1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
4 INTEROPERATE with 1.4.x versions automatically. In many cases a
5 special upgrade step is needed. Please read the user documentation
6 before upgrading any part of a 1.4.x system.
7 * WARNING: Lustre configuration and startup changes are required with
8 1.6.x releases. See https://mail.clusterfs.com/wikis/lustre/MountConf
10 * Support for kernels:
11 2.6.9-42.0.10.EL (RHEL 4)
13 2.4.21-47.0.1.EL (RHEL 3)
14 2.6.12.6 vanilla (kernel.org)
15 2.6.16.27-0.9 (SLES 10)
16 * Client support for unpatched kernels:
17 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
18 2.6.16 - 2.6.19 vanilla (kernel.org)
19 2.6.9-42.0.8EL (RHEL 4)
20 * Recommended e2fsprogs version: 1.39.cfs6
22 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
25 Severity : enhancement
27 Description: lfs setstripe use optional parameters instead of postional
30 Severity : enhancement
32 Description: Nanosecond timestamp support for ldiskfs
33 Details : The on-disk ldiskfs filesystem has added support for nanosecond
34 resolution timestamps. There is not yet support for this at
35 the Lustre filesystem level.
38 Frequency : during server recovery
40 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
41 Details : request with rq_no_resend flag not awake l_wait_event if they get a
45 Frequency : nfs export on patchless client
47 Description: connectathon hang when test nfs export over patchless client
48 Details : Disconnected dentry cannot be found with lookup, so we do not need
49 to unhash it or make it invalid
53 Description: ENOENT returned for valid filehandle during dbench.
54 Details: : Check if a directory has children when invalidating dentries
55 associated with an inode during lock cancellation. This fixes
56 an incorrect ENOENT sometimes seen for valid filehandles during
59 --------------------------------------------------------------------------------
61 tbd Cluster File Systems, Inc. <info@clusterfs.com>
63 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
64 INTEROPERATE with older versions automatically. In many cases a
65 special upgrade step is needed. Please read the
66 user documentation before upgrading any part of a 1.4.x system.
67 * WARNING: Lustre configuration and startup changes are required with
68 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
70 * Support for kernels:
71 2.6.9-42.0.10.EL (RHEL 4)
73 2.4.21-47.0.1.EL (RHEL 3)
74 2.6.12.6 vanilla (kernel.org)
75 2.6.16.27-0.9 (SLES 10)
76 * Client support for unpatched kernels:
77 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
78 2.6.16 - 2.6.19 vanilla (kernel.org)
79 2.6.9-42.0.8EL (RHEL 4)
80 * Recommended e2fsprogs version: 1.39.cfs5
82 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
85 Severity : enhancement
87 Description: MountConf
88 Details : Lustre configuration is now managed via mkfs and mount
89 commands instead of lmc and lconf. New obd types (MGS, MGC)
90 are added for dynamic configuration management. See
91 https://mail.clusterfs.com/wikis/lustre/MountConf for
94 Severity : enhancement
96 Description: dynamic OST addition
97 Details : OSTs can now be added to a live filesystem
99 Severity : enhancement
101 Description: startup order invariance
102 Details : MDTs and OSTs can be started in any order. Clients only
103 require the MDT to complete startup.
105 Severity : enhancement
107 Description: parallel, asynchronous orphan cleanup
108 Details : orphan cleanup is now performed in separate threads for each
109 OST, allowing parallel non-blocking operation.
111 Severity : enhancement
113 Description: optimized stripe assignment
114 Details : stripe assignments are now made based on ost space available,
115 ost previous usage, and OSS previous usage, in order to try
116 to optimize storage space and networking resources.
118 Severity : enhancement
120 Description: Permanently set tunables
121 Details : All writable /proc/fs/lustre tunables can now be permanently
122 set on a per-server basis, at mkfs time or on a live
125 Severity : enhancement
127 Description: Lustre message v2
128 Details : Add lustre message format v2.
130 Severity : enhancement
132 Description: client OST exclusion list
133 Details : Clients can be started with a list of OSTs that should be
134 declared "inactive" for known non-responsive OSTs.
137 Frequency : SFS test only (otherwise harmless)
139 Description: SPEC SFS validation failure on NFS v2 over lustre.
140 Details : Changes the blocksize for regular files to be 2x RPC size,
141 and not depend on stripe size.
143 Severity : enhancement
145 Description: fine-grained SMP locking inside DLM
146 Details : Improve DLM performance on SMP systems by removing the single
147 per-namespace lock and replace it with per-resource locks.
149 Severity : enhancement
151 Description: don't hold multiple extent locks at one time
152 Details : To avoid client eviction during large writes, locks are not
153 held on multiple stripes at one time or for very large writes.
154 Otherwise, clients can block waiting for a lock on a failed OST
155 while holding locks on other OSTs and be evicted.
157 Severity : enhancement
159 Description: Multiple MD RPCs in flight.
160 Details : Further unserialise some read-only MDT RPCs - learn about intents.
161 To avoid overly-overloading MDT, introduce a limit on number of
162 MDT RPCs in flight for a single client and add /proc controls
163 to adjust this limit.
165 Severity : enhancement
167 Description: client read/write statistics
168 Details : Add client read/write call usage stats for performance
169 analysis of user processes.
170 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
171 file access. extents_stats shows chunk size distribution.
172 extents_stats_per_process show chunk size distribution per
175 Severity : enhancement
177 Description: per-client statistics on server
178 Details : Add ldlm and operations statistics for each client in
179 /proc/fs/lustre/mds|obdfilter/*/exports/
181 Severity : enhancement
183 Description: improved MDT statistics
184 Details : Add detailed MDT operations statistics in
185 /proc/fs/lustre/mds/*/stats
187 Severity : enhancement
189 Description: VFS operations stats
190 Details : Add client VFS call stats, trackable by pid, ppid, or gid
191 /proc/fs/lustre/llite/*/vfs_ops_stats
192 /proc/fs/lustre/llite/*/vfs_track_[pid|ppid|gid]
197 Description: Fix client-side osc byte counters
198 Details : The osc read/write byte counters in
199 /proc/fs/lustre/osc/*/stats are now working
202 Frequency : always as root on SLES
204 Description: Failure of copying files with lustre special EAs.
205 Details : Client side always return success for setxattr call for lustre
206 special xattr (currently only "trusted.lov").
211 Description: Refcount LNET uuids
212 Details : The global LNET uuid list grew linearly with every startup;
213 refcount repeated list entries instead of always adding to
216 Severity : enhancement
218 Description: Dynamic service threads
219 Details : Within a small range, start extra service threads
220 automatically when the request queue builds up.
223 Frequency : mixed-endian client/server environments
225 Description: mixed-endian crashes
226 Details : The new msg_v2 system had some failures in mixed-endian
229 Severity : enhancement
231 Description: Easy OST removal
232 Details : OSTs can be permanently deactivated with e.g. 'lctl
233 conf_param lustre-OST0001.osc.active=0'
235 Severity : enhancement
237 Description: MGS proc entries
238 Details : Added basic proc entries for the MGS showing what filesystems
241 Severity : enhancement
243 Description: provide MGS failover
244 Details : Added config lock reacquisition after MGS server failover.
246 Severity : enhancement
248 Description: add Linux 2.4 support
249 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
253 Description: a large application tries to do I/O to the same resource and dies
255 Details : Check the req->rq_arrival time after the call to
256 ost_brw_lock_get(), but before we do anything about
257 processing it & sending the BULK transfer request. This
258 should help move old stale pending locks off the queue as
259 quickly as obd_timeout.
262 Frequency : when an incorrect nid is specified during startup
264 Description: ptlrpc connect to non-existant node causes kernel crash
265 Details : LNET can't be re-entered from an event callback, which
266 happened when we expire a message after the export has been
267 cleaned up. Instead, hand the zombie cleanup off to another
270 Severity : enhancement
272 Description: plain/inodebits lock performance improvement
273 Details : Grouping plain/inodebits in granted list by their request modes
274 and bits policy, thus improving the performance of search through
278 Frequency : only for kernels with patches from Lustre below 1.4.3
280 Description: Remove old rdonly API
281 Details : Remove old rdonly API which unsed from at least lustre 1.4.3
284 Frequency : only for devices with external journals
286 Description: Set external device read-only also
287 Details : During a commanded failover stop, we set the disk device
288 read-only while the server shuts down. We now also set any
289 external journal device read-only at the same time.
292 Frequency : when upgrading from 1.4 while trying to change parameters
294 Description: The wrong (new) MDC name was used when setting parameters for
295 upgraded MDT's. Also allows changing of OSC (and MDC)
296 parameters if --writeconf is specified at tunefs upgrade time.
299 Frequency : when setting specific ost indicies
301 Description: QOS code breaks on skipped indicies
302 Details : Add checks for missing OST indicies in the QOS code, so OSTs
303 created with --index need not be sequential.
305 Severity : enhancement
307 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
308 Details : The uninit_groups feature works in conjunction with the kernel
309 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
310 up the pass1 processing of e2fsck. This is a read-only feature
311 in ldiskfs2 only, so older kernels and current ldiskfs cannot
312 mount filesystems that have had this feature enabled.
314 Severity : enhancement
316 Description: Improve multi-block allocation algorithm to avoid fragmentation
317 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
318 allocation locality and avoid filesystem fragmentation.
320 ------------------------------------------------------------------------------
322 TBD Cluster File Systems, Inc. <info@clusterfs.com>
324 * Support for kernels:
325 2.4.21-47.0.1.EL (RHEL 3)
327 2.6.9-42.0.10.EL (RHEL 4)
328 2.6.12.6 vanilla (kernel.org)
329 2.6.16.27-0.9 (SLES 10)
330 * Recommended e2fsprogs version: 1.39.cfs6
331 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
334 ------------------------------------------------------------------------------
336 2007-04-30 Cluster File Systems, Inc. <info@clusterfs.com>
338 * Support for kernels:
339 2.4.21-47.0.1.EL (RHEL 3)
341 2.6.9-42.0.10.EL (RHEL 4)
342 2.6.12.6 vanilla (kernel.org)
343 2.6.16.27-0.9 (SLES 10)
344 * Recommended e2fsprogs version: 1.39.cfs6
345 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
349 Frequency : occasional, depends on client load and configuration
350 Bugzilla : 12181, 12203
351 Description: data loss for recently-modified files
353 Details : In some cases it is possible that recently written or created
354 files may not be written to disk in a timely manner (this should
355 normally be within 30s unless client IO load is very high).
356 The problem appears as zero-length files or files that are a
357 multiple of 1MB in size after a client crash or client eviction
358 that are missing data at the end of the file.
360 This problem is more likely to be hit on clients where files are
361 repeatedly created and unlinked in the same directory, clients
362 have a large amount of RAM, have many CPUs, the filesystem has
363 many OSTs, the clients are rebooted frequently, and/or the files
364 are not accessed by other nodes after being written.
366 The presence of the problem can be detected by looking at
367 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
368 smaller than the second (nr_unused) then dirty files will not
369 be flushed automatically to disk. "sync; sleep 10" should be
370 run several times on the node before unmounting it to update
371 Lustre (this is also safe to run on nodes without this problem).
373 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
374 that can cause this same problem, so customers using that kernel
375 also need to update the kernel in addition to Lustre. In order
376 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
378 It is normal that files written just before a client crash (less
379 than 30s) may not yet have been flushed to disk, even for local
383 Frequency : frequent on thin XT3 nodes
385 Description: UUID collision on thin XT3 Linux nodes
386 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
387 randomly, since we relied on an insufficiently-seeded PRNG.
392 Description: OSS hangs after "All ost request buffers busy"
393 Details : A deadlock between quota and journal operations caused OSS
394 hangs after printing "All ost request buffers busy."
397 Frequency : always on liblustre builds
399 Description: Cleanup compiler warnings on liblustre
402 Frequency : always on liblustre builds on XT3
404 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
406 ------------------------------------------------------------------------------
408 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
410 * Support for kernels:
411 2.6.16.21-0.8 (SLES10)
412 2.6.9-42.0.8EL (RHEL 4)
414 2.4.21-47.0.1.EL (RHEL 3)
415 2.6.12.6 vanilla (kernel.org)
416 * Recommended e2fsprogs version: 1.39.cfs5
419 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
421 Description: System hang on startup
422 Details : This bug allowed the liblustre (e.g. catamount) client to
423 return to the app before handling all startup RPCs. This
424 could leave the node unresponsive to lustre network traffic
425 and manifested as a server ptllnd timeout.
427 Severity : enhancement
429 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
430 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
431 boolean to control whether to print peer debug info when a
432 client's RPC times out.
437 Description: make O_SYNC working on 2.6 kernels
438 Details : 2.6 kernels use different method for mark pages for write,
439 so need add a code to lustre for O_SYNC work.
444 Description: Failure to close file and release space on NFS
445 Details : Put inode details into lock acquired in ll_intent_file_open.
446 Use mdc_intent_lock in ll_intent_open to properly
447 detect all kind of errors unhandled by mdc_enqueue
452 Description: proc file read during shutdown sometimes raced obd removal,
454 Details : Add lock to prevent obd access after proc file removal.
457 Frequency : Only for files larger than 4GB on 32-bit clients.
459 Description: improperly doing page alignment of locks
460 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
461 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
466 Description: RPCs being resent when they shouldn't be
467 Details : Some RPCs that should not be resent are being resent. This
468 can cause inconsistencies in the RPC state machine. Do not
469 resend such requests.
472 Frequency : rare, only with NFS export
474 Description: Crash on NFS re-export node
475 Details : under very unusual load conditions an assertion is hit in
476 ll_intent_file_open()
479 Frequency : only if OST filesystem is corrupted
481 Description: client incorrectly hits assertion in ptlrpc_replay_req()
482 Details : for a short time RPCs with bulk IO are in the replay list,
483 but replay of bulk IOs is unimplemented. If the OST filesystem
484 is corrupted due to disk cache incoherency and then replay is
485 started it is possible to trip an assertion. Avoid putting
486 committed RPCs into the replay list at all to avoid this issue.
491 Description: large O_DIRECT requests fail under memory pressure/fragmentation
492 Details : Large single O_DIRECT read and write calls can fail to allocate
493 a sufficiently large buffer to process the request. In case of
494 allocation failure the allocation is retried with a smaller
495 buffer and broken into smaller requests.
500 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
501 > 32000 subdirectories
502 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
503 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
504 subdirectory count crosses 32000. This will aid e2fsck to
505 correctly handle more than 32000 subdirectories.
510 Description: versioning check is incomplete
511 Details : Checking the version difference of client vs. server, report
512 error if the gap is too big.
517 Description: add support PG_writeback bit
518 Details : add support for PG_writeback bit for Lustre, for more carefull
519 work with page cache in 2.6 kernel. This also fix some deadlocks
520 and remove hack for work O_SYNC with 2.6 kernel.
522 ------------------------------------------------------------------------------
524 2006-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
526 * Support for kernels:
527 2.6.16.21-0.8 (SLES10)
528 2.6.9-42.0.3EL (RHEL 4)
530 2.4.21-47.0.1.EL (RHEL 3)
531 2.6.12.6 vanilla (kernel.org)
534 * The backwards-compatible /proc/sys/portals symlink has been removed
535 in this release. Before upgrading, please ensure that you change
536 any configuration scripts or /etc/sysctl.conf files that access
537 /proc/sys/portals/* or sysctl portals.* to use the corresponding
538 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
539 in advance of the upgrade on any system running Lustre 1.4.6 or
540 newer, since /proc/sys/lnet was added in that version.
541 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
545 Frequency : MDS failover only, very rarely
547 Description: "went back in time" messages on mds failover
548 Details : The greatest transno may be lost when the current operation
549 finishes with an error (transno==0) and the client's last_rcvd
550 record is over-written. Save the greatest transno in the
551 mds_last_transno for this case.
554 Frequency : always for specific kernels and striping counts
556 Description: client may get "Matching packet too big" without ACL support
557 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
558 when trying to access files in certain configurations. The
559 clients should in fact be denied when mounting because they do
563 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
565 Description: many clients connecting with IO in progress causes connect timeouts
566 Details : Avoid synchronous journal commits to avoid delays caused by many
567 clients connecting/disconnecting when bulk IO is in progress.
568 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
569 OST_IO_PORTAL to avoid delays behind potentially many pending
573 Frequency : occasionally with multiple writers to a single file
575 Description: shared writes to file may result in wrong size reported by stat()
576 Details : Allow growing of kms when extent lock is cancelled
579 Frequency : always with random mmap IO to multi-striped file
581 Description: mmap write might be lost if we are writing to a 'hole' in stripe
582 Details : Only if the hole is at the end of OST object so that kms is too
583 small. Fix is to increase kms accordingly in ll_nopage.
586 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
588 Description: writes to a missing object would leak memory on the OST
589 Details : If there is an inconsistency between the MDS and OST filesystems,
590 such that the MDS references an object that doesn't exist, writes
591 to that object will leak memory due to incorrect cleanup in the
592 error handling path, eventually running out of memory on the OST.
597 Description: Creating too long symlink causes lustre errors
598 Details : Check symlink and name lengths before sending requests to MDS.
601 Frequency : only if flock is enabled (not on by default)
603 Description: posix locks not released on fd closure on 2.6.9+
604 Details : We failed to add posix locks to list of inode locks on 2.6.9+
605 kernels, this caused such locks not to be released on fd close and
606 then assertions on fs unmount about still used locks.
609 Frequency : MDS failover only, very rarely
611 Description: clients may get ASSERTION(granted_lock != NULL)
612 Details : When request was taking a long time, and a client was resending
613 a getattr by name lock request. The were multiple lock requests
614 with the same client lock handle and
615 mds_getattr_name->fixup_handle_for_resent_request found one of the
616 lock handles but later failed with ASSERTION(granted_lock != NULL).
621 Description: handle->h_buffer_credits > 0, assertion failure
622 Details : h_buffer_credits is zero after truncate, causing assertion
623 failure. This patch extends the transaction or creates a new
627 Frequency : NFS re-export or patchless client
628 Bugzilla : 11179, 10796
629 Description: Crash on NFS re-export node (__d_move)
630 Details : We do not want to hash the dentry if we don't have a lock.
631 But if this dentry is later used in d_move, we'd hit uninitialised
632 list head d_hash, so we just do this to init d_hash field but
633 leave dentry unhashed.
636 Frequency : NFS re-export or patchless client
638 Description: NFS exports has problem with symbolic link
639 Details : lustre client didn't properly install dentry when re-exported
640 to NFS or running patchless client.
643 Frequency : NFS re-export or patchless client
645 Description: Various nfs/patchless fixes.
646 Details : fixes reuse disconected alias for lookup process - this fixes
647 warning "find_exported_dentry: npd != pd",
648 fix permission error with open files at nfs.
649 fix apply umaks when do revalidate.
652 Frequency : occasional
654 Description: Crash on NFS re-export node
655 Details : calling clear_page() on the wrong pointer triggered oops in
656 generic_mapping_read().
659 Frequency : rarely, using O_DIRECT IO
661 Description: unaligned directio crashes client with LASSERT
662 Details : check for unaligned buffers before trying any requests.
665 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
667 Description: stale data returned from RAID cache
668 Details : If only a small amount of IO is done to the RAID device before
669 reading it again it is possible to get stale data from the RAID
670 cache instead of reading it from disk.
673 Frequency : depends on arch, kernel and compiler version, always on sles10
676 Description: recursive or deep enough symlinks cause stack overflow
677 Details : getting rid of large stack-allocated variable in
681 Frequency : depends on hardware
683 Description: lustre write performance loss in the SLES10 kernel
684 Details : the performance loss is caused by using of write barriers in the
685 ext3 code. The SLES10 kernel turns barrier support on by
686 default. The fix is to undo that change for ldiskfs.
688 ------------------------------------------------------------------------------
690 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
692 * Support for kernels:
693 2.6.9-42.0.3EL (RHEL 4)
695 2.4.21-47.0.1.EL (RHEL 3)
696 2.6.12.6 vanilla (kernel.org)
700 Frequency : quota enabled and large files being deleted
702 Description: releasing more than 4GB of quota at once hangs OST
703 Details : If a user deletes more than 4GB of files on a single OST it
704 will cause the OST to spin in an infinite loop. Release
705 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
710 Description: statfs data retrieved from /proc may be stale or zero
711 Details : When reading per-device statfs data from /proc, in the
712 {kbytes,files}_{total,free,avail} files, it may appear
713 as zero or be out of date.
716 Frequency : systems with MD RAID1 external journal devices
718 Description: lconf's call to blkid is confused by RAID1 journal devices
719 Details : Use the "blkid -l" flag to locate the MD RAID device instead
720 of returning all block devices that match the journal UUID.
723 Frequency : always, for aggregate stripe size over 4GB
725 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
726 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
727 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
730 Frequency : always if "lfs find" used on a local file/directory
732 Description: "lfs find" segfaults if used on a local file/directory
733 Details : The case where a directory component was not specified wasn't
734 handled correctly. Handle this properly.
737 Frequency : always on ppc64
739 Description: the write to an ext3 filesystem mounted with mballoc got stuck
740 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
741 perform endianness conversion.
744 Frequency : rarely (truncate to non-zero file size after write under load)
745 Bugzilla : 10730, 10687
746 Description: Files padded with zeros to next 4K multiple
747 Details : With filesystems mounted using the "extents" option (2.6 kernels)
748 it is possible that files that are truncated to a non-zero size
749 immediately after being written are filled with zero bytes beyond
750 the truncated size. No file data is lost.
752 Severity : enhancement
754 Description: Allow recovery/failover for liblustre clients.
755 Details : liblustre clients were unaware of failover configurations until
758 Severity : enhancement
760 Description: user file locks should fail when not mounting with flock option
761 Details : Set up an error-returning stub in ll_file_operations.lock field
762 to prevent incorrect behaviour when client is mounted without
763 flock option. Also, set up properly f_op->flock field for
767 Frequency : always on ia64
769 Description: "lfs df" loops on printing out MDS statfs information
770 Details : The obd_ioctl_data was not initialized and in some systems
771 this caused a failure during the ioctl that did not return
772 an error. Initialize the struct and return an error on failure.
775 Frequency : SLES 9 only
777 Description: Error of copying files with lustre special EAs as root
778 Details : Client side always return success for setxattr call for lustre
779 special xattr (currently only "trusted.lov").
782 Frequency : rarely on clusters with both ia64+i386 clients
784 Description: ia64+i686 clients doing shared IO on the same file may LBUG
785 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
786 clients are doing concurrent writes to the same file it is
787 possible that the ia64 clients may LASSERT because the OST
788 extent locks are not PAGE_SIZE aligned. Ensure that grown
789 locks are always aligned on the request boundary.
792 Frequency : specific use, occasional
794 Description: Overwriting in use executable truncates on-disk binary image
795 Details : If one node attempts to overwrite an executable in use by
796 another node, we now correctly return ETXTBSY instead of
802 Description: chmod on Lustre root is propagated to other clients
803 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
804 invalid by the follow_mount time.
807 Frequency : liblustre clients only
809 Description: Race in 'instant cancel' lock handling could lead to such locks
810 never to be granted in case of SMP MDS
811 Details : Do not destroy not yet granted but cbpending locks in
815 Frequency : replay/resend of open
817 Description: non null lock assetion failure in mds_intent_policy
818 Details : Trying to replay/resend lockless open requests resulted in
819 mds_open() returning 0 with no lock. Now it sets a flag if
820 a lock is going to be returned.
822 Severity : enhancement
824 Description: Checksum enhancements
825 Details : New checksum enhancements allow for resending RPCs that failed
828 Severity : enhancement
830 Description: Tunables on number of dirty pages in cacche
831 Details : Allow to set limit on number of dirty pages cached.
836 Description: client crash on unmount - lock still has references
837 Details : In some error handling cases it was possible to leak a lock
838 reference on a client while accessing a file. This was not
839 harmful to the client during operation, but would cause the
840 client to crash when the filesystem is unmounted.
843 Frequency : specific case, rare
845 Description: ETXTBSY on mds though file not in use by client
846 Details : ETXTBSY is no longer incorrectly returned when attempting to
847 chmod or chown a directory that the user previously tried to
848 execute or a currently-executing binary.
851 Frequency : extremely rare except on liblustre-based clients
853 Description: Lustre space not freed when files are deleted
854 Details : Clean up open-unlinked files after client eviction. Previously
855 the unlink was skipped and the files remained as orphans.
860 Description: OST failure "would be an LBUG" in waiting_locks_callback()
861 Details : In some cases it was possible to send a blocking callback to
862 a client doing a glimpse, even though that client didn't get
863 a lock granted. When the glimpse lock is cancelled on the OST
864 the freed lock is left on the waiting list and corrupted the list.
867 Frequency : all core dumps
869 Description: Broke core dumps to lustre
870 Details : Negative dentry may be unhashed if parent does not have UPDATE
871 lock, but some callers, e.g. do_coredump, expect dentry to be
872 hashed after successful create, hash it in ll_create_it.
874 ------------------------------------------------------------------------------
876 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
878 * Support for kernels:
879 2.6.9-42.0.2.EL (RHEL 4)
881 2.4.21-40.EL (RHEL 3)
882 2.6.12.6 vanilla (kernel.org)
886 Frequency : always on RHEL 3
888 Description: Number of open files grows over time
889 Details : The number of open files grows over time, whether or not
890 Lustre is started. This was due to a filp leak introduced
891 by one of our kernel patches.
893 ------------------------------------------------------------------------------
895 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
897 * Support for kernels:
900 2.4.21-40.EL (RHEL 3)
901 2.6.12.6 vanilla (kernel.org)
906 Bugzilla : 5719, 9635, 9792, 9684
907 Description: OST (or MDS) trips assertions in (re)connection under heavy load
908 Details : If a server is under heavy load and cannot reply to new
909 connection requests before the client resends the (re)connect,
910 the connection handling code can behave badly if two service
911 threads are concurrently handing separate (re)connections from
912 the same client. Add better locking to the connection handling
913 code, and ensure that only a single connection will be processed
914 for a given client UUID, even if the lock is dropped.
916 Severity : enhancement
918 Description: add TCP zero-copy support to kernel
919 Details : Add support to the kernel TCP stack to allow zero-copy bulk
920 sends if the hardware supports scatter-gather and checksumming.
921 This allows socklnd to do client-write and server-read more
922 efficiently and reduce CPU utilization from skbuf copying.
925 Frequency : only if NFS exporting from client
927 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
928 Details : If setting attributes on a file created under NFS that had
929 never been opened it would be possible to oops the client
930 if the file had no objects.
934 Bugzilla : 9326, 10402, 10897
935 Description: client crash in ptlrpcd_wake() thread when sending async RPC
936 Details : It is possible that ptlrpcd_wake() dereferences a freed async
937 RPC. In rare cases the ptlrpcd thread alread processed the RPC
938 before ptlrpcd_wake() was called and the request was freed.
941 Frequency : always for liblustre
943 Description: liblustre client does MDS+OSTs setattr RPC for each write
944 Details : When doing a write from a liblustre client, the client
945 incorrectly issued an RPC to the MDS and each OST the file was
946 striped over in order to update the timestamps. When writing
947 with small chunks and many clients this could overwhelm the MDS
948 with RPCs. In all cases it would slow down the write because
949 these RPCs are unnecessary.
951 Severity : enhancement
953 Description: allow number of MDS service threads to be changed at module load
954 Details : It is now possible to change the number of MDS service threads
955 running. Adding "options mds mds_num_threads={N}" to the MDS's
956 /etc/modprobe.conf will set the number of threads for the next
957 time Lustre is restarted (assuming the "mds" module is also
958 reloaded at that time). The default number of threads will
959 stay the same, 32 for most systems.
964 Description: OST crash if filesystem is unformatted or corrupt
965 Details : If an OST is started on a device that has never been formatted
966 or if the filesystem is corrupt and cannot even mount then the
967 error handling cleanup routines would dereference a NULL pointer.
972 Description: NULL pointer deref in llap_from_page.
973 Details : get_cache_page_nowait can return a page with NULL (or otherwise
974 incorrect) mapping if the page was truncated/reclaimed while it was
975 searched for. Check for this condition and skip such pages when
976 doing readahead. Introduce extra check to llap_from_page() to
977 verify page->mapping->host is non-NULL (so page is not anonymous).
980 Frequency : Sometimes when using sys_sendfile
982 Description: "page not covered by a lock" warnings from ll_readpage
983 Details : sendfile called ll_readpage without right page locks present.
984 Now we introduced ll_file_sendfile that does necessary locking
985 around call to generic_file_sendfile() much like we do in
989 Frequency : with certain MDS communication failures at client mount time
991 Description: NULL pointer deref after failed client mount
992 Details : a client connection request may delayed by the network layer
993 and not be sent until after the PTLRPC layer has timed out the
994 request. If the client fails the mount immediately it will try
995 to clean up before the network times out the request. Add a
996 reference from the request import to the obd device and delay
997 the cleanup until the network drops the request.
1000 Frequency : occasionally during client (re)connect
1002 Description: assertion failure during client (re)connect
1003 Details : processing a client connection request may be delayed by the
1004 client or server longer than the client connect timeout. This
1005 causes the client to resend the connection request. If the
1006 original connection request is replied in this interval, the
1007 client may trip an assertion failure in ptlrpc_connect_interpret()
1008 which thought it would be the only running connect process.
1011 Frequency : only with obd_echo servers and clients that are rebooted
1013 Description: kernel BUG accessing uninitialized data structure
1014 Details : When running an obd_echo server it did not start the ping_evictor
1015 thread, and when a client was evicted an uninitialized data
1016 structure was accessed. Start the ping_evictor in the RPC
1017 service startup instead of the OBD startup.
1019 Severity : enhancement
1020 Bugzilla : 10193 (patchless)
1021 Description: Remove dependency on various unexported kernel interfaces.
1022 Details : No longer need reparent_to_init, exit_mm, exit_files,
1023 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1026 Frequency : rare (only users of deprecated and unsupported LDAP config)
1028 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1029 Details : LDAP apparently contains 'lustreName' attributes instead of
1030 'name'. A simple remapping of the name is sufficient.
1033 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1035 Description: waiting_locks_callback trips kernel BUG if client is evicted
1036 Details : Running with the dump_on_timeout debug flag turned on makes
1037 it possible that the waiting_locks_callback() can try to dump
1038 the Lustre kernel debug logs from an interrupt handler. Defer
1039 this log dumping to the expired_lock_main() thread.
1041 Severity : enhancement
1043 Description: Support NFS exporting on 2.6 kernels.
1044 Details : Implement non-rawops metadata methods for NFS server to use without
1045 changing NFS server code.
1048 Frequency : very rare (synthetic metadata workload only)
1050 Description: two racing renames might cause an MDS thread to deadlock
1051 Details : Running the "racer" program may cause one MDS thread to rename
1052 a file from being the source of a rename to being the target of
1053 a rename at exactly the same time that another thread is doing
1054 so, and the second thread has already enqueued these locks after
1055 doing a lookup of the target and is trying to relock them in
1056 order. Ensure that we don't try to re-lock the same resource.
1059 Frequency : only very large systems with liblustre clients
1061 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1062 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1063 all OSTs in parallel. This allows the eviction of stale liblustre
1064 clients to proceed much faster than if they were done in series,
1065 and also offers similar improvements for other set_info RPCs.
1070 Description: excessive CPU usage during initial read phase on client
1071 Details : During the initial read phase on a client, it would agressively
1072 retry readahead on the file, consuming too much CPU and impacting
1073 performance (since 1.4.5.8). Improve the readahead algorithm
1074 to avoid this, and also improve some other common cases (read
1075 of small files in particular, where "small" is files smaller than
1076 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1081 Description: MDS crash when receiving packet with unknown intent.
1082 Details : Do not LBUG in unknown intent case, just return -EFAULT
1084 Severity : enhancement
1085 Bugzilla : 9293, 9385
1086 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1087 Details : Do not serialize getattr (non-intent version) and statfs.
1090 Frequency : occasional, when OST network is overloaded/intermittent
1092 Description: client evicted by OST after bulk IO timeout
1093 Details : If a client sends a bulk IO request (read or write) the OST
1094 may evict the client if it is unresposive to its data GET/PUT
1095 request. This is incorrect if the network is overloaded (takes
1096 too long to transfer the RPC data) or dropped the OST GET/PUT
1097 request. There is no need to evict the client at all, since
1098 the pinger and/or lock callbacks will handle this, and the
1099 client can restart the bulk request.
1102 Frequency : Always when mmapping file with no objects
1104 Description: client crashes when mmapping file with no objects
1105 Details : Check that we actually have objects in a file before doing any
1106 operations on objects in ll_vm_open, ll_vm_close and
1112 Description: Request leak when working with deleted CWD
1113 Details : Introduce advanced request refcount tracking for requests
1114 referenced from lustre intent.
1116 Severity : Enhancement
1118 Description: Cache open file handles on client.
1119 Details : MDS now will return special lock along with openhandle, if
1120 requested and client is allowed to hold openhandle, even if unused,
1121 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1122 closing files for every read/write openration.
1124 Severity : Enhancement
1126 Description: Cache open negative dentries on client when possible.
1127 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1128 negative dentries on lock revocation.
1133 Description: Remounting a client read-only wasn't possible with a zconf mount
1134 Details : It wasn't possible to remount a client read-only with llmount.
1136 Severity : enhancement
1137 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1138 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1139 MPICH is included to add Lustre-specific ADIO interfaces.
1140 This is based closely on the UFS ADIO layer and only differs
1141 in file creation, in order to allow the OST striping to be set.
1142 This is user-contributed code and not supported by CFS.
1147 Description: extended inode attributes (immutable, append-only) work improperly
1148 when 2.4 and 2.6 kernels are used on client/server or vice versa
1149 Details : Introduce kernel-independent values for these flags.
1151 Severity : enhancement
1154 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1155 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1156 granularity. Now, Fractional MB granularity is be supported,
1157 this is very useful for low memory system.
1159 Severity : enhancement
1161 Description: Getattr by fid
1162 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1163 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1164 have locks and avoids getattr rpc requests for it on every stat.
1167 Frequency : Always, for filesystems larger than 2TB
1169 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1170 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1171 filesystems larger than 2TB to be mounted with mballoc enabled.
1174 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1176 Description: filesystem corruption for non-standard kernels and very large OSTs
1177 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1178 filesystems larger than 2TB is mounted then the kernel will
1179 silently corrupt the start of the filesystem. CONFIG_LBD is
1180 enabled for all CFS-supported kernels, but the possibility of
1181 this happening with a modified kernel config exists.
1183 Severity : enhancement
1185 Description: add client O_DIRECT support for 2.6 kernels
1186 Details : It is now possible to do O_DIRECT reads and writes to files
1187 in the Lustre client mountpoint on 2.6 kernel clients.
1189 Severity : enhancement
1191 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1192 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1193 parallel, not waiting for response from every OST before sending
1194 a rpc to the next OST.
1199 Description: setattr vs write race when updating file timestamps
1200 Details : Client processes that update a file timestamp into the past
1201 right after writing to the file (e.g. tar) it is possible that
1202 the updated file modification time can be reset to the current
1203 time due to a race between processing the setattr and write RPC.
1205 Severity : enhancement
1207 Description: Bring 'lfs find' closer in line with regular Linux find.
1208 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1209 -print0 options and obtains all the needed info through the lustre
1212 Severity : enhancement
1214 Description: support up to 1024 configured devices on one node
1215 Details : change obd_dev array from statically allocated to dynamically
1216 allocated structs as they are first used to reduce memory usage
1221 Description: Flush dirty partially truncated pages during truncate
1222 Details : Immediatelly flush partially truncated pages in filter_setattr,
1223 this way we completely avoid having any pages in page cache on OST
1224 and can retire ugly workarounds during writes to flush such pages.
1229 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1230 Details : protect inode from truncation within vfs_unlink() context
1231 just take a reference before calling vfs_unlink() and release it
1232 when parent's i_sem is free.
1235 Frequency : always, if extents are used on OSTs
1237 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1238 Details : OSTs using the extents format would not zero the high 16 bits of
1239 the index physical block number. This is not a problem for any
1240 OST filesystems smaller than 16TB, and no kernels support ext3
1241 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1242 new/modified files) and can be fixed for existing filesystems
1243 with e2fsprogs-1.39-cfs1.
1248 Description: import connection selection may be incorrect if timer wraps
1249 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1250 import connections to be ignored if the 32-bit jiffies counter
1251 wraps. Use a 64-bit jiffies counter.
1254 Frequency : during server recovery
1256 Description: crash after server is denying duplicate export
1257 Details : If clients are resending connect requests to the server, the
1258 server refuses to allow a client to connect multiple times.
1259 Fixed a bug in the handling of this case.
1262 Frequency : very large clusters immediately after boot
1264 Description: LNET request buffers exhausted under heavy short-term load
1265 Details : If a large number of client requests are generated on a service
1266 that has previously never seen so many requests it is possible
1267 that the request buffer growth cannot keep up with the spike in
1268 demand. Instead of dropping incoming requests, they are held in
1269 the LND until the RPC service can accept more requests.
1272 Frequency : Sometimes during replay
1274 Description: Assertion failure in ll_local_open after replay.
1275 Details : If replay happened on an open request reply before we were able
1276 to set replay handler, reply will become not swabbed tripping the
1277 assertion in ll_local_open. Now we set the handler right after
1278 recognising of open request
1281 Frequency : very rare
1283 Description: kernel reports "badness in vsnprintf"
1284 Details : Reading from the "recovery_status" /proc file in small chunks
1285 may cause a negative length in lprocfs_obd_rd_recovery_status()
1286 call to vsnprintf() (which is otherwise harmless). Exit early
1287 if there is no more space in the output buffer.
1289 Severity : enhancement
1291 Description: clear OBD RPC statistics by writing to them
1292 Details : It is now possible to clear the OBD RPC statistics by writing
1293 to the "stats" file.
1298 Description: Client mtime is not the same on different clients after utimes
1299 Details : In some cases, the client was using the utimes() syscall on
1300 a file cached on another node. The clients now validate the
1301 ctime from the MDS + OSTs to determine which one is right.
1306 Description: Inability to activate failout mode
1307 Details : lconf script incorrectly assumed that in python string's numeric
1308 value is used in comparisons.
1311 Frequency : always with multiple stripes per file
1313 Description: Inefficient object allocation for mutli-stripe files
1314 Details : When selecting which OSTs to stripe files over, for files with
1315 a stripe count that divides evenly into the number of OSTs,
1316 the MDS is always picking the same starting OST for each file.
1317 Return the OST selection heuristic to the original design.
1322 Description: mount failures may take full timeout to return an error
1323 Details : Under some heavy load conditions it is possible that a
1324 failed mount can wait for the full obd_timeout interval,
1325 possibly several minutes, before reporting an error.
1326 Instead return an error as soon as the status is known.
1328 ------------------------------------------------------------------------------
1330 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
1332 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1333 INTEROPERATE with older versions automatically. Please read the
1334 user documentation before upgrading any part of a live system.
1335 * WARNING: Lustre networking configuration changes are required with
1336 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1339 * Support for kernels:
1340 2.6.9-22.0.2.EL (RHEL 4)
1341 2.6.5-7.244 (SLES 9)
1342 2.6.12.6 vanilla (kernel.org)
1345 Severity : enhancement
1346 Bugzilla : 7981/8208
1347 Description: Introduced Lustre Networking (LNET)
1348 Details : LNET is new networking infrastructure for Lustre, it includes
1349 a reorganized network configuration mode (see the user
1350 documentation for full details) as well as support for routing
1351 between different network fabrics. Lustre Networking Devices
1352 (LNDs) for the supported network fabrics have also been
1353 created for this new infrastructure.
1355 Severity : enhancement
1356 Description: Introduced Access control lists
1357 Details : clients can set ACLs on files and directories in order to have
1358 more fine-grained permissions than the standard Unix UGO+RWX.
1359 The MDS must be started with the "-o acl" mount option.
1361 Severity : enhancement
1362 Description: Introduced filesystem quotas
1363 Details : Administrators may now establish per-user quotas on the
1366 Severity : enhancement
1368 Description: Configuration change for the XT3
1369 The PTLLND is now used to run Lustre over Portals on the XT3
1370 The configure option(s) --with-cray-portals are no longer used.
1371 Rather --with-portals=<path-to-portals-includes> is used to
1372 enable building on the XT3. In addition to enable XT3 specific
1373 features the option --enable-cray-xt3 must be used.
1378 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1379 Details : A race between allocating a new llcd and re-getting the llcd_lock
1380 allowed another thread to grab newly-allocated llcd.
1382 Severity : enhancement
1384 Description: 2.6 OST async journal commit and locking fix to improve performance
1385 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1386 2.6 kernels are now async as they already were in 2.4 kernels so
1387 that they can commit concurrently with the network bulk transfer.
1388 For block-allocated files the filter allocation semaphore is held
1389 to avoid filesystem fragmentation during allocation. BKL lock
1390 removed for 2.6 xattr operations where it is no longer needed.
1395 Description: lconf incorrectly determined whether two IP networks could talk
1396 Details : In some more complicated routing and multiple-network
1397 configurations, lconf will avoid trying to make a network
1398 connection to a disjoint part of the IP space. It was doing the
1399 math incorrectly for one set of cases.
1404 Description: Fix for potential infinite loop processing records in an llog.
1405 Details : If an llog record is corrupted/zeroed, it is possible to loop
1406 forever in llog_process(). Validate the llog record length
1407 and skip the remainder of the block on error.
1410 Frequency : occasional (liblustre only)
1412 Description: liblustre could not open files whose last component is a symlink
1413 Details : sysio_path_walk() would incorrectly pass the open intent to
1414 intermediate path components.
1417 Frequency : rare (liblustre only with non-standard tuning)
1418 Bugzilla : 7201 (7350)
1419 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1420 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1421 after it has been released, only do this for referenced locks
1423 Severity : enhancement
1425 Description: specifying an (invalid) directory default stripe_size of -1
1426 would reset the directory default striping
1427 Details : stripe_size -1 was used internally to signal directory stripe
1428 removal, now use "all default" to signal dir stripe removal
1429 as a directory striping of "all default" is not useful
1432 Frequency : common for large clusters running liblustre clients
1434 Description: doing an ls when liblustre clients are running is slow
1435 Details : sending a glimpse AST to a liblustre client waits for every AST
1436 to time out, as liblustre clients will not respond. Since they
1437 cannot cache data we refresh the OST lock LVB from disk instead.
1439 Severity : enhancement
1441 Description: doing an ls at the same time as file IO can be slow
1442 Details : enqueue and other "small" requests can be blocked behind many
1443 large IO requests. Create a new OST IO portal for non-IO
1444 requests so they can be processed faster.
1447 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1449 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1450 created file with mode 000
1451 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1452 allow an ftruncate() as a non-root user to a file with mode 000.
1453 Lustre now respects this flag to disable mode checks when
1454 truncating a file owned by the user
1457 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1460 Description: Revoking locks from clients that went dead or catatonic might take
1462 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1463 cancellation of such locks instant on servers without waiting for
1464 any reply from clients. Clients drops these locks when cancel
1465 notification from server is received without replying.
1468 Frequency : liblustre-only, when liblustre client dies or becomes busy
1470 Description: Doing ls on Linux clients can take a long time with active
1472 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1473 granting such locks to it in the first place if possible. Locks
1474 are taken by proxy on the OST during the read or write and
1475 dropped immediately afterward. Add connect flags handling, do
1476 not grant locks to liblustre clients for glimpse ASTs.
1478 Severity : enhancement
1480 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1481 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1482 sequential reads and behaves badly with applications doing random
1483 reads. Improve it by reading ahead at least read region, and
1484 avoiding excessive large RPC for small reads.
1486 Severity : enhancement
1488 Description: Creating more than 1000 files for a single job may cause a load
1489 imbalance on the OSTs if there are also a large number of OSTs.
1490 Details : qos_prep_create() uses an OST index reseed value that is an
1491 even multiple of the number of available OSTs so that if the
1492 reseed happens in the middle of the object allocation it will
1493 still utilize the OSTs as uniformly as possible.
1498 Description: OST or MDS may oops in ping_evictor_main()
1499 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1500 but doesn't restart at beginning of obd_exports_timed list
1503 Severity : enhancement
1505 Description: improve by-nid export eviction on the MDS and OST
1506 Details : allow multiple exports with the same NID to be evicted at one
1507 time without re-searching the exports list.
1510 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1512 Description: MDS may oops in groups_free()
1513 Details : in rare race conditions a newly allocated group_info struct is
1514 freed again, and this can be NULL. The 2.4 compatibility code
1515 for groups_free() checked for a NULL pointer, but 2.6 did not.
1518 Frequency : common for liblustre clients doing little filesystem IO
1519 Bugzilla : 9352, 7313
1520 Description: server may evict liblustre clients accessing contended locks
1521 Details : if a client is granted a lock or receives a completion AST
1522 with a blocking AST already set it would not reply to the AST
1523 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1526 Frequency : lfs setstripe, only systems with more than 160 OSTs
1528 Description: unable to set striping with a starting offset beyond OST 160
1529 Details : llapi_create_file() incorrectly limited the starting stripe
1530 index to the maximum single-file stripe count.
1533 Frequency : LDAP users only
1535 Description: lconf did not handle in-kernel recovery with LDAP properly
1536 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1538 Severity : enhancement
1540 Description: bind OST threads to NUMA nodes to improve performance
1541 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1542 node and do their allocations there to localize memory access
1544 Severity : enhancement
1546 Description: llmount can determine client NID directly from Myrinet (GM)
1547 Details : the client NID code from gmnalnid was moved directly into
1548 llmount, removing the need to use this or specifying the
1549 client NID explicitly when mounting GM clients with zeroconf
1552 Frequency : if client is started with down MDS
1554 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1555 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1556 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1557 ptlrpc_import_delayed_req() because we didn't check if the
1558 request was interrupted, and we also didn't break out of the
1559 event loop if there was no timeout
1564 Description: data loss during non-page-aligned writes to a single file from
1565 both multiple nodes and multiple threads on one node at same time
1566 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1567 inconsistency led to false short-reads, that were cached and later
1568 used by ->prepare_write() to fill in partially written page,
1569 leading to data loss.
1572 Frequency : always, if lconf --abort_recovery used
1574 Description: lconf --abort_recovery fails with 'Operation not supported'
1575 Details : lconf was attempting to abort recovery on the MDT device and not
1578 Severity : enhancement
1580 Description: remove cleanup logs
1581 Details : replace lconf-generated cleanup logs with lustre internal
1582 cleanup routines. Eliminates the need for client-cleanup and
1585 Severity : enhancement
1587 Description: add support for EAs (user and system) on lustre filesystems
1588 Details : it is now possible to store extended attributes in the Lustre
1589 client filesystem, and with the user_xattr mount option it
1590 is possible to allow users to store EAs on their files also
1592 Severity : enhancement
1594 Description: Add possibility (config option) to show minimal available OST free
1596 Details : When compiled with --enable-mindf configure option, statfs(2)
1597 (and so, df) will return least minimal free space available from
1598 all OSTs as amount of free space on FS, instead of summary of
1599 free spaces of all OSTs.
1601 Severity : enhancement
1603 Description: do not expand extent locks acquired on OST-side
1604 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1605 by server: they are not cached anyway.
1608 Frequency : when mmap is used/binaries executed from Lustre
1610 Description: Unmmap pages before throwing them away from read cache.
1611 Details : llap_shrink cache now attempts to unmap pages before discarding
1612 them (if unmapping failed - do not discard). SLES9 kernel has
1613 extra checks that trigger if this unmapping is not done first.
1618 Description: lconf didn't resolve symlinks before checking to see whether a
1619 given mountpoint was already in use
1622 Frequency : when migrating failover services
1623 Bugzilla : 6395, 9514
1624 Description: When migrating a subset of services from a node (e.g. failback
1625 from a failover service node) the remaining services would
1626 time out and evict clients.
1627 Details : lconf --force (implied by --failover) sets the global obd_timeout
1628 to 5 seconds in order to quickly disconnect, but this caused
1629 other RPCs to time out too quickly. Do not change the global
1630 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1632 Severity : enhancement
1633 Frequency : if MDS is started with down OST
1634 Bugzilla : 9439,5706
1635 Description: Allow startup/shutdown of an MDS without depending on the
1636 availability of the OSTs.
1637 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1638 Add appropriate locking and lov-osc refcounts for safe
1639 cleaning. Add osc abort_inflight calls in case the
1640 synchronize never started.
1643 Frequency : occasional (Cray XT3 only)
1645 Description: root not authorized to access files in CRAY_PORTALS environment
1646 Details : The client process capabilities were not honoured on the MDS in
1647 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1648 been accessed by an authorized user then root was able to access
1649 the file on the local client also. The root user capabilities
1650 are now allowed on the MDS, as this environment has secure UID.
1653 Frequency : occasional
1655 Description: ldiskfs "too long searching" message happens too often
1656 Details : A debugging message (otherwise harmless) prints too often on
1657 the OST console. This has been reduced to only happen when
1658 there are fragmentation problems on the filesystem.
1663 Description: Division by zero in statfs when all OSCs are inactive
1664 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
1665 lov_statfs divides by value returned by lov_get_stripecnt().
1669 Bugzilla : 9489, 3273
1670 Description: First write from each client to each OST was only 4kB in size,
1671 to initialize client writeback cache, which caused sub-optimal
1672 RPCs and poor layout on disk for the first writen file.
1673 Details : Clients now request an initial cache grant at (re)connect time
1674 and so that they can start streaming writes to the cache right
1675 away and always do full-sized RPCs if there is enough data.
1676 If the OST is rebooted the client also re-establishes its grant
1677 so that client cached writes will be honoured under the grant.
1682 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
1683 Details : Now I/O RPCs go to different portal number and (presumably) fast
1684 lock requests (and glimses) and other RPCs get their own service
1685 threads pool that should be able to service those RPCs
1688 Severity : enhancement
1690 Description: Ability to exchange lustre version between client and servers and
1691 issue warnings at client side if client is too old. Also for
1692 liblustre clients there is ability to refuse connection of too old
1694 Details : New 'version' field is added to connect data structure that is
1695 filled with version info. That info is later checked by server and
1699 Frequency : rare, liblustre only.
1700 Bugzilla : 9296, 9581
1701 Description: Two simultaneous writes from liblustre at offset within same page
1702 might proceed at the same time overwriting eachother with stale
1704 Details : I/O lock withing llu_file_prwv was released too early, before data
1705 actually was hitting the wire. Extended lock-holding time until
1706 server acknowledges receiving data.
1709 Frequency : extremely rare. Never observed in practice.
1711 Description: avoid generating lustre_handle cookie of 0.
1712 Details : class_handle_hash() generates handle cookies by incrementing
1713 global counter, and can hit 0 occasionaly (this is unlikely, but
1714 not impossible, because initial value of cookie counter is
1715 selected randonly). Value of 0 is used as a sentinel meaning
1716 "unassigned handle" --- avoid it. Also coalesce two critical
1717 sections in this function into one.
1719 Severity : enhancement
1721 Description: allow liblustre clients to delegate truncate locking to OST
1722 Details : To avoid overhead of locking, liblustre client instructs OST to
1723 take extent lock in ost_punch() on client's behalf. New connection
1724 flag is added to handle backward compatibility.
1726 Severity : enhancement
1727 Bugzilla : 4928, 7341, 9758
1728 Description: allow number of OST service threads to be specified
1729 Details : a module parameter allows the number of OST service threads
1730 to be specified via "options ost ost_num_threads={N}" in the
1731 OSS's /etc/modules.conf or /etc/modprobe.conf.
1735 Bugzilla : 6146, 9635, 9895
1736 Description: servers crash with bad pointer in target_handle_connect()
1737 Details : In rare cases when a client is reconnecting it was possible that
1738 the connection request was the last reference for that export.
1739 We would temporarily drop the export reference and get a new
1740 one, but this may have been the last reference and the export
1741 was just destroyed. Get new reference before dropping old one.
1743 Severity : enhancement
1744 Frequency : if client is started with failover MDS
1746 Description: Allow multiple MDS hostnames in the mount command
1747 Details : Try to read the configuration from all specified MDS
1748 hostnames during a client mount in case the "primary"
1751 Severity : enhancement
1753 Description: Stop sending data to evicted clients as soon as possible.
1754 Details : Check if the client we are about to send or are sending data to
1755 was evicted already. (Check is done every second of waiting,
1756 for which l_wait_event interface was extended to allow checking
1757 of exit condition at specified intervals).
1760 Frequency : rare, normally only when NFS exporting is done from client
1762 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
1764 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
1767 Frequency : always, if srand() is called before liblustre initialization
1769 Description: Liblustre uses system PRNG disturbing its usage by user application
1770 Details : Introduce internal to lustre fast and high-quality PRNG for
1771 lustre usage and make liblustre and some other places in generic
1772 lustre code to use it.
1774 Severity : enhancement
1775 Bugzilla : 9477, 9557, 9870
1776 Description: Verify that the MDS configuration logs are updated when xml is
1777 Details : Check if the .xml configuration logs are newer than the config
1778 logs stored on the MDS and report an error if this is the case.
1779 Request --write-conf, or allow starting with --old_conf.
1781 Severity : enhancement
1783 Description: Handle symlinks in the path when checking if Lustre is mounted.
1784 Details : Resolve intermediate symlinks when checking if a client has
1785 mounted a filesystem to avoid duplicate client mounts.
1790 Description: lconf can hit an error exception but still return success.
1791 Details : The lconf command catches the Command error exception at the top
1792 level script context and will exit with the associated exit
1793 status, but doesn't ensure that this exit status is non-zero.
1798 Description: failure of ptlrpc thread startup can cause oops
1799 Details : Starting a ptlrpc service thread can fail if there are a large
1800 number of threads or the server memory is very fragmented.
1801 Handle this without oopsing.
1804 Frequency : always, only if liblustre and non-default acceptor port was used
1806 Description: liblustre cannot connect to servers with non-default acceptor port
1807 Details : tcpnal_set_default_params() was not called and was therefore
1808 ignoring the environment varaible TCPNAL_PORT, as well as other
1809 TCPNAL_ environment variables
1814 Description: two objects could be created on the same OST for a single file
1815 Details : If an OST is down, in some cases it was possible to create two
1816 objects on a single OST for a single file. No problems other
1817 than potential performance impact and spurious error messages.
1821 Bugzilla : 5681, 9562
1822 Description: Client may oops in ll_unhash_aliases
1823 Details : Client dcache may become inconsistent in race condition.
1824 In some cases "getcwd" can fail if the current directory is
1830 Description: Inode refcounting problems in NFS export code
1831 Details : link_raw functions used to call d_instantiate without obtaining
1832 extra inode reference first.
1836 Bugzilla : 9942, 9903
1837 Description: Referencing freed requests leading to crash, memleaks with NFS.
1838 Details : We used to require that call to ll_revalidate_it was always
1839 followed by ll_lookup_it. Also with revalidate_special() it is
1840 possible to call ll_revalidate_it() twice for the same dentry
1841 even if first occurence returned success. This fix changes semantic
1842 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
1843 reference on a request referred from the intent.
1844 ll_intent_release() then releases such a request.
1847 Frequency : rare, normally benchmark loads only
1849 Description: unlinked inodes were kept in memory on the client
1850 Details : If a client is repeatedly creating and unlinking files it
1851 can accumulate a lot of stale inodes in the inode slab cache.
1852 If there is no other client load running this can cause the
1853 client node to run out of memory. Instead flush old inodes
1854 from client cache that have the same inode number as a new inode.
1857 Frequency : SLES9 2.6.5 kernel and long filenames only
1858 Bugzilla : 9969, 10379
1859 Description: utime reports stale NFS file handle
1860 Details : SLES9 uses out-of-dentry names in some cases, which confused
1861 the lustre dentry revalidation. Change it to always use the
1865 Frequency : rare, unless heavy write-truncate concurrency is continuous
1866 Bugzilla : 4180, 6984, 7171, 9963, 9331
1867 Description: OST becomes very slow and/or deadlocked during object unlink
1868 Details : filter_destroy() was holding onto the parent directory lock
1869 while truncating+unlinking objects. For very large objects this
1870 may block other threads for a long time and slow overall OST
1871 responsiveness. It may also be possible to get a lock ordering
1872 deadlock in this case, or run out of journal credits because of
1873 the combined truncate+unlink. Solution is to do object truncate
1874 first in one transaction without parent lock, and then do the
1875 final unlink in a new transaction with the parent lock. This
1876 reduces the lock hold time dramatically.
1879 Frequency : rare, 2.4 kernels only
1881 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
1882 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
1883 before unmounting their filesystem in order to avoid deadlock.
1884 The kernel_locked() function in 2.4 kernels only checks whether
1885 the kernel lock is held, not whether it is this process that is
1886 holding it as 2.6 kernels do.
1891 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
1892 Details : The client ptlrpc code may be trying to reconnect to a down
1893 server before a previous connection attempt has timed out.
1894 Increase the reconnect interval to be longer than the connection
1895 timeout interval to avoid sending duplicate connections to
1899 Frequency : echo_client brw_test command
1901 Description: fix echo_client to work with OST preallocated code
1902 Details : OST preallocation code (5137) didn't take echo_client IO path
1903 into account: echo_client calls filter methods outside of any
1904 OST thread and, hence, there is no per-thread preallocated
1905 pages and buffers to use. Solution: hijack pga pages for IO. As
1906 a byproduct, this avoids unnecessary data copying.
1910 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
1911 Description: Client can oops in mdc_commit_close() after open replay
1912 Details : It was possible for the MDS to return an open request with no
1913 transaction number in mds_finish_transno() if the client was
1914 evicted, but without actually returning an error. Clients
1915 would later try to replay that open and may trip an assertion
1916 Simplify the client close codepath, and always return an error
1917 from the MDS in case the open is not successful.
1920 Frequency : rare, 2.6 OSTs only
1922 Description: OST may deadlock under high load on fragmented files
1923 Details : If there was a heavy load and highly-fragmented OST filesystems
1924 it was possible to have all the OST threads deadlock waiting on
1925 allocation of biovecs, because the biovecs were not released
1926 until the entire RPC IO was completed. Instead, release biovecs
1927 as soon as they are complete to ensure forward IO progress.
1929 Severity : enhancement
1931 Description: Support for specifying external journal device at mount
1932 Details : If an OST or MDS device is formatted with an external journal
1933 device, this device major/minor is stored in the ext3 superblock
1934 and may not be valid for failover. Allow detecting and
1935 specifying the external journal at mount time.
1940 Description: Mounting an MDS with pending unlinked files may cause oops
1941 Details : target_finish_recovery() calls mds_postrecov() which returned
1942 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
1943 considers this an error and immediately begins cleaning up the
1944 lov, just after starting the mds_lov process
1946 Severity : enhancement
1948 Description: Implement 'lfs df' to report actual free space on per-OST basis
1949 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
1950 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
1951 usage of objects; '-h' to report in human readable format.
1953 ------------------------------------------------------------------------------
1955 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
1962 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
1963 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
1964 at startup during orphan unlinking. Instead, skip buddy update
1965 before setup, as it will be regenerated after recovery is complete.
1970 Description: If an OST is inactive, its locks might reference stale inodes.
1971 Details : lov_change_cbdata() must iterate over all namespaces, even if
1972 they are inactive to clear inode references from the lock.
1974 Severity : enhancement
1975 Frequency : occasional, if non-standard max_dirty_mb used
1977 Description: Client will block write RPCs if not enough grant
1978 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
1979 then the client will block writes while waiting for another RPC
1980 to complete instead of consuming its dirty limit. With change
1981 we get improved performance when max_dirty_mb is small.
1983 Severity : enhancement
1984 Bugzilla : 3389, 6253
1985 Description: Add support for supplementary groups on the MDS.
1986 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
1987 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
1988 lookups for user supplementary groups into a cache.
1992 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
1993 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
1994 for liblustre O_EXCL is enforced by MDS.
1998 Description: readdir never returns NULL in liblustre.
1999 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2000 dentry in struct dirent.
2004 Frequency : liblustre only, depends on application IO pattern
2005 Description: liblustre clients evicted if not contacting servers
2006 Details : Don't put liblustre clients into the ping_evictor list, so
2007 they will not be evicted by the pinger ever.
2009 Severity : enhancement
2011 Description: Add ability to evict clients by NID from MDS.
2012 Details : By echoing "nid:$NID" string into
2013 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2014 $NID would be instantly evicted from this MDS and from all active
2015 OSTs connected to it.
2019 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2020 Details : lookup_it_finish() used to query file size from OSTs that was not
2025 Description: service threads change working directory to that of init
2026 Details : Starting lustre service threads may pin the working directory
2027 of the parent thread, making that filesystem busy. Threads
2028 now change to the working directory of init to avoid this.
2032 Frequency : during shutdown only
2033 Description: shutdown with a failed MDS or OST can cause unmount to hang
2034 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2039 Frequency : 2.6 only
2040 Description: chmod/chown may include an extra supplementary group
2041 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2042 supplementary group and if none is specified this is used.
2045 Bugzilla : 5479 (6816)
2047 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2048 Details : If lookup is in progress on a file that is unlinked we might try
2049 to revalidate the inode and fail in revalidate after lookup is
2050 complete and ll_file_open() enqueues the open again but
2051 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2054 Frequency : always, if lconf --abort_recovery used
2056 Description: lconf --abort_recovery fails with 'Operation not supported'
2057 Details : lconf was attempting to abort recovery on the MDT device and not
2060 ------------------------------------------------------------------------------
2062 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2067 Frequency : rare (only unsupported configurations with a node running as an
2069 Bugzilla : 6514, 5137
2070 Description: Mounting a Lustre file system on a node running as an OST could
2072 Details : OSTs now preallocates memory needed to write out data at
2073 startup, instead of when needed, to avoid having to
2074 allocate memory in possibly low memory situations.
2075 Specifically, if the file system is mounted on on OST,
2076 memory pressure could force it to try to write out data,
2077 which it needed to allocate memory to do. Due to the low
2078 memory, it would be unable to do so and the node would
2079 become unresponsive.
2081 Severity : enhancement
2083 Description: Addition of lconf --service command line option
2084 Details : lconf now accepts a '--service <arg>' option, which is
2085 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2087 Severity : enhancement
2089 Description: Failover mode is now the default for OSTs.
2090 Details : By default, OSTs will now run in failover mode. To return to
2091 the old behaviour, add '--failout' to the lmc line for OSTs.
2093 Severity : enhancement
2095 Description: Health checks are now provided for MDS and OSTs
2096 Details : Additional detailed health check information on MSD and OSTs
2097 is now provided through the procfs health_check value.
2100 Frequency : occasional, depends on IO load
2102 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2103 after numerous create/delete cycles
2104 Details : The ext3 inode allocation policy would not allocate new inodes
2105 very well on the OSTs because there are no new directories
2106 being created. Instead we look for groups with free space if
2107 the parent directories are nearly full.
2112 Description: Network or server problems during mount may cause partially
2113 mounted clients instead of returning an error.
2114 Details : The config llog parsing code may overwrite the error return
2115 code during mount error handling, returning success instead
2121 Description: MDS can fail to allocate large reply buffers
2122 Details : After long uptimes the MDS can fail to allocate large reply
2123 buffers (e.g. zconf client mount config records) due to memory
2124 fragmentation or consumption by the buffer cache. Preallocate
2125 some large reply buffers so that these replies can be sent even
2126 under memory pressure.
2130 Frequency : rare (liblustre)
2131 Description: fsx running with liblustre complained that using truncate() to
2132 extend the file doesn't work. This patch corrects that issue.
2133 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2134 ATTR_SIZE and lsm use in llu_setattr_raw.
2138 Frequency : rare, only 2.6 kernels
2139 Description: Unusual file access patterns on the MDS may result in inode
2140 data being lost in very rare circumstances.
2141 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2142 code in the 2.6 kernel caused the inode and/or EA data not to
2143 be read from disk, causing single-file corruption.
2147 Frequency : rare, only 2.6 filesystems using extents
2148 Description: Heavy concurrent write and delete load may cause data corruption.
2149 Details : It was possible under high-load situations to have an extent
2150 metadata block in the block device cache from a just-unlinked
2151 file overwrite a newly-allocated data block. We now unmap any
2152 metadata buffers that alias just-allocated data blocks.
2156 Frequency : filesystems with default stripe_count larger than 77
2157 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2158 Details : lconf specifies an inode size of 4096 bytes when the default
2159 stripe_count is larger than 77. This conflicts with the default
2160 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2161 this case to avoid pinning too much memory for large EAs.
2163 ------------------------------------------------------------------------------
2165 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2170 Frequency : rare (extremely heavy IO load with hundreds of clients)
2172 Description: Client is evicted, gets IO error writing to file
2173 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2174 caused clients to be evicted for AST timeouts. The fixes in
2175 bug 5192 mean we no longer need to have such short AST timeouts
2176 so ldlm_timeout has been increased.
2179 Frequency : occasional during --force or --failover shutdown under load
2180 Bugzilla : 5949, 4834
2181 Description: Server oops/LBUG if stopped with --force or --failover under load
2182 Details : a collection of import/export refcount and cleanup ordering
2183 issues fixed for safer force cleanup
2186 Frequency : only filesystems larger than 120 OSTs
2187 Bugzilla : 5990, 6223
2188 Description: lfs getstripe would oops on a very large filesystem
2189 Details : lov_getconfig used kfree on vmalloc'd memory
2192 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2193 Bugzilla : 6242, 6243
2194 Description: reading from files that had been truncated to a non-zero size
2195 but never opened returned no data
2196 Details : ll_file_read() reads zeros from no-object files to EOF
2201 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2202 Details : There's little harm in aborting MDS/OSS recovery and letting it
2203 try again, so I removed the LASSERT and return an error instead.
2205 Severity : enhancement
2207 Description: New debugging infrastructure for tracking down data corruption
2208 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2209 (b) cover more of the client-side code path, and (c) try to narrow
2210 down where problems occurred
2214 Bugzilla : 3819, 4364, 4397, 6313
2215 Description: Racing close and eviction MDS could cause assertion in mds_close
2216 Details : It was possible to get multiple mfd references during close and
2217 client eviction, leading to one thread referencing a freed mfd.
2219 Severity: : enhancement
2220 Bugzilla : 3262, 6359
2221 Description: Attempts to reconnect to servers are now more aggressive.
2222 Details : This builds on the enhanced upcall-less recovery that was added
2223 in 1.4.2. When trying to reconnect to servers, clients will
2224 now try each server in the failover group every 10 seconds. By
2225 default, clients would previously try one server every 25 seconds.
2230 Description: After recovery, certain operations trigger a failed
2231 assertion on a client.
2232 Details : Failing over an mds, using lconf -d --failover, while a
2233 client was doing a readdir() call would cause the client to
2234 LBUG after recovery completed and the readdir() was resent.
2236 Severity : enhancement
2238 Description: Default groups are now added by lconf
2239 Details : You can now run lconf --group <servicename> without having to
2240 manually add groups with lmc.
2243 Frequency : occasional
2245 Description: Nodes with an elan id of 0 trigger a failed assertion
2248 Frequency : always when accessing e.g. tty/console device nodes
2250 Description: tty and some other devices nodes cannot be used on lustre
2251 Details : file's private_data field is used by device data and lustre
2252 values in there got lost. New field was added to struct file to
2253 store fs-specific private data.
2256 Frequency : when exporting Lustre via NFS
2258 Description: NFSD failed occasionally when looking up a path component
2259 Details : NFSD is looking up ".." which was broken in ext3 directories
2260 that had grown large enough to become hashed.
2263 Frequency : Clusters with multiple interfaces not on the same subnet
2265 Description: Nodes will repeatedly try to reconnect to an interface which it
2266 cannot reach and report an error to the log.
2267 Details : Extra peer list entries will be created by lconf with some peers
2268 unreachable. lconf now validates the peer before adding it.
2271 Frequency : Only if a default stripe is set on the filesystem root.
2273 Description: Setting a default stripe on the filesystem root prevented the
2274 filesystem from being remounted.
2275 Details : The client was sending extra request flags in the root getattr
2276 request and did not allocate a reply buffer for the dir EA.
2279 Frequency : occasional, higher if lots of files are accessed by one client
2280 Bugzilla : 6159, 6097
2281 Description: Client trips assertion regarding lsm mismatch/magic
2282 Details : While revalidating inodes the VFS looks up inodes with ifind()
2283 and in rare cases can find an inode that is being freed.
2284 The ll_test_inode() code will free the lsm during ifind()
2285 when it finds an existing inode and then the VFS later attaches
2286 this free lsm to a new inode.
2290 Bugzilla : 6422, 7030
2291 Description: MDS deadlock between mkdir and client eviction
2292 Details : Creating a new file via mkdir or mknod (starting a transaction
2293 and getting the ns lock) can deadlock with client eviction
2294 (gets ns lock and trying to finish a synchronous transaction).
2297 Frequency : occasional
2298 Description: While starting a server, the fsfilt_ext3 module could not be
2300 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2301 kernels. Previously, lconf would still use the ext3 name
2302 when trying to load modules. Now, it will correctly use
2303 ext3 on 2.4 and ldiskfs on 2.6.
2305 Severity : enhancement
2306 Description: The default stripe count has been changed to 1
2307 Details : The interpretation of the default stripe count (0, to lfs
2308 or lmc) has been changed to mean striping across a single
2309 OST, rather than all available. For general usage we have
2310 found a stripe count of 1 or 2 works best.
2312 Severity : enhancement
2313 Description: Add support for compiling against Cray portals.
2314 Details : Conditional compiling for some areas that are different
2318 Frequency : occasional
2319 Bugzilla : 6409, 6834
2320 Description: Creating files with an explicit stripe count may lead to
2321 a failed assertion on the MDS
2322 Details : If some OSTs are full or unavailable, creating files may
2323 trigger a failed assertion on the MDS. Now, Lustre will
2324 try to use other servers or return an error to the
2328 Frequency : occasional
2330 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2331 caused lock timeout during readahead (since 1.4.2).
2332 Details : Processes doing readahead might match a lock that hasn't been
2333 granted yet if there are overlapping and conflicting lock
2334 requests. The readahead process waits on ungranted lock
2335 (original lock is CBPENDING), while OST waits for that process
2336 to cancel CBPENDING read lock and eventually evicts client.
2338 Severity : enhancement
2340 Description: Initial enabling of flock support for clients
2341 Details : Implements fcntl advisory locking and file status functions.
2342 This feature is provided as an optional mount flag (default
2343 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2344 locking are implemented yet, and those that are are not guaranteed
2345 to be completely correct in production environments.
2346 mount -t lustre -o [flock|noflock] ...
2349 Frequency : occasional
2351 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2352 assertion in the ext3 JBD (journaling) layer.
2353 Details : The b_committed_data struct is protected by the big kernel lock
2354 in 2.4 kernels, serializing journal_commit_transaction() and
2355 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2356 access to this struct. In 2.6 kernels there is finer grained
2357 locking to improve SMP performance of the JBD layer.
2361 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2363 -----------------------------------------------------------------------------
2365 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2367 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2368 versions of Lustre. Please update all servers and clients to
2369 version 1.4.2 or later at the same time. You must also run
2370 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2371 to update the configuration logs.
2373 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2374 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2375 - swab llog records of type '0' so we get proper header size/idx (5861)
2376 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2377 - fix rename of one directory over another leaking an inode (5953)
2378 - avoid SetPageDirty on 2.6 (5981)
2379 - don't re-add just-being-destroyed locks to the waiting list (5653)
2380 - when creating new directories, inherit the parent's custom
2381 striping settings if present parent (3048)
2382 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2383 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2384 - don't hold client locks on temporary worklist from l_lru (5666)
2385 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2386 - automatically evict dead clients (5921)
2387 - Update file size properly in create+truncate+fstat case (6196)
2388 - Do not unhash mountpoint dentries, do not allow removal of
2390 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2391 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2392 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2393 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2394 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2395 - Correctly handle reads of files with no objects (6243)
2396 - lctl recover will also mark a device active if deactivate used (5933)
2398 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2399 - allow --write-conf on an MDS with different nettype than client (5619)
2400 - don't write config llogs to MDS for mounts not from that MDS (5617)
2401 - lconf should create multiple TCP connections from a client (5201)
2402 - init scripts are now turned off by default; run chkconfig --on
2403 lustre and chkconfig --on lustrefs to use them
2404 - upcalls are no longer needed for clients to recover to failover
2406 - add --abort-recovery option to lconf to abort recovery on device
2408 - add support for an arbitrary number of OSTs (3026)
2409 - Quota support protocol changes.
2410 - forward compatibility changes to wire structs (6007)
2411 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2412 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2413 - improved "lustre status" script
2414 - initialize blocksize for non-regular files (6062)
2415 - added --disable-server and --disable-client configure options (5782)
2416 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2417 - Vanilla 2.4.29 support
2418 - increase maximum number of obd devices to 520 (6242)
2419 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2420 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2421 SLES 9 2.6.5 kernels
2422 - limit stripes per file to 160 (the maximum EA size) (6093)
2424 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2427 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2428 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2429 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2430 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2431 - small ext3 extents cleanups and fixes (5733)
2432 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2433 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2434 - avoid needless client->OST connect, fix handle mismatch (5317)
2435 - fix DLM error path that led to out-of-sync client, long delays (5779)
2436 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2437 - fix several locking issues related to i_size (5492,5624,5654,5672)
2438 - don't move pending lock onto export if it is already evicted (5683)
2439 - fix kernel oops when creating .foo in unlinked directory (5548)
2440 - fix deadlock in obdfilter statistics vs. object create (5811)
2441 - use time_{before,after} to avoid timer jiffies wrap (5882)
2442 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2443 - Do not leak request if server was not able to process it (5154)
2444 - If mds_open unable to find parent dir, make that negative lookup(5154)
2445 - don't create new directories with extent-mapping (5909, 5936)
2447 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2448 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2449 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2450 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2451 - support for RHEL4 kernel on i686 (5773)
2452 - provide error messages when incompatible logs are encountered (5898)
2454 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2455 * version 1.4.0.10 (1.4.1 release candidate 1)
2457 - don't keep a lock reference when lock is not granted (4238)
2458 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2459 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2460 - drop import inflight refcount on signal_completed_replay error (5255)
2461 - unlock page after async write error during send (3677)
2462 - handle missing objects in filter_preprw_read properly (5265)
2463 - no transno return for symlink open, don't save no-trasno open (3440)
2464 - don't try to complete elan receive that already failed (4012)
2465 - free RPC server reply state on error (5406)
2466 - clean up thread from ptlrpc_start_thread() on error (5160)
2467 - readahead could read extra page into cache that wasn't ejected (5388)
2468 - prevent races in class_attach/setup/cleanup/detach (5260)
2469 - don't dereference de->d_inode after l_dput of de (5458)
2470 - use "int" for stripe value returned from lock_to_stripe (5544)
2471 - mballoc allocation and error-checking fixes in 2.6 (5504)
2472 - block device patches to fix I/O request sizes in 2.6 (5482)
2473 - look up hostnames for IB nals (5602)
2474 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2475 - don't start multiple acceptors for the same port (5277)
2476 - fix incorrect LASSERT in mds_getattr_name (5635)
2477 - export a proc file for general "ping" checking (5628)
2478 - fix "lfs check" to not block when the MDS is down (5628)
2480 - service request history (4965)
2481 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2482 - create an "evict_client" /proc entry on OSTs, like the MDS has
2483 - fix mount usage message, return errors per mount(8) (5168)
2484 - change grep [] to grep "[]" in tests so they work in more UMLs
2485 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2486 - remove ancient LOV_MAGIC_V0 EA support (5047)
2487 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2488 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2489 - allow readahead allocations to fail when low on memory (5383)
2490 - mmap locking landed again, after considerable improvement (2828)
2491 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2493 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2496 - send OST transaction number in read/write reply to free req (4966)
2497 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2498 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2499 - the watchdog thread now runs as interruptible (5246)
2500 - flock/lockf fixes (but it's still disabled, pending 5135)
2501 - don't use EXT3 constants in llite code (5094)
2502 - memory shortage at startup could cause assertion (5176)
2504 - reorganization of lov code
2505 - single portals codebase
2507 - add extents/mballoc support (5025)
2508 - direct I/O reads in the obdfilter (4048)
2509 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2511 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2514 - send OST transaction number in read/write reply to free req (4966)
2515 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2516 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2517 - the watchdog thread now runs as interruptible (5246)
2518 - handle missing objects in filter_preprw_read properly (5265)
2519 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2520 - drop import inflight refcount on signal_completed_replay error (5255)
2521 - unlock page after async write error during send (3677)
2522 - return original error code on reconstructed replies (3761)
2523 - no transno return for symlink open, don't save no-trasno open (3440)
2525 - add pid to ldlm debugging output (4922)
2526 - bump the watchdog timeouts -- we can't handle 30sec yet
2527 - extra debugging for orphan dentry/inode bug (5259)
2529 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2532 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2533 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2534 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2535 - fix NULL dereference in /proc/sys/portals/routes (4827)
2536 - allow failed mdc_close() operations to be interrupted (4561)
2537 - stop precreate on OST before MDS would time out on it (4778)
2538 - don't send partial-page writes before EOF from client (4410)
2539 - discard client grant for sub-page writes on large-page clients (4520)
2540 - don't free dentries not owned by NFS code, check generation (4806)
2541 - fix lsm leak if mds_create_objects() fails (4801)
2542 - limit debug_daemon file size, always print CERROR messages (4789)
2543 - use transno after validating reply (3892)
2544 - process timed out requests if import state changes (3754)
2545 - update mtime on OST during writes, return in glimpse (4829)
2546 - add mkfsoptions to LDAP (4679)
2547 - use ->max_readahead method instead of zapping global ra (5039)
2548 - don't interrupt __l_wait_event() during strace
2550 - add software watchdogs to catch hung threads quickly (4941)
2551 - make lustrefs init script start after nfs is mounted
2552 - fix CWARN/ERROR duplication (4930)
2553 - return async write errors to application if possible (2248)
2554 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2555 - print NAL number in %x format (4645)
2556 - update barely-supported suse-2.4.21-171 series (4842)
2557 - support for sles 9 %post scripts
2558 - support for building 2.6 kernel-source packages
2559 - support for sles km_* packages
2561 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2564 - ignore -ENOENT errors in osc_destroy (3639)
2565 - notify osc create thread that OSC is being cleaned up (4600)
2566 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2567 - reconstruct ost_handle() like mds_handle() (4657)
2568 - create a new thread to do import eviction to avoid deadlock (3969)
2569 - let lconf resolve symlinked-to devices (4629)
2570 - don't unlink "objects" from directory with default EA (4554)
2571 - hold socknal file ref over connect in case target is down (4394)
2572 - allow more than 32000 subdirectories in a single directory (3244)
2573 - fix blocks count for O_DIRECT writes (3751)
2574 - OST returns ENOSPC from object create when no space left (4539)
2575 - don't send truncate RPC if file size isn't changing (4410)
2576 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2577 - bind to privileged port in socknal and tcpnal (3689)
2579 - rate limit CERROR/CWARN console message to avoid overload (4519)
2580 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2581 - basic mmap support (3918)
2582 - kernel patch series update from b1_4 (4711)
2584 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2587 - avoid crash during MDS cleanup with OST shut down (2775)
2588 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2589 - don't use bad inodes on the MDS (3744)
2590 - dynamic object preallocation to improve recovery speed (4236)
2591 - don't hold spinlock over lock dumping or change debug flags (4401)
2592 - don't zero obd_dev when it is force cleaned (3651)
2593 - print grants to console if they go negative (4431)
2594 - "lctl deactivate" will stop automatic recovery attempts (3406)
2595 - look for existing locks in ldlm_handle_enqueue() (3764)
2596 - don't resolve lock handle twice in recovery avoiding race (4401)
2597 - revalidate should check working dir is a directory (4134)
2599 - don't always mark "slow" obdfilter messages as errors (4418)
2601 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2604 - don't close LustreDB during write_conf until it is done (3860)
2605 - fix typo in lconf for_each_profile (3821)
2606 - allow dumping logs from multiple threads at one time (3820)
2607 - don't allow multiple threads in OSC recovery (3812)
2608 - fix debug_size parameters (3864)
2609 - fix mds_postrecov to initialize import for llog ctxt (3121)
2610 - replace config semaphore with spinlock (3306)
2611 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2612 - don't allow enqueue to complete on a destroyed export (3822)
2613 - down write_lock before checking llog header bitmap (3825)
2614 - recover from lock replay timeout (3764)
2615 - up llog sem before sending rpc (3652)
2616 - reduce ns lock hold times when setting kms (3267)
2617 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2618 - fix NULL deref and obd_dev leak on setup error (3312)
2619 - replace some LBUG about llog ops with error handling (3841)
2620 - don't match INVALID dentries from d_lookup and spin (3784)
2621 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2622 - fix invalid assertion in ptlrpc_set_wait (3880)
2624 - add libwrap support for the TCP acceptor (3996)
2625 - add /proc/sys/portals/routes for non-root route listing (3994)
2626 - allow setting MDS UUID in .xml (2580)
2627 - print the stack of a process that LBUGs (4228)
2629 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2632 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2633 - make sure to unset replay flag from failed open requests (3440)
2634 - if default stripe count is 0, use OST count for inode size (3636)
2635 - update parent mtime/ctime on client for create/unlink (2611)
2636 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2637 - free recovery state on server during a forced cleanup (3571)
2638 - unregister_reply for resent reqs (3063)
2639 - loop back devices mounting and status check on 2.6 (3563)
2640 - fix resource-creation race that can provoke i_size == 0 (3513)
2641 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2642 - more debugging for page-accounting assertion (3746)
2643 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2644 - avoid deadlock after precreation failure (3758)
2645 - fix race and lock order deadlock in orphan handling (3450, 3750)
2646 - add validity checks when grabbing inodes from l_ast_data (3599)
2648 - add /proc/.../recovery_status to obdfilter (3428)
2649 - lightweight CDEBUG infrastructure, debug daemon (3668)
2650 - change default OSC RPC parameters to be better on small clusters
2651 - turn off OST read cache for files smaller than 32MB
2652 - install man pages and include them in rpms (3100)
2653 - add new init script for (un)mounting lustre filesystems (2593)
2654 - run chkconfig in %post for init scripts (3701)
2655 - drop scimac NAL (unmaintained)
2657 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
2660 - clean kiobufs before and after use (3485)
2661 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
2662 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
2663 - change init script to fail more clearly if not run as root (1528)
2664 - allow clients to reconnect during replay (1742)
2665 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
2666 - don't do DNS lookups on NIDs too small for IP addresses (3442)
2667 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
2668 - fix cond_resched (3554)
2669 - only evict unfinished clients after recovery (3515)
2670 - allow bulk resend, prevent data loss (3570)
2671 - dynamic ptlrpc request buffer allocation (2102)
2672 - don't allow unlinking open directory if it isn't empty (2904)
2673 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
2674 - remove extraneous obd dereference causing LASSERT failure (3334)
2675 - don't use get_cycles() when creating temp. files on the mds (3156)
2676 - hold i_sem when setting i_size in ll_extent_lock() (3564)
2677 - handle EEXIST for set-stripe, set proper directory name (3336)
2679 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
2680 - fix ksocknal_fmb_callback() error messages (2918)
2682 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
2685 - don't copy lvb into (possibly NULL) reply on error (2983)
2686 - don't deref dentry after dput, don't free lvb on error (2922)
2687 - use the kms to determine writeback rpc length (2947)
2688 - increment oti_logcookies when osc is inactive (2948)
2689 - update client's i_blocks count via lvb messages (2543)
2690 - handle intent open/close of special files properly (1557)
2691 - mount MDS with errors=remount-ro, like obdfilter (2009)
2692 - initialize lock handle to avoid ASSERT on error cleanup (3057)
2693 - don't use cancelling-locks' kms values (2947)
2694 - use highest lock extent for kms, not last one (2925)
2695 - don't dereference ERR_PTR() dentry in error handling path (3107)
2696 - fix thread race in portals_debug_dumplog() (3122)
2697 - create lprocfs device entries at setup instead of at attach (1519)
2698 - common AST error handler, don't evict client on completion race (3145)
2699 - zero nameidata in detach_mnt in 2.6 (3118)
2700 - verify d_inode after revalidate_special is valid in 2.6 (3116)
2701 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
2702 - initialize RPC timeout timer earlier for 2.6 (3219)
2703 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
2704 - print nal/nid for unknown nid (3258)
2705 - additional checks for oscc recovery before doing precreate (3284)
2706 - fix ll_extent_lock() error return code for 64-bit systems (3043)
2707 - don't crash in mdc_close for bad permissions on open (3285)
2708 - zero i_rdev for non-device files (3147)
2709 - clear page->private before handing to FS, better assertion (3119)
2710 - tune the read pipeline (3236)
2711 - fix incorrect decref of invalidated dentry (2350)
2712 - provide read-ahead stats and refine rpc in flight stats (3328)
2713 - don't hold journal transaction open across create RPC (3313)
2714 - update atime on MDS at close time (3265)
2715 - close LDAP connection when recovering to avoid server load (3315)
2716 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
2717 - don't leak open file on MDS after open resend (3325)
2718 - serialize filter_precreate and filter_destroy_precreated (3329)
2719 - loop device shouldn't call sync_dev() for nul device (3092)
2720 - clear page cache after eviction (2766)
2721 - resynchronize MDS->OST in background (2824)
2722 - refuse to mount the same filesystem twice on same mountpoint (3394)
2723 - allow llmount to create routes for mounting behind routers (3320)
2724 - push lock cancellation to blocking thread for glimpse ASTs (3409)
2725 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
2726 - fix rare problem with rename on htree directories (3417)
2728 - allow default OST striping configuration per directory (1414)
2729 - fix compilation for qswnal for 2.6 kernels (3125)
2730 - increase maximum number of MDS request buffers for large systems
2731 - change liblustreapi to be useful for external progs like lfsck (3098)
2732 - increase local configuration timeout for slow disks (3353)
2733 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
2735 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2738 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
2739 - don't overwrite extent policy data in reply if lock was blocked (2901)
2740 - drop filter export grants atomically with removal from device (2663)
2741 - del obd_self_export from work_list in class_disconnect_exports (2908)
2742 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
2743 - swab reply message in mdc_close, other PPC fixes (2464)
2744 - fix destroying of named logs (2325)
2745 - overwrite old logs when running lconf --write_conf (2264)
2746 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
2747 - fix race in target_handle_connect (2898)
2748 - mds_reint_create() should take same inode create lock (2926)
2749 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
2750 - don't close files for self_export to avoid uninitialized obd (2936)
2751 - allow MDS with the same name as client node (2939)
2752 - hold dentry reference for closed log files for unlink (2325)
2753 - reserve space for all logs during transactions (2059)
2754 - don't evict page beyond end of stripe extent (2925)
2755 - don't oops on a deleted current working directory (2399)
2756 - handle hard links to targets without a parent properly (2517)
2757 - don't dereference NULL lock when racing during eviction (2867)
2758 - don't grow lock extents when lots of conflicting locks (2919)
2760 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
2763 - account for cache space usage on clients to avoid data loss (974)
2764 - lfsck support in lustre kernel code (2349)
2765 - reduce journal credits needed for BRW writes (2370)
2766 - orphan handling to avoid losing space on client/server crashes
2767 - ptlrpcd can be blocked, stopping ALL progress (2477)
2768 - use lock value blocks to assist in proper KMS, faster stat (1021)
2769 - takes i_sem instead of DLM locks internally on obdfilter (2720)
2770 - recovery for initial connections (2355)
2771 - fixes for mds_cleanup_orphans (1934)
2772 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
2773 - block all file creations until orphan recovery completes (1901)
2774 - client remove rq_connection from request struct (2423)
2775 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
2776 - recovery timer race (2670)
2777 - mdc_close recovey bug (2532)
2778 - ptlrpc cleanup bug (2710)
2779 - mds timeout on local locks (2588)
2780 - namespace lock held during RPCs (2431)
2781 - handle interrupted sync write properly (2503)
2782 - don't try to handle a message that hasn't been replied to (2699)
2783 - client assert failure during cleanup after abort recovery (2701)
2784 - leak mdc device after failed mount (2712)
2785 - ptlrpc_check_set allows timedout requests to complete (2714)
2786 - wait for inflight reqs when ptlrpcd finishes (2710)
2787 - make sure unregistered services are removed from the srv_list
2788 - reset bulk XID's when resending them (caught by 1138 test)
2789 - unregister_bulk after timeout
2790 - fix lconf error (2694)
2791 - handle write after unfinished setstripe, stripe-only getstripe (2388)
2792 - readahead locks pages, leaves pending causing memory pressure (2673)
2793 - increase OST request buffers to 4096 on large machines (2729)
2794 - fix up permission of existing directories in simple_mkdir (2661)
2795 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
2796 - don't assign transno to errored transactions (2742)
2797 - don't delete objects on OST if given a bogus objid from MDS (2751)
2798 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
2799 - if rq_no_resend, then timeout request after recovery (2432)
2800 - fix MDS llog_logid record size, 64-bit array alignment (2733)
2801 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
2802 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
2803 - MDS recovery shouldn't delete objects that it has given out (2730)
2804 - if enqueue arrives after completion, don't clobber LVB (2819)
2805 - don't unlock pages twice when trigger_group_io returns error (2814)
2806 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
2807 - don't write pages to disk if there was an error (1450)
2808 - don't ping imports that have recovery disabled (2676)
2809 - take buffered bytes into account when balancing socknal conn (2817)
2810 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
2811 - reconnect unlink llog connection after MDS reconnects to OST (2816)
2812 - remove little-endian swabbing of llog records (1987)
2813 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
2814 - retry reposting request buffers if they fail (1191)
2815 - grow extent at grant time to avoid granting a revoked lock (2809)
2816 - lock revoke doesn't evict page if covered by a second lock (2765)
2817 - disable VM readahead to avoid reading outside lock extents (2805)
2819 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
2820 - updated kernel patches for hp-2.4.20 kernel (2681)
2822 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
2825 - fix truncated write corruption (2366)
2826 - fix for failed assertion in iopen_connect_dentry (1792,2517)
2828 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
2829 - ensure len(uuid) < 37 in lmc (1171)
2830 - fix ia64 OOPS in llog_test (2255)
2831 - zero end of page at obdfilter for partial page writes (2648)
2832 - don't leave stale dentries around after renames (bug 2428)
2833 - fix timeouts when evicting a client with a single lock held (2642)
2834 - set deadline for the initial HELLO message to drain (2634)
2835 - print out dotted-quad IP addresses in the socknal (2302)
2837 - additional debugging for MDS client eviction problem (2443)
2838 - fix mkfsoptions support for osts (2603, 2604)
2840 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
2843 - add series for the vanilla 2.6.0 kernel
2844 - add series for the vanilla 2.4.24 kernel
2845 - add series for a cray x86/64 UL kernel drop
2846 - fix xattr patches for the vanilla 2.4.19 series
2848 - generate true UUIDs in lmc (1171)
2849 - have portals stack dumping break in UML (2466)
2850 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
2851 - allocate with _NOFS in ldlm to avoid deadlock (1933)
2852 - wake callback waiting threads on client eviction (2460)
2853 - Add --ptldebug and --subsystem to lmc (1719)
2854 - update assertion to allow safe interrupt allocation
2855 - set rq_no_resend for cancel requests (2432)
2856 - recalculate ptlrpcd timeout after resend (2494)
2857 - call vfs_rmdir when removing pending directories (2368)
2858 - fix renaming a file to itself (2429)
2859 - lmc creates a default one-stripe lov (2454)
2860 - expand procfs space to handle large clusters (2326)
2861 - increase UML stack to avoid overflow
2862 - update lconf's list of debug and subsystem masks
2863 - fix lfs find --obd (2510)
2864 - /proc tunable for disabling filter read caching (2591)
2865 - stop rpm packages from altering slapd.conf (2301)
2866 - disable nagle in the socknal under 0conf (2578)
2867 - choose mds inode size based on stripe count (2572)
2868 - fix kernel-source rpm problems (2516)
2870 - add --disable-doc to avoid pdf generation (2421)
2871 - update documentation, tests, type-os, comments
2872 - avoid format warnings on ia64
2873 - remove the TOE NAL
2874 - tiny code cleanups by removing unused fields
2876 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
2879 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
2880 - remove the most blatant lies from BUILDING file (2371)
2881 - change default debug level to reasonable production setting
2882 - reduce client side cache size to reduce cache flush time
2883 - reduce max RPCs in flight to avoid unnecessary file fragmentation
2884 - make TCP zerocopy and pinger support enabled by default (2476)
2885 - sync writes completed after process exits caused crashes (2319)
2886 - maintain correct mount count on the MDS (2356)
2887 - backout 1557, because 2316 wasn't really fixed
2888 - better file I/O statistics gathering in /proc
2889 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
2890 - another kernel patch to fix zero-copy TCP function export
2891 - don't take duplicate lock when processing re-sent getattr (2420)
2892 - lctl uses obd_self_export instead of creating new conn (2353)
2893 - MDS/OST recovery case which requires object creation asserted (2425)
2894 - move lfs from /usr/sbin to /usr/bin in packages
2895 - fix race between mds_client_add and mds_client_free (2417)
2896 - use kmalloc instead of slabs in portals (2430)
2897 - don't create duplicate records when a failover MDS is present (2442)
2898 - remove unnecessary mount age check (2332)
2899 - don't remove directory inodes from locks prematurely (2451)
2900 - don't break if MDS service name is the same as hostname (2103)
2901 - fix races in client write RPC generation when cache full (2482)
2903 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
2906 - remove now-unused request->rq_obd (278)
2907 - if an allocation fails, print out how much memory we've used (1933)
2908 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
2909 - add the "configurable stack size" patch to most series files (1256)
2910 - ability to write large log records, for 100+ OST configs (2306)
2911 - fix NULL deref when filter_prep fails (2314)
2912 - fix operator precedence error in filter_sync
2913 - dynamic allocation of socknal TX descriptors (2315)
2914 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
2915 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
2916 - ensure that CWARN messages in Portals always get to the syslog
2917 - __init/__exit are not for prototype decls (ldlm_init/exit)
2918 - x86-64 compile warning fixes
2919 - fix gateway LMC keyword conflict (2318)
2920 - fix MDS lock inversions in getattr/reint paths (1844)
2921 - fix a rare lock re-ordering bug, which caused deadlock (2322)
2922 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
2923 - DLM race condition prevented some lock evictions (2328)
2924 - ENOMEM detection and retry on socknal sends (2230)
2925 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
2926 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
2927 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
2928 - better error messages when a client is rejected during recovery (1505)
2929 - avoid cancelling locks which were never granted, after failure (2330)
2930 - fix i_sem/journal inversion in mds_client_add (2333)
2931 - fix truncate/getattr lock cycle deadlock (2334)
2932 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
2933 - fix two rare exit paths which could leak an l_lock() ref (2321)
2934 - fencepost error in MDS/OST orphan recovery (2226)
2935 - make log record alignment 8 bytes (1988)
2936 - lstripe now fails when requested offset > ost_count (2237)
2937 - ensure that all kernel series have a complete list.h (1607)
2938 - fix crashes in special-file operations (2316)
2939 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
2941 - allow configurable automake binary, for testing new versions
2942 - small update to the lfs documentation
2944 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
2946 * fix negative export reference count in fsfilt_sync (2312)
2948 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
2949 * release candidate 0.9.1
2951 - orphans are moved into the PENDING directory for possible recovery
2952 - replayed opens now open by fid for orphan/rename safety (1042)
2953 - last close of an orphan inode generates a transno (683)
2954 - chdir() and mount() now pin the directory entry (1020)
2955 - avoid CERROR in normal ll_setattr_raw() error case (1500)
2956 - discard very old requests without processing them (1502)
2957 - remove some common, well-understood CERRORs (1505)
2958 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
2959 - clear "grant" flags in OST replies until OST grant code lands (1644)
2960 - fix read performance by not clobbering i_blksize on client (1598)
2961 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
2962 - make LRU size a /proc tunable, clears locks when reduced (707)
2963 - fix some lprocfs rot that prevented ptlbd from loading (1732)
2964 - server locks take references on exports now (1558)
2965 - build fixes for 2.4.20-rh trees (1663)
2966 - return an error from lov_create if all OSCs are inactive (1751)
2967 - fix import levels when a reconnect happens without a timeout (1597)
2968 - exit early from mds_open if we get a lookup error (1749)
2969 - partial page read at EOF wouldn't wait for disk before sending (1642)
2970 - avoid NULL deref in obdfilter when reading page past EOF (1592)
2971 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
2972 - fix LBUG in ll_it_open_error with rc = -2 (1861)
2973 - write/truncate lock inversion (1639)
2974 - Don't auto-load obdclass, portals modules during cleanup (1495)
2975 - fix timestamps from jumping to "now" (1763)
2976 - extra journal assertions (1648)
2977 - add an extra multiunlink test (1771)
2978 - fix read_record/write_record API (1776)
2979 - fix leak of offset_extent, possible incorrect i_size later (1772)
2980 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
2981 - Debugging for the kqswnal_get_idle_tx problems (1820)
2982 - Allow recovery to be attempted multiple times (1536)
2983 - Write out MDS last_rcvd file after it is first created (1600)
2984 - Fix tx_descriptor leak in failed transmit situations (1827)
2985 - ext3 journaling fixes for assertion failure after IO error (1871)
2986 - class_export_put() on freed export after completion AST error (1896)
2987 - Fix revalidate looping in VFS (1322)
2988 - Don't access a freed export during MDS_REINT timeout (1521)
2989 - Add open-unlink recovery support on the MDS (1673,1764)
2990 - Return an error if no MDS data was read from last_rcvd (1946)
2991 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
2992 - Don't setup a disk device that doesn't match exported UUID (317)
2993 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
2994 - avoid committing NULL handle in force close
2995 - local.sh is now a one-stripe LOV configuration
2996 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
2997 - fix ext3 htree duplicate directory entry corruption (1516)
2998 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
2999 - update to new LOV EA format (2097)
3000 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3002 2003-06-15 Phil Schwan <phil@clusterfs.com>
3005 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3006 - per-import/export recovery handling (958, 931, 959)
3007 - multiple last-rcvd slots, for serving multiple FSes (949)
3008 - connections are again shared between multiple imp/exports (963, 964)
3009 - "umount -f" would hang if any requests needed to be sent (393, 978)
3010 - avoid pinning large req buffer by copying for queued messages (989)
3011 - add "uuid" to "lctl device" command to help upcalls (991)
3012 - "open" RPCs with transnos would confuse recovery counters (1037)
3013 - do proper endian conversion of all wire messages (288, 340, 891)
3014 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3015 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3016 - fix ptlrpc_connection leak in target_handle_connect (1174)
3017 - fix import refcounting bug in OST and MDS cleanup (1134)
3018 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3019 - fix very unlikely obd_types race condition (501)
3020 - remove osc_open hack for echo_client (1187)
3021 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3022 - a failure in read_inode2 leads to deadlock (1139)
3023 - cancel ack-locks as soon as transaction is committed (1072)
3024 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3025 - make sure to commitrw after any preprw to avoid deadlock (1162)
3026 - failing to execute a file in a lustre FS would lock inode (1203)
3027 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3028 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3029 - obd_uuid2tgt would walk off the end of the list (1255)
3030 - on IA64 the osc would give portals incorrect bulk size (1258)
3031 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3032 - fix lock inversion caused by new llite matching code (1282)
3033 - limit the number of dirty pages on a client to 10MB (1286)
3034 - timed out locks were not being corrected cancelled (1289)
3035 - fix O_DIRECT above 4GB on IA-32 (1292)
3036 * major user-visible changes
3037 - fail out/fail over policy now controlled by the upcall (993)
3039 - add OBD_PING to check server availability and failure (954)
3040 - lustre messages are now sent in sending host order (288, 340, 891)
3041 - add eadatalen to MDS getattr reply (340)
3042 - OST read replies may contain second buffer, with per-page status (593)
3044 2003-03-11 Phil Schwan <phil@clusterfs.com>
3047 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3048 - failed open()s could cause deadlock; fixed (867, 869)
3049 - stop cancelling OST locks when files are closed (481)
3050 - overlapping XID spaces caused network corruption (851, 853)
3051 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3052 - setattr_raw added, to do single-RPC, server-side setattrs
3053 - lmc/lconf syntax change for OST UUIDs
3054 - fix crashy race condition between ptlrpc_free_req and osc_close
3055 - don't use request in mdc_enqueue if we hit a timeout (889)
3056 - don't set the inode i_size for regular files from the MDS (896)
3057 - handle out of order completion AST (842)
3058 - don't LBUG if a lock request times out after receiving AST (913)
3059 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3060 - if a bad lock AST arrives, send an error instead of dropping entirely
3061 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3062 - fix leak in bulk IO when only partially completed (899, 900, 926)
3063 - fix O_DIRECT for ia64 (55)
3064 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3065 - C-z after timeout could hang a process forever; fixed (977)
3067 - client-side I/O cache (678, 924, 929, 941, 970)
3069 - READPAGE and SETATTRs which don't take server-side locks get
3072 2003-02-11 Phil Schwan <phil@clusterfs.com>
3075 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3076 - Fix fsfilt_extN_readpage() to read a full page of directory
3077 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3078 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3079 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3080 - Most intent operations were removed, in favour of a new RPC mode
3081 that does a single RPC to the server and bypasses most of the VFS
3082 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3083 - Aggressively cancel local locks on DLM servers
3084 - mds_reint_unlink sends EA to the client if it's the last nlink.
3085 client uses that EA to unlink OST objects.
3086 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3087 - recursive symlinks were fixed (439)
3088 - fixed NULL deref in DEBUG_REQ
3089 - filter_update_lastobjid no longer calls sync, which annoyed extN
3090 - fixed multi-client small-writes to a single file problem (445)
3091 - fixed mtime updates during file writes (607)
3092 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3093 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3094 - fixed Linux OST size reporting problem (444, 656)
3095 - OST now updates object mtime with writes or setattr (607, 619)
3096 - client verifies file size before zeroing page past EOF (445)
3097 - OST now writes last allocated objid to disk with allocation (108)
3098 - LOV on echo now works (409)
3100 - mds_reint_unlink sends a new buffer, with the EA included. this
3101 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3102 set if a regular file was being unlinked, and it was the last link
3103 - use PtlGet from the target for bulk writes (315)
3104 - OST now updates object mtime with writes or setattr (607, 619)
3105 - LDLM now has a grant-time callback to revalidate locked items, if
3107 - Many MDS operations were reorganized to combat race conditions
3109 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3110 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3112 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3115 - Fully reactivate OST imports after reconnection (512, others)
3116 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3117 - More graceful error handling for truncating on dead OST (515)
3118 - Don't error out unless we're actually accessing dead stripes (474)
3119 - Fix garbage sizes when stripes are missing (410)
3120 - LRU counters were broken, causing constant lock purge (433, 432)
3121 - garbage on read from stripes with failed OSTs (441)
3122 - mark OSCs as active before reconnecting during recovery (438)
3123 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3124 - lfind did not preserve OST order in output (443)
3125 - symlinks cause hung clients, incorrect data (439)
3126 - stop dereferencing request after dropping refcount (457)
3127 - don't LASSERT(spin_is_locked) on non-SMP (455)
3128 - fixes for many rename() bugs
3129 - fstat didn't correctly synchronize attributes (399)
3130 - server must handle lock cancellation during blocking AST prep (487)
3131 - bulk descriptors were free()d too soon (511)
3132 - fix paths in lconf, which would load incorrect modules (451, 507)
3133 - fix confusing lconf 'host not found' error message (386)
3134 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3135 - fix race condition in mdc_blocking_ast() for inode access (526)
3136 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3137 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3138 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3139 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3140 - fix lstripe in several ways (kernel side) (527)
3141 - fix request leak in ldlm_cli_enqueue (262)
3142 - incorrect OSC was marked inactive after OST failure
3143 - call mds_fs_cleanup before unmounting filesystem (524)
3144 - fix races between taking ns_lock and ldlm_lock_change_resource
3145 - fix races updating LOV export open file list
3146 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3147 - fix recovery NULL deref in ldlm_cli_cancel_unused
3148 - fix some DLM races by using new hash table for lock handles (419)
3149 - permit the client to specify desired inodes, at replay
3150 - duplicate requests when we queue them for replay reintegration
3151 - fix last_rcvd offset calculation
3152 - sync after each recovered transaction, so we always make progress
3153 - never, not always, ERESTART requests without transnos
3154 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3156 - skip replay if the MDS says that the client is already connected
3157 - don't check for a recovery-enabled export to match lctl's UUID
3158 - don't INC_USE_COUNT for phantom exports
3159 - don't crash when cleaning up phantom exports (567)
3160 - don't double-finish or set replay data for errored mdc_open requests
3161 - abort requests when they time out, so we don't get old replies
3162 - send/receive replies for AST messages again
3163 - if the client says that it doesn't have the lock, cancel it on the
3165 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3166 mark it as destroyed, it will all work out when decref is called
3167 - fix module use counts (22, 581)
3169 - ASTs now expect a reply (server cancels lock on error reply)
3171 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3174 - fix many simultaneous client startup (392)
3175 - fix dentry->d_it clobbering
3176 - credentials weren't being shipped for readdir/getattr operations
3177 - remove invalid assertions triggered during some concurrent MD
3179 - proper Lustre versions added (336, 389)
3180 - fix memory leak for create error case (398)
3181 - fix LOV locking bug that would get cli/srv out of sync
3182 - fix echo client over LOV (409)
3183 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3184 - fix double-O_EXCL intent crash (424)
3185 - avoid sending multiple lock CANCELs (352)
3187 - MDS can do multi-client recovery (modulo bugs in new code)
3189 - many updates, edits, cleanups
3191 2002-11-18 Phil Schwan <phil@clusterfs.com>
3194 - fix null d_it dereference (346)
3195 - fix full OST/dbench hang (333)
3196 - fix permission problem with file removal (286)
3197 - fix removal of OSCs from LOV when they fail
3198 - fix NULL deref during bulk timeout (214)
3199 - fix problems related to multiple filesystems on one MDS (241)
3200 - fixed serious subtle metadata locking bugs
3201 - free locks on clients when inodes are removed due to memory
3203 - fix inode pointer in lock data (285)
3204 - partial support for multiple MDS on a single host (241)
3205 - data locks weren't cancelled at clear_inode time (290, 311)
3206 - intent locks could lead to unbounded lock growth (205)
3207 - added a maximum lock count, an LRU list, and a flusher
3208 - fix multiple rename (365)
3209 - properly abstracted the echo client
3210 - OSC locked 1 byte too many; fixed
3211 - rewrote brw callback code:
3212 - fixed recovery bugs related to LOVs (306)
3213 - fixed too-many-pages-in-one-write crash (191)
3214 - fixed (again) crash in sync_io_timeout (214)
3215 - probably fixed callback-related race (385)
3217 - Add capability to MDS protocol
3218 - LDLM cancellations and callbacks on different portals
3220 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3223 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3224 - defer open object destruction to close time (601981, 138)
3225 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3226 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3227 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3228 - don't call obd_setattr in ll_file_release for destroyed objects
3229 * protocol change to lustre_msg: move |version| and add |flags|
3230 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3231 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3232 * added replay of create, unlink, link and rename operations during
3233 MDS failover; recovery should be much more robust now
3234 * remove failed OSCs from LOVs (only lov_create uses this so far)
3235 * the lustre-HOWTO was brought (more) up to date (582544)
3237 2002-10-23 Phil Schwan <phil@clusterfs.com>
3240 - in-use dentries weren't being reused properly (617851)
3241 - prevent multiple LDLM setup (599178)
3242 - fix LOV size calculations for truncate (617853)
3243 - fix client handling of MDS intent errors (POSIX)
3244 - fix permission bug in lovstripe.c test (624321)
3245 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3246 - truncate past end of file could corrupt data
3247 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3248 - a race in recovery could return ETIMEDOUT to apps (623947)
3249 - building outside the source directory was fixed
3250 * the lustre-HOWTO was brought (more) up to date (582544)
3251 * major progress was made on recovery functionality
3253 2002-10-10 Phil Schwan <phil@clusterfs.com>
3256 - recovery deadlock fix
3257 - rm -rf causes LBUG fix (617817)
3258 - file open by multiple tasks fix (618962)
3259 - directory permissions bugs (602707 and 620007)
3260 - journal_stop fixed with locking (611313)
3261 - O_APPEND failures resolved (618273, perhaps 614459)
3262 - lconf PATH fix (619770)
3263 - IA64 build fix (621450)
3264 - RPC buffer sizes scale with amount of memory
3266 2002-10-01 Phil Schwan <phil@clusterfs.com>
3269 - locks would be cancelled without throwing away data pages,
3270 resulting in inconsistent data (605627)
3271 - inode attributes were not always being refreshed (605627, 612449)
3272 - lconf now continues to cleanup after lctl reports an error
3273 - MDS now enforces user permissions (602707)
3274 - lprocfs cleanup fixed, but not yet enabled (614157)
3275 - fixed infinite server hang, should a client not respond to an AST
3276 - avoid going into recovery if user calls readlink() with a buffer
3277 that's too small (613941)
3278 - AST RPCs no longer require replies (614867) -- this may be changed
3279 - don't crash server if client sends an IOV that's too big (611336)
3280 - fixed lock conversion deadlock (611892)
3281 - fixed the following of symlinks (614622)
3282 * recovery: the server can remove locks from a client that dies, other
3283 clients can make progress
3284 * more extN patch fixes
3285 * compile-time configurable ptlrpc buffer allocations
3287 - collaborative read cache document
3288 - Lustre Lite Performance CDR document-in-progress
3290 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3293 - fix typo in patch-2.4.18
3295 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3298 - clear ptlrpc request each time in handle_incoming_request()
3299 - unlink of files now destroys the object on the OST
3301 2002-09-19 Peter Braam <braam@clusterfs.com>
3303 * add hard link support
3304 * change obdfile creation method
3305 * kernel patch changed
3307 2002-09-19 Peter Braam <braam@clusterfs.com>
3310 - stack overflow bug in extN fixed
3312 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3314 * documentation updates
3315 - add man pages for config tools
3316 - update tests/README to describe testing with new config tools
3317 - finish metadata API descriptions
3318 * bug fixes and cleanups
3319 - statfs workaround for 16TB limit
3320 - LOV stripe allocation improved, can stripe on subset of OSTs
3321 - LOV file size/IO offset was wrong for files > 4GB in size
3322 - object EA data was being dropped, caused files to be unreadable
3323 - memory overflow with non-LOV OST caused memory corruption
3324 - fixed regression tests to work with new config tools, obdfilter
3325 - fixed bug when directory size became larger than 1 block
3326 - fixed bug (for single client case) when PWD was deleted
3327 - invalidate local directory pages when doing intent-based ops
3328 - avoid LDLM oops when lock callback contained bad data
3330 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3332 * documentation updates
3333 * bug fixes and cleanups
3334 - configuration tools
3337 - 64-bit compile warnings
3338 - 64-bit internal statfs data
3340 * test_brw on persistent OST devices
3342 * lprocfs (disabled)
3344 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3346 * documentation updates
3347 * bug fixes and cleanups
3348 * configuration tools
3350 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3353 * many small fixes to 0_5_4
3354 * io/network handling
3355 * thinkos in MDS operations
3357 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3360 * crucial basic fixes to 0.5.3
3361 * IOR, Iozone work over Elan
3364 2002-08-07 Phil Schwan <phil@clusterfs.com>
3365 * version 0_5_3, our first alpha
3366 * we use the new Portals iovs
3367 * documentation updates
3368 * bug fixes and cleanups
3369 * small changes in the DLM wire protocol
3371 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3372 * version 0_5_1 with some initial stability,
3373 * locking on MD and file I/O.
3374 * documentation updates
3375 * several bug fixes since 0.5.0
3376 * small changes in wire protocol
3378 2002-07-18 Phil Schwan <phil@clusterfs.com>
3380 * delivered as Lustre Light Alpha
3381 * fixed a crash after handling invalid MDS requests
3382 * fixed directory pages for architectures with non-4k pages sizes
3384 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3385 * release version v0_4_4
3386 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3388 * Fixes a number of interruption problems with OST operations.
3389 * Update documentation for portals header changes
3390 * Move all wire protocol structs/defines to lustre_idl.h
3391 * Fixes symlink length bug.
3392 * Add tcpdump to repository.
3394 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3395 * release version v0_4_3
3396 * Fixes statfs for inodes on extN.
3397 * Fixes bug in runtests which would delete /etc/hosts.
3398 * Use 64-bit object IDs wherever possible (not into VFS though)
3399 Remove ost_get_info, which is unused by lustre, and out of date.
3401 2002-07-03 Peter Braam <braam@clusterfs.com>
3402 * release version v0_4_2 Fixes a lookup error (type not passed)
3403 * move forward to head of Portals
3404 * move forward to latest Lustre kernel
3406 2002-06-25 Peter Braam <braam@clusterfs.com>
3407 * release version v0_4_1. Hopefully stable on single node use.