1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4 2.4.21-47.0.1.EL (RHEL 3)
6 2.6.9-42.0.10.EL (RHEL 4)
7 2.6.12.6 vanilla (kernel.org)
8 2.6.16.27-0.9 (SLES 10)
9 * Client support for unpatched kernels:
10 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
11 2.6.16 - 2.6.19 vanilla (kernel.org)
12 2.6.9-42.0.8.EL (RHEL 4)
13 * Recommended e2fsprogs version: 1.39.cfs6
14 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
17 Severity : enhancement
19 Description: Add LNET router traceability for debug purposes
20 Details : If a checksum failure occurs with a router as part of the
21 IO path, the NID of the last router that forwarded the bulk data
22 is printed so it can be identified.
27 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
28 Details : Due to a race condition, liblustre clients were occasionally
31 Severity : enhancement
33 Description: lfs setstripe use optional parameters instead of postional
36 Severity : enhancement
38 Description: Nanosecond timestamp support for ldiskfs
39 Details : The on-disk ldiskfs filesystem has added support for nanosecond
40 resolution timestamps. There is not yet support for this at
41 the Lustre filesystem level.
44 Frequency : during server recovery
46 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
47 Details : request with rq_no_resend flag not awake l_wait_event if they get a
51 Frequency : nfs export on patchless client
53 Description: connectathon hang when test nfs export over patchless client
54 Details : Disconnected dentry cannot be found with lookup, so we do not need
55 to unhash it or make it invalid
58 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
59 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
60 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
65 Description: open req refcounting wrong on reconnect
66 Details : If reconnect happened between getting open reply from server and
67 call to mdc_set_replay_data in ll_file_open, we will schedule
68 replay for unreferenced request that we are about to free.
69 Subsequent close will crash in variety of ways.
70 Check that request is still eligible for replay in
71 mdc_set_replay_data().
76 Description: disable writes to filesystem when reading health_check file
77 Details : the default for reading the health_check proc file has changed
78 to NOT do a journal transaction and write to disk, because this
79 can cause reads of the /proc file to hang and block HA state
80 checking on a healthy but otherwise heavily loaded system. It
81 is possible to return to the previous behaviour during configure
82 with --enable-health-write.
84 --------------------------------------------------------------------------------
86 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
91 Frequency : on some architectures
93 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
94 Details : Uninitialized flags sometimes cause configuration commands to
97 --------------------------------------------------------------------------------
99 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
101 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
102 INTEROPERATE with older versions automatically. In many cases a
103 special upgrade step is needed. Please read the
104 user documentation before upgrading any part of a 1.4.x system.
105 * WARNING: Lustre configuration and startup changes are required with
106 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
108 * Support for kernels:
109 2.4.21-47.0.1.EL (RHEL 3)
111 2.6.9-42.0.10.EL (RHEL 4)
112 2.6.12.6 vanilla (kernel.org)
113 2.6.16.27-0.9 (SLES10)
114 * Client support for unpatched kernels:
115 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
116 2.6.16 - 2.6.19 vanilla (kernel.org)
117 2.6.9-42.0.8EL (RHEL 4)
118 * Recommended e2fsprogs version: 1.39.cfs6
119 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
122 Severity : enhancement
124 Description: MountConf
125 Details : Lustre configuration is now managed via mkfs and mount
126 commands instead of lmc and lconf. New obd types (MGS, MGC)
127 are added for dynamic configuration management. See
128 https://mail.clusterfs.com/wikis/lustre/MountConf for
131 Severity : enhancement
133 Description: dynamic OST addition
134 Details : OSTs can now be added to a live filesystem
136 Severity : enhancement
138 Description: startup order invariance
139 Details : MDTs and OSTs can be started in any order. Clients only
140 require the MDT to complete startup.
142 Severity : enhancement
144 Description: parallel, asynchronous orphan cleanup
145 Details : orphan cleanup is now performed in separate threads for each
146 OST, allowing parallel non-blocking operation.
148 Severity : enhancement
150 Description: optimized stripe assignment
151 Details : stripe assignments are now made based on ost space available,
152 ost previous usage, and OSS previous usage, in order to try
153 to optimize storage space and networking resources.
155 Severity : enhancement
157 Description: Permanently set tunables
158 Details : All writable /proc/fs/lustre tunables can now be permanently
159 set on a per-server basis, at mkfs time or on a live
162 Severity : enhancement
164 Description: Lustre message v2
165 Details : Add lustre message format v2.
167 Severity : enhancement
169 Description: client OST exclusion list
170 Details : Clients can be started with a list of OSTs that should be
171 declared "inactive" for known non-responsive OSTs.
175 Description: ENOENT returned for valid filehandle during dbench.
176 Details : Check if a directory has children when invalidating dentries
177 associated with an inode during lock cancellation. This fixes
178 an incorrect ENOENT sometimes seen for valid filehandles during
182 Frequency : SFS test only (otherwise harmless)
184 Description: SPEC SFS validation failure on NFS v2 over lustre.
185 Details : Changes the blocksize for regular files to be 2x RPC size,
186 and not depend on stripe size.
188 Severity : enhancement
190 Description: fine-grained SMP locking inside DLM
191 Details : Improve DLM performance on SMP systems by removing the single
192 per-namespace lock and replace it with per-resource locks.
194 Severity : enhancement
196 Description: don't hold multiple extent locks at one time
197 Details : To avoid client eviction during large writes, locks are not
198 held on multiple stripes at one time or for very large writes.
199 Otherwise, clients can block waiting for a lock on a failed OST
200 while holding locks on other OSTs and be evicted.
202 Severity : enhancement
204 Description: Multiple MD RPCs in flight.
205 Details : Further unserialise some read-only MDT RPCs - learn about intents.
206 To avoid overly-overloading MDT, introduce a limit on number of
207 MDT RPCs in flight for a single client and add /proc controls
208 to adjust this limit.
210 Severity : enhancement
212 Description: client read/write statistics
213 Details : Add client read/write call usage stats for performance
214 analysis of user processes.
215 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
216 file access. extents_stats shows chunk size distribution.
217 extents_stats_per_process show chunk size distribution per
220 Severity : enhancement
222 Description: per-client statistics on server
223 Details : Add ldlm and operations statistics for each client in
224 /proc/fs/lustre/mds|obdfilter/*/exports/
226 Severity : enhancement
228 Description: improved MDT statistics
229 Details : Add detailed MDT operations statistics in
230 /proc/fs/lustre/mds/*/stats
232 Severity : enhancement
234 Description: VFS operations stats
235 Details : Add client VFS call stats, trackable by pid, ppid, or gid
236 /proc/fs/lustre/llite/*/vfs_ops_stats
237 /proc/fs/lustre/llite/*/vfs_track_[pid|ppid|gid]
242 Description: Fix client-side osc byte counters
243 Details : The osc read/write byte counters in
244 /proc/fs/lustre/osc/*/stats are now working
247 Frequency : always as root on SLES
249 Description: Failure of copying files with lustre special EAs.
250 Details : Client side always return success for setxattr call for lustre
251 special xattr (currently only "trusted.lov").
256 Description: Refcount LNET uuids
257 Details : The global LNET uuid list grew linearly with every startup;
258 refcount repeated list entries instead of always adding to
261 Severity : enhancement
263 Description: Dynamic service threads
264 Details : Within a small range, start extra service threads
265 automatically when the request queue builds up.
268 Frequency : mixed-endian client/server environments
270 Description: mixed-endian crashes
271 Details : The new msg_v2 system had some failures in mixed-endian
274 Severity : enhancement
276 Description: Easy OST removal
277 Details : OSTs can be permanently deactivated with e.g. 'lctl
278 conf_param lustre-OST0001.osc.active=0'
280 Severity : enhancement
282 Description: MGS proc entries
283 Details : Added basic proc entries for the MGS showing what filesystems
286 Severity : enhancement
288 Description: provide MGS failover
289 Details : Added config lock reacquisition after MGS server failover.
291 Severity : enhancement
293 Description: add Linux 2.4 support
294 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
298 Description: a large application tries to do I/O to the same resource and dies
300 Details : Check the req->rq_arrival time after the call to
301 ost_brw_lock_get(), but before we do anything about
302 processing it & sending the BULK transfer request. This
303 should help move old stale pending locks off the queue as
304 quickly as obd_timeout.
307 Frequency : when an incorrect nid is specified during startup
309 Description: ptlrpc connect to non-existant node causes kernel crash
310 Details : LNET can't be re-entered from an event callback, which
311 happened when we expire a message after the export has been
312 cleaned up. Instead, hand the zombie cleanup off to another
315 Severity : enhancement
317 Description: plain/inodebits lock performance improvement
318 Details : Grouping plain/inodebits in granted list by their request modes
319 and bits policy, thus improving the performance of search through
323 Frequency : only if OST filesystem is corrupted
325 Description: client incorrectly hits assertion in ptlrpc_replay_req()
326 Details : for a short time RPCs with bulk IO are in the replay list,
327 but replay of bulk IOs is unimplemented. If the OST filesystem
328 is corrupted due to disk cache incoherency and then replay is
329 started it is possible to trip an assertion. Avoid putting
330 committed RPCs into the replay list at all to avoid this issue.
333 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
335 Description: System hang on startup
336 Details : This bug allowed the liblustre (e.g. catamount) client to
337 return to the app before handling all startup RPCs. This
338 could leave the node unresponsive to lustre network traffic
339 and manifested as a server ptllnd timeout.
341 Severity : enhancement
343 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
344 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
345 boolean to control whether to print peer debug info when a
346 client's RPC times out.
349 Frequency : only for kernels with patches from Lustre below 1.4.3
351 Description: Remove old rdonly API
352 Details : Remove old rdonly API which unused from at least lustre 1.4.3
355 Frequency : only for devices with external journals
357 Description: Set external device read-only also
358 Details : During a commanded failover stop, we set the disk device
359 read-only while the server shuts down. We now also set any
360 external journal device read-only at the same time.
363 Frequency : when upgrading from 1.4 while trying to change parameters
365 Description: The wrong (new) MDC name was used when setting parameters for
366 upgraded MDT's. Also allows changing of OSC (and MDC)
367 parameters if --writeconf is specified at tunefs upgrade time.
370 Frequency : when setting specific ost indicies
372 Description: QOS code breaks on skipped indicies
373 Details : Add checks for missing OST indicies in the QOS code, so OSTs
374 created with --index need not be sequential.
376 Severity : enhancement
378 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
379 Details : The uninit_groups feature works in conjunction with the kernel
380 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
381 up the pass1 processing of e2fsck. This is a read-only feature
382 in ldiskfs2 only, so older kernels and current ldiskfs cannot
383 mount filesystems that have had this feature enabled.
385 Severity : enhancement
387 Description: Improve multi-block allocation algorithm to avoid fragmentation
388 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
389 allocation locality and avoid filesystem fragmentation.
391 ------------------------------------------------------------------------------
393 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
395 * Support for kernels:
396 2.4.21-47.0.1.EL (RHEL 3)
398 2.6.9-42.0.10.EL (RHEL 4)
399 2.6.12.6 vanilla (kernel.org)
400 2.6.16.27-0.9 (SLES 10)
401 * Recommended e2fsprogs version: 1.39.cfs5
403 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
407 Frequency : occasional, depends on client load and configuration
408 Bugzilla : 12181, 12203
409 Description: data loss for recently-modified files
411 Details : In some cases it is possible that recently written or created
412 files may not be written to disk in a timely manner (this should
413 normally be within 30s unless client IO load is very high).
414 The problem appears as zero-length files or files that are a
415 multiple of 1MB in size after a client crash or client eviction
416 that are missing data at the end of the file.
418 This problem is more likely to be hit on clients where files are
419 repeatedly created and unlinked in the same directory, clients
420 have a large amount of RAM, have many CPUs, the filesystem has
421 many OSTs, the clients are rebooted frequently, and/or the files
422 are not accessed by other nodes after being written.
424 The presence of the problem can be detected by looking at
425 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
426 smaller than the second (nr_unused) then dirty files will not
427 be flushed automatically to disk. "sync; sleep 10" should be
428 run several times on the node before unmounting it to update
429 Lustre (this is also safe to run on nodes without this problem).
431 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
432 that can cause this same problem, so customers using that kernel
433 also need to update the kernel in addition to Lustre. In order
434 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
436 It is normal that files written just before a client crash (less
437 than 30s) may not yet have been flushed to disk, even for local
441 Frequency : frequent on thin XT3 nodes
443 Description: UUID collision on thin XT3 Linux nodes
444 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
445 randomly, since we relied on an insufficiently-seeded PRNG.
450 Description: OSS hangs after "All ost request buffers busy"
451 Details : A deadlock between quota and journal operations caused OSS
452 hangs after printing "All ost request buffers busy."
455 Frequency : always on liblustre builds
457 Description: Cleanup compiler warnings on liblustre
460 Frequency : always on liblustre builds on XT3
462 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
466 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
467 > 32000 subdirectories
468 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
469 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
470 subdirectory count crosses 32000. This will aid e2fsck to
471 correctly handle more than 32000 subdirectories.
474 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
476 Description: System hang on startup
477 Details : This bug allowed the liblustre (e.g. catamount) client to
478 return to the app before handling all startup RPCs. This
479 could leave the node unresponsive to lustre network traffic
480 and manifested as a server ptllnd timeout.
482 Severity : enhancement
484 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
485 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
486 boolean to control whether to print peer debug info when a
487 client's RPC times out.
492 Description: make O_SYNC working on 2.6 kernels
493 Details : 2.6 kernels use different method for mark pages for write,
494 so need add a code to lustre for O_SYNC work.
499 Description: Failure to close file and release space on NFS
500 Details : Put inode details into lock acquired in ll_intent_file_open.
501 Use mdc_intent_lock in ll_intent_open to properly
502 detect all kind of errors unhandled by mdc_enqueue.
507 Description: proc file read during shutdown sometimes raced obd removal,
509 Details : Add lock to prevent obd access after proc file removal.
512 Frequency : Only for files larger than 4GB on 32-bit clients.
514 Description: improperly doing page alignment of locks
515 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
516 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
521 Description: RPCs being resent when they shouldn't be
522 Details : Some RPCs that should not be resent are being resent. This
523 can cause inconsistencies in the RPC state machine. Do not
524 resend such requests.
527 Frequency : rare, only with NFS export
529 Description: Crash on NFS re-export node
530 Details : under very unusual load conditions an assertion is hit in
531 ll_intent_file_open()
534 Frequency : only if OST filesystem is corrupted
536 Description: client incorrectly hits assertion in ptlrpc_replay_req()
537 Details : for a short time RPCs with bulk IO are in the replay list,
538 but replay of bulk IOs is unimplemented. If the OST filesystem
539 is corrupted due to disk cache incoherency and then replay is
540 started it is possible to trip an assertion. Avoid putting
541 committed RPCs into the replay list at all to avoid this issue.
546 Description: large O_DIRECT requests fail under memory pressure/fragmentation
547 Details : Large single O_DIRECT read and write calls can fail to allocate
548 a sufficiently large buffer to process the request. In case of
549 allocation failure the allocation is retried with a smaller
550 buffer and broken into smaller requests.
552 Severity : enhancement
554 Description: Add -o localflock option to simulate old noflock behaviour.
555 Details : This will achieve local-only flock/fcntl locks coherentness.
560 Description: versioning check is incomplete
561 Details : Checking the version difference of client vs. server, report
562 error if the gap is too big.
567 Description: add support PG_writeback bit
568 Details : add support for PG_writeback bit for Lustre, for more carefull
569 work with page cache in 2.6 kernel. This also fix some deadlocks
570 and remove hack for work O_SYNC with 2.6 kernel.
572 Severity : enhancement
574 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
575 Details : The uninit_groups feature works in conjunction with the kernel
576 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
577 up the pass1 processing of e2fsck. This is a read-only feature
578 in ldiskfs2 only, so older kernels and current ldiskfs cannot
579 mount filesystems that have had this feature enabled.
581 Severity : enhancement
583 Description: Improve multi-block allocation algorithm to avoid fragmentation
584 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
585 allocation locality and avoid filesystem fragmentation.
587 ------------------------------------------------------------------------------
589 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
591 * Support for kernels:
592 2.6.9-42.0.3.EL (RHEL 4)
594 2.4.21-47.0.1.EL (RHEL 3)
595 2.6.12.6 vanilla (kernel.org)
596 2.6.16.21-0.8 (SLES10)
597 * Recommended e2fsprogs version: 1.39.cfs2-0
599 * The backwards-compatible /proc/sys/portals symlink has been removed
600 in this release. Before upgrading, please ensure that you change
601 any configuration scripts or /etc/sysctl.conf files that access
602 /proc/sys/portals/* or sysctl portals.* to use the corresponding
603 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
604 in advance of the upgrade on any system running Lustre 1.4.6 or
605 newer, since /proc/sys/lnet was added in that version.
606 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
610 Frequency : only when quota is used
612 Description: avoid scanning export list for quota master
613 Details : Change the algorithms to avoid scanning export list in order
614 to improve the efficiency.
617 Frequency : MDS failover only, very rarely
619 Description: "went back in time" messages on mds failover
620 Details : The greatest transno may be lost when the current operation
621 finishes with an error (transno==0) and the client's last_rcvd
622 record is over-written. Save the greatest transno in the
623 mds_last_transno for this case.
626 Frequency : always for specific kernels and striping counts
628 Description: client may get "Matching packet too big" without ACL support
629 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
630 when trying to access files in certain configurations. The
631 clients should in fact be denied when mounting because they do
635 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
637 Description: many clients connecting with IO in progress causes connect timeouts
638 Details : Avoid synchronous journal commits to avoid delays caused by many
639 clients connecting/disconnecting when bulk IO is in progress.
640 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
641 OST_IO_PORTAL to avoid delays behind potentially many pending
645 Frequency : occasionally with multiple writers to a single file
647 Description: shared writes to file may result in wrong size reported by stat()
648 Details : Allow growing of kms when extent lock is cancelled
651 Frequency : always with random mmap IO to multi-striped file
653 Description: mmap write might be lost if we are writing to a 'hole' in stripe
654 Details : Only if the hole is at the end of OST object so that kms is too
655 small. Fix is to increase kms accordingly in ll_nopage.
658 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
660 Description: writes to a missing object would leak memory on the OST
661 Details : If there is an inconsistency between the MDS and OST filesystems,
662 such that the MDS references an object that doesn't exist, writes
663 to that object will leak memory due to incorrect cleanup in the
664 error handling path, eventually running out of memory on the OST.
669 Description: Creating too long symlink causes lustre errors
670 Details : Check symlink and name lengths before sending requests to MDS.
673 Frequency : only if flock is enabled (not on by default)
675 Description: posix locks not released on fd closure on 2.6.9+
676 Details : We failed to add posix locks to list of inode locks on 2.6.9+
677 kernels, this caused such locks not to be released on fd close and
678 then assertions on fs unmount about still used locks.
681 Frequency : MDS failover only, very rarely
683 Description: clients may get ASSERTION(granted_lock != NULL)
684 Details : When request was taking a long time, and a client was resending
685 a getattr by name lock request. The were multiple lock requests
686 with the same client lock handle and
687 mds_getattr_name->fixup_handle_for_resent_request found one of the
688 lock handles but later failed with ASSERTION(granted_lock != NULL).
693 Description: handle->h_buffer_credits > 0, assertion failure
694 Details : h_buffer_credits is zero after truncate, causing assertion
695 failure. This patch extends the transaction or creates a new
699 Frequency : NFS re-export or patchless client
700 Bugzilla : 11179, 10796
701 Description: Crash on NFS re-export node (__d_move)
702 Details : We do not want to hash the dentry if we don't have a lock.
703 But if this dentry is later used in d_move, we'd hit uninitialised
704 list head d_hash, so we just do this to init d_hash field but
705 leave dentry unhashed.
708 Frequency : NFS re-export or patchless client
710 Description: NFS exports has problem with symbolic link
711 Details : lustre client didn't properly install dentry when re-exported
712 to NFS or running patchless client.
715 Frequency : NFS re-export or patchless client
717 Description: Various nfs/patchless fixes.
718 Details : fixes reuse disconected alias for lookup process - this fixes
719 warning "find_exported_dentry: npd != pd",
720 fix permission error with open files at nfs.
721 fix apply umask when do revalidate.
724 Frequency : occasional
726 Description: Crash on NFS re-export node
727 Details : calling clear_page() on the wrong pointer triggered oops in
728 generic_mapping_read().
731 Frequency : rarely, using O_DIRECT IO
733 Description: unaligned directio crashes client with LASSERT
734 Details : check for unaligned buffers before trying any requests.
737 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
739 Description: stale data returned from RAID cache
740 Details : If only a small amount of IO is done to the RAID device before
741 reading it again it is possible to get stale data from the RAID
742 cache instead of reading it from disk.
745 Frequency : always for sles10 kernel
747 Description: sles10 support
748 Details : ll_follow_link: compile fixes and using of nd_set_link
752 Frequency : depends on arch, kernel and compiler version, always on sles10
755 Description: recursive or deep enough symlinks cause stack overflow
756 Details : getting rid of large stack-allocated variable in
760 Frequency : depends on hardware
762 Description: lustre write performance loss in the SLES10 kernel
763 Details : the performance loss is caused by using of write barriers in the
764 ext3 code. The SLES10 kernel turns barrier support on by
765 default. The fix is to undo that change for ldiskfs.
768 ------------------------------------------------------------------------------
770 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
772 * Support for kernels:
773 2.6.9-42.0.3EL (RHEL 4)
775 2.4.21-47.0.1.EL (RHEL 3)
776 2.6.12.6 vanilla (kernel.org)
780 Frequency : quota enabled and large files being deleted
782 Description: releasing more than 4GB of quota at once hangs OST
783 Details : If a user deletes more than 4GB of files on a single OST it
784 will cause the OST to spin in an infinite loop. Release
785 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
790 Description: statfs data retrieved from /proc may be stale or zero
791 Details : When reading per-device statfs data from /proc, in the
792 {kbytes,files}_{total,free,avail} files, it may appear
793 as zero or be out of date.
796 Frequency : systems with MD RAID1 external journal devices
798 Description: lconf's call to blkid is confused by RAID1 journal devices
799 Details : Use the "blkid -l" flag to locate the MD RAID device instead
800 of returning all block devices that match the journal UUID.
803 Frequency : always, for aggregate stripe size over 4GB
805 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
806 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
807 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
810 Frequency : always if "lfs find" used on a local file/directory
812 Description: "lfs find" segfaults if used on a local file/directory
813 Details : The case where a directory component was not specified wasn't
814 handled correctly. Handle this properly.
817 Frequency : always on ppc64
819 Description: the write to an ext3 filesystem mounted with mballoc got stuck
820 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
821 perform endianness conversion.
824 Frequency : rarely (truncate to non-zero file size after write under load)
825 Bugzilla : 10730, 10687
826 Description: Files padded with zeros to next 4K multiple
827 Details : With filesystems mounted using the "extents" option (2.6 kernels)
828 it is possible that files that are truncated to a non-zero size
829 immediately after being written are filled with zero bytes beyond
830 the truncated size. No file data is lost.
832 Severity : enhancement
834 Description: Allow recovery/failover for liblustre clients.
835 Details : liblustre clients were unaware of failover configurations until
838 Severity : enhancement
840 Description: user file locks should fail when not mounting with flock option
841 Details : Set up an error-returning stub in ll_file_operations.lock field
842 to prevent incorrect behaviour when client is mounted without
843 flock option. Also, set up properly f_op->flock field for
847 Frequency : always on ia64
849 Description: "lfs df" loops on printing out MDS statfs information
850 Details : The obd_ioctl_data was not initialized and in some systems
851 this caused a failure during the ioctl that did not return
852 an error. Initialize the struct and return an error on failure.
855 Frequency : SLES 9 only
857 Description: Error of copying files with lustre special EAs as root
858 Details : Client side always return success for setxattr call for lustre
859 special xattr (currently only "trusted.lov").
862 Frequency : rarely on clusters with both ia64+i386 clients
864 Description: ia64+i686 clients doing shared IO on the same file may LBUG
865 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
866 clients are doing concurrent writes to the same file it is
867 possible that the ia64 clients may LASSERT because the OST
868 extent locks are not PAGE_SIZE aligned. Ensure that grown
869 locks are always aligned on the request boundary.
872 Frequency : specific use, occasional
874 Description: Overwriting in use executable truncates on-disk binary image
875 Details : If one node attempts to overwrite an executable in use by
876 another node, we now correctly return ETXTBSY instead of
879 Severity : enhancement
881 Description: Async OSC create to avoid the blocking unnecessarily.
882 Details : If a OST has no remain object, system will block on the creating
883 when need to create a new object on this OST. Now, ways use
884 pre-created objects when available, instead of blocking on an
885 empty osc while others are not empty. If we must block, we block
886 for the shortest possible period of time.
891 Description: chmod on Lustre root is propagated to other clients
892 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
893 invalid by the follow_mount time.
896 Frequency : liblustre clients only
898 Description: Race in 'instant cancel' lock handling could lead to such locks
899 never to be granted in case of SMP MDS
900 Details : Do not destroy not yet granted but cbpending locks in
904 Frequency : replay/resend of open
906 Description: non null lock assetion failure in mds_intent_policy
907 Details : Trying to replay/resend lockless open requests resulted in
908 mds_open() returning 0 with no lock. Now it sets a flag if
909 a lock is going to be returned.
911 Severity : enhancement
913 Description: Checksum enhancements
914 Details : New checksum enhancements allow for resending RPCs that failed
917 Severity : enhancement
919 Description: Tunables on number of dirty pages in cacche
920 Details : Allow to set limit on number of dirty pages cached.
925 Description: client crash on unmount - lock still has references
926 Details : In some error handling cases it was possible to leak a lock
927 reference on a client while accessing a file. This was not
928 harmful to the client during operation, but would cause the
929 client to crash when the filesystem is unmounted.
932 Frequency : specific case, rare
934 Description: ETXTBSY on mds though file not in use by client
935 Details : ETXTBSY is no longer incorrectly returned when attempting to
936 chmod or chown a directory that the user previously tried to
937 execute or a currently-executing binary.
940 Frequency : extremely rare except on liblustre-based clients
942 Description: Lustre space not freed when files are deleted
943 Details : Clean up open-unlinked files after client eviction. Previously
944 the unlink was skipped and the files remained as orphans.
949 Description: OST failure "would be an LBUG" in waiting_locks_callback()
950 Details : In some cases it was possible to send a blocking callback to
951 a client doing a glimpse, even though that client didn't get
952 a lock granted. When the glimpse lock is cancelled on the OST
953 the freed lock is left on the waiting list and corrupted the list.
956 Frequency : all core dumps
958 Description: Broke core dumps to lustre
959 Details : Negative dentry may be unhashed if parent does not have UPDATE
960 lock, but some callers, e.g. do_coredump, expect dentry to be
961 hashed after successful create, hash it in ll_create_it.
963 ------------------------------------------------------------------------------
965 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
967 * Support for kernels:
968 2.6.9-42.0.2.EL (RHEL 4)
970 2.4.21-40.EL (RHEL 3)
971 2.6.12.6 vanilla (kernel.org)
975 Frequency : always on RHEL 3
977 Description: Number of open files grows over time
978 Details : The number of open files grows over time, whether or not
979 Lustre is started. This was due to a filp leak introduced
980 by one of our kernel patches.
982 ------------------------------------------------------------------------------
984 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
986 * Support for kernels:
989 2.4.21-40.EL (RHEL 3)
990 2.6.12.6 vanilla (kernel.org)
995 Bugzilla : 5719, 9635, 9792, 9684
996 Description: OST (or MDS) trips assertions in (re)connection under heavy load
997 Details : If a server is under heavy load and cannot reply to new
998 connection requests before the client resends the (re)connect,
999 the connection handling code can behave badly if two service
1000 threads are concurrently handing separate (re)connections from
1001 the same client. Add better locking to the connection handling
1002 code, and ensure that only a single connection will be processed
1003 for a given client UUID, even if the lock is dropped.
1005 Severity : enhancement
1007 Description: add TCP zero-copy support to kernel
1008 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1009 sends if the hardware supports scatter-gather and checksumming.
1010 This allows socklnd to do client-write and server-read more
1011 efficiently and reduce CPU utilization from skbuf copying.
1014 Frequency : only if NFS exporting from client
1016 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1017 Details : If setting attributes on a file created under NFS that had
1018 never been opened it would be possible to oops the client
1019 if the file had no objects.
1023 Bugzilla : 9326, 10402, 10897
1024 Description: client crash in ptlrpcd_wake() thread when sending async RPC
1025 Details : It is possible that ptlrpcd_wake() dereferences a freed async
1026 RPC. In rare cases the ptlrpcd thread alread processed the RPC
1027 before ptlrpcd_wake() was called and the request was freed.
1030 Frequency : always for liblustre
1032 Description: liblustre client does MDS+OSTs setattr RPC for each write
1033 Details : When doing a write from a liblustre client, the client
1034 incorrectly issued an RPC to the MDS and each OST the file was
1035 striped over in order to update the timestamps. When writing
1036 with small chunks and many clients this could overwhelm the MDS
1037 with RPCs. In all cases it would slow down the write because
1038 these RPCs are unnecessary.
1040 Severity : enhancement
1042 Description: allow number of MDS service threads to be changed at module load
1043 Details : It is now possible to change the number of MDS service threads
1044 running. Adding "options mds mds_num_threads={N}" to the MDS's
1045 /etc/modprobe.conf will set the number of threads for the next
1046 time Lustre is restarted (assuming the "mds" module is also
1047 reloaded at that time). The default number of threads will
1048 stay the same, 32 for most systems.
1053 Description: OST crash if filesystem is unformatted or corrupt
1054 Details : If an OST is started on a device that has never been formatted
1055 or if the filesystem is corrupt and cannot even mount then the
1056 error handling cleanup routines would dereference a NULL pointer.
1061 Description: NULL pointer deref in llap_from_page.
1062 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1063 incorrect) mapping if the page was truncated/reclaimed while it was
1064 searched for. Check for this condition and skip such pages when
1065 doing readahead. Introduce extra check to llap_from_page() to
1066 verify page->mapping->host is non-NULL (so page is not anonymous).
1069 Frequency : Sometimes when using sys_sendfile
1071 Description: "page not covered by a lock" warnings from ll_readpage
1072 Details : sendfile called ll_readpage without right page locks present.
1073 Now we introduced ll_file_sendfile that does necessary locking
1074 around call to generic_file_sendfile() much like we do in
1078 Frequency : with certain MDS communication failures at client mount time
1080 Description: NULL pointer deref after failed client mount
1081 Details : a client connection request may delayed by the network layer
1082 and not be sent until after the PTLRPC layer has timed out the
1083 request. If the client fails the mount immediately it will try
1084 to clean up before the network times out the request. Add a
1085 reference from the request import to the obd device and delay
1086 the cleanup until the network drops the request.
1089 Frequency : occasionally during client (re)connect
1091 Description: assertion failure during client (re)connect
1092 Details : processing a client connection request may be delayed by the
1093 client or server longer than the client connect timeout. This
1094 causes the client to resend the connection request. If the
1095 original connection request is replied in this interval, the
1096 client may trip an assertion failure in ptlrpc_connect_interpret()
1097 which thought it would be the only running connect process.
1100 Frequency : only with obd_echo servers and clients that are rebooted
1102 Description: kernel BUG accessing uninitialized data structure
1103 Details : When running an obd_echo server it did not start the ping_evictor
1104 thread, and when a client was evicted an uninitialized data
1105 structure was accessed. Start the ping_evictor in the RPC
1106 service startup instead of the OBD startup.
1108 Severity : enhancement
1109 Bugzilla : 10193 (patchless)
1110 Description: Remove dependency on various unexported kernel interfaces.
1111 Details : No longer need reparent_to_init, exit_mm, exit_files,
1112 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1115 Frequency : rare (only users of deprecated and unsupported LDAP config)
1117 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1118 Details : LDAP apparently contains 'lustreName' attributes instead of
1119 'name'. A simple remapping of the name is sufficient.
1122 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1124 Description: waiting_locks_callback trips kernel BUG if client is evicted
1125 Details : Running with the dump_on_timeout debug flag turned on makes
1126 it possible that the waiting_locks_callback() can try to dump
1127 the Lustre kernel debug logs from an interrupt handler. Defer
1128 this log dumping to the expired_lock_main() thread.
1130 Severity : enhancement
1132 Description: Support NFS exporting on 2.6 kernels.
1133 Details : Implement non-rawops metadata methods for NFS server to use without
1134 changing NFS server code.
1137 Frequency : very rare (synthetic metadata workload only)
1139 Description: two racing renames might cause an MDS thread to deadlock
1140 Details : Running the "racer" program may cause one MDS thread to rename
1141 a file from being the source of a rename to being the target of
1142 a rename at exactly the same time that another thread is doing
1143 so, and the second thread has already enqueued these locks after
1144 doing a lookup of the target and is trying to relock them in
1145 order. Ensure that we don't try to re-lock the same resource.
1148 Frequency : only very large systems with liblustre clients
1150 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1151 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1152 all OSTs in parallel. This allows the eviction of stale liblustre
1153 clients to proceed much faster than if they were done in series,
1154 and also offers similar improvements for other set_info RPCs.
1159 Description: excessive CPU usage during initial read phase on client
1160 Details : During the initial read phase on a client, it would agressively
1161 retry readahead on the file, consuming too much CPU and impacting
1162 performance (since 1.4.5.8). Improve the readahead algorithm
1163 to avoid this, and also improve some other common cases (read
1164 of small files in particular, where "small" is files smaller than
1165 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1170 Description: MDS crash when receiving packet with unknown intent.
1171 Details : Do not LBUG in unknown intent case, just return -EFAULT
1173 Severity : enhancement
1174 Bugzilla : 9293, 9385
1175 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1176 Details : Do not serialize getattr (non-intent version) and statfs.
1179 Frequency : occasional, when OST network is overloaded/intermittent
1181 Description: client evicted by OST after bulk IO timeout
1182 Details : If a client sends a bulk IO request (read or write) the OST
1183 may evict the client if it is unresposive to its data GET/PUT
1184 request. This is incorrect if the network is overloaded (takes
1185 too long to transfer the RPC data) or dropped the OST GET/PUT
1186 request. There is no need to evict the client at all, since
1187 the pinger and/or lock callbacks will handle this, and the
1188 client can restart the bulk request.
1191 Frequency : Always when mmapping file with no objects
1193 Description: client crashes when mmapping file with no objects
1194 Details : Check that we actually have objects in a file before doing any
1195 operations on objects in ll_vm_open, ll_vm_close and
1201 Description: Request leak when working with deleted CWD
1202 Details : Introduce advanced request refcount tracking for requests
1203 referenced from lustre intent.
1205 Severity : Enhancement
1207 Description: Cache open file handles on client.
1208 Details : MDS now will return special lock along with openhandle, if
1209 requested and client is allowed to hold openhandle, even if unused,
1210 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1211 closing files for every read/write openration.
1213 Severity : Enhancement
1215 Description: Cache open negative dentries on client when possible.
1216 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1217 negative dentries on lock revocation.
1222 Description: Remounting a client read-only wasn't possible with a zconf mount
1223 Details : It wasn't possible to remount a client read-only with llmount.
1225 Severity : enhancement
1226 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1227 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1228 MPICH is included to add Lustre-specific ADIO interfaces.
1229 This is based closely on the UFS ADIO layer and only differs
1230 in file creation, in order to allow the OST striping to be set.
1231 This is user-contributed code and not supported by CFS.
1236 Description: extended inode attributes (immutable, append-only) work improperly
1237 when 2.4 and 2.6 kernels are used on client/server or vice versa
1238 Details : Introduce kernel-independent values for these flags.
1240 Severity : enhancement
1243 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1244 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1245 granularity. Now, Fractional MB granularity is be supported,
1246 this is very useful for low memory system.
1248 Severity : enhancement
1250 Description: Getattr by fid
1251 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1252 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1253 have locks and avoids getattr rpc requests for it on every stat.
1256 Frequency : Always, for filesystems larger than 2TB
1258 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1259 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1260 filesystems larger than 2TB to be mounted with mballoc enabled.
1263 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1265 Description: filesystem corruption for non-standard kernels and very large OSTs
1266 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1267 filesystems larger than 2TB is mounted then the kernel will
1268 silently corrupt the start of the filesystem. CONFIG_LBD is
1269 enabled for all CFS-supported kernels, but the possibility of
1270 this happening with a modified kernel config exists.
1272 Severity : enhancement
1274 Description: add client O_DIRECT support for 2.6 kernels
1275 Details : It is now possible to do O_DIRECT reads and writes to files
1276 in the Lustre client mountpoint on 2.6 kernel clients.
1278 Severity : enhancement
1280 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1281 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1282 parallel, not waiting for response from every OST before sending
1283 a rpc to the next OST.
1288 Description: setattr vs write race when updating file timestamps
1289 Details : Client processes that update a file timestamp into the past
1290 right after writing to the file (e.g. tar) it is possible that
1291 the updated file modification time can be reset to the current
1292 time due to a race between processing the setattr and write RPC.
1294 Severity : enhancement
1296 Description: Bring 'lfs find' closer in line with regular Linux find.
1297 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1298 -print0 options and obtains all the needed info through the lustre
1301 Severity : enhancement
1303 Description: support up to 1024 configured devices on one node
1304 Details : change obd_dev array from statically allocated to dynamically
1305 allocated structs as they are first used to reduce memory usage
1310 Description: Flush dirty partially truncated pages during truncate
1311 Details : Immediatelly flush partially truncated pages in filter_setattr,
1312 this way we completely avoid having any pages in page cache on OST
1313 and can retire ugly workarounds during writes to flush such pages.
1318 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1319 Details : protect inode from truncation within vfs_unlink() context
1320 just take a reference before calling vfs_unlink() and release it
1321 when parent's i_sem is free.
1324 Frequency : always, if extents are used on OSTs
1326 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1327 Details : OSTs using the extents format would not zero the high 16 bits of
1328 the index physical block number. This is not a problem for any
1329 OST filesystems smaller than 16TB, and no kernels support ext3
1330 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1331 new/modified files) and can be fixed for existing filesystems
1332 with e2fsprogs-1.39-cfs1.
1337 Description: import connection selection may be incorrect if timer wraps
1338 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1339 import connections to be ignored if the 32-bit jiffies counter
1340 wraps. Use a 64-bit jiffies counter.
1343 Frequency : during server recovery
1345 Description: crash after server is denying duplicate export
1346 Details : If clients are resending connect requests to the server, the
1347 server refuses to allow a client to connect multiple times.
1348 Fixed a bug in the handling of this case.
1351 Frequency : very large clusters immediately after boot
1353 Description: LNET request buffers exhausted under heavy short-term load
1354 Details : If a large number of client requests are generated on a service
1355 that has previously never seen so many requests it is possible
1356 that the request buffer growth cannot keep up with the spike in
1357 demand. Instead of dropping incoming requests, they are held in
1358 the LND until the RPC service can accept more requests.
1361 Frequency : Sometimes during replay
1363 Description: Assertion failure in ll_local_open after replay.
1364 Details : If replay happened on an open request reply before we were able
1365 to set replay handler, reply will become not swabbed tripping the
1366 assertion in ll_local_open. Now we set the handler right after
1367 recognising of open request
1370 Frequency : very rare
1372 Description: kernel reports "badness in vsnprintf"
1373 Details : Reading from the "recovery_status" /proc file in small chunks
1374 may cause a negative length in lprocfs_obd_rd_recovery_status()
1375 call to vsnprintf() (which is otherwise harmless). Exit early
1376 if there is no more space in the output buffer.
1378 Severity : enhancement
1380 Description: clear OBD RPC statistics by writing to them
1381 Details : It is now possible to clear the OBD RPC statistics by writing
1382 to the "stats" file.
1387 Description: Client mtime is not the same on different clients after utimes
1388 Details : In some cases, the client was using the utimes() syscall on
1389 a file cached on another node. The clients now validate the
1390 ctime from the MDS + OSTs to determine which one is right.
1395 Description: Inability to activate failout mode
1396 Details : lconf script incorrectly assumed that in python string's numeric
1397 value is used in comparisons.
1400 Frequency : always with multiple stripes per file
1402 Description: Inefficient object allocation for mutli-stripe files
1403 Details : When selecting which OSTs to stripe files over, for files with
1404 a stripe count that divides evenly into the number of OSTs,
1405 the MDS is always picking the same starting OST for each file.
1406 Return the OST selection heuristic to the original design.
1411 Description: mount failures may take full timeout to return an error
1412 Details : Under some heavy load conditions it is possible that a
1413 failed mount can wait for the full obd_timeout interval,
1414 possibly several minutes, before reporting an error.
1415 Instead return an error as soon as the status is known.
1417 ------------------------------------------------------------------------------
1419 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
1421 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1422 INTEROPERATE with older versions automatically. Please read the
1423 user documentation before upgrading any part of a live system.
1424 * WARNING: Lustre networking configuration changes are required with
1425 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1428 * Support for kernels:
1429 2.6.9-22.0.2.EL (RHEL 4)
1430 2.6.5-7.244 (SLES 9)
1431 2.6.12.6 vanilla (kernel.org)
1434 Severity : enhancement
1435 Bugzilla : 7981/8208
1436 Description: Introduced Lustre Networking (LNET)
1437 Details : LNET is new networking infrastructure for Lustre, it includes
1438 a reorganized network configuration mode (see the user
1439 documentation for full details) as well as support for routing
1440 between different network fabrics. Lustre Networking Devices
1441 (LNDs) for the supported network fabrics have also been
1442 created for this new infrastructure.
1444 Severity : enhancement
1445 Description: Introduced Access control lists
1446 Details : clients can set ACLs on files and directories in order to have
1447 more fine-grained permissions than the standard Unix UGO+RWX.
1448 The MDS must be started with the "-o acl" mount option.
1450 Severity : enhancement
1451 Description: Introduced filesystem quotas
1452 Details : Administrators may now establish per-user quotas on the
1455 Severity : enhancement
1457 Description: Configuration change for the XT3
1458 The PTLLND is now used to run Lustre over Portals on the XT3
1459 The configure option(s) --with-cray-portals are no longer used.
1460 Rather --with-portals=<path-to-portals-includes> is used to
1461 enable building on the XT3. In addition to enable XT3 specific
1462 features the option --enable-cray-xt3 must be used.
1467 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1468 Details : A race between allocating a new llcd and re-getting the llcd_lock
1469 allowed another thread to grab newly-allocated llcd.
1471 Severity : enhancement
1473 Description: 2.6 OST async journal commit and locking fix to improve performance
1474 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1475 2.6 kernels are now async as they already were in 2.4 kernels so
1476 that they can commit concurrently with the network bulk transfer.
1477 For block-allocated files the filter allocation semaphore is held
1478 to avoid filesystem fragmentation during allocation. BKL lock
1479 removed for 2.6 xattr operations where it is no longer needed.
1484 Description: lconf incorrectly determined whether two IP networks could talk
1485 Details : In some more complicated routing and multiple-network
1486 configurations, lconf will avoid trying to make a network
1487 connection to a disjoint part of the IP space. It was doing the
1488 math incorrectly for one set of cases.
1493 Description: Fix for potential infinite loop processing records in an llog.
1494 Details : If an llog record is corrupted/zeroed, it is possible to loop
1495 forever in llog_process(). Validate the llog record length
1496 and skip the remainder of the block on error.
1499 Frequency : occasional (liblustre only)
1501 Description: liblustre could not open files whose last component is a symlink
1502 Details : sysio_path_walk() would incorrectly pass the open intent to
1503 intermediate path components.
1506 Frequency : rare (liblustre only with non-standard tuning)
1507 Bugzilla : 7201 (7350)
1508 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1509 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1510 after it has been released, only do this for referenced locks
1512 Severity : enhancement
1514 Description: specifying an (invalid) directory default stripe_size of -1
1515 would reset the directory default striping
1516 Details : stripe_size -1 was used internally to signal directory stripe
1517 removal, now use "all default" to signal dir stripe removal
1518 as a directory striping of "all default" is not useful
1521 Frequency : common for large clusters running liblustre clients
1523 Description: doing an ls when liblustre clients are running is slow
1524 Details : sending a glimpse AST to a liblustre client waits for every AST
1525 to time out, as liblustre clients will not respond. Since they
1526 cannot cache data we refresh the OST lock LVB from disk instead.
1528 Severity : enhancement
1530 Description: doing an ls at the same time as file IO can be slow
1531 Details : enqueue and other "small" requests can be blocked behind many
1532 large IO requests. Create a new OST IO portal for non-IO
1533 requests so they can be processed faster.
1536 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1538 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1539 created file with mode 000
1540 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1541 allow an ftruncate() as a non-root user to a file with mode 000.
1542 Lustre now respects this flag to disable mode checks when
1543 truncating a file owned by the user
1546 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1549 Description: Revoking locks from clients that went dead or catatonic might take
1551 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1552 cancellation of such locks instant on servers without waiting for
1553 any reply from clients. Clients drops these locks when cancel
1554 notification from server is received without replying.
1557 Frequency : liblustre-only, when liblustre client dies or becomes busy
1559 Description: Doing ls on Linux clients can take a long time with active
1561 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1562 granting such locks to it in the first place if possible. Locks
1563 are taken by proxy on the OST during the read or write and
1564 dropped immediately afterward. Add connect flags handling, do
1565 not grant locks to liblustre clients for glimpse ASTs.
1567 Severity : enhancement
1569 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1570 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1571 sequential reads and behaves badly with applications doing random
1572 reads. Improve it by reading ahead at least read region, and
1573 avoiding excessive large RPC for small reads.
1575 Severity : enhancement
1577 Description: Creating more than 1000 files for a single job may cause a load
1578 imbalance on the OSTs if there are also a large number of OSTs.
1579 Details : qos_prep_create() uses an OST index reseed value that is an
1580 even multiple of the number of available OSTs so that if the
1581 reseed happens in the middle of the object allocation it will
1582 still utilize the OSTs as uniformly as possible.
1587 Description: OST or MDS may oops in ping_evictor_main()
1588 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1589 but doesn't restart at beginning of obd_exports_timed list
1592 Severity : enhancement
1594 Description: improve by-nid export eviction on the MDS and OST
1595 Details : allow multiple exports with the same NID to be evicted at one
1596 time without re-searching the exports list.
1599 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1601 Description: MDS may oops in groups_free()
1602 Details : in rare race conditions a newly allocated group_info struct is
1603 freed again, and this can be NULL. The 2.4 compatibility code
1604 for groups_free() checked for a NULL pointer, but 2.6 did not.
1607 Frequency : common for liblustre clients doing little filesystem IO
1608 Bugzilla : 9352, 7313
1609 Description: server may evict liblustre clients accessing contended locks
1610 Details : if a client is granted a lock or receives a completion AST
1611 with a blocking AST already set it would not reply to the AST
1612 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1615 Frequency : lfs setstripe, only systems with more than 160 OSTs
1617 Description: unable to set striping with a starting offset beyond OST 160
1618 Details : llapi_create_file() incorrectly limited the starting stripe
1619 index to the maximum single-file stripe count.
1622 Frequency : LDAP users only
1624 Description: lconf did not handle in-kernel recovery with LDAP properly
1625 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1627 Severity : enhancement
1629 Description: bind OST threads to NUMA nodes to improve performance
1630 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1631 node and do their allocations there to localize memory access
1633 Severity : enhancement
1635 Description: llmount can determine client NID directly from Myrinet (GM)
1636 Details : the client NID code from gmnalnid was moved directly into
1637 llmount, removing the need to use this or specifying the
1638 client NID explicitly when mounting GM clients with zeroconf
1641 Frequency : if client is started with down MDS
1643 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1644 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1645 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1646 ptlrpc_import_delayed_req() because we didn't check if the
1647 request was interrupted, and we also didn't break out of the
1648 event loop if there was no timeout
1653 Description: data loss during non-page-aligned writes to a single file from
1654 both multiple nodes and multiple threads on one node at same time
1655 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1656 inconsistency led to false short-reads, that were cached and later
1657 used by ->prepare_write() to fill in partially written page,
1658 leading to data loss.
1661 Frequency : always, if lconf --abort_recovery used
1663 Description: lconf --abort_recovery fails with 'Operation not supported'
1664 Details : lconf was attempting to abort recovery on the MDT device and not
1667 Severity : enhancement
1669 Description: remove cleanup logs
1670 Details : replace lconf-generated cleanup logs with lustre internal
1671 cleanup routines. Eliminates the need for client-cleanup and
1674 Severity : enhancement
1676 Description: add support for EAs (user and system) on lustre filesystems
1677 Details : it is now possible to store extended attributes in the Lustre
1678 client filesystem, and with the user_xattr mount option it
1679 is possible to allow users to store EAs on their files also
1681 Severity : enhancement
1683 Description: Add possibility (config option) to show minimal available OST free
1685 Details : When compiled with --enable-mindf configure option, statfs(2)
1686 (and so, df) will return least minimal free space available from
1687 all OSTs as amount of free space on FS, instead of summary of
1688 free spaces of all OSTs.
1690 Severity : enhancement
1692 Description: do not expand extent locks acquired on OST-side
1693 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1694 by server: they are not cached anyway.
1697 Frequency : when mmap is used/binaries executed from Lustre
1699 Description: Unmmap pages before throwing them away from read cache.
1700 Details : llap_shrink cache now attempts to unmap pages before discarding
1701 them (if unmapping failed - do not discard). SLES9 kernel has
1702 extra checks that trigger if this unmapping is not done first.
1707 Description: lconf didn't resolve symlinks before checking to see whether a
1708 given mountpoint was already in use
1711 Frequency : when migrating failover services
1712 Bugzilla : 6395, 9514
1713 Description: When migrating a subset of services from a node (e.g. failback
1714 from a failover service node) the remaining services would
1715 time out and evict clients.
1716 Details : lconf --force (implied by --failover) sets the global obd_timeout
1717 to 5 seconds in order to quickly disconnect, but this caused
1718 other RPCs to time out too quickly. Do not change the global
1719 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1721 Severity : enhancement
1722 Frequency : if MDS is started with down OST
1723 Bugzilla : 9439,5706
1724 Description: Allow startup/shutdown of an MDS without depending on the
1725 availability of the OSTs.
1726 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1727 Add appropriate locking and lov-osc refcounts for safe
1728 cleaning. Add osc abort_inflight calls in case the
1729 synchronize never started.
1732 Frequency : occasional (Cray XT3 only)
1734 Description: root not authorized to access files in CRAY_PORTALS environment
1735 Details : The client process capabilities were not honoured on the MDS in
1736 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1737 been accessed by an authorized user then root was able to access
1738 the file on the local client also. The root user capabilities
1739 are now allowed on the MDS, as this environment has secure UID.
1742 Frequency : occasional
1744 Description: ldiskfs "too long searching" message happens too often
1745 Details : A debugging message (otherwise harmless) prints too often on
1746 the OST console. This has been reduced to only happen when
1747 there are fragmentation problems on the filesystem.
1752 Description: Division by zero in statfs when all OSCs are inactive
1753 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
1754 lov_statfs divides by value returned by lov_get_stripecnt().
1758 Bugzilla : 9489, 3273
1759 Description: First write from each client to each OST was only 4kB in size,
1760 to initialize client writeback cache, which caused sub-optimal
1761 RPCs and poor layout on disk for the first writen file.
1762 Details : Clients now request an initial cache grant at (re)connect time
1763 and so that they can start streaming writes to the cache right
1764 away and always do full-sized RPCs if there is enough data.
1765 If the OST is rebooted the client also re-establishes its grant
1766 so that client cached writes will be honoured under the grant.
1771 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
1772 Details : Now I/O RPCs go to different portal number and (presumably) fast
1773 lock requests (and glimses) and other RPCs get their own service
1774 threads pool that should be able to service those RPCs
1777 Severity : enhancement
1779 Description: Ability to exchange lustre version between client and servers and
1780 issue warnings at client side if client is too old. Also for
1781 liblustre clients there is ability to refuse connection of too old
1783 Details : New 'version' field is added to connect data structure that is
1784 filled with version info. That info is later checked by server and
1788 Frequency : rare, liblustre only.
1789 Bugzilla : 9296, 9581
1790 Description: Two simultaneous writes from liblustre at offset within same page
1791 might proceed at the same time overwriting eachother with stale
1793 Details : I/O lock withing llu_file_prwv was released too early, before data
1794 actually was hitting the wire. Extended lock-holding time until
1795 server acknowledges receiving data.
1798 Frequency : extremely rare. Never observed in practice.
1800 Description: avoid generating lustre_handle cookie of 0.
1801 Details : class_handle_hash() generates handle cookies by incrementing
1802 global counter, and can hit 0 occasionaly (this is unlikely, but
1803 not impossible, because initial value of cookie counter is
1804 selected randonly). Value of 0 is used as a sentinel meaning
1805 "unassigned handle" --- avoid it. Also coalesce two critical
1806 sections in this function into one.
1808 Severity : enhancement
1810 Description: allow liblustre clients to delegate truncate locking to OST
1811 Details : To avoid overhead of locking, liblustre client instructs OST to
1812 take extent lock in ost_punch() on client's behalf. New connection
1813 flag is added to handle backward compatibility.
1815 Severity : enhancement
1816 Bugzilla : 4928, 7341, 9758
1817 Description: allow number of OST service threads to be specified
1818 Details : a module parameter allows the number of OST service threads
1819 to be specified via "options ost ost_num_threads={N}" in the
1820 OSS's /etc/modules.conf or /etc/modprobe.conf.
1824 Bugzilla : 6146, 9635, 9895
1825 Description: servers crash with bad pointer in target_handle_connect()
1826 Details : In rare cases when a client is reconnecting it was possible that
1827 the connection request was the last reference for that export.
1828 We would temporarily drop the export reference and get a new
1829 one, but this may have been the last reference and the export
1830 was just destroyed. Get new reference before dropping old one.
1832 Severity : enhancement
1833 Frequency : if client is started with failover MDS
1835 Description: Allow multiple MDS hostnames in the mount command
1836 Details : Try to read the configuration from all specified MDS
1837 hostnames during a client mount in case the "primary"
1840 Severity : enhancement
1842 Description: Stop sending data to evicted clients as soon as possible.
1843 Details : Check if the client we are about to send or are sending data to
1844 was evicted already. (Check is done every second of waiting,
1845 for which l_wait_event interface was extended to allow checking
1846 of exit condition at specified intervals).
1849 Frequency : rare, normally only when NFS exporting is done from client
1851 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
1853 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
1856 Frequency : always, if srand() is called before liblustre initialization
1858 Description: Liblustre uses system PRNG disturbing its usage by user application
1859 Details : Introduce internal to lustre fast and high-quality PRNG for
1860 lustre usage and make liblustre and some other places in generic
1861 lustre code to use it.
1863 Severity : enhancement
1864 Bugzilla : 9477, 9557, 9870
1865 Description: Verify that the MDS configuration logs are updated when xml is
1866 Details : Check if the .xml configuration logs are newer than the config
1867 logs stored on the MDS and report an error if this is the case.
1868 Request --write-conf, or allow starting with --old_conf.
1870 Severity : enhancement
1872 Description: Handle symlinks in the path when checking if Lustre is mounted.
1873 Details : Resolve intermediate symlinks when checking if a client has
1874 mounted a filesystem to avoid duplicate client mounts.
1879 Description: lconf can hit an error exception but still return success.
1880 Details : The lconf command catches the Command error exception at the top
1881 level script context and will exit with the associated exit
1882 status, but doesn't ensure that this exit status is non-zero.
1887 Description: failure of ptlrpc thread startup can cause oops
1888 Details : Starting a ptlrpc service thread can fail if there are a large
1889 number of threads or the server memory is very fragmented.
1890 Handle this without oopsing.
1893 Frequency : always, only if liblustre and non-default acceptor port was used
1895 Description: liblustre cannot connect to servers with non-default acceptor port
1896 Details : tcpnal_set_default_params() was not called and was therefore
1897 ignoring the environment varaible TCPNAL_PORT, as well as other
1898 TCPNAL_ environment variables
1903 Description: two objects could be created on the same OST for a single file
1904 Details : If an OST is down, in some cases it was possible to create two
1905 objects on a single OST for a single file. No problems other
1906 than potential performance impact and spurious error messages.
1910 Bugzilla : 5681, 9562
1911 Description: Client may oops in ll_unhash_aliases
1912 Details : Client dcache may become inconsistent in race condition.
1913 In some cases "getcwd" can fail if the current directory is
1919 Description: Inode refcounting problems in NFS export code
1920 Details : link_raw functions used to call d_instantiate without obtaining
1921 extra inode reference first.
1925 Bugzilla : 9942, 9903
1926 Description: Referencing freed requests leading to crash, memleaks with NFS.
1927 Details : We used to require that call to ll_revalidate_it was always
1928 followed by ll_lookup_it. Also with revalidate_special() it is
1929 possible to call ll_revalidate_it() twice for the same dentry
1930 even if first occurence returned success. This fix changes semantic
1931 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
1932 reference on a request referred from the intent.
1933 ll_intent_release() then releases such a request.
1936 Frequency : rare, normally benchmark loads only
1938 Description: unlinked inodes were kept in memory on the client
1939 Details : If a client is repeatedly creating and unlinking files it
1940 can accumulate a lot of stale inodes in the inode slab cache.
1941 If there is no other client load running this can cause the
1942 client node to run out of memory. Instead flush old inodes
1943 from client cache that have the same inode number as a new inode.
1946 Frequency : SLES9 2.6.5 kernel and long filenames only
1947 Bugzilla : 9969, 10379
1948 Description: utime reports stale NFS file handle
1949 Details : SLES9 uses out-of-dentry names in some cases, which confused
1950 the lustre dentry revalidation. Change it to always use the
1954 Frequency : rare, unless heavy write-truncate concurrency is continuous
1955 Bugzilla : 4180, 6984, 7171, 9963, 9331
1956 Description: OST becomes very slow and/or deadlocked during object unlink
1957 Details : filter_destroy() was holding onto the parent directory lock
1958 while truncating+unlinking objects. For very large objects this
1959 may block other threads for a long time and slow overall OST
1960 responsiveness. It may also be possible to get a lock ordering
1961 deadlock in this case, or run out of journal credits because of
1962 the combined truncate+unlink. Solution is to do object truncate
1963 first in one transaction without parent lock, and then do the
1964 final unlink in a new transaction with the parent lock. This
1965 reduces the lock hold time dramatically.
1968 Frequency : rare, 2.4 kernels only
1970 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
1971 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
1972 before unmounting their filesystem in order to avoid deadlock.
1973 The kernel_locked() function in 2.4 kernels only checks whether
1974 the kernel lock is held, not whether it is this process that is
1975 holding it as 2.6 kernels do.
1980 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
1981 Details : The client ptlrpc code may be trying to reconnect to a down
1982 server before a previous connection attempt has timed out.
1983 Increase the reconnect interval to be longer than the connection
1984 timeout interval to avoid sending duplicate connections to
1988 Frequency : echo_client brw_test command
1990 Description: fix echo_client to work with OST preallocated code
1991 Details : OST preallocation code (5137) didn't take echo_client IO path
1992 into account: echo_client calls filter methods outside of any
1993 OST thread and, hence, there is no per-thread preallocated
1994 pages and buffers to use. Solution: hijack pga pages for IO. As
1995 a byproduct, this avoids unnecessary data copying.
1999 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2000 Description: Client can oops in mdc_commit_close() after open replay
2001 Details : It was possible for the MDS to return an open request with no
2002 transaction number in mds_finish_transno() if the client was
2003 evicted, but without actually returning an error. Clients
2004 would later try to replay that open and may trip an assertion
2005 Simplify the client close codepath, and always return an error
2006 from the MDS in case the open is not successful.
2009 Frequency : rare, 2.6 OSTs only
2011 Description: OST may deadlock under high load on fragmented files
2012 Details : If there was a heavy load and highly-fragmented OST filesystems
2013 it was possible to have all the OST threads deadlock waiting on
2014 allocation of biovecs, because the biovecs were not released
2015 until the entire RPC IO was completed. Instead, release biovecs
2016 as soon as they are complete to ensure forward IO progress.
2018 Severity : enhancement
2020 Description: Support for specifying external journal device at mount
2021 Details : If an OST or MDS device is formatted with an external journal
2022 device, this device major/minor is stored in the ext3 superblock
2023 and may not be valid for failover. Allow detecting and
2024 specifying the external journal at mount time.
2029 Description: Mounting an MDS with pending unlinked files may cause oops
2030 Details : target_finish_recovery() calls mds_postrecov() which returned
2031 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2032 considers this an error and immediately begins cleaning up the
2033 lov, just after starting the mds_lov process
2035 Severity : enhancement
2037 Description: Implement 'lfs df' to report actual free space on per-OST basis
2038 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2039 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2040 usage of objects; '-h' to report in human readable format.
2042 ------------------------------------------------------------------------------
2044 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
2051 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2052 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2053 at startup during orphan unlinking. Instead, skip buddy update
2054 before setup, as it will be regenerated after recovery is complete.
2059 Description: If an OST is inactive, its locks might reference stale inodes.
2060 Details : lov_change_cbdata() must iterate over all namespaces, even if
2061 they are inactive to clear inode references from the lock.
2063 Severity : enhancement
2064 Frequency : occasional, if non-standard max_dirty_mb used
2066 Description: Client will block write RPCs if not enough grant
2067 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2068 then the client will block writes while waiting for another RPC
2069 to complete instead of consuming its dirty limit. With change
2070 we get improved performance when max_dirty_mb is small.
2072 Severity : enhancement
2073 Bugzilla : 3389, 6253
2074 Description: Add support for supplementary groups on the MDS.
2075 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2076 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2077 lookups for user supplementary groups into a cache.
2081 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2082 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2083 for liblustre O_EXCL is enforced by MDS.
2087 Description: readdir never returns NULL in liblustre.
2088 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2089 dentry in struct dirent.
2093 Frequency : liblustre only, depends on application IO pattern
2094 Description: liblustre clients evicted if not contacting servers
2095 Details : Don't put liblustre clients into the ping_evictor list, so
2096 they will not be evicted by the pinger ever.
2098 Severity : enhancement
2100 Description: Add ability to evict clients by NID from MDS.
2101 Details : By echoing "nid:$NID" string into
2102 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2103 $NID would be instantly evicted from this MDS and from all active
2104 OSTs connected to it.
2108 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2109 Details : lookup_it_finish() used to query file size from OSTs that was not
2114 Description: service threads change working directory to that of init
2115 Details : Starting lustre service threads may pin the working directory
2116 of the parent thread, making that filesystem busy. Threads
2117 now change to the working directory of init to avoid this.
2121 Frequency : during shutdown only
2122 Description: shutdown with a failed MDS or OST can cause unmount to hang
2123 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2128 Frequency : 2.6 only
2129 Description: chmod/chown may include an extra supplementary group
2130 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2131 supplementary group and if none is specified this is used.
2134 Bugzilla : 5479 (6816)
2136 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2137 Details : If lookup is in progress on a file that is unlinked we might try
2138 to revalidate the inode and fail in revalidate after lookup is
2139 complete and ll_file_open() enqueues the open again but
2140 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2143 Frequency : always, if lconf --abort_recovery used
2145 Description: lconf --abort_recovery fails with 'Operation not supported'
2146 Details : lconf was attempting to abort recovery on the MDT device and not
2149 ------------------------------------------------------------------------------
2151 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2156 Frequency : rare (only unsupported configurations with a node running as an
2158 Bugzilla : 6514, 5137
2159 Description: Mounting a Lustre file system on a node running as an OST could
2161 Details : OSTs now preallocates memory needed to write out data at
2162 startup, instead of when needed, to avoid having to
2163 allocate memory in possibly low memory situations.
2164 Specifically, if the file system is mounted on on OST,
2165 memory pressure could force it to try to write out data,
2166 which it needed to allocate memory to do. Due to the low
2167 memory, it would be unable to do so and the node would
2168 become unresponsive.
2170 Severity : enhancement
2172 Description: Addition of lconf --service command line option
2173 Details : lconf now accepts a '--service <arg>' option, which is
2174 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2176 Severity : enhancement
2178 Description: Failover mode is now the default for OSTs.
2179 Details : By default, OSTs will now run in failover mode. To return to
2180 the old behaviour, add '--failout' to the lmc line for OSTs.
2182 Severity : enhancement
2184 Description: Health checks are now provided for MDS and OSTs
2185 Details : Additional detailed health check information on MSD and OSTs
2186 is now provided through the procfs health_check value.
2189 Frequency : occasional, depends on IO load
2191 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2192 after numerous create/delete cycles
2193 Details : The ext3 inode allocation policy would not allocate new inodes
2194 very well on the OSTs because there are no new directories
2195 being created. Instead we look for groups with free space if
2196 the parent directories are nearly full.
2201 Description: Network or server problems during mount may cause partially
2202 mounted clients instead of returning an error.
2203 Details : The config llog parsing code may overwrite the error return
2204 code during mount error handling, returning success instead
2210 Description: MDS can fail to allocate large reply buffers
2211 Details : After long uptimes the MDS can fail to allocate large reply
2212 buffers (e.g. zconf client mount config records) due to memory
2213 fragmentation or consumption by the buffer cache. Preallocate
2214 some large reply buffers so that these replies can be sent even
2215 under memory pressure.
2219 Frequency : rare (liblustre)
2220 Description: fsx running with liblustre complained that using truncate() to
2221 extend the file doesn't work. This patch corrects that issue.
2222 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2223 ATTR_SIZE and lsm use in llu_setattr_raw.
2227 Frequency : rare, only 2.6 kernels
2228 Description: Unusual file access patterns on the MDS may result in inode
2229 data being lost in very rare circumstances.
2230 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2231 code in the 2.6 kernel caused the inode and/or EA data not to
2232 be read from disk, causing single-file corruption.
2236 Frequency : rare, only 2.6 filesystems using extents
2237 Description: Heavy concurrent write and delete load may cause data corruption.
2238 Details : It was possible under high-load situations to have an extent
2239 metadata block in the block device cache from a just-unlinked
2240 file overwrite a newly-allocated data block. We now unmap any
2241 metadata buffers that alias just-allocated data blocks.
2245 Frequency : filesystems with default stripe_count larger than 77
2246 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2247 Details : lconf specifies an inode size of 4096 bytes when the default
2248 stripe_count is larger than 77. This conflicts with the default
2249 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2250 this case to avoid pinning too much memory for large EAs.
2252 ------------------------------------------------------------------------------
2254 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2259 Frequency : rare (extremely heavy IO load with hundreds of clients)
2261 Description: Client is evicted, gets IO error writing to file
2262 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2263 caused clients to be evicted for AST timeouts. The fixes in
2264 bug 5192 mean we no longer need to have such short AST timeouts
2265 so ldlm_timeout has been increased.
2268 Frequency : occasional during --force or --failover shutdown under load
2269 Bugzilla : 5949, 4834
2270 Description: Server oops/LBUG if stopped with --force or --failover under load
2271 Details : a collection of import/export refcount and cleanup ordering
2272 issues fixed for safer force cleanup
2275 Frequency : only filesystems larger than 120 OSTs
2276 Bugzilla : 5990, 6223
2277 Description: lfs getstripe would oops on a very large filesystem
2278 Details : lov_getconfig used kfree on vmalloc'd memory
2281 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2282 Bugzilla : 6242, 6243
2283 Description: reading from files that had been truncated to a non-zero size
2284 but never opened returned no data
2285 Details : ll_file_read() reads zeros from no-object files to EOF
2290 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2291 Details : There's little harm in aborting MDS/OSS recovery and letting it
2292 try again, so I removed the LASSERT and return an error instead.
2294 Severity : enhancement
2296 Description: New debugging infrastructure for tracking down data corruption
2297 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2298 (b) cover more of the client-side code path, and (c) try to narrow
2299 down where problems occurred
2303 Bugzilla : 3819, 4364, 4397, 6313
2304 Description: Racing close and eviction MDS could cause assertion in mds_close
2305 Details : It was possible to get multiple mfd references during close and
2306 client eviction, leading to one thread referencing a freed mfd.
2308 Severity: : enhancement
2309 Bugzilla : 3262, 6359
2310 Description: Attempts to reconnect to servers are now more aggressive.
2311 Details : This builds on the enhanced upcall-less recovery that was added
2312 in 1.4.2. When trying to reconnect to servers, clients will
2313 now try each server in the failover group every 10 seconds. By
2314 default, clients would previously try one server every 25 seconds.
2319 Description: After recovery, certain operations trigger a failed
2320 assertion on a client.
2321 Details : Failing over an mds, using lconf -d --failover, while a
2322 client was doing a readdir() call would cause the client to
2323 LBUG after recovery completed and the readdir() was resent.
2325 Severity : enhancement
2327 Description: Default groups are now added by lconf
2328 Details : You can now run lconf --group <servicename> without having to
2329 manually add groups with lmc.
2332 Frequency : occasional
2334 Description: Nodes with an elan id of 0 trigger a failed assertion
2337 Frequency : always when accessing e.g. tty/console device nodes
2339 Description: tty and some other devices nodes cannot be used on lustre
2340 Details : file's private_data field is used by device data and lustre
2341 values in there got lost. New field was added to struct file to
2342 store fs-specific private data.
2345 Frequency : when exporting Lustre via NFS
2347 Description: NFSD failed occasionally when looking up a path component
2348 Details : NFSD is looking up ".." which was broken in ext3 directories
2349 that had grown large enough to become hashed.
2352 Frequency : Clusters with multiple interfaces not on the same subnet
2354 Description: Nodes will repeatedly try to reconnect to an interface which it
2355 cannot reach and report an error to the log.
2356 Details : Extra peer list entries will be created by lconf with some peers
2357 unreachable. lconf now validates the peer before adding it.
2360 Frequency : Only if a default stripe is set on the filesystem root.
2362 Description: Setting a default stripe on the filesystem root prevented the
2363 filesystem from being remounted.
2364 Details : The client was sending extra request flags in the root getattr
2365 request and did not allocate a reply buffer for the dir EA.
2368 Frequency : occasional, higher if lots of files are accessed by one client
2369 Bugzilla : 6159, 6097
2370 Description: Client trips assertion regarding lsm mismatch/magic
2371 Details : While revalidating inodes the VFS looks up inodes with ifind()
2372 and in rare cases can find an inode that is being freed.
2373 The ll_test_inode() code will free the lsm during ifind()
2374 when it finds an existing inode and then the VFS later attaches
2375 this free lsm to a new inode.
2379 Bugzilla : 6422, 7030
2380 Description: MDS deadlock between mkdir and client eviction
2381 Details : Creating a new file via mkdir or mknod (starting a transaction
2382 and getting the ns lock) can deadlock with client eviction
2383 (gets ns lock and trying to finish a synchronous transaction).
2386 Frequency : occasional
2387 Description: While starting a server, the fsfilt_ext3 module could not be
2389 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2390 kernels. Previously, lconf would still use the ext3 name
2391 when trying to load modules. Now, it will correctly use
2392 ext3 on 2.4 and ldiskfs on 2.6.
2394 Severity : enhancement
2395 Description: The default stripe count has been changed to 1
2396 Details : The interpretation of the default stripe count (0, to lfs
2397 or lmc) has been changed to mean striping across a single
2398 OST, rather than all available. For general usage we have
2399 found a stripe count of 1 or 2 works best.
2401 Severity : enhancement
2402 Description: Add support for compiling against Cray portals.
2403 Details : Conditional compiling for some areas that are different
2407 Frequency : occasional
2408 Bugzilla : 6409, 6834
2409 Description: Creating files with an explicit stripe count may lead to
2410 a failed assertion on the MDS
2411 Details : If some OSTs are full or unavailable, creating files may
2412 trigger a failed assertion on the MDS. Now, Lustre will
2413 try to use other servers or return an error to the
2417 Frequency : occasional
2419 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2420 caused lock timeout during readahead (since 1.4.2).
2421 Details : Processes doing readahead might match a lock that hasn't been
2422 granted yet if there are overlapping and conflicting lock
2423 requests. The readahead process waits on ungranted lock
2424 (original lock is CBPENDING), while OST waits for that process
2425 to cancel CBPENDING read lock and eventually evicts client.
2427 Severity : enhancement
2429 Description: Initial enabling of flock support for clients
2430 Details : Implements fcntl advisory locking and file status functions.
2431 This feature is provided as an optional mount flag (default
2432 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2433 locking are implemented yet, and those that are are not guaranteed
2434 to be completely correct in production environments.
2435 mount -t lustre -o [flock|noflock] ...
2438 Frequency : occasional
2440 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2441 assertion in the ext3 JBD (journaling) layer.
2442 Details : The b_committed_data struct is protected by the big kernel lock
2443 in 2.4 kernels, serializing journal_commit_transaction() and
2444 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2445 access to this struct. In 2.6 kernels there is finer grained
2446 locking to improve SMP performance of the JBD layer.
2450 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2452 -----------------------------------------------------------------------------
2454 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2456 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2457 versions of Lustre. Please update all servers and clients to
2458 version 1.4.2 or later at the same time. You must also run
2459 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2460 to update the configuration logs.
2462 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2463 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2464 - swab llog records of type '0' so we get proper header size/idx (5861)
2465 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2466 - fix rename of one directory over another leaking an inode (5953)
2467 - avoid SetPageDirty on 2.6 (5981)
2468 - don't re-add just-being-destroyed locks to the waiting list (5653)
2469 - when creating new directories, inherit the parent's custom
2470 striping settings if present parent (3048)
2471 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2472 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2473 - don't hold client locks on temporary worklist from l_lru (5666)
2474 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2475 - automatically evict dead clients (5921)
2476 - Update file size properly in create+truncate+fstat case (6196)
2477 - Do not unhash mountpoint dentries, do not allow removal of
2479 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2480 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2481 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2482 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2483 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2484 - Correctly handle reads of files with no objects (6243)
2485 - lctl recover will also mark a device active if deactivate used (5933)
2487 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2488 - allow --write-conf on an MDS with different nettype than client (5619)
2489 - don't write config llogs to MDS for mounts not from that MDS (5617)
2490 - lconf should create multiple TCP connections from a client (5201)
2491 - init scripts are now turned off by default; run chkconfig --on
2492 lustre and chkconfig --on lustrefs to use them
2493 - upcalls are no longer needed for clients to recover to failover
2495 - add --abort-recovery option to lconf to abort recovery on device
2497 - add support for an arbitrary number of OSTs (3026)
2498 - Quota support protocol changes.
2499 - forward compatibility changes to wire structs (6007)
2500 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2501 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2502 - improved "lustre status" script
2503 - initialize blocksize for non-regular files (6062)
2504 - added --disable-server and --disable-client configure options (5782)
2505 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2506 - Vanilla 2.4.29 support
2507 - increase maximum number of obd devices to 520 (6242)
2508 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2509 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2510 SLES 9 2.6.5 kernels
2511 - limit stripes per file to 160 (the maximum EA size) (6093)
2513 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2516 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2517 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2518 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2519 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2520 - small ext3 extents cleanups and fixes (5733)
2521 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2522 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2523 - avoid needless client->OST connect, fix handle mismatch (5317)
2524 - fix DLM error path that led to out-of-sync client, long delays (5779)
2525 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2526 - fix several locking issues related to i_size (5492,5624,5654,5672)
2527 - don't move pending lock onto export if it is already evicted (5683)
2528 - fix kernel oops when creating .foo in unlinked directory (5548)
2529 - fix deadlock in obdfilter statistics vs. object create (5811)
2530 - use time_{before,after} to avoid timer jiffies wrap (5882)
2531 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2532 - Do not leak request if server was not able to process it (5154)
2533 - If mds_open unable to find parent dir, make that negative lookup(5154)
2534 - don't create new directories with extent-mapping (5909, 5936)
2536 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2537 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2538 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2539 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2540 - support for RHEL4 kernel on i686 (5773)
2541 - provide error messages when incompatible logs are encountered (5898)
2543 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2544 * version 1.4.0.10 (1.4.1 release candidate 1)
2546 - don't keep a lock reference when lock is not granted (4238)
2547 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2548 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2549 - drop import inflight refcount on signal_completed_replay error (5255)
2550 - unlock page after async write error during send (3677)
2551 - handle missing objects in filter_preprw_read properly (5265)
2552 - no transno return for symlink open, don't save no-trasno open (3440)
2553 - don't try to complete elan receive that already failed (4012)
2554 - free RPC server reply state on error (5406)
2555 - clean up thread from ptlrpc_start_thread() on error (5160)
2556 - readahead could read extra page into cache that wasn't ejected (5388)
2557 - prevent races in class_attach/setup/cleanup/detach (5260)
2558 - don't dereference de->d_inode after l_dput of de (5458)
2559 - use "int" for stripe value returned from lock_to_stripe (5544)
2560 - mballoc allocation and error-checking fixes in 2.6 (5504)
2561 - block device patches to fix I/O request sizes in 2.6 (5482)
2562 - look up hostnames for IB nals (5602)
2563 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2564 - don't start multiple acceptors for the same port (5277)
2565 - fix incorrect LASSERT in mds_getattr_name (5635)
2566 - export a proc file for general "ping" checking (5628)
2567 - fix "lfs check" to not block when the MDS is down (5628)
2569 - service request history (4965)
2570 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2571 - create an "evict_client" /proc entry on OSTs, like the MDS has
2572 - fix mount usage message, return errors per mount(8) (5168)
2573 - change grep [] to grep "[]" in tests so they work in more UMLs
2574 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2575 - remove ancient LOV_MAGIC_V0 EA support (5047)
2576 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2577 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2578 - allow readahead allocations to fail when low on memory (5383)
2579 - mmap locking landed again, after considerable improvement (2828)
2580 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2582 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2585 - send OST transaction number in read/write reply to free req (4966)
2586 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2587 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2588 - the watchdog thread now runs as interruptible (5246)
2589 - flock/lockf fixes (but it's still disabled, pending 5135)
2590 - don't use EXT3 constants in llite code (5094)
2591 - memory shortage at startup could cause assertion (5176)
2593 - reorganization of lov code
2594 - single portals codebase
2596 - add extents/mballoc support (5025)
2597 - direct I/O reads in the obdfilter (4048)
2598 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2600 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2603 - send OST transaction number in read/write reply to free req (4966)
2604 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2605 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2606 - the watchdog thread now runs as interruptible (5246)
2607 - handle missing objects in filter_preprw_read properly (5265)
2608 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2609 - drop import inflight refcount on signal_completed_replay error (5255)
2610 - unlock page after async write error during send (3677)
2611 - return original error code on reconstructed replies (3761)
2612 - no transno return for symlink open, don't save no-trasno open (3440)
2614 - add pid to ldlm debugging output (4922)
2615 - bump the watchdog timeouts -- we can't handle 30sec yet
2616 - extra debugging for orphan dentry/inode bug (5259)
2618 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2621 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2622 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2623 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2624 - fix NULL dereference in /proc/sys/portals/routes (4827)
2625 - allow failed mdc_close() operations to be interrupted (4561)
2626 - stop precreate on OST before MDS would time out on it (4778)
2627 - don't send partial-page writes before EOF from client (4410)
2628 - discard client grant for sub-page writes on large-page clients (4520)
2629 - don't free dentries not owned by NFS code, check generation (4806)
2630 - fix lsm leak if mds_create_objects() fails (4801)
2631 - limit debug_daemon file size, always print CERROR messages (4789)
2632 - use transno after validating reply (3892)
2633 - process timed out requests if import state changes (3754)
2634 - update mtime on OST during writes, return in glimpse (4829)
2635 - add mkfsoptions to LDAP (4679)
2636 - use ->max_readahead method instead of zapping global ra (5039)
2637 - don't interrupt __l_wait_event() during strace
2639 - add software watchdogs to catch hung threads quickly (4941)
2640 - make lustrefs init script start after nfs is mounted
2641 - fix CWARN/ERROR duplication (4930)
2642 - return async write errors to application if possible (2248)
2643 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2644 - print NAL number in %x format (4645)
2645 - update barely-supported suse-2.4.21-171 series (4842)
2646 - support for sles 9 %post scripts
2647 - support for building 2.6 kernel-source packages
2648 - support for sles km_* packages
2650 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2653 - ignore -ENOENT errors in osc_destroy (3639)
2654 - notify osc create thread that OSC is being cleaned up (4600)
2655 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2656 - reconstruct ost_handle() like mds_handle() (4657)
2657 - create a new thread to do import eviction to avoid deadlock (3969)
2658 - let lconf resolve symlinked-to devices (4629)
2659 - don't unlink "objects" from directory with default EA (4554)
2660 - hold socknal file ref over connect in case target is down (4394)
2661 - allow more than 32000 subdirectories in a single directory (3244)
2662 - fix blocks count for O_DIRECT writes (3751)
2663 - OST returns ENOSPC from object create when no space left (4539)
2664 - don't send truncate RPC if file size isn't changing (4410)
2665 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2666 - bind to privileged port in socknal and tcpnal (3689)
2668 - rate limit CERROR/CWARN console message to avoid overload (4519)
2669 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2670 - basic mmap support (3918)
2671 - kernel patch series update from b1_4 (4711)
2673 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2676 - avoid crash during MDS cleanup with OST shut down (2775)
2677 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2678 - don't use bad inodes on the MDS (3744)
2679 - dynamic object preallocation to improve recovery speed (4236)
2680 - don't hold spinlock over lock dumping or change debug flags (4401)
2681 - don't zero obd_dev when it is force cleaned (3651)
2682 - print grants to console if they go negative (4431)
2683 - "lctl deactivate" will stop automatic recovery attempts (3406)
2684 - look for existing locks in ldlm_handle_enqueue() (3764)
2685 - don't resolve lock handle twice in recovery avoiding race (4401)
2686 - revalidate should check working dir is a directory (4134)
2688 - don't always mark "slow" obdfilter messages as errors (4418)
2690 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2693 - don't close LustreDB during write_conf until it is done (3860)
2694 - fix typo in lconf for_each_profile (3821)
2695 - allow dumping logs from multiple threads at one time (3820)
2696 - don't allow multiple threads in OSC recovery (3812)
2697 - fix debug_size parameters (3864)
2698 - fix mds_postrecov to initialize import for llog ctxt (3121)
2699 - replace config semaphore with spinlock (3306)
2700 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2701 - don't allow enqueue to complete on a destroyed export (3822)
2702 - down write_lock before checking llog header bitmap (3825)
2703 - recover from lock replay timeout (3764)
2704 - up llog sem before sending rpc (3652)
2705 - reduce ns lock hold times when setting kms (3267)
2706 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2707 - fix NULL deref and obd_dev leak on setup error (3312)
2708 - replace some LBUG about llog ops with error handling (3841)
2709 - don't match INVALID dentries from d_lookup and spin (3784)
2710 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2711 - fix invalid assertion in ptlrpc_set_wait (3880)
2713 - add libwrap support for the TCP acceptor (3996)
2714 - add /proc/sys/portals/routes for non-root route listing (3994)
2715 - allow setting MDS UUID in .xml (2580)
2716 - print the stack of a process that LBUGs (4228)
2718 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2721 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2722 - make sure to unset replay flag from failed open requests (3440)
2723 - if default stripe count is 0, use OST count for inode size (3636)
2724 - update parent mtime/ctime on client for create/unlink (2611)
2725 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2726 - free recovery state on server during a forced cleanup (3571)
2727 - unregister_reply for resent reqs (3063)
2728 - loop back devices mounting and status check on 2.6 (3563)
2729 - fix resource-creation race that can provoke i_size == 0 (3513)
2730 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2731 - more debugging for page-accounting assertion (3746)
2732 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2733 - avoid deadlock after precreation failure (3758)
2734 - fix race and lock order deadlock in orphan handling (3450, 3750)
2735 - add validity checks when grabbing inodes from l_ast_data (3599)
2737 - add /proc/.../recovery_status to obdfilter (3428)
2738 - lightweight CDEBUG infrastructure, debug daemon (3668)
2739 - change default OSC RPC parameters to be better on small clusters
2740 - turn off OST read cache for files smaller than 32MB
2741 - install man pages and include them in rpms (3100)
2742 - add new init script for (un)mounting lustre filesystems (2593)
2743 - run chkconfig in %post for init scripts (3701)
2744 - drop scimac NAL (unmaintained)
2746 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
2749 - clean kiobufs before and after use (3485)
2750 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
2751 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
2752 - change init script to fail more clearly if not run as root (1528)
2753 - allow clients to reconnect during replay (1742)
2754 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
2755 - don't do DNS lookups on NIDs too small for IP addresses (3442)
2756 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
2757 - fix cond_resched (3554)
2758 - only evict unfinished clients after recovery (3515)
2759 - allow bulk resend, prevent data loss (3570)
2760 - dynamic ptlrpc request buffer allocation (2102)
2761 - don't allow unlinking open directory if it isn't empty (2904)
2762 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
2763 - remove extraneous obd dereference causing LASSERT failure (3334)
2764 - don't use get_cycles() when creating temp. files on the mds (3156)
2765 - hold i_sem when setting i_size in ll_extent_lock() (3564)
2766 - handle EEXIST for set-stripe, set proper directory name (3336)
2768 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
2769 - fix ksocknal_fmb_callback() error messages (2918)
2771 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
2774 - don't copy lvb into (possibly NULL) reply on error (2983)
2775 - don't deref dentry after dput, don't free lvb on error (2922)
2776 - use the kms to determine writeback rpc length (2947)
2777 - increment oti_logcookies when osc is inactive (2948)
2778 - update client's i_blocks count via lvb messages (2543)
2779 - handle intent open/close of special files properly (1557)
2780 - mount MDS with errors=remount-ro, like obdfilter (2009)
2781 - initialize lock handle to avoid ASSERT on error cleanup (3057)
2782 - don't use cancelling-locks' kms values (2947)
2783 - use highest lock extent for kms, not last one (2925)
2784 - don't dereference ERR_PTR() dentry in error handling path (3107)
2785 - fix thread race in portals_debug_dumplog() (3122)
2786 - create lprocfs device entries at setup instead of at attach (1519)
2787 - common AST error handler, don't evict client on completion race (3145)
2788 - zero nameidata in detach_mnt in 2.6 (3118)
2789 - verify d_inode after revalidate_special is valid in 2.6 (3116)
2790 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
2791 - initialize RPC timeout timer earlier for 2.6 (3219)
2792 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
2793 - print nal/nid for unknown nid (3258)
2794 - additional checks for oscc recovery before doing precreate (3284)
2795 - fix ll_extent_lock() error return code for 64-bit systems (3043)
2796 - don't crash in mdc_close for bad permissions on open (3285)
2797 - zero i_rdev for non-device files (3147)
2798 - clear page->private before handing to FS, better assertion (3119)
2799 - tune the read pipeline (3236)
2800 - fix incorrect decref of invalidated dentry (2350)
2801 - provide read-ahead stats and refine rpc in flight stats (3328)
2802 - don't hold journal transaction open across create RPC (3313)
2803 - update atime on MDS at close time (3265)
2804 - close LDAP connection when recovering to avoid server load (3315)
2805 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
2806 - don't leak open file on MDS after open resend (3325)
2807 - serialize filter_precreate and filter_destroy_precreated (3329)
2808 - loop device shouldn't call sync_dev() for nul device (3092)
2809 - clear page cache after eviction (2766)
2810 - resynchronize MDS->OST in background (2824)
2811 - refuse to mount the same filesystem twice on same mountpoint (3394)
2812 - allow llmount to create routes for mounting behind routers (3320)
2813 - push lock cancellation to blocking thread for glimpse ASTs (3409)
2814 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
2815 - fix rare problem with rename on htree directories (3417)
2817 - allow default OST striping configuration per directory (1414)
2818 - fix compilation for qswnal for 2.6 kernels (3125)
2819 - increase maximum number of MDS request buffers for large systems
2820 - change liblustreapi to be useful for external progs like lfsck (3098)
2821 - increase local configuration timeout for slow disks (3353)
2822 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
2824 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2827 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
2828 - don't overwrite extent policy data in reply if lock was blocked (2901)
2829 - drop filter export grants atomically with removal from device (2663)
2830 - del obd_self_export from work_list in class_disconnect_exports (2908)
2831 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
2832 - swab reply message in mdc_close, other PPC fixes (2464)
2833 - fix destroying of named logs (2325)
2834 - overwrite old logs when running lconf --write_conf (2264)
2835 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
2836 - fix race in target_handle_connect (2898)
2837 - mds_reint_create() should take same inode create lock (2926)
2838 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
2839 - don't close files for self_export to avoid uninitialized obd (2936)
2840 - allow MDS with the same name as client node (2939)
2841 - hold dentry reference for closed log files for unlink (2325)
2842 - reserve space for all logs during transactions (2059)
2843 - don't evict page beyond end of stripe extent (2925)
2844 - don't oops on a deleted current working directory (2399)
2845 - handle hard links to targets without a parent properly (2517)
2846 - don't dereference NULL lock when racing during eviction (2867)
2847 - don't grow lock extents when lots of conflicting locks (2919)
2849 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
2852 - account for cache space usage on clients to avoid data loss (974)
2853 - lfsck support in lustre kernel code (2349)
2854 - reduce journal credits needed for BRW writes (2370)
2855 - orphan handling to avoid losing space on client/server crashes
2856 - ptlrpcd can be blocked, stopping ALL progress (2477)
2857 - use lock value blocks to assist in proper KMS, faster stat (1021)
2858 - takes i_sem instead of DLM locks internally on obdfilter (2720)
2859 - recovery for initial connections (2355)
2860 - fixes for mds_cleanup_orphans (1934)
2861 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
2862 - block all file creations until orphan recovery completes (1901)
2863 - client remove rq_connection from request struct (2423)
2864 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
2865 - recovery timer race (2670)
2866 - mdc_close recovey bug (2532)
2867 - ptlrpc cleanup bug (2710)
2868 - mds timeout on local locks (2588)
2869 - namespace lock held during RPCs (2431)
2870 - handle interrupted sync write properly (2503)
2871 - don't try to handle a message that hasn't been replied to (2699)
2872 - client assert failure during cleanup after abort recovery (2701)
2873 - leak mdc device after failed mount (2712)
2874 - ptlrpc_check_set allows timedout requests to complete (2714)
2875 - wait for inflight reqs when ptlrpcd finishes (2710)
2876 - make sure unregistered services are removed from the srv_list
2877 - reset bulk XID's when resending them (caught by 1138 test)
2878 - unregister_bulk after timeout
2879 - fix lconf error (2694)
2880 - handle write after unfinished setstripe, stripe-only getstripe (2388)
2881 - readahead locks pages, leaves pending causing memory pressure (2673)
2882 - increase OST request buffers to 4096 on large machines (2729)
2883 - fix up permission of existing directories in simple_mkdir (2661)
2884 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
2885 - don't assign transno to errored transactions (2742)
2886 - don't delete objects on OST if given a bogus objid from MDS (2751)
2887 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
2888 - if rq_no_resend, then timeout request after recovery (2432)
2889 - fix MDS llog_logid record size, 64-bit array alignment (2733)
2890 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
2891 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
2892 - MDS recovery shouldn't delete objects that it has given out (2730)
2893 - if enqueue arrives after completion, don't clobber LVB (2819)
2894 - don't unlock pages twice when trigger_group_io returns error (2814)
2895 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
2896 - don't write pages to disk if there was an error (1450)
2897 - don't ping imports that have recovery disabled (2676)
2898 - take buffered bytes into account when balancing socknal conn (2817)
2899 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
2900 - reconnect unlink llog connection after MDS reconnects to OST (2816)
2901 - remove little-endian swabbing of llog records (1987)
2902 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
2903 - retry reposting request buffers if they fail (1191)
2904 - grow extent at grant time to avoid granting a revoked lock (2809)
2905 - lock revoke doesn't evict page if covered by a second lock (2765)
2906 - disable VM readahead to avoid reading outside lock extents (2805)
2908 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
2909 - updated kernel patches for hp-2.4.20 kernel (2681)
2911 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
2914 - fix truncated write corruption (2366)
2915 - fix for failed assertion in iopen_connect_dentry (1792,2517)
2917 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
2918 - ensure len(uuid) < 37 in lmc (1171)
2919 - fix ia64 OOPS in llog_test (2255)
2920 - zero end of page at obdfilter for partial page writes (2648)
2921 - don't leave stale dentries around after renames (bug 2428)
2922 - fix timeouts when evicting a client with a single lock held (2642)
2923 - set deadline for the initial HELLO message to drain (2634)
2924 - print out dotted-quad IP addresses in the socknal (2302)
2926 - additional debugging for MDS client eviction problem (2443)
2927 - fix mkfsoptions support for osts (2603, 2604)
2929 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
2932 - add series for the vanilla 2.6.0 kernel
2933 - add series for the vanilla 2.4.24 kernel
2934 - add series for a cray x86/64 UL kernel drop
2935 - fix xattr patches for the vanilla 2.4.19 series
2937 - generate true UUIDs in lmc (1171)
2938 - have portals stack dumping break in UML (2466)
2939 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
2940 - allocate with _NOFS in ldlm to avoid deadlock (1933)
2941 - wake callback waiting threads on client eviction (2460)
2942 - Add --ptldebug and --subsystem to lmc (1719)
2943 - update assertion to allow safe interrupt allocation
2944 - set rq_no_resend for cancel requests (2432)
2945 - recalculate ptlrpcd timeout after resend (2494)
2946 - call vfs_rmdir when removing pending directories (2368)
2947 - fix renaming a file to itself (2429)
2948 - lmc creates a default one-stripe lov (2454)
2949 - expand procfs space to handle large clusters (2326)
2950 - increase UML stack to avoid overflow
2951 - update lconf's list of debug and subsystem masks
2952 - fix lfs find --obd (2510)
2953 - /proc tunable for disabling filter read caching (2591)
2954 - stop rpm packages from altering slapd.conf (2301)
2955 - disable nagle in the socknal under 0conf (2578)
2956 - choose mds inode size based on stripe count (2572)
2957 - fix kernel-source rpm problems (2516)
2959 - add --disable-doc to avoid pdf generation (2421)
2960 - update documentation, tests, type-os, comments
2961 - avoid format warnings on ia64
2962 - remove the TOE NAL
2963 - tiny code cleanups by removing unused fields
2965 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
2968 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
2969 - remove the most blatant lies from BUILDING file (2371)
2970 - change default debug level to reasonable production setting
2971 - reduce client side cache size to reduce cache flush time
2972 - reduce max RPCs in flight to avoid unnecessary file fragmentation
2973 - make TCP zerocopy and pinger support enabled by default (2476)
2974 - sync writes completed after process exits caused crashes (2319)
2975 - maintain correct mount count on the MDS (2356)
2976 - backout 1557, because 2316 wasn't really fixed
2977 - better file I/O statistics gathering in /proc
2978 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
2979 - another kernel patch to fix zero-copy TCP function export
2980 - don't take duplicate lock when processing re-sent getattr (2420)
2981 - lctl uses obd_self_export instead of creating new conn (2353)
2982 - MDS/OST recovery case which requires object creation asserted (2425)
2983 - move lfs from /usr/sbin to /usr/bin in packages
2984 - fix race between mds_client_add and mds_client_free (2417)
2985 - use kmalloc instead of slabs in portals (2430)
2986 - don't create duplicate records when a failover MDS is present (2442)
2987 - remove unnecessary mount age check (2332)
2988 - don't remove directory inodes from locks prematurely (2451)
2989 - don't break if MDS service name is the same as hostname (2103)
2990 - fix races in client write RPC generation when cache full (2482)
2992 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
2995 - remove now-unused request->rq_obd (278)
2996 - if an allocation fails, print out how much memory we've used (1933)
2997 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
2998 - add the "configurable stack size" patch to most series files (1256)
2999 - ability to write large log records, for 100+ OST configs (2306)
3000 - fix NULL deref when filter_prep fails (2314)
3001 - fix operator precedence error in filter_sync
3002 - dynamic allocation of socknal TX descriptors (2315)
3003 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3004 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3005 - ensure that CWARN messages in Portals always get to the syslog
3006 - __init/__exit are not for prototype decls (ldlm_init/exit)
3007 - x86-64 compile warning fixes
3008 - fix gateway LMC keyword conflict (2318)
3009 - fix MDS lock inversions in getattr/reint paths (1844)
3010 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3011 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3012 - DLM race condition prevented some lock evictions (2328)
3013 - ENOMEM detection and retry on socknal sends (2230)
3014 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3015 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3016 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3017 - better error messages when a client is rejected during recovery (1505)
3018 - avoid cancelling locks which were never granted, after failure (2330)
3019 - fix i_sem/journal inversion in mds_client_add (2333)
3020 - fix truncate/getattr lock cycle deadlock (2334)
3021 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3022 - fix two rare exit paths which could leak an l_lock() ref (2321)
3023 - fencepost error in MDS/OST orphan recovery (2226)
3024 - make log record alignment 8 bytes (1988)
3025 - lstripe now fails when requested offset > ost_count (2237)
3026 - ensure that all kernel series have a complete list.h (1607)
3027 - fix crashes in special-file operations (2316)
3028 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3030 - allow configurable automake binary, for testing new versions
3031 - small update to the lfs documentation
3033 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3035 * fix negative export reference count in fsfilt_sync (2312)
3037 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3038 * release candidate 0.9.1
3040 - orphans are moved into the PENDING directory for possible recovery
3041 - replayed opens now open by fid for orphan/rename safety (1042)
3042 - last close of an orphan inode generates a transno (683)
3043 - chdir() and mount() now pin the directory entry (1020)
3044 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3045 - discard very old requests without processing them (1502)
3046 - remove some common, well-understood CERRORs (1505)
3047 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3048 - clear "grant" flags in OST replies until OST grant code lands (1644)
3049 - fix read performance by not clobbering i_blksize on client (1598)
3050 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3051 - make LRU size a /proc tunable, clears locks when reduced (707)
3052 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3053 - server locks take references on exports now (1558)
3054 - build fixes for 2.4.20-rh trees (1663)
3055 - return an error from lov_create if all OSCs are inactive (1751)
3056 - fix import levels when a reconnect happens without a timeout (1597)
3057 - exit early from mds_open if we get a lookup error (1749)
3058 - partial page read at EOF wouldn't wait for disk before sending (1642)
3059 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3060 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3061 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3062 - write/truncate lock inversion (1639)
3063 - Don't auto-load obdclass, portals modules during cleanup (1495)
3064 - fix timestamps from jumping to "now" (1763)
3065 - extra journal assertions (1648)
3066 - add an extra multiunlink test (1771)
3067 - fix read_record/write_record API (1776)
3068 - fix leak of offset_extent, possible incorrect i_size later (1772)
3069 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3070 - Debugging for the kqswnal_get_idle_tx problems (1820)
3071 - Allow recovery to be attempted multiple times (1536)
3072 - Write out MDS last_rcvd file after it is first created (1600)
3073 - Fix tx_descriptor leak in failed transmit situations (1827)
3074 - ext3 journaling fixes for assertion failure after IO error (1871)
3075 - class_export_put() on freed export after completion AST error (1896)
3076 - Fix revalidate looping in VFS (1322)
3077 - Don't access a freed export during MDS_REINT timeout (1521)
3078 - Add open-unlink recovery support on the MDS (1673,1764)
3079 - Return an error if no MDS data was read from last_rcvd (1946)
3080 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3081 - Don't setup a disk device that doesn't match exported UUID (317)
3082 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3083 - avoid committing NULL handle in force close
3084 - local.sh is now a one-stripe LOV configuration
3085 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3086 - fix ext3 htree duplicate directory entry corruption (1516)
3087 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3088 - update to new LOV EA format (2097)
3089 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3091 2003-06-15 Phil Schwan <phil@clusterfs.com>
3094 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3095 - per-import/export recovery handling (958, 931, 959)
3096 - multiple last-rcvd slots, for serving multiple FSes (949)
3097 - connections are again shared between multiple imp/exports (963, 964)
3098 - "umount -f" would hang if any requests needed to be sent (393, 978)
3099 - avoid pinning large req buffer by copying for queued messages (989)
3100 - add "uuid" to "lctl device" command to help upcalls (991)
3101 - "open" RPCs with transnos would confuse recovery counters (1037)
3102 - do proper endian conversion of all wire messages (288, 340, 891)
3103 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3104 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3105 - fix ptlrpc_connection leak in target_handle_connect (1174)
3106 - fix import refcounting bug in OST and MDS cleanup (1134)
3107 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3108 - fix very unlikely obd_types race condition (501)
3109 - remove osc_open hack for echo_client (1187)
3110 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3111 - a failure in read_inode2 leads to deadlock (1139)
3112 - cancel ack-locks as soon as transaction is committed (1072)
3113 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3114 - make sure to commitrw after any preprw to avoid deadlock (1162)
3115 - failing to execute a file in a lustre FS would lock inode (1203)
3116 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3117 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3118 - obd_uuid2tgt would walk off the end of the list (1255)
3119 - on IA64 the osc would give portals incorrect bulk size (1258)
3120 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3121 - fix lock inversion caused by new llite matching code (1282)
3122 - limit the number of dirty pages on a client to 10MB (1286)
3123 - timed out locks were not being corrected cancelled (1289)
3124 - fix O_DIRECT above 4GB on IA-32 (1292)
3125 * major user-visible changes
3126 - fail out/fail over policy now controlled by the upcall (993)
3128 - add OBD_PING to check server availability and failure (954)
3129 - lustre messages are now sent in sending host order (288, 340, 891)
3130 - add eadatalen to MDS getattr reply (340)
3131 - OST read replies may contain second buffer, with per-page status (593)
3133 2003-03-11 Phil Schwan <phil@clusterfs.com>
3136 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3137 - failed open()s could cause deadlock; fixed (867, 869)
3138 - stop cancelling OST locks when files are closed (481)
3139 - overlapping XID spaces caused network corruption (851, 853)
3140 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3141 - setattr_raw added, to do single-RPC, server-side setattrs
3142 - lmc/lconf syntax change for OST UUIDs
3143 - fix crashy race condition between ptlrpc_free_req and osc_close
3144 - don't use request in mdc_enqueue if we hit a timeout (889)
3145 - don't set the inode i_size for regular files from the MDS (896)
3146 - handle out of order completion AST (842)
3147 - don't LBUG if a lock request times out after receiving AST (913)
3148 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3149 - if a bad lock AST arrives, send an error instead of dropping entirely
3150 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3151 - fix leak in bulk IO when only partially completed (899, 900, 926)
3152 - fix O_DIRECT for ia64 (55)
3153 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3154 - C-z after timeout could hang a process forever; fixed (977)
3156 - client-side I/O cache (678, 924, 929, 941, 970)
3158 - READPAGE and SETATTRs which don't take server-side locks get
3161 2003-02-11 Phil Schwan <phil@clusterfs.com>
3164 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3165 - Fix fsfilt_extN_readpage() to read a full page of directory
3166 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3167 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3168 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3169 - Most intent operations were removed, in favour of a new RPC mode
3170 that does a single RPC to the server and bypasses most of the VFS
3171 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3172 - Aggressively cancel local locks on DLM servers
3173 - mds_reint_unlink sends EA to the client if it's the last nlink.
3174 client uses that EA to unlink OST objects.
3175 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3176 - recursive symlinks were fixed (439)
3177 - fixed NULL deref in DEBUG_REQ
3178 - filter_update_lastobjid no longer calls sync, which annoyed extN
3179 - fixed multi-client small-writes to a single file problem (445)
3180 - fixed mtime updates during file writes (607)
3181 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3182 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3183 - fixed Linux OST size reporting problem (444, 656)
3184 - OST now updates object mtime with writes or setattr (607, 619)
3185 - client verifies file size before zeroing page past EOF (445)
3186 - OST now writes last allocated objid to disk with allocation (108)
3187 - LOV on echo now works (409)
3189 - mds_reint_unlink sends a new buffer, with the EA included. this
3190 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3191 set if a regular file was being unlinked, and it was the last link
3192 - use PtlGet from the target for bulk writes (315)
3193 - OST now updates object mtime with writes or setattr (607, 619)
3194 - LDLM now has a grant-time callback to revalidate locked items, if
3196 - Many MDS operations were reorganized to combat race conditions
3198 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3199 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3201 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3204 - Fully reactivate OST imports after reconnection (512, others)
3205 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3206 - More graceful error handling for truncating on dead OST (515)
3207 - Don't error out unless we're actually accessing dead stripes (474)
3208 - Fix garbage sizes when stripes are missing (410)
3209 - LRU counters were broken, causing constant lock purge (433, 432)
3210 - garbage on read from stripes with failed OSTs (441)
3211 - mark OSCs as active before reconnecting during recovery (438)
3212 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3213 - lfind did not preserve OST order in output (443)
3214 - symlinks cause hung clients, incorrect data (439)
3215 - stop dereferencing request after dropping refcount (457)
3216 - don't LASSERT(spin_is_locked) on non-SMP (455)
3217 - fixes for many rename() bugs
3218 - fstat didn't correctly synchronize attributes (399)
3219 - server must handle lock cancellation during blocking AST prep (487)
3220 - bulk descriptors were free()d too soon (511)
3221 - fix paths in lconf, which would load incorrect modules (451, 507)
3222 - fix confusing lconf 'host not found' error message (386)
3223 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3224 - fix race condition in mdc_blocking_ast() for inode access (526)
3225 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3226 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3227 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3228 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3229 - fix lstripe in several ways (kernel side) (527)
3230 - fix request leak in ldlm_cli_enqueue (262)
3231 - incorrect OSC was marked inactive after OST failure
3232 - call mds_fs_cleanup before unmounting filesystem (524)
3233 - fix races between taking ns_lock and ldlm_lock_change_resource
3234 - fix races updating LOV export open file list
3235 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3236 - fix recovery NULL deref in ldlm_cli_cancel_unused
3237 - fix some DLM races by using new hash table for lock handles (419)
3238 - permit the client to specify desired inodes, at replay
3239 - duplicate requests when we queue them for replay reintegration
3240 - fix last_rcvd offset calculation
3241 - sync after each recovered transaction, so we always make progress
3242 - never, not always, ERESTART requests without transnos
3243 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3245 - skip replay if the MDS says that the client is already connected
3246 - don't check for a recovery-enabled export to match lctl's UUID
3247 - don't INC_USE_COUNT for phantom exports
3248 - don't crash when cleaning up phantom exports (567)
3249 - don't double-finish or set replay data for errored mdc_open requests
3250 - abort requests when they time out, so we don't get old replies
3251 - send/receive replies for AST messages again
3252 - if the client says that it doesn't have the lock, cancel it on the
3254 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3255 mark it as destroyed, it will all work out when decref is called
3256 - fix module use counts (22, 581)
3258 - ASTs now expect a reply (server cancels lock on error reply)
3260 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3263 - fix many simultaneous client startup (392)
3264 - fix dentry->d_it clobbering
3265 - credentials weren't being shipped for readdir/getattr operations
3266 - remove invalid assertions triggered during some concurrent MD
3268 - proper Lustre versions added (336, 389)
3269 - fix memory leak for create error case (398)
3270 - fix LOV locking bug that would get cli/srv out of sync
3271 - fix echo client over LOV (409)
3272 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3273 - fix double-O_EXCL intent crash (424)
3274 - avoid sending multiple lock CANCELs (352)
3276 - MDS can do multi-client recovery (modulo bugs in new code)
3278 - many updates, edits, cleanups
3280 2002-11-18 Phil Schwan <phil@clusterfs.com>
3283 - fix null d_it dereference (346)
3284 - fix full OST/dbench hang (333)
3285 - fix permission problem with file removal (286)
3286 - fix removal of OSCs from LOV when they fail
3287 - fix NULL deref during bulk timeout (214)
3288 - fix problems related to multiple filesystems on one MDS (241)
3289 - fixed serious subtle metadata locking bugs
3290 - free locks on clients when inodes are removed due to memory
3292 - fix inode pointer in lock data (285)
3293 - partial support for multiple MDS on a single host (241)
3294 - data locks weren't cancelled at clear_inode time (290, 311)
3295 - intent locks could lead to unbounded lock growth (205)
3296 - added a maximum lock count, an LRU list, and a flusher
3297 - fix multiple rename (365)
3298 - properly abstracted the echo client
3299 - OSC locked 1 byte too many; fixed
3300 - rewrote brw callback code:
3301 - fixed recovery bugs related to LOVs (306)
3302 - fixed too-many-pages-in-one-write crash (191)
3303 - fixed (again) crash in sync_io_timeout (214)
3304 - probably fixed callback-related race (385)
3306 - Add capability to MDS protocol
3307 - LDLM cancellations and callbacks on different portals
3309 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3312 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3313 - defer open object destruction to close time (601981, 138)
3314 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3315 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3316 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3317 - don't call obd_setattr in ll_file_release for destroyed objects
3318 * protocol change to lustre_msg: move |version| and add |flags|
3319 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3320 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3321 * added replay of create, unlink, link and rename operations during
3322 MDS failover; recovery should be much more robust now
3323 * remove failed OSCs from LOVs (only lov_create uses this so far)
3324 * the lustre-HOWTO was brought (more) up to date (582544)
3326 2002-10-23 Phil Schwan <phil@clusterfs.com>
3329 - in-use dentries weren't being reused properly (617851)
3330 - prevent multiple LDLM setup (599178)
3331 - fix LOV size calculations for truncate (617853)
3332 - fix client handling of MDS intent errors (POSIX)
3333 - fix permission bug in lovstripe.c test (624321)
3334 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3335 - truncate past end of file could corrupt data
3336 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3337 - a race in recovery could return ETIMEDOUT to apps (623947)
3338 - building outside the source directory was fixed
3339 * the lustre-HOWTO was brought (more) up to date (582544)
3340 * major progress was made on recovery functionality
3342 2002-10-10 Phil Schwan <phil@clusterfs.com>
3345 - recovery deadlock fix
3346 - rm -rf causes LBUG fix (617817)
3347 - file open by multiple tasks fix (618962)
3348 - directory permissions bugs (602707 and 620007)
3349 - journal_stop fixed with locking (611313)
3350 - O_APPEND failures resolved (618273, perhaps 614459)
3351 - lconf PATH fix (619770)
3352 - IA64 build fix (621450)
3353 - RPC buffer sizes scale with amount of memory
3355 2002-10-01 Phil Schwan <phil@clusterfs.com>
3358 - locks would be cancelled without throwing away data pages,
3359 resulting in inconsistent data (605627)
3360 - inode attributes were not always being refreshed (605627, 612449)
3361 - lconf now continues to cleanup after lctl reports an error
3362 - MDS now enforces user permissions (602707)
3363 - lprocfs cleanup fixed, but not yet enabled (614157)
3364 - fixed infinite server hang, should a client not respond to an AST
3365 - avoid going into recovery if user calls readlink() with a buffer
3366 that's too small (613941)
3367 - AST RPCs no longer require replies (614867) -- this may be changed
3368 - don't crash server if client sends an IOV that's too big (611336)
3369 - fixed lock conversion deadlock (611892)
3370 - fixed the following of symlinks (614622)
3371 * recovery: the server can remove locks from a client that dies, other
3372 clients can make progress
3373 * more extN patch fixes
3374 * compile-time configurable ptlrpc buffer allocations
3376 - collaborative read cache document
3377 - Lustre Lite Performance CDR document-in-progress
3379 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3382 - fix typo in patch-2.4.18
3384 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3387 - clear ptlrpc request each time in handle_incoming_request()
3388 - unlink of files now destroys the object on the OST
3390 2002-09-19 Peter Braam <braam@clusterfs.com>
3392 * add hard link support
3393 * change obdfile creation method
3394 * kernel patch changed
3396 2002-09-19 Peter Braam <braam@clusterfs.com>
3399 - stack overflow bug in extN fixed
3401 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3403 * documentation updates
3404 - add man pages for config tools
3405 - update tests/README to describe testing with new config tools
3406 - finish metadata API descriptions
3407 * bug fixes and cleanups
3408 - statfs workaround for 16TB limit
3409 - LOV stripe allocation improved, can stripe on subset of OSTs
3410 - LOV file size/IO offset was wrong for files > 4GB in size
3411 - object EA data was being dropped, caused files to be unreadable
3412 - memory overflow with non-LOV OST caused memory corruption
3413 - fixed regression tests to work with new config tools, obdfilter
3414 - fixed bug when directory size became larger than 1 block
3415 - fixed bug (for single client case) when PWD was deleted
3416 - invalidate local directory pages when doing intent-based ops
3417 - avoid LDLM oops when lock callback contained bad data
3419 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3421 * documentation updates
3422 * bug fixes and cleanups
3423 - configuration tools
3426 - 64-bit compile warnings
3427 - 64-bit internal statfs data
3429 * test_brw on persistent OST devices
3431 * lprocfs (disabled)
3433 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3435 * documentation updates
3436 * bug fixes and cleanups
3437 * configuration tools
3439 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3442 * many small fixes to 0_5_4
3443 * io/network handling
3444 * thinkos in MDS operations
3446 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3449 * crucial basic fixes to 0.5.3
3450 * IOR, Iozone work over Elan
3453 2002-08-07 Phil Schwan <phil@clusterfs.com>
3454 * version 0_5_3, our first alpha
3455 * we use the new Portals iovs
3456 * documentation updates
3457 * bug fixes and cleanups
3458 * small changes in the DLM wire protocol
3460 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3461 * version 0_5_1 with some initial stability,
3462 * locking on MD and file I/O.
3463 * documentation updates
3464 * several bug fixes since 0.5.0
3465 * small changes in wire protocol
3467 2002-07-18 Phil Schwan <phil@clusterfs.com>
3469 * delivered as Lustre Light Alpha
3470 * fixed a crash after handling invalid MDS requests
3471 * fixed directory pages for architectures with non-4k pages sizes
3473 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3474 * release version v0_4_4
3475 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3477 * Fixes a number of interruption problems with OST operations.
3478 * Update documentation for portals header changes
3479 * Move all wire protocol structs/defines to lustre_idl.h
3480 * Fixes symlink length bug.
3481 * Add tcpdump to repository.
3483 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3484 * release version v0_4_3
3485 * Fixes statfs for inodes on extN.
3486 * Fixes bug in runtests which would delete /etc/hosts.
3487 * Use 64-bit object IDs wherever possible (not into VFS though)
3488 Remove ost_get_info, which is unused by lustre, and out of date.
3490 2002-07-03 Peter Braam <braam@clusterfs.com>
3491 * release version v0_4_2 Fixes a lookup error (type not passed)
3492 * move forward to head of Portals
3493 * move forward to latest Lustre kernel
3495 2002-06-25 Peter Braam <braam@clusterfs.com>
3496 * release version v0_4_1. Hopefully stable on single node use.