 32-bit jiffies counter
1240 wraps. Use a 64-bit jiffies counter.
1243 Frequency : during server recovery
1245 Description: crash after server is denying duplicate export
1246 Details : If clients are resending connect requests to the server, the
1247 server refuses to allow a client to connect multiple times.
1248 Fixed a bug in the handling of this case.
1251 Frequency : very large clusters immediately after boot
1253 Description: LNET request buffers exhausted under heavy short-term load
1254 Details : If a large number of client requests are generated on a service
1255 that has previously never seen so many requests it is possible
1256 that the request buffer growth cannot keep up with the spike in
1257 demand. Instead of dropping incoming requests, they are held in
1258 the LND until the RPC service can accept more requests.
1261 Frequency : Sometimes during replay
1263 Description: Assertion failure in ll_local_open after replay.
1264 Details : If replay happened on an open request reply before we were able
1265 to set replay handler, reply will become not swabbed tripping the
1266 assertion in ll_local_open. Now we set the handler right after
1267 recognising of open request
1270 Frequency : very rare
1272 Description: kernel reports "badness in vsnprintf"
1273 Details : Reading from the "recovery_status" /proc file in small chunks
1274 may cause a negative length in lprocfs_obd_rd_recovery_status()
1275 call to vsnprintf() (which is otherwise harmless). Exit early
1276 if there is no more space in the output buffer.
1278 Severity : enhancement
1280 Description: clear OBD RPC statistics by writing to them
1281 Details : It is now possible to clear the OBD RPC statistics by writing
1282 to the "stats" file.
1287 Description: Client mtime is not the same on different clients after utimes
1288 Details : In some cases, the client was using the utimes() syscall on
1289 a file cached on another node. The clients now validate the
1290 ctime from the MDS + OSTs to determine which one is right.
1295 Description: Inability to activate failout mode
1296 Details : lconf script incorrectly assumed that in python string's numeric
1297 value is used in comparisons.
1300 Frequency : always with multiple stripes per file
1302 Description: Inefficient object allocation for mutli-stripe files
1303 Details : When selecting which OSTs to stripe files over, for files with
1304 a stripe count that divides evenly into the number of OSTs,
1305 the MDS is always picking the same starting OST for each file.
1306 Return the OST selection heuristic to the original design.
1311 Description: mount failures may take full timeout to return an error
1312 Details : Under some heavy load conditions it is possible that a
1313 failed mount can wait for the full obd_timeout interval,
1314 possibly several minutes, before reporting an error.
1315 Instead return an error as soon as the status is known.
1317 ------------------------------------------------------------------------------
1319 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
1321 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1322 INTEROPERATE with older versions automatically. Please read the
1323 user documentation before upgrading any part of a live system.
1324 * WARNING: Lustre networking configuration changes are required with
1325 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1328 * Support for kernels:
1329 2.6.9-22.0.2.EL (RHEL 4)
1330 2.6.5-7.244 (SLES 9)
1331 2.6.12.6 vanilla (kernel.org)
1334 Severity : enhancement
1335 Bugzilla : 7981/8208
1336 Description: Introduced Lustre Networking (LNET)
1337 Details : LNET is new networking infrastructure for Lustre, it includes
1338 a reorganized network configuration mode (see the user
1339 documentation for full details) as well as support for routing
1340 between different network fabrics. Lustre Networking Devices
1341 (LNDs) for the supported network fabrics have also been
1342 created for this new infrastructure.
1344 Severity : enhancement
1345 Description: Introduced Access control lists
1346 Details : clients can set ACLs on files and directories in order to have
1347 more fine-grained permissions than the standard Unix UGO+RWX.
1348 The MDS must be started with the "-o acl" mount option.
1350 Severity : enhancement
1351 Description: Introduced filesystem quotas
1352 Details : Administrators may now establish per-user quotas on the
1355 Severity : enhancement
1357 Description: Configuration change for the XT3
1358 The PTLLND is now used to run Lustre over Portals on the XT3
1359 The configure option(s) --with-cray-portals are no longer used.
1360 Rather --with-portals=<path-to-portals-includes> is used to
1361 enable building on the XT3. In addition to enable XT3 specific
1362 features the option --enable-cray-xt3 must be used.
1367 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1368 Details : A race between allocating a new llcd and re-getting the llcd_lock
1369 allowed another thread to grab newly-allocated llcd.
1371 Severity : enhancement
1373 Description: 2.6 OST async journal commit and locking fix to improve performance
1374 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1375 2.6 kernels are now async as they already were in 2.4 kernels so
1376 that they can commit concurrently with the network bulk transfer.
1377 For block-allocated files the filter allocation semaphore is held
1378 to avoid filesystem fragmentation during allocation. BKL lock
1379 removed for 2.6 xattr operations where it is no longer needed.
1384 Description: lconf incorrectly determined whether two IP networks could talk
1385 Details : In some more complicated routing and multiple-network
1386 configurations, lconf will avoid trying to make a network
1387 connection to a disjoint part of the IP space. It was doing the
1388 math incorrectly for one set of cases.
1393 Description: Fix for potential infinite loop processing records in an llog.
1394 Details : If an llog record is corrupted/zeroed, it is possible to loop
1395 forever in llog_process(). Validate the llog record length
1396 and skip the remainder of the block on error.
1399 Frequency : occasional (liblustre only)
1401 Description: liblustre could not open files whose last component is a symlink
1402 Details : sysio_path_walk() would incorrectly pass the open intent to
1403 intermediate path components.
1406 Frequency : rare (liblustre only with non-standard tuning)
1407 Bugzilla : 7201 (7350)
1408 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1409 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1410 after it has been released, only do this for referenced locks
1412 Severity : enhancement
1414 Description: specifying an (invalid) directory default stripe_size of -1
1415 would reset the directory default striping
1416 Details : stripe_size -1 was used internally to signal directory stripe
1417 removal, now use "all default" to signal dir stripe removal
1418 as a directory striping of "all default" is not useful
1421 Frequency : common for large clusters running liblustre clients
1423 Description: doing an ls when liblustre clients are running is slow
1424 Details : sending a glimpse AST to a liblustre client waits for every AST
1425 to time out, as liblustre clients will not respond. Since they
1426 cannot cache data we refresh the OST lock LVB from disk instead.
1428 Severity : enhancement
1430 Description: doing an ls at the same time as file IO can be slow
1431 Details : enqueue and other "small" requests can be blocked behind many
1432 large IO requests. Create a new OST IO portal for non-IO
1433 requests so they can be processed faster.
1436 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1438 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1439 created file with mode 000
1440 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1441 allow an ftruncate() as a non-root user to a file with mode 000.
1442 Lustre now respects this flag to disable mode checks when
1443 truncating a file owned by the user
1446 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1449 Description: Revoking locks from clients that went dead or catatonic might take
1451 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1452 cancellation of such locks instant on servers without waiting for
1453 any reply from clients. Clients drops these locks when cancel
1454 notification from server is received without replying.
1457 Frequency : liblustre-only, when liblustre client dies or becomes busy
1459 Description: Doing ls on Linux clients can take a long time with active
1461 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1462 granting such locks to it in the first place if possible. Locks
1463 are taken by proxy on the OST during the read or write and
1464 dropped immediately afterward. Add connect flags handling, do
1465 not grant locks to liblustre clients for glimpse ASTs.
1467 Severity : enhancement
1469 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1470 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1471 sequential reads and behaves badly with applications doing random
1472 reads. Improve it by reading ahead at least read region, and
1473 avoiding excessive large RPC for small reads.
1475 Severity : enhancement
1477 Description: Creating more than 1000 files for a single job may cause a load
1478 imbalance on the OSTs if there are also a large number of OSTs.
1479 Details : qos_prep_create() uses an OST index reseed value that is an
1480 even multiple of the number of available OSTs so that if the
1481 reseed happens in the middle of the object allocation it will
1482 still utilize the OSTs as uniformly as possible.
1487 Description: OST or MDS may oops in ping_evictor_main()
1488 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1489 but doesn't restart at beginning of obd_exports_timed list
1492 Severity : enhancement
1494 Description: improve by-nid export eviction on the MDS and OST
1495 Details : allow multiple exports with the same NID to be evicted at one
1496 time without re-searching the exports list.
1499 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1501 Description: MDS may oops in groups_free()
1502 Details : in rare race conditions a newly allocated group_info struct is
1503 freed again, and this can be NULL. The 2.4 compatibility code
1504 for groups_free() checked for a NULL pointer, but 2.6 did not.
1507 Frequency : common for liblustre clients doing little filesystem IO
1508 Bugzilla : 9352, 7313
1509 Description: server may evict liblustre clients accessing contended locks
1510 Details : if a client is granted a lock or receives a completion AST
1511 with a blocking AST already set it would not reply to the AST
1512 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1515 Frequency : lfs setstripe, only systems with more than 160 OSTs
1517 Description: unable to set striping with a starting offset beyond OST 160
1518 Details : llapi_create_file() incorrectly limited the starting stripe
1519 index to the maximum single-file stripe count.
1522 Frequency : LDAP users only
1524 Description: lconf did not handle in-kernel recovery with LDAP properly
1525 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1527 Severity : enhancement
1529 Description: bind OST threads to NUMA nodes to improve performance
1530 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1531 node and do their allocations there to localize memory access
1533 Severity : enhancement
1535 Description: llmount can determine client NID directly from Myrinet (GM)
1536 Details : the client NID code from gmnalnid was moved directly into
1537 llmount, removing the need to use this or specifying the
1538 client NID explicitly when mounting GM clients with zeroconf
1541 Frequency : if client is started with down MDS
1543 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1544 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1545 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1546 ptlrpc_import_delayed_req() because we didn't check if the
1547 request was interrupted, and we also didn't break out of the
1548 event loop if there was no timeout
1553 Description: data loss during non-page-aligned writes to a single file from
1554 both multiple nodes and multiple threads on one node at same time
1555 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1556 inconsistency led to false short-reads, that were cached and later
1557 used by ->prepare_write() to fill in partially written page,
1558 leading to data loss.
1561 Frequency : always, if lconf --abort_recovery used
1563 Description: lconf --abort_recovery fails with 'Operation not supported'
1564 Details : lconf was attempting to abort recovery on the MDT device and not
1567 Severity : enhancement
1569 Description: remove cleanup logs
1570 Details : replace lconf-generated cleanup logs with lustre internal
1571 cleanup routines. Eliminates the need for client-cleanup and
1574 Severity : enhancement
1576 Description: add support for EAs (user and system) on lustre filesystems
1577 Details : it is now possible to store extended attributes in the Lustre
1578 client filesystem, and with the user_xattr mount option it
1579 is possible to allow users to store EAs on their files also
1581 Severity : enhancement
1583 Description: Add possibility (config option) to show minimal available OST free
1585 Details : When compiled with --enable-mindf configure option, statfs(2)
1586 (and so, df) will return least minimal free space available from
1587 all OSTs as amount of free space on FS, instead of summary of
1588 free spaces of all OSTs.
1590 Severity : enhancement
1592 Description: do not expand extent locks acquired on OST-side
1593 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1594 by server: they are not cached anyway.
1597 Frequency : when mmap is used/binaries executed from Lustre
1599 Description: Unmmap pages before throwing them away from read cache.
1600 Details : llap_shrink cache now attempts to unmap pages before discarding
1601 them (if unmapping failed - do not discard). SLES9 kernel has
1602 extra checks that trigger if this unmapping is not done first.
1607 Description: lconf didn't resolve symlinks before checking to see whether a
1608 given mountpoint was already in use
1611 Frequency : when migrating failover services
1612 Bugzilla : 6395, 9514
1613 Description: When migrating a subset of services from a node (e.g. failback
1614 from a failover service node) the remaining services would
1615 time out and evict clients.
1616 Details : lconf --force (implied by --failover) sets the global obd_timeout
1617 to 5 seconds in order to quickly disconnect, but this caused
1618 other RPCs to time out too quickly. Do not change the global
1619 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1621 Severity : enhancement
1622 Frequency : if MDS is started with down OST
1623 Bugzilla : 9439,5706
1624 Description: Allow startup/shutdown of an MDS without depending on the
1625 availability of the OSTs.
1626 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1627 Add appropriate locking and lov-osc refcounts for safe
1628 cleaning. Add osc abort_inflight calls in case the
1629 synchronize never started.
1632 Frequency : occasional (Cray XT3 only)
1634 Description: root not authorized to access files in CRAY_PORTALS environment
1635 Details : The client process capabilities were not honoured on the MDS in
1636 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1637 been accessed by an authorized user then root was able to access
1638 the file on the local client also. The root user capabilities
1639 are now allowed on the MDS, as this environment has secure UID.
1642 Frequency : occasional
1644 Description: ldiskfs "too long searching" message happens too often
1645 Details : A debugging message (otherwise harmless) prints too often on
1646 the OST console. This has been reduced to only happen when
1647 there are fragmentation problems on the filesystem.
1652 Description: Division by zero in statfs when all OSCs are inactive
1653 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
1654 lov_statfs divides by value returned by lov_get_stripecnt().
1658 Bugzilla : 9489, 3273
1659 Description: First write from each client to each OST was only 4kB in size,
1660 to initialize client writeback cache, which caused sub-optimal
1661 RPCs and poor layout on disk for the first writen file.
1662 Details : Clients now request an initial cache grant at (re)connect time
1663 and so that they can start streaming writes to the cache right
1664 away and always do full-sized RPCs if there is enough data.
1665 If the OST is rebooted the client also re-establishes its grant
1666 so that client cached writes will be honoured under the grant.
1671 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
1672 Details : Now I/O RPCs go to different portal number and (presumably) fast
1673 lock requests (and glimses) and other RPCs get their own service
1674 threads pool that should be able to service those RPCs
1677 Severity : enhancement
1679 Description: Ability to exchange lustre version between client and servers and
1680 issue warnings at client side if client is too old. Also for
1681 liblustre clients there is ability to refuse connection of too old
1683 Details : New 'version' field is added to connect data structure that is
1684 filled with version info. That info is later checked by server and
1688 Frequency : rare, liblustre only.
1689 Bugzilla : 9296, 9581
1690 Description: Two simultaneous writes from liblustre at offset within same page
1691 might proceed at the same time overwriting eachother with stale
1693 Details : I/O lock withing llu_file_prwv was released too early, before data
1694 actually was hitting the wire. Extended lock-holding time until
1695 server acknowledges receiving data.
1698 Frequency : extremely rare. Never observed in practice.
1700 Description: avoid generating lustre_handle cookie of 0.
1701 Details : class_handle_hash() generates handle cookies by incrementing
1702 global counter, and can hit 0 occasionaly (this is unlikely, but
1703 not impossible, because initial value of cookie counter is
1704 selected randonly). Value of 0 is used as a sentinel meaning
1705 "unassigned handle" --- avoid it. Also coalesce two critical
1706 sections in this function into one.
1708 Severity : enhancement
1710 Description: allow liblustre clients to delegate truncate locking to OST
1711 Details : To avoid overhead of locking, liblustre client instructs OST to
1712 take extent lock in ost_punch() on client's behalf. New connection
1713 flag is added to handle backward compatibility.
1715 Severity : enhancement
1716 Bugzilla : 4928, 7341, 9758
1717 Description: allow number of OST service threads to be specified
1718 Details : a module parameter allows the number of OST service threads
1719 to be specified via "options ost ost_num_threads={N}" in the
1720 OSS's /etc/modules.conf or /etc/modprobe.conf.
1724 Bugzilla : 6146, 9635, 9895
1725 Description: servers crash with bad pointer in target_handle_connect()
1726 Details : In rare cases when a client is reconnecting it was possible that
1727 the connection request was the last reference for that export.
1728 We would temporarily drop the export reference and get a new
1729 one, but this may have been the last reference and the export
1730 was just destroyed. Get new reference before dropping old one.
1732 Severity : enhancement
1733 Frequency : if client is started with failover MDS
1735 Description: Allow multiple MDS hostnames in the mount command
1736 Details : Try to read the configuration from all specified MDS
1737 hostnames during a client mount in case the "primary"
1740 Severity : enhancement
1742 Description: Stop sending data to evicted clients as soon as possible.
1743 Details : Check if the client we are about to send or are sending data to
1744 was evicted already. (Check is done every second of waiting,
1745 for which l_wait_event interface was extended to allow checking
1746 of exit condition at specified intervals).
1749 Frequency : rare, normally only when NFS exporting is done from client
1751 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
1753 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
1756 Frequency : always, if srand() is called before liblustre initialization
1758 Description: Liblustre uses system PRNG disturbing its usage by user application
1759 Details : Introduce internal to lustre fast and high-quality PRNG for
1760 lustre usage and make liblustre and some other places in generic
1761 lustre code to use it.
1763 Severity : enhancement
1764 Bugzilla : 9477, 9557, 9870
1765 Description: Verify that the MDS configuration logs are updated when xml is
1766 Details : Check if the .xml configuration logs are newer than the config
1767 logs stored on the MDS and report an error if this is the case.
1768 Request --write-conf, or allow starting with --old_conf.
1770 Severity : enhancement
1772 Description: Handle symlinks in the path when checking if Lustre is mounted.
1773 Details : Resolve intermediate symlinks when checking if a client has
1774 mounted a filesystem to avoid duplicate client mounts.
1779 Description: lconf can hit an error exception but still return success.
1780 Details : The lconf command catches the Command error exception at the top
1781 level script context and will exit with the associated exit
1782 status, but doesn't ensure that this exit status is non-zero.
1787 Description: failure of ptlrpc thread startup can cause oops
1788 Details : Starting a ptlrpc service thread can fail if there are a large
1789 number of threads or the server memory is very fragmented.
1790 Handle this without oopsing.
1793 Frequency : always, only if liblustre and non-default acceptor port was used
1795 Description: liblustre cannot connect to servers with non-default acceptor port
1796 Details : tcpnal_set_default_params() was not called and was therefore
1797 ignoring the environment varaible TCPNAL_PORT, as well as other
1798 TCPNAL_ environment variables
1803 Description: two objects could be created on the same OST for a single file
1804 Details : If an OST is down, in some cases it was possible to create two
1805 objects on a single OST for a single file. No problems other
1806 than potential performance impact and spurious error messages.
1810 Bugzilla : 5681, 9562
1811 Description: Client may oops in ll_unhash_aliases
1812 Details : Client dcache may become inconsistent in race condition.
1813 In some cases "getcwd" can fail if the current directory is
1819 Description: Inode refcounting problems in NFS export code
1820 Details : link_raw functions used to call d_instantiate without obtaining
1821 extra inode reference first.
1825 Bugzilla : 9942, 9903
1826 Description: Referencing freed requests leading to crash, memleaks with NFS.
1827 Details : We used to require that call to ll_revalidate_it was always
1828 followed by ll_lookup_it. Also with revalidate_special() it is
1829 possible to call ll_revalidate_it() twice for the same dentry
1830 even if first occurence returned success. This fix changes semantic
1831 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
1832 reference on a request referred from the intent.
1833 ll_intent_release() then releases such a request.
1836 Frequency : rare, normally benchmark loads only
1838 Description: unlinked inodes were kept in memory on the client
1839 Details : If a client is repeatedly creating and unlinking files it
1840 can accumulate a lot of stale inodes in the inode slab cache.
1841 If there is no other client load running this can cause the
1842 client node to run out of memory. Instead flush old inodes
1843 from client cache that have the same inode number as a new inode.
1846 Frequency : SLES9 2.6.5 kernel and long filenames only
1847 Bugzilla : 9969, 10379
1848 Description: utime reports stale NFS file handle
1849 Details : SLES9 uses out-of-dentry names in some cases, which confused
1850 the lustre dentry revalidation. Change it to always use the
1854 Frequency : rare, unless heavy write-truncate concurrency is continuous
1855 Bugzilla : 4180, 6984, 7171, 9963, 9331
1856 Description: OST becomes very slow and/or deadlocked during object unlink
1857 Details : filter_destroy() was holding onto the parent directory lock
1858 while truncating+unlinking objects. For very large objects this
1859 may block other threads for a long time and slow overall OST
1860 responsiveness. It may also be possible to get a lock ordering
1861 deadlock in this case, or run out of journal credits because of
1862 the combined truncate+unlink. Solution is to do object truncate
1863 first in one transaction without parent lock, and then do the
1864 final unlink in a new transaction with the parent lock. This
1865 reduces the lock hold time dramatically.
1868 Frequency : rare, 2.4 kernels only
1870 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
1871 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
1872 before unmounting their filesystem in order to avoid deadlock.
1873 The kernel_locked() function in 2.4 kernels only checks whether
1874 the kernel lock is held, not whether it is this process that is
1875 holding it as 2.6 kernels do.
1880 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
1881 Details : The client ptlrpc code may be trying to reconnect to a down
1882 server before a previous connection attempt has timed out.
1883 Increase the reconnect interval to be longer than the connection
1884 timeout interval to avoid sending duplicate connections to
1888 Frequency : echo_client brw_test command
1890 Description: fix echo_client to work with OST preallocated code
1891 Details : OST preallocation code (5137) didn't take echo_client IO path
1892 into account: echo_client calls filter methods outside of any
1893 OST thread and, hence, there is no per-thread preallocated
1894 pages and buffers to use. Solution: hijack pga pages for IO. As
1895 a byproduct, this avoids unnecessary data copying.
1899 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
1900 Description: Client can oops in mdc_commit_close() after open replay
1901 Details : It was possible for the MDS to return an open request with no
1902 transaction number in mds_finish_transno() if the client was
1903 evicted, but without actually returning an error. Clients
1904 would later try to replay that open and may trip an assertion
1905 Simplify the client close codepath, and always return an error
1906 from the MDS in case the open is not successful.
1909 Frequency : rare, 2.6 OSTs only
1911 Description: OST may deadlock under high load on fragmented files
1912 Details : If there was a heavy load and highly-fragmented OST filesystems
1913 it was possible to have all the OST threads deadlock waiting on
1914 allocation of biovecs, because the biovecs were not released
1915 until the entire RPC IO was completed. Instead, release biovecs
1916 as soon as they are complete to ensure forward IO progress.
1918 Severity : enhancement
1920 Description: Support for specifying external journal device at mount
1921 Details : If an OST or MDS device is formatted with an external journal
1922 device, this device major/minor is stored in the ext3 superblock
1923 and may not be valid for failover. Allow detecting and
1924 specifying the external journal at mount time.
1929 Description: Mounting an MDS with pending unlinked files may cause oops
1930 Details : target_finish_recovery() calls mds_postrecov() which returned
1931 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
1932 considers this an error and immediately begins cleaning up the
1933 lov, just after starting the mds_lov process
1935 Severity : enhancement
1937 Description: Implement 'lfs df' to report actual free space on per-OST basis
1938 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
1939 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
1940 usage of objects; '-h' to report in human readable format.
1942 ------------------------------------------------------------------------------
1944 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
1951 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
1952 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
1953 at startup during orphan unlinking. Instead, skip buddy update
1954 before setup, as it will be regenerated after recovery is complete.
1959 Description: If an OST is inactive, its locks might reference stale inodes.
1960 Details : lov_change_cbdata() must iterate over all namespaces, even if
1961 they are inactive to clear inode references from the lock.
1963 Severity : enhancement
1964 Frequency : occasional, if non-standard max_dirty_mb used
1966 Description: Client will block write RPCs if not enough grant
1967 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
1968 then the client will block writes while waiting for another RPC
1969 to complete instead of consuming its dirty limit. With change
1970 we get improved performance when max_dirty_mb is small.
1972 Severity : enhancement
1973 Bugzilla : 3389, 6253
1974 Description: Add support for supplementary groups on the MDS.
1975 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
1976 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
1977 lookups for user supplementary groups into a cache.
1981 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
1982 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
1983 for liblustre O_EXCL is enforced by MDS.
1987 Description: readdir never returns NULL in liblustre.
1988 Details : Corrected llu_iop_getdirentries logic, to return offset of next
1989 dentry in struct dirent.
1993 Frequency : liblustre only, depends on application IO pattern
1994 Description: liblustre clients evicted if not contacting servers
1995 Details : Don't put liblustre clients into the ping_evictor list, so
1996 they will not be evicted by the pinger ever.
1998 Severity : enhancement
2000 Description: Add ability to evict clients by NID from MDS.
2001 Details : By echoing "nid:$NID" string into
2002 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2003 $NID would be instantly evicted from this MDS and from all active
2004 OSTs connected to it.
2008 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2009 Details : lookup_it_finish() used to query file size from OSTs that was not
2014 Description: service threads change working directory to that of init
2015 Details : Starting lustre service threads may pin the working directory
2016 of the parent thread, making that filesystem busy. Threads
2017 now change to the working directory of init to avoid this.
2021 Frequency : during shutdown only
2022 Description: shutdown with a failed MDS or OST can cause unmount to hang
2023 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2028 Frequency : 2.6 only
2029 Description: chmod/chown may include an extra supplementary group
2030 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2031 supplementary group and if none is specified this is used.
2034 Bugzilla : 5479 (6816)
2036 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2037 Details : If lookup is in progress on a file that is unlinked we might try
2038 to revalidate the inode and fail in revalidate after lookup is
2039 complete and ll_file_open() enqueues the open again but
2040 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2043 Frequency : always, if lconf --abort_recovery used
2045 Description: lconf --abort_recovery fails with 'Operation not supported'
2046 Details : lconf was attempting to abort recovery on the MDT device and not
2049 ------------------------------------------------------------------------------
2051 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2056 Frequency : rare (only unsupported configurations with a node running as an
2058 Bugzilla : 6514, 5137
2059 Description: Mounting a Lustre file system on a node running as an OST could
2061 Details : OSTs now preallocates memory needed to write out data at
2062 startup, instead of when needed, to avoid having to
2063 allocate memory in possibly low memory situations.
2064 Specifically, if the file system is mounted on on OST,
2065 memory pressure could force it to try to write out data,
2066 which it needed to allocate memory to do. Due to the low
2067 memory, it would be unable to do so and the node would
2068 become unresponsive.
2070 Severity : enhancement
2072 Description: Addition of lconf --service command line option
2073 Details : lconf now accepts a '--service <arg>' option, which is
2074 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2076 Severity : enhancement
2078 Description: Failover mode is now the default for OSTs.
2079 Details : By default, OSTs will now run in failover mode. To return to
2080 the old behaviour, add '--failout' to the lmc line for OSTs.
2082 Severity : enhancement
2084 Description: Health checks are now provided for MDS and OSTs
2085 Details : Additional detailed health check information on MSD and OSTs
2086 is now provided through the procfs health_check value.
2089 Frequency : occasional, depends on IO load
2091 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2092 after numerous create/delete cycles
2093 Details : The ext3 inode allocation policy would not allocate new inodes
2094 very well on the OSTs because there are no new directories
2095 being created. Instead we look for groups with free space if
2096 the parent directories are nearly full.
2101 Description: Network or server problems during mount may cause partially
2102 mounted clients instead of returning an error.
2103 Details : The config llog parsing code may overwrite the error return
2104 code during mount error handling, returning success instead
2110 Description: MDS can fail to allocate large reply buffers
2111 Details : After long uptimes the MDS can fail to allocate large reply
2112 buffers (e.g. zconf client mount config records) due to memory
2113 fragmentation or consumption by the buffer cache. Preallocate
2114 some large reply buffers so that these replies can be sent even
2115 under memory pressure.
2119 Frequency : rare (liblustre)
2120 Description: fsx running with liblustre complained that using truncate() to
2121 extend the file doesn't work. This patch corrects that issue.
2122 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2123 ATTR_SIZE and lsm use in llu_setattr_raw.
2127 Frequency : rare, only 2.6 kernels
2128 Description: Unusual file access patterns on the MDS may result in inode
2129 data being lost in very rare circumstances.
2130 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2131 code in the 2.6 kernel caused the inode and/or EA data not to
2132 be read from disk, causing single-file corruption.
2136 Frequency : rare, only 2.6 filesystems using extents
2137 Description: Heavy concurrent write and delete load may cause data corruption.
2138 Details : It was possible under high-load situations to have an extent
2139 metadata block in the block device cache from a just-unlinked
2140 file overwrite a newly-allocated data block. We now unmap any
2141 metadata buffers that alias just-allocated data blocks.
2145 Frequency : filesystems with default stripe_count larger than 77
2146 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2147 Details : lconf specifies an inode size of 4096 bytes when the default
2148 stripe_count is larger than 77. This conflicts with the default
2149 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2150 this case to avoid pinning too much memory for large EAs.
2152 ------------------------------------------------------------------------------
2154 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2159 Frequency : rare (extremely heavy IO load with hundreds of clients)
2161 Description: Client is evicted, gets IO error writing to file
2162 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2163 caused clients to be evicted for AST timeouts. The fixes in
2164 bug 5192 mean we no longer need to have such short AST timeouts
2165 so ldlm_timeout has been increased.
2168 Frequency : occasional during --force or --failover shutdown under load
2169 Bugzilla : 5949, 4834
2170 Description: Server oops/LBUG if stopped with --force or --failover under load
2171 Details : a collection of import/export refcount and cleanup ordering
2172 issues fixed for safer force cleanup
2175 Frequency : only filesystems larger than 120 OSTs
2176 Bugzilla : 5990, 6223
2177 Description: lfs getstripe would oops on a very large filesystem
2178 Details : lov_getconfig used kfree on vmalloc'd memory
2181 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2182 Bugzilla : 6242, 6243
2183 Description: reading from files that had been truncated to a non-zero size
2184 but never opened returned no data
2185 Details : ll_file_read() reads zeros from no-object files to EOF
2190 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2191 Details : There's little harm in aborting MDS/OSS recovery and letting it
2192 try again, so I removed the LASSERT and return an error instead.
2194 Severity : enhancement
2196 Description: New debugging infrastructure for tracking down data corruption
2197 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2198 (b) cover more of the client-side code path, and (c) try to narrow
2199 down where problems occurred
2203 Bugzilla : 3819, 4364, 4397, 6313
2204 Description: Racing close and eviction MDS could cause assertion in mds_close
2205 Details : It was possible to get multiple mfd references during close and
2206 client eviction, leading to one thread referencing a freed mfd.
2208 Severity: : enhancement
2209 Bugzilla : 3262, 6359
2210 Description: Attempts to reconnect to servers are now more aggressive.
2211 Details : This builds on the enhanced upcall-less recovery that was added
2212 in 1.4.2. When trying to reconnect to servers, clients will
2213 now try each server in the failover group every 10 seconds. By
2214 default, clients would previously try one server every 25 seconds.
2219 Description: After recovery, certain operations trigger a failed
2220 assertion on a client.
2221 Details : Failing over an mds, using lconf -d --failover, while a
2222 client was doing a readdir() call would cause the client to
2223 LBUG after recovery completed and the readdir() was resent.
2225 Severity : enhancement
2227 Description: Default groups are now added by lconf
2228 Details : You can now run lconf --group <servicename> without having to
2229 manually add groups with lmc.
2232 Frequency : occasional
2234 Description: Nodes with an elan id of 0 trigger a failed assertion
2237 Frequency : always when accessing e.g. tty/console device nodes
2239 Description: tty and some other devices nodes cannot be used on lustre
2240 Details : file's private_data field is used by device data and lustre
2241 values in there got lost. New field was added to struct file to
2242 store fs-specific private data.
2245 Frequency : when exporting Lustre via NFS
2247 Description: NFSD failed occasionally when looking up a path component
2248 Details : NFSD is looking up ".." which was broken in ext3 directories
2249 that had grown large enough to become hashed.
2252 Frequency : Clusters with multiple interfaces not on the same subnet
2254 Description: Nodes will repeatedly try to reconnect to an interface which it
2255 cannot reach and report an error to the log.
2256 Details : Extra peer list entries will be created by lconf with some peers
2257 unreachable. lconf now validates the peer before adding it.
2260 Frequency : Only if a default stripe is set on the filesystem root.
2262 Description: Setting a default stripe on the filesystem root prevented the
2263 filesystem from being remounted.
2264 Details : The client was sending extra request flags in the root getattr
2265 request and did not allocate a reply buffer for the dir EA.
2268 Frequency : occasional, higher if lots of files are accessed by one client
2269 Bugzilla : 6159, 6097
2270 Description: Client trips assertion regarding lsm mismatch/magic
2271 Details : While revalidating inodes the VFS looks up inodes with ifind()
2272 and in rare cases can find an inode that is being freed.
2273 The ll_test_inode() code will free the lsm during ifind()
2274 when it finds an existing inode and then the VFS later attaches
2275 this free lsm to a new inode.
2279 Bugzilla : 6422, 7030
2280 Description: MDS deadlock between mkdir and client eviction
2281 Details : Creating a new file via mkdir or mknod (starting a transaction
2282 and getting the ns lock) can deadlock with client eviction
2283 (gets ns lock and trying to finish a synchronous transaction).
2286 Frequency : occasional
2287 Description: While starting a server, the fsfilt_ext3 module could not be
2289 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2290 kernels. Previously, lconf would still use the ext3 name
2291 when trying to load modules. Now, it will correctly use
2292 ext3 on 2.4 and ldiskfs on 2.6.
2294 Severity : enhancement
2295 Description: The default stripe count has been changed to 1
2296 Details : The interpretation of the default stripe count (0, to lfs
2297 or lmc) has been changed to mean striping across a single
2298 OST, rather than all available. For general usage we have
2299 found a stripe count of 1 or 2 works best.
2301 Severity : enhancement
2302 Description: Add support for compiling against Cray portals.
2303 Details : Conditional compiling for some areas that are different
2307 Frequency : occasional
2308 Bugzilla : 6409, 6834
2309 Description: Creating files with an explicit stripe count may lead to
2310 a failed assertion on the MDS
2311 Details : If some OSTs are full or unavailable, creating files may
2312 trigger a failed assertion on the MDS. Now, Lustre will
2313 try to use other servers or return an error to the
2317 Frequency : occasional
2319 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2320 caused lock timeout during readahead (since 1.4.2).
2321 Details : Processes doing readahead might match a lock that hasn't been
2322 granted yet if there are overlapping and conflicting lock
2323 requests. The readahead process waits on ungranted lock
2324 (original lock is CBPENDING), while OST waits for that process
2325 to cancel CBPENDING read lock and eventually evicts client.
2327 Severity : enhancement
2329 Description: Initial enabling of flock support for clients
2330 Details : Implements fcntl advisory locking and file status functions.
2331 This feature is provided as an optional mount flag (default
2332 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2333 locking are implemented yet, and those that are are not guaranteed
2334 to be completely correct in production environments.
2335 mount -t lustre -o [flock|noflock] ...
2338 Frequency : occasional
2340 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2341 assertion in the ext3 JBD (journaling) layer.
2342 Details : The b_committed_data struct is protected by the big kernel lock
2343 in 2.4 kernels, serializing journal_commit_transaction() and
2344 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2345 access to this struct. In 2.6 kernels there is finer grained
2346 locking to improve SMP performance of the JBD layer.
2350 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2352 -----------------------------------------------------------------------------
2354 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2356 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2357 versions of Lustre. Please update all servers and clients to
2358 version 1.4.2 or later at the same time. You must also run
2359 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2360 to update the configuration logs.
2362 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2363 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2364 - swab llog records of type '0' so we get proper header size/idx (5861)
2365 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2366 - fix rename of one directory over another leaking an inode (5953)
2367 - avoid SetPageDirty on 2.6 (5981)
2368 - don't re-add just-being-destroyed locks to the waiting list (5653)
2369 - when creating new directories, inherit the parent's custom
2370 striping settings if present parent (3048)
2371 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2372 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2373 - don't hold client locks on temporary worklist from l_lru (5666)
2374 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2375 - automatically evict dead clients (5921)
2376 - Update file size properly in create+truncate+fstat case (6196)
2377 - Do not unhash mountpoint dentries, do not allow removal of
2379 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2380 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2381 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2382 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2383 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2384 - Correctly handle reads of files with no objects (6243)
2385 - lctl recover will also mark a device active if deactivate used (5933)
2387 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2388 - allow --write-conf on an MDS with different nettype than client (5619)
2389 - don't write config llogs to MDS for mounts not from that MDS (5617)
2390 - lconf should create multiple TCP connections from a client (5201)
2391 - init scripts are now turned off by default; run chkconfig --on
2392 lustre and chkconfig --on lustrefs to use them
2393 - upcalls are no longer needed for clients to recover to failover
2395 - add --abort-recovery option to lconf to abort recovery on device
2397 - add support for an arbitrary number of OSTs (3026)
2398 - Quota support protocol changes.
2399 - forward compatibility changes to wire structs (6007)
2400 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2401 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2402 - improved "lustre status" script
2403 - initialize blocksize for non-regular files (6062)
2404 - added --disable-server and --disable-client configure options (5782)
2405 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2406 - Vanilla 2.4.29 support
2407 - increase maximum number of obd devices to 520 (6242)
2408 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2409 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2410 SLES 9 2.6.5 kernels
2411 - limit stripes per file to 160 (the maximum EA size) (6093)
2413 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2416 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2417 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2418 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2419 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2420 - small ext3 extents cleanups and fixes (5733)
2421 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2422 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2423 - avoid needless client->OST connect, fix handle mismatch (5317)
2424 - fix DLM error path that led to out-of-sync client, long delays (5779)
2425 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2426 - fix several locking issues related to i_size (5492,5624,5654,5672)
2427 - don't move pending lock onto export if it is already evicted (5683)
2428 - fix kernel oops when creating .foo in unlinked directory (5548)
2429 - fix deadlock in obdfilter statistics vs. object create (5811)
2430 - use time_{before,after} to avoid timer jiffies wrap (5882)
2431 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2432 - Do not leak request if server was not able to process it (5154)
2433 - If mds_open unable to find parent dir, make that negative lookup(5154)
2434 - don't create new directories with extent-mapping (5909, 5936)
2436 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2437 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2438 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2439 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2440 - support for RHEL4 kernel on i686 (5773)
2441 - provide error messages when incompatible logs are encountered (5898)
2443 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2444 * version 1.4.0.10 (1.4.1 release candidate 1)
2446 - don't keep a lock reference when lock is not granted (4238)
2447 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2448 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2449 - drop import inflight refcount on signal_completed_replay error (5255)
2450 - unlock page after async write error during send (3677)
2451 - handle missing objects in filter_preprw_read properly (5265)
2452 - no transno return for symlink open, don't save no-trasno open (3440)
2453 - don't try to complete elan receive that already failed (4012)
2454 - free RPC server reply state on error (5406)
2455 - clean up thread from ptlrpc_start_thread() on error (5160)
2456 - readahead could read extra page into cache that wasn't ejected (5388)
2457 - prevent races in class_attach/setup/cleanup/detach (5260)
2458 - don't dereference de->d_inode after l_dput of de (5458)
2459 - use "int" for stripe value returned from lock_to_stripe (5544)
2460 - mballoc allocation and error-checking fixes in 2.6 (5504)
2461 - block device patches to fix I/O request sizes in 2.6 (5482)
2462 - look up hostnames for IB nals (5602)
2463 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2464 - don't start multiple acceptors for the same port (5277)
2465 - fix incorrect LASSERT in mds_getattr_name (5635)
2466 - export a proc file for general "ping" checking (5628)
2467 - fix "lfs check" to not block when the MDS is down (5628)
2469 - service request history (4965)
2470 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2471 - create an "evict_client" /proc entry on OSTs, like the MDS has
2472 - fix mount usage message, return errors per mount(8) (5168)
2473 - change grep [] to grep "[]" in tests so they work in more UMLs
2474 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2475 - remove ancient LOV_MAGIC_V0 EA support (5047)
2476 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2477 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2478 - allow readahead allocations to fail when low on memory (5383)
2479 - mmap locking landed again, after considerable improvement (2828)
2480 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2482 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2485 - send OST transaction number in read/write reply to free req (4966)
2486 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2487 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2488 - the watchdog thread now runs as interruptible (5246)
2489 - flock/lockf fixes (but it's still disabled, pending 5135)
2490 - don't use EXT3 constants in llite code (5094)
2491 - memory shortage at startup could cause assertion (5176)
2493 - reorganization of lov code
2494 - single portals codebase
2496 - add extents/mballoc support (5025)
2497 - direct I/O reads in the obdfilter (4048)
2498 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2500 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2503 - send OST transaction number in read/write reply to free req (4966)
2504 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2505 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2506 - the watchdog thread now runs as interruptible (5246)
2507 - handle missing objects in filter_preprw_read properly (5265)
2508 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2509 - drop import inflight refcount on signal_completed_replay error (5255)
2510 - unlock page after async write error during send (3677)
2511 - return original error code on reconstructed replies (3761)
2512 - no transno return for symlink open, don't save no-trasno open (3440)
2514 - add pid to ldlm debugging output (4922)
2515 - bump the watchdog timeouts -- we can't handle 30sec yet
2516 - extra debugging for orphan dentry/inode bug (5259)
2518 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2521 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2522 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2523 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2524 - fix NULL dereference in /proc/sys/portals/routes (4827)
2525 - allow failed mdc_close() operations to be interrupted (4561)
2526 - stop precreate on OST before MDS would time out on it (4778)
2527 - don't send partial-page writes before EOF from client (4410)
2528 - discard client grant for sub-page writes on large-page clients (4520)
2529 - don't free dentries not owned by NFS code, check generation (4806)
2530 - fix lsm leak if mds_create_objects() fails (4801)
2531 - limit debug_daemon file size, always print CERROR messages (4789)
2532 - use transno after validating reply (3892)
2533 - process timed out requests if import state changes (3754)
2534 - update mtime on OST during writes, return in glimpse (4829)
2535 - add mkfsoptions to LDAP (4679)
2536 - use ->max_readahead method instead of zapping global ra (5039)
2537 - don't interrupt __l_wait_event() during strace
2539 - add software watchdogs to catch hung threads quickly (4941)
2540 - make lustrefs init script start after nfs is mounted
2541 - fix CWARN/ERROR duplication (4930)
2542 - return async write errors to application if possible (2248)
2543 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2544 - print NAL number in %x format (4645)
2545 - update barely-supported suse-2.4.21-171 series (4842)
2546 - support for sles 9 %post scripts
2547 - support for building 2.6 kernel-source packages
2548 - support for sles km_* packages
2550 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2553 - ignore -ENOENT errors in osc_destroy (3639)
2554 - notify osc create thread that OSC is being cleaned up (4600)
2555 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2556 - reconstruct ost_handle() like mds_handle() (4657)
2557 - create a new thread to do import eviction to avoid deadlock (3969)
2558 - let lconf resolve symlinked-to devices (4629)
2559 - don't unlink "objects" from directory with default EA (4554)
2560 - hold socknal file ref over connect in case target is down (4394)
2561 - allow more than 32000 subdirectories in a single directory (3244)
2562 - fix blocks count for O_DIRECT writes (3751)
2563 - OST returns ENOSPC from object create when no space left (4539)
2564 - don't send truncate RPC if file size isn't changing (4410)
2565 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2566 - bind to privileged port in socknal and tcpnal (3689)
2568 - rate limit CERROR/CWARN console message to avoid overload (4519)
2569 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2570 - basic mmap support (3918)
2571 - kernel patch series update from b1_4 (4711)
2573 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2576 - avoid crash during MDS cleanup with OST shut down (2775)
2577 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2578 - don't use bad inodes on the MDS (3744)
2579 - dynamic object preallocation to improve recovery speed (4236)
2580 - don't hold spinlock over lock dumping or change debug flags (4401)
2581 - don't zero obd_dev when it is force cleaned (3651)
2582 - print grants to console if they go negative (4431)
2583 - "lctl deactivate" will stop automatic recovery attempts (3406)
2584 - look for existing locks in ldlm_handle_enqueue() (3764)
2585 - don't resolve lock handle twice in recovery avoiding race (4401)
2586 - revalidate should check working dir is a directory (4134)
2588 - don't always mark "slow" obdfilter messages as errors (4418)
2590 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2593 - don't close LustreDB during write_conf until it is done (3860)
2594 - fix typo in lconf for_each_profile (3821)
2595 - allow dumping logs from multiple threads at one time (3820)
2596 - don't allow multiple threads in OSC recovery (3812)
2597 - fix debug_size parameters (3864)
2598 - fix mds_postrecov to initialize import for llog ctxt (3121)
2599 - replace config semaphore with spinlock (3306)
2600 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2601 - don't allow enqueue to complete on a destroyed export (3822)
2602 - down write_lock before checking llog header bitmap (3825)
2603 - recover from lock replay timeout (3764)
2604 - up llog sem before sending rpc (3652)
2605 - reduce ns lock hold times when setting kms (3267)
2606 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2607 - fix NULL deref and obd_dev leak on setup error (3312)
2608 - replace some LBUG about llog ops with error handling (3841)
2609 - don't match INVALID dentries from d_lookup and spin (3784)
2610 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2611 - fix invalid assertion in ptlrpc_set_wait (3880)
2613 - add libwrap support for the TCP acceptor (3996)
2614 - add /proc/sys/portals/routes for non-root route listing (3994)
2615 - allow setting MDS UUID in .xml (2580)
2616 - print the stack of a process that LBUGs (4228)
2618 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2621 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2622 - make sure to unset replay flag from failed open requests (3440)
2623 - if default stripe count is 0, use OST count for inode size (3636)
2624 - update parent mtime/ctime on client for create/unlink (2611)
2625 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2626 - free recovery state on server during a forced cleanup (3571)
2627 - unregister_reply for resent reqs (3063)
2628 - loop back devices mounting and status check on 2.6 (3563)
2629 - fix resource-creation race that can provoke i_size == 0 (3513)
2630 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2631 - more debugging for page-accounting assertion (3746)
2632 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2633 - avoid deadlock after precreation failure (3758)
2634 - fix race and lock order deadlock in orphan handling (3450, 3750)
2635 - add validity checks when grabbing inodes from l_ast_data (3599)
2637 - add /proc/.../recovery_status to obdfilter (3428)
2638 - lightweight CDEBUG infrastructure, debug daemon (3668)
2639 - change default OSC RPC parameters to be better on small clusters
2640 - turn off OST read cache for files smaller than 32MB
2641 - install man pages and include them in rpms (3100)
2642 - add new init script for (un)mounting lustre filesystems (2593)
2643 - run chkconfig in %post for init scripts (3701)
2644 - drop scimac NAL (unmaintained)
2646 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
2649 - clean kiobufs before and after use (3485)
2650 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
2651 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
2652 - change init script to fail more clearly if not run as root (1528)
2653 - allow clients to reconnect during replay (1742)
2654 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
2655 - don't do DNS lookups on NIDs too small for IP addresses (3442)
2656 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
2657 - fix cond_resched (3554)
2658 - only evict unfinished clients after recovery (3515)
2659 - allow bulk resend, prevent data loss (3570)
2660 - dynamic ptlrpc request buffer allocation (2102)
2661 - don't allow unlinking open directory if it isn't empty (2904)
2662 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
2663 - remove extraneous obd dereference causing LASSERT failure (3334)
2664 - don't use get_cycles() when creating temp. files on the mds (3156)
2665 - hold i_sem when setting i_size in ll_extent_lock() (3564)
2666 - handle EEXIST for set-stripe, set proper directory name (3336)
2668 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
2669 - fix ksocknal_fmb_callback() error messages (2918)
2671 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
2674 - don't copy lvb into (possibly NULL) reply on error (2983)
2675 - don't deref dentry after dput, don't free lvb on error (2922)
2676 - use the kms to determine writeback rpc length (2947)
2677 - increment oti_logcookies when osc is inactive (2948)
2678 - update client's i_blocks count via lvb messages (2543)
2679 - handle intent open/close of special files properly (1557)
2680 - mount MDS with errors=remount-ro, like obdfilter (2009)
2681 - initialize lock handle to avoid ASSERT on error cleanup (3057)
2682 - don't use cancelling-locks' kms values (2947)
2683 - use highest lock extent for kms, not last one (2925)
2684 - don't dereference ERR_PTR() dentry in error handling path (3107)
2685 - fix thread race in portals_debug_dumplog() (3122)
2686 - create lprocfs device entries at setup instead of at attach (1519)
2687 - common AST error handler, don't evict client on completion race (3145)
2688 - zero nameidata in detach_mnt in 2.6 (3118)
2689 - verify d_inode after revalidate_special is valid in 2.6 (3116)
2690 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
2691 - initialize RPC timeout timer earlier for 2.6 (3219)
2692 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
2693 - print nal/nid for unknown nid (3258)
2694 - additional checks for oscc recovery before doing precreate (3284)
2695 - fix ll_extent_lock() error return code for 64-bit systems (3043)
2696 - don't crash in mdc_close for bad permissions on open (3285)
2697 - zero i_rdev for non-device files (3147)
2698 - clear page->private before handing to FS, better assertion (3119)
2699 - tune the read pipeline (3236)
2700 - fix incorrect decref of invalidated dentry (2350)
2701 - provide read-ahead stats and refine rpc in flight stats (3328)
2702 - don't hold journal transaction open across create RPC (3313)
2703 - update atime on MDS at close time (3265)
2704 - close LDAP connection when recovering to avoid server load (3315)
2705 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
2706 - don't leak open file on MDS after open resend (3325)
2707 - serialize filter_precreate and filter_destroy_precreated (3329)
2708 - loop device shouldn't call sync_dev() for nul device (3092)
2709 - clear page cache after eviction (2766)
2710 - resynchronize MDS->OST in background (2824)
2711 - refuse to mount the same filesystem twice on same mountpoint (3394)
2712 - allow llmount to create routes for mounting behind routers (3320)
2713 - push lock cancellation to blocking thread for glimpse ASTs (3409)
2714 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
2715 - fix rare problem with rename on htree directories (3417)
2717 - allow default OST striping configuration per directory (1414)
2718 - fix compilation for qswnal for 2.6 kernels (3125)
2719 - increase maximum number of MDS request buffers for large systems
2720 - change liblustreapi to be useful for external progs like lfsck (3098)
2721 - increase local configuration timeout for slow disks (3353)
2722 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
2724 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2727 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
2728 - don't overwrite extent policy data in reply if lock was blocked (2901)
2729 - drop filter export grants atomically with removal from device (2663)
2730 - del obd_self_export from work_list in class_disconnect_exports (2908)
2731 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
2732 - swab reply message in mdc_close, other PPC fixes (2464)
2733 - fix destroying of named logs (2325)
2734 - overwrite old logs when running lconf --write_conf (2264)
2735 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
2736 - fix race in target_handle_connect (2898)
2737 - mds_reint_create() should take same inode create lock (2926)
2738 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
2739 - don't close files for self_export to avoid uninitialized obd (2936)
2740 - allow MDS with the same name as client node (2939)
2741 - hold dentry reference for closed log files for unlink (2325)
2742 - reserve space for all logs during transactions (2059)
2743 - don't evict page beyond end of stripe extent (2925)
2744 - don't oops on a deleted current working directory (2399)
2745 - handle hard links to targets without a parent properly (2517)
2746 - don't dereference NULL lock when racing during eviction (2867)
2747 - don't grow lock extents when lots of conflicting locks (2919)
2749 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
2752 - account for cache space usage on clients to avoid data loss (974)
2753 - lfsck support in lustre kernel code (2349)
2754 - reduce journal credits needed for BRW writes (2370)
2755 - orphan handling to avoid losing space on client/server crashes
2756 - ptlrpcd can be blocked, stopping ALL progress (2477)
2757 - use lock value blocks to assist in proper KMS, faster stat (1021)
2758 - takes i_sem instead of DLM locks internally on obdfilter (2720)
2759 - recovery for initial connections (2355)
2760 - fixes for mds_cleanup_orphans (1934)
2761 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
2762 - block all file creations until orphan recovery completes (1901)
2763 - client remove rq_connection from request struct (2423)
2764 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
2765 - recovery timer race (2670)
2766 - mdc_close recovey bug (2532)
2767 - ptlrpc cleanup bug (2710)
2768 - mds timeout on local locks (2588)
2769 - namespace lock held during RPCs (2431)
2770 - handle interrupted sync write properly (2503)
2771 - don't try to handle a message that hasn't been replied to (2699)
2772 - client assert failure during cleanup after abort recovery (2701)
2773 - leak mdc device after failed mount (2712)
2774 - ptlrpc_check_set allows timedout requests to complete (2714)
2775 - wait for inflight reqs when ptlrpcd finishes (2710)
2776 - make sure unregistered services are removed from the srv_list
2777 - reset bulk XID's when resending them (caught by 1138 test)
2778 - unregister_bulk after timeout
2779 - fix lconf error (2694)
2780 - handle write after unfinished setstripe, stripe-only getstripe (2388)
2781 - readahead locks pages, leaves pending causing memory pressure (2673)
2782 - increase OST request buffers to 4096 on large machines (2729)
2783 - fix up permission of existing directories in simple_mkdir (2661)
2784 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
2785 - don't assign transno to errored transactions (2742)
2786 - don't delete objects on OST if given a bogus objid from MDS (2751)
2787 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
2788 - if rq_no_resend, then timeout request after recovery (2432)
2789 - fix MDS llog_logid record size, 64-bit array alignment (2733)
2790 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
2791 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
2792 - MDS recovery shouldn't delete objects that it has given out (2730)
2793 - if enqueue arrives after completion, don't clobber LVB (2819)
2794 - don't unlock pages twice when trigger_group_io returns error (2814)
2795 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
2796 - don't write pages to disk if there was an error (1450)
2797 - don't ping imports that have recovery disabled (2676)
2798 - take buffered bytes into account when balancing socknal conn (2817)
2799 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
2800 - reconnect unlink llog connection after MDS reconnects to OST (2816)
2801 - remove little-endian swabbing of llog records (1987)
2802 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
2803 - retry reposting request buffers if they fail (1191)
2804 - grow extent at grant time to avoid granting a revoked lock (2809)
2805 - lock revoke doesn't evict page if covered by a second lock (2765)
2806 - disable VM readahead to avoid reading outside lock extents (2805)
2808 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
2809 - updated kernel patches for hp-2.4.20 kernel (2681)
2811 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
2814 - fix truncated write corruption (2366)
2815 - fix for failed assertion in iopen_connect_dentry (1792,2517)
2817 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
2818 - ensure len(uuid) < 37 in lmc (1171)
2819 - fix ia64 OOPS in llog_test (2255)
2820 - zero end of page at obdfilter for partial page writes (2648)
2821 - don't leave stale dentries around after renames (bug 2428)
2822 - fix timeouts when evicting a client with a single lock held (2642)
2823 - set deadline for the initial HELLO message to drain (2634)
2824 - print out dotted-quad IP addresses in the socknal (2302)
2826 - additional debugging for MDS client eviction problem (2443)
2827 - fix mkfsoptions support for osts (2603, 2604)
2829 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
2832 - add series for the vanilla 2.6.0 kernel
2833 - add series for the vanilla 2.4.24 kernel
2834 - add series for a cray x86/64 UL kernel drop
2835 - fix xattr patches for the vanilla 2.4.19 series
2837 - generate true UUIDs in lmc (1171)
2838 - have portals stack dumping break in UML (2466)
2839 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
2840 - allocate with _NOFS in ldlm to avoid deadlock (1933)
2841 - wake callback waiting threads on client eviction (2460)
2842 - Add --ptldebug and --subsystem to lmc (1719)
2843 - update assertion to allow safe interrupt allocation
2844 - set rq_no_resend for cancel requests (2432)
2845 - recalculate ptlrpcd timeout after resend (2494)
2846 - call vfs_rmdir when removing pending directories (2368)
2847 - fix renaming a file to itself (2429)
2848 - lmc creates a default one-stripe lov (2454)
2849 - expand procfs space to handle large clusters (2326)
2850 - increase UML stack to avoid overflow
2851 - update lconf's list of debug and subsystem masks
2852 - fix lfs find --obd (2510)
2853 - /proc tunable for disabling filter read caching (2591)
2854 - stop rpm packages from altering slapd.conf (2301)
2855 - disable nagle in the socknal under 0conf (2578)
2856 - choose mds inode size based on stripe count (2572)
2857 - fix kernel-source rpm problems (2516)
2859 - add --disable-doc to avoid pdf generation (2421)
2860 - update documentation, tests, type-os, comments
2861 - avoid format warnings on ia64
2862 - remove the TOE NAL
2863 - tiny code cleanups by removing unused fields
2865 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
2868 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
2869 - remove the most blatant lies from BUILDING file (2371)
2870 - change default debug level to reasonable production setting
2871 - reduce client side cache size to reduce cache flush time
2872 - reduce max RPCs in flight to avoid unnecessary file fragmentation
2873 - make TCP zerocopy and pinger support enabled by default (2476)
2874 - sync writes completed after process exits caused crashes (2319)
2875 - maintain correct mount count on the MDS (2356)
2876 - backout 1557, because 2316 wasn't really fixed
2877 - better file I/O statistics gathering in /proc
2878 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
2879 - another kernel patch to fix zero-copy TCP function export
2880 - don't take duplicate lock when processing re-sent getattr (2420)
2881 - lctl uses obd_self_export instead of creating new conn (2353)
2882 - MDS/OST recovery case which requires object creation asserted (2425)
2883 - move lfs from /usr/sbin to /usr/bin in packages
2884 - fix race between mds_client_add and mds_client_free (2417)
2885 - use kmalloc instead of slabs in portals (2430)
2886 - don't create duplicate records when a failover MDS is present (2442)
2887 - remove unnecessary mount age check (2332)
2888 - don't remove directory inodes from locks prematurely (2451)
2889 - don't break if MDS service name is the same as hostname (2103)
2890 - fix races in client write RPC generation when cache full (2482)
2892 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
2895 - remove now-unused request->rq_obd (278)
2896 - if an allocation fails, print out how much memory we've used (1933)
2897 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
2898 - add the "configurable stack size" patch to most series files (1256)
2899 - ability to write large log records, for 100+ OST configs (2306)
2900 - fix NULL deref when filter_prep fails (2314)
2901 - fix operator precedence error in filter_sync
2902 - dynamic allocation of socknal TX descriptors (2315)
2903 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
2904 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
2905 - ensure that CWARN messages in Portals always get to the syslog
2906 - __init/__exit are not for prototype decls (ldlm_init/exit)
2907 - x86-64 compile warning fixes
2908 - fix gateway LMC keyword conflict (2318)
2909 - fix MDS lock inversions in getattr/reint paths (1844)
2910 - fix a rare lock re-ordering bug, which caused deadlock (2322)
2911 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
2912 - DLM race condition prevented some lock evictions (2328)
2913 - ENOMEM detection and retry on socknal sends (2230)
2914 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
2915 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
2916 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
2917 - better error messages when a client is rejected during recovery (1505)
2918 - avoid cancelling locks which were never granted, after failure (2330)
2919 - fix i_sem/journal inversion in mds_client_add (2333)
2920 - fix truncate/getattr lock cycle deadlock (2334)
2921 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
2922 - fix two rare exit paths which could leak an l_lock() ref (2321)
2923 - fencepost error in MDS/OST orphan recovery (2226)
2924 - make log record alignment 8 bytes (1988)
2925 - lstripe now fails when requested offset > ost_count (2237)
2926 - ensure that all kernel series have a complete list.h (1607)
2927 - fix crashes in special-file operations (2316)
2928 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
2930 - allow configurable automake binary, for testing new versions
2931 - small update to the lfs documentation
2933 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
2935 * fix negative export reference count in fsfilt_sync (2312)
2937 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
2938 * release candidate 0.9.1
2940 - orphans are moved into the PENDING directory for possible recovery
2941 - replayed opens now open by fid for orphan/rename safety (1042)
2942 - last close of an orphan inode generates a transno (683)
2943 - chdir() and mount() now pin the directory entry (1020)
2944 - avoid CERROR in normal ll_setattr_raw() error case (1500)
2945 - discard very old requests without processing them (1502)
2946 - remove some common, well-understood CERRORs (1505)
2947 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
2948 - clear "grant" flags in OST replies until OST grant code lands (1644)
2949 - fix read performance by not clobbering i_blksize on client (1598)
2950 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
2951 - make LRU size a /proc tunable, clears locks when reduced (707)
2952 - fix some lprocfs rot that prevented ptlbd from loading (1732)
2953 - server locks take references on exports now (1558)
2954 - build fixes for 2.4.20-rh trees (1663)
2955 - return an error from lov_create if all OSCs are inactive (1751)
2956 - fix import levels when a reconnect happens without a timeout (1597)
2957 - exit early from mds_open if we get a lookup error (1749)
2958 - partial page read at EOF wouldn't wait for disk before sending (1642)
2959 - avoid NULL deref in obdfilter when reading page past EOF (1592)
2960 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
2961 - fix LBUG in ll_it_open_error with rc = -2 (1861)
2962 - write/truncate lock inversion (1639)
2963 - Don't auto-load obdclass, portals modules during cleanup (1495)
2964 - fix timestamps from jumping to "now" (1763)
2965 - extra journal assertions (1648)
2966 - add an extra multiunlink test (1771)
2967 - fix read_record/write_record API (1776)
2968 - fix leak of offset_extent, possible incorrect i_size later (1772)
2969 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
2970 - Debugging for the kqswnal_get_idle_tx problems (1820)
2971 - Allow recovery to be attempted multiple times (1536)
2972 - Write out MDS last_rcvd file after it is first created (1600)
2973 - Fix tx_descriptor leak in failed transmit situations (1827)
2974 - ext3 journaling fixes for assertion failure after IO error (1871)
2975 - class_export_put() on freed export after completion AST error (1896)
2976 - Fix revalidate looping in VFS (1322)
2977 - Don't access a freed export during MDS_REINT timeout (1521)
2978 - Add open-unlink recovery support on the MDS (1673,1764)
2979 - Return an error if no MDS data was read from last_rcvd (1946)
2980 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
2981 - Don't setup a disk device that doesn't match exported UUID (317)
2982 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
2983 - avoid committing NULL handle in force close
2984 - local.sh is now a one-stripe LOV configuration
2985 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
2986 - fix ext3 htree duplicate directory entry corruption (1516)
2987 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
2988 - update to new LOV EA format (2097)
2989 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
2991 2003-06-15 Phil Schwan <phil@clusterfs.com>
2994 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
2995 - per-import/export recovery handling (958, 931, 959)
2996 - multiple last-rcvd slots, for serving multiple FSes (949)
2997 - connections are again shared between multiple imp/exports (963, 964)
2998 - "umount -f" would hang if any requests needed to be sent (393, 978)
2999 - avoid pinning large req buffer by copying for queued messages (989)
3000 - add "uuid" to "lctl device" command to help upcalls (991)
3001 - "open" RPCs with transnos would confuse recovery counters (1037)
3002 - do proper endian conversion of all wire messages (288, 340, 891)
3003 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3004 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3005 - fix ptlrpc_connection leak in target_handle_connect (1174)
3006 - fix import refcounting bug in OST and MDS cleanup (1134)
3007 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3008 - fix very unlikely obd_types race condition (501)
3009 - remove osc_open hack for echo_client (1187)
3010 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3011 - a failure in read_inode2 leads to deadlock (1139)
3012 - cancel ack-locks as soon as transaction is committed (1072)
3013 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3014 - make sure to commitrw after any preprw to avoid deadlock (1162)
3015 - failing to execute a file in a lustre FS would lock inode (1203)
3016 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3017 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3018 - obd_uuid2tgt would walk off the end of the list (1255)
3019 - on IA64 the osc would give portals incorrect bulk size (1258)
3020 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3021 - fix lock inversion caused by new llite matching code (1282)
3022 - limit the number of dirty pages on a client to 10MB (1286)
3023 - timed out locks were not being corrected cancelled (1289)
3024 - fix O_DIRECT above 4GB on IA-32 (1292)
3025 * major user-visible changes
3026 - fail out/fail over policy now controlled by the upcall (993)
3028 - add OBD_PING to check server availability and failure (954)
3029 - lustre messages are now sent in sending host order (288, 340, 891)
3030 - add eadatalen to MDS getattr reply (340)
3031 - OST read replies may contain second buffer, with per-page status (593)
3033 2003-03-11 Phil Schwan <phil@clusterfs.com>
3036 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3037 - failed open()s could cause deadlock; fixed (867, 869)
3038 - stop cancelling OST locks when files are closed (481)
3039 - overlapping XID spaces caused network corruption (851, 853)
3040 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3041 - setattr_raw added, to do single-RPC, server-side setattrs
3042 - lmc/lconf syntax change for OST UUIDs
3043 - fix crashy race condition between ptlrpc_free_req and osc_close
3044 - don't use request in mdc_enqueue if we hit a timeout (889)
3045 - don't set the inode i_size for regular files from the MDS (896)
3046 - handle out of order completion AST (842)
3047 - don't LBUG if a lock request times out after receiving AST (913)
3048 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3049 - if a bad lock AST arrives, send an error instead of dropping entirely
3050 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3051 - fix leak in bulk IO when only partially completed (899, 900, 926)
3052 - fix O_DIRECT for ia64 (55)
3053 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3054 - C-z after timeout could hang a process forever; fixed (977)
3056 - client-side I/O cache (678, 924, 929, 941, 970)
3058 - READPAGE and SETATTRs which don't take server-side locks get
3061 2003-02-11 Phil Schwan <phil@clusterfs.com>
3064 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3065 - Fix fsfilt_extN_readpage() to read a full page of directory
3066 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3067 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3068 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3069 - Most intent operations were removed, in favour of a new RPC mode
3070 that does a single RPC to the server and bypasses most of the VFS
3071 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3072 - Aggressively cancel local locks on DLM servers
3073 - mds_reint_unlink sends EA to the client if it's the last nlink.
3074 client uses that EA to unlink OST objects.
3075 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3076 - recursive symlinks were fixed (439)
3077 - fixed NULL deref in DEBUG_REQ
3078 - filter_update_lastobjid no longer calls sync, which annoyed extN
3079 - fixed multi-client small-writes to a single file problem (445)
3080 - fixed mtime updates during file writes (607)
3081 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3082 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3083 - fixed Linux OST size reporting problem (444, 656)
3084 - OST now updates object mtime with writes or setattr (607, 619)
3085 - client verifies file size before zeroing page past EOF (445)
3086 - OST now writes last allocated objid to disk with allocation (108)
3087 - LOV on echo now works (409)
3089 - mds_reint_unlink sends a new buffer, with the EA included. this
3090 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3091 set if a regular file was being unlinked, and it was the last link
3092 - use PtlGet from the target for bulk writes (315)
3093 - OST now updates object mtime with writes or setattr (607, 619)
3094 - LDLM now has a grant-time callback to revalidate locked items, if
3096 - Many MDS operations were reorganized to combat race conditions
3098 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3099 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3101 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3104 - Fully reactivate OST imports after reconnection (512, others)
3105 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3106 - More graceful error handling for truncating on dead OST (515)
3107 - Don't error out unless we're actually accessing dead stripes (474)
3108 - Fix garbage sizes when stripes are missing (410)
3109 - LRU counters were broken, causing constant lock purge (433, 432)
3110 - garbage on read from stripes with failed OSTs (441)
3111 - mark OSCs as active before reconnecting during recovery (438)
3112 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3113 - lfind did not preserve OST order in output (443)
3114 - symlinks cause hung clients, incorrect data (439)
3115 - stop dereferencing request after dropping refcount (457)
3116 - don't LASSERT(spin_is_locked) on non-SMP (455)
3117 - fixes for many rename() bugs
3118 - fstat didn't correctly synchronize attributes (399)
3119 - server must handle lock cancellation during blocking AST prep (487)
3120 - bulk descriptors were free()d too soon (511)
3121 - fix paths in lconf, which would load incorrect modules (451, 507)
3122 - fix confusing lconf 'host not found' error message (386)
3123 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3124 - fix race condition in mdc_blocking_ast() for inode access (526)
3125 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3126 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3127 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3128 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3129 - fix lstripe in several ways (kernel side) (527)
3130 - fix request leak in ldlm_cli_enqueue (262)
3131 - incorrect OSC was marked inactive after OST failure
3132 - call mds_fs_cleanup before unmounting filesystem (524)
3133 - fix races between taking ns_lock and ldlm_lock_change_resource
3134 - fix races updating LOV export open file list
3135 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3136 - fix recovery NULL deref in ldlm_cli_cancel_unused
3137 - fix some DLM races by using new hash table for lock handles (419)
3138 - permit the client to specify desired inodes, at replay
3139 - duplicate requests when we queue them for replay reintegration
3140 - fix last_rcvd offset calculation
3141 - sync after each recovered transaction, so we always make progress
3142 - never, not always, ERESTART requests without transnos
3143 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3145 - skip replay if the MDS says that the client is already connected
3146 - don't check for a recovery-enabled export to match lctl's UUID
3147 - don't INC_USE_COUNT for phantom exports
3148 - don't crash when cleaning up phantom exports (567)
3149 - don't double-finish or set replay data for errored mdc_open requests
3150 - abort requests when they time out, so we don't get old replies
3151 - send/receive replies for AST messages again
3152 - if the client says that it doesn't have the lock, cancel it on the
3154 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3155 mark it as destroyed, it will all work out when decref is called
3156 - fix module use counts (22, 581)
3158 - ASTs now expect a reply (server cancels lock on error reply)
3160 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3163 - fix many simultaneous client startup (392)
3164 - fix dentry->d_it clobbering
3165 - credentials weren't being shipped for readdir/getattr operations
3166 - remove invalid assertions triggered during some concurrent MD
3168 - proper Lustre versions added (336, 389)
3169 - fix memory leak for create error case (398)
3170 - fix LOV locking bug that would get cli/srv out of sync
3171 - fix echo client over LOV (409)
3172 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3173 - fix double-O_EXCL intent crash (424)
3174 - avoid sending multiple lock CANCELs (352)
3176 - MDS can do multi-client recovery (modulo bugs in new code)
3178 - many updates, edits, cleanups
3180 2002-11-18 Phil Schwan <phil@clusterfs.com>
3183 - fix null d_it dereference (346)
3184 - fix full OST/dbench hang (333)
3185 - fix permission problem with file removal (286)
3186 - fix removal of OSCs from LOV when they fail
3187 - fix NULL deref during bulk timeout (214)
3188 - fix problems related to multiple filesystems on one MDS (241)
3189 - fixed serious subtle metadata locking bugs
3190 - free locks on clients when inodes are removed due to memory
3192 - fix inode pointer in lock data (285)
3193 - partial support for multiple MDS on a single host (241)
3194 - data locks weren't cancelled at clear_inode time (290, 311)
3195 - intent locks could lead to unbounded lock growth (205)
3196 - added a maximum lock count, an LRU list, and a flusher
3197 - fix multiple rename (365)
3198 - properly abstracted the echo client
3199 - OSC locked 1 byte too many; fixed
3200 - rewrote brw callback code:
3201 - fixed recovery bugs related to LOVs (306)
3202 - fixed too-many-pages-in-one-write crash (191)
3203 - fixed (again) crash in sync_io_timeout (214)
3204 - probably fixed callback-related race (385)
3206 - Add capability to MDS protocol
3207 - LDLM cancellations and callbacks on different portals
3209 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3212 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3213 - defer open object destruction to close time (601981, 138)
3214 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3215 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3216 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3217 - don't call obd_setattr in ll_file_release for destroyed objects
3218 * protocol change to lustre_msg: move |version| and add |flags|
3219 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3220 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3221 * added replay of create, unlink, link and rename operations during
3222 MDS failover; recovery should be much more robust now
3223 * remove failed OSCs from LOVs (only lov_create uses this so far)
3224 * the lustre-HOWTO was brought (more) up to date (582544)
3226 2002-10-23 Phil Schwan <phil@clusterfs.com>
3229 - in-use dentries weren't being reused properly (617851)
3230 - prevent multiple LDLM setup (599178)
3231 - fix LOV size calculations for truncate (617853)
3232 - fix client handling of MDS intent errors (POSIX)
3233 - fix permission bug in lovstripe.c test (624321)
3234 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3235 - truncate past end of file could corrupt data
3236 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3237 - a race in recovery could return ETIMEDOUT to apps (623947)
3238 - building outside the source directory was fixed
3239 * the lustre-HOWTO was brought (more) up to date (582544)
3240 * major progress was made on recovery functionality
3242 2002-10-10 Phil Schwan <phil@clusterfs.com>
3245 - recovery deadlock fix
3246 - rm -rf causes LBUG fix (617817)
3247 - file open by multiple tasks fix (618962)
3248 - directory permissions bugs (602707 and 620007)
3249 - journal_stop fixed with locking (611313)
3250 - O_APPEND failures resolved (618273, perhaps 614459)
3251 - lconf PATH fix (619770)
3252 - IA64 build fix (621450)
3253 - RPC buffer sizes scale with amount of memory
3255 2002-10-01 Phil Schwan <phil@clusterfs.com>
3258 - locks would be cancelled without throwing away data pages,
3259 resulting in inconsistent data (605627)
3260 - inode attributes were not always being refreshed (605627, 612449)
3261 - lconf now continues to cleanup after lctl reports an error
3262 - MDS now enforces user permissions (602707)
3263 - lprocfs cleanup fixed, but not yet enabled (614157)
3264 - fixed infinite server hang, should a client not respond to an AST
3265 - avoid going into recovery if user calls readlink() with a buffer
3266 that's too small (613941)
3267 - AST RPCs no longer require replies (614867) -- this may be changed
3268 - don't crash server if client sends an IOV that's too big (611336)
3269 - fixed lock conversion deadlock (611892)
3270 - fixed the following of symlinks (614622)
3271 * recovery: the server can remove locks from a client that dies, other
3272 clients can make progress
3273 * more extN patch fixes
3274 * compile-time configurable ptlrpc buffer allocations
3276 - collaborative read cache document
3277 - Lustre Lite Performance CDR document-in-progress
3279 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3282 - fix typo in patch-2.4.18
3284 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3287 - clear ptlrpc request each time in handle_incoming_request()
3288 - unlink of files now destroys the object on the OST
3290 2002-09-19 Peter Braam <braam@clusterfs.com>
3292 * add hard link support
3293 * change obdfile creation method
3294 * kernel patch changed
3296 2002-09-19 Peter Braam <braam@clusterfs.com>
3299 - stack overflow bug in extN fixed
3301 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3303 * documentation updates
3304 - add man pages for config tools
3305 - update tests/README to describe testing with new config tools
3306 - finish metadata API descriptions
3307 * bug fixes and cleanups
3308 - statfs workaround for 16TB limit
3309 - LOV stripe allocation improved, can stripe on subset of OSTs
3310 - LOV file size/IO offset was wrong for files > 4GB in size
3311 - object EA data was being dropped, caused files to be unreadable
3312 - memory overflow with non-LOV OST caused memory corruption
3313 - fixed regression tests to work with new config tools, obdfilter
3314 - fixed bug when directory size became larger than 1 block
3315 - fixed bug (for single client case) when PWD was deleted
3316 - invalidate local directory pages when doing intent-based ops
3317 - avoid LDLM oops when lock callback contained bad data
3319 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3321 * documentation updates
3322 * bug fixes and cleanups
3323 - configuration tools
3326 - 64-bit compile warnings
3327 - 64-bit internal statfs data
3329 * test_brw on persistent OST devices
3331 * lprocfs (disabled)
3333 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3335 * documentation updates
3336 * bug fixes and cleanups
3337 * configuration tools
3339 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3342 * many small fixes to 0_5_4
3343 * io/network handling
3344 * thinkos in MDS operations
3346 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3349 * crucial basic fixes to 0.5.3
3350 * IOR, Iozone work over Elan
3353 2002-08-07 Phil Schwan <phil@clusterfs.com>
3354 * version 0_5_3, our first alpha
3355 * we use the new Portals iovs
3356 * documentation updates
3357 * bug fixes and cleanups
3358 * small changes in the DLM wire protocol
3360 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3361 * version 0_5_1 with some initial stability,
3362 * locking on MD and file I/O.
3363 * documentation updates
3364 * several bug fixes since 0.5.0
3365 * small changes in wire protocol
3367 2002-07-18 Phil Schwan <phil@clusterfs.com>
3369 * delivered as Lustre Light Alpha
3370 * fixed a crash after handling invalid MDS requests
3371 * fixed directory pages for architectures with non-4k pages sizes
3373 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3374 * release version v0_4_4
3375 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3377 * Fixes a number of interruption problems with OST operations.
3378 * Update documentation for portals header changes
3379 * Move all wire protocol structs/defines to lustre_idl.h
3380 * Fixes symlink length bug.
3381 * Add tcpdump to repository.
3383 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3384 * release version v0_4_3
3385 * Fixes statfs for inodes on extN.
3386 * Fixes bug in runtests which would delete /etc/hosts.
3387 * Use 64-bit object IDs wherever possible (not into VFS though)
3388 Remove ost_get_info, which is unused by lustre, and out of date.
3390 2002-07-03 Peter Braam <braam@clusterfs.com>
3391 * release version v0_4_2 Fixes a lookup error (type not passed)
3392 * move forward to head of Portals
3393 * move forward to latest Lustre kernel
3395 2002-06-25 Peter Braam <braam@clusterfs.com>
3396 * release version v0_4_1. Hopefully stable on single node use.