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
51 --------------------------------------------------------------------------------
53 tbd Cluster File Systems, Inc. <info@clusterfs.com>
55 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
56 INTEROPERATE with older versions automatically. In many cases a
57 special upgrade step is needed. Please read the
58 user documentation before upgrading any part of a 1.4.x system.
59 * WARNING: Lustre configuration and startup changes are required with
60 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
62 * Support for kernels:
63 2.6.9-42.0.10.EL (RHEL 4)
65 2.4.21-47.0.1.EL (RHEL 3)
66 2.6.12.6 vanilla (kernel.org)
67 2.6.16.27-0.9 (SLES 10)
68 * Client support for unpatched kernels:
69 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
70 2.6.16 - 2.6.19 vanilla (kernel.org)
71 2.6.9-42.0.8EL (RHEL 4)
72 * Recommended e2fsprogs version: 1.39.cfs5
74 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
77 Severity : enhancement
79 Description: MountConf
80 Details : Lustre configuration is now managed via mkfs and mount
81 commands instead of lmc and lconf. New obd types (MGS, MGC)
82 are added for dynamic configuration management. See
83 https://mail.clusterfs.com/wikis/lustre/MountConf for
86 Severity : enhancement
88 Description: dynamic OST addition
89 Details : OSTs can now be added to a live filesystem
91 Severity : enhancement
93 Description: startup order invariance
94 Details : MDTs and OSTs can be started in any order. Clients only
95 require the MDT to complete startup.
97 Severity : enhancement
99 Description: parallel, asynchronous orphan cleanup
100 Details : orphan cleanup is now performed in separate threads for each
101 OST, allowing parallel non-blocking operation.
103 Severity : enhancement
105 Description: optimized stripe assignment
106 Details : stripe assignments are now made based on ost space available,
107 ost previous usage, and OSS previous usage, in order to try
108 to optimize storage space and networking resources.
110 Severity : enhancement
112 Description: Permanently set tunables
113 Details : All writable /proc/fs/lustre tunables can now be permanently
114 set on a per-server basis, at mkfs time or on a live
117 Severity : enhancement
119 Description: Lustre message v2
120 Details : Add lustre message format v2.
122 Severity : enhancement
124 Description: client OST exclusion list
125 Details : Clients can be started with a list of OSTs that should be
126 declared "inactive" for known non-responsive OSTs.
129 Frequency : SFS test only (otherwise harmless)
131 Description: SPEC SFS validation failure on NFS v2 over lustre.
132 Details : Changes the blocksize for regular files to be 2x RPC size,
133 and not depend on stripe size.
135 Severity : enhancement
137 Description: Multiple MD RPCs in flight.
138 Details : Further unserialise some read-only MDT RPCs - learn about intents.
139 To avoid overly-overloading MDT, introduce a limit on number of
140 MDT RPCs in flight for a single client and add /proc controls
141 to adjust this limit.
143 Severity : enhancement
145 Description: client read/write statistics
146 Details : Add client read/write call usage stats for performance
147 analysis of user processes.
148 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
149 file access. extents_stats shows chunk size distribution.
150 extents_stats_per_process show chunk size distribution per
153 Severity : enhancement
155 Description: per-client statistics on server
156 Details : Add ldlm and operations statistics for each client in
157 /proc/fs/lustre/mds|obdfilter/*/exports/
159 Severity : enhancement
161 Description: improved MDT statistics
162 Details : Add detailed MDT operations statistics in
163 /proc/fs/lustre/mds/*/stats
165 Severity : enhancement
167 Description: VFS operations stats
168 Details : Add client VFS call stats, trackable by pid, ppid, or gid
169 /proc/fs/lustre/llite/*/vfs_ops_stats
170 /proc/fs/lustre/llite/*/vfs_track_[pid|ppid|gid]
175 Description: Fix client-side osc byte counters
176 Details : The osc read/write byte counters in
177 /proc/fs/lustre/osc/*/stats are now working
180 Frequency : always as root on SLES
182 Description: Failure of copying files with lustre special EAs.
183 Details : Client side always return success for setxattr call for lustre
184 special xattr (currently only "trusted.lov").
189 Description: Refcount LNET uuids
190 Details : The global LNET uuid list grew linearly with every startup;
191 refcount repeated list entries instead of always adding to
194 Severity : enhancement
196 Description: Dynamic service threads
197 Details : Within a small range, start extra service threads
198 automatically when the request queue builds up.
201 Frequency : mixed-endian client/server environments
203 Description: mixed-endian crashes
204 Details : The new msg_v2 system had some failures in mixed-endian
207 Severity : enhancement
209 Description: Easy OST removal
210 Details : OSTs can be permanently deactivated with e.g. 'lctl
211 conf_param lustre-OST0001.osc.active=0'
213 Severity : enhancement
215 Description: MGS proc entries
216 Details : Added basic proc entries for the MGS showing what filesystems
219 Severity : enhancement
221 Description: provide MGS failover
222 Details : Added config lock reacquisition after MGS server failover.
224 Severity : enhancement
226 Description: add Linux 2.4 support
227 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
231 Description: a large application tries to do I/O to the same resource and dies
233 Details : Check the req->rq_arrival time after the call to
234 ost_brw_lock_get(), but before we do anything about
235 processing it & sending the BULK transfer request. This
236 should help move old stale pending locks off the queue as
237 quickly as obd_timeout.
240 Frequency : when an incorrect nid is specified during startup
242 Description: ptlrpc connect to non-existant node causes kernel crash
243 Details : LNET can't be re-entered from an event callback, which
244 happened when we expire a message after the export has been
245 cleaned up. Instead, hand the zombie cleanup off to another
248 Severity : enhancement
250 Description: plain/inodebits lock performance improvement
251 Details : Grouping plain/inodebits in granted list by their request modes
252 and bits policy, thus improving the performance of search through
256 Frequency : only for kernels with patches from Lustre below 1.4.3
258 Description: Remove old rdonly API
259 Details : Remove old rdonly API which unsed from at least lustre 1.4.3
262 Frequency : only for devices with external journals
264 Description: Set external device read-only also
265 Details : During a commanded failover stop, we set the disk device
266 read-only while the server shuts down. We now also set any
267 external journal device read-only at the same time.
270 Frequency : when upgrading from 1.4 while trying to change parameters
272 Description: The wrong (new) MDC name was used when setting parameters for
273 upgraded MDT's. Also allows changing of OSC (and MDC)
274 parameters if --writeconf is specified at tunefs upgrade time.
277 Frequency : when setting specific ost indicies
279 Description: QOS code breaks on skipped indicies
280 Details : Add checks for missing OST indicies in the QOS code, so OSTs
281 created with --index need not be sequential.
283 Severity : enhancement
285 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
286 Details : The uninit_groups feature works in conjunction with the kernel
287 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
288 up the pass1 processing of e2fsck. This is a read-only feature
289 in ldiskfs2 only, so older kernels and current ldiskfs cannot
290 mount filesystems that have had this feature enabled.
292 Severity : enhancement
294 Description: Improve multi-block allocation algorithm to avoid fragmentation
295 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
296 allocation locality and avoid filesystem fragmentation.
298 ------------------------------------------------------------------------------
300 TBD Cluster File Systems, Inc. <info@clusterfs.com>
302 * Support for kernels:
303 2.4.21-47.0.1.EL (RHEL 3)
305 2.6.9-42.0.10.EL (RHEL 4)
306 2.6.12.6 vanilla (kernel.org)
307 2.6.16.27-0.9 (SLES 10)
308 * Recommended e2fsprogs version: 1.39.cfs6
309 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
312 ------------------------------------------------------------------------------
314 2007-04-30 Cluster File Systems, Inc. <info@clusterfs.com>
316 * Support for kernels:
317 2.4.21-47.0.1.EL (RHEL 3)
319 2.6.9-42.0.10.EL (RHEL 4)
320 2.6.12.6 vanilla (kernel.org)
321 2.6.16.27-0.9 (SLES 10)
322 * Recommended e2fsprogs version: 1.39.cfs6
323 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
327 Frequency : occasional, depends on client load and configuration
328 Bugzilla : 12181, 12203
329 Description: data loss for recently-modified files
331 Details : In some cases it is possible that recently written or created
332 files may not be written to disk in a timely manner (this should
333 normally be within 30s unless client IO load is very high).
334 The problem appears as zero-length files or files that are a
335 multiple of 1MB in size after a client crash or client eviction
336 that are missing data at the end of the file.
338 This problem is more likely to be hit on clients where files are
339 repeatedly created and unlinked in the same directory, clients
340 have a large amount of RAM, have many CPUs, the filesystem has
341 many OSTs, the clients are rebooted frequently, and/or the files
342 are not accessed by other nodes after being written.
344 The presence of the problem can be detected by looking at
345 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
346 smaller than the second (nr_unused) then dirty files will not
347 be flushed automatically to disk. "sync; sleep 10" should be
348 run several times on the node before unmounting it to update
349 Lustre (this is also safe to run on nodes without this problem).
351 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
352 that can cause this same problem, so customers using that kernel
353 also need to update the kernel in addition to Lustre. In order
354 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
356 It is normal that files written just before a client crash (less
357 than 30s) may not yet have been flushed to disk, even for local
361 Frequency : frequent on thin XT3 nodes
363 Description: UUID collision on thin XT3 Linux nodes
364 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
365 randomly, since we relied on an insufficiently-seeded PRNG.
370 Description: OSS hangs after "All ost request buffers busy"
371 Details : A deadlock between quota and journal operations caused OSS
372 hangs after printing "All ost request buffers busy."
375 Frequency : always on liblustre builds
377 Description: Cleanup compiler warnings on liblustre
379 Severity : enhancement
381 Description: Move random uuid functions to prng.c.
383 ------------------------------------------------------------------------------
385 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
387 * Support for kernels:
388 2.6.16.21-0.8 (SLES10)
389 2.6.9-42.0.8EL (RHEL 4)
391 2.4.21-47.0.1.EL (RHEL 3)
392 2.6.12.6 vanilla (kernel.org)
393 * Recommended e2fsprogs version: 1.39.cfs5
396 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
398 Description: System hang on startup
399 Details : This bug allowed the liblustre (e.g. catamount) client to
400 return to the app before handling all startup RPCs. This
401 could leave the node unresponsive to lustre network traffic
402 and manifested as a server ptllnd timeout.
404 Severity : enhancement
406 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
407 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
408 boolean to control whether to print peer debug info when a
409 client's RPC times out.
414 Description: make O_SYNC working on 2.6 kernels
415 Details : 2.6 kernels use different method for mark pages for write,
416 so need add a code to lustre for O_SYNC work.
421 Description: Failure to close file and release space on NFS
422 Details : Put inode details into lock acquired in ll_intent_file_open.
423 Use mdc_intent_lock in ll_intent_open to properly
424 detect all kind of errors unhandled by mdc_enqueue
429 Description: proc file read during shutdown sometimes raced obd removal,
431 Details : Add lock to prevent obd access after proc file removal.
434 Frequency : Only for files larger than 4GB on 32-bit clients.
436 Description: improperly doing page alignment of locks
437 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
438 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
443 Description: RPCs being resent when they shouldn't be
444 Details : Some RPCs that should not be resent are being resent. This
445 can cause inconsistencies in the RPC state machine. Do not
446 resend such requests.
449 Frequency : rare, only with NFS export
451 Description: Crash on NFS re-export node
452 Details : under very unusual load conditions an assertion is hit in
453 ll_intent_file_open()
456 Frequency : only if OST filesystem is corrupted
458 Description: client incorrectly hits assertion in ptlrpc_replay_req()
459 Details : for a short time RPCs with bulk IO are in the replay list,
460 but replay of bulk IOs is unimplemented. If the OST filesystem
461 is corrupted due to disk cache incoherency and then replay is
462 started it is possible to trip an assertion. Avoid putting
463 committed RPCs into the replay list at all to avoid this issue.
468 Description: large O_DIRECT requests fail under memory pressure/fragmentation
469 Details : Large single O_DIRECT read and write calls can fail to allocate
470 a sufficiently large buffer to process the request. In case of
471 allocation failure the allocation is retried with a smaller
472 buffer and broken into smaller requests.
477 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
478 > 32000 subdirectories
479 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
480 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
481 subdirectory count crosses 32000. This will aid e2fsck to
482 correctly handle more than 32000 subdirectories.
487 Description: versioning check is incomplete
488 Details : Checking the version difference of client vs. server, report
489 error if the gap is too big.
494 Description: add support PG_writeback bit
495 Details : add support for PG_writeback bit for Lustre, for more carefull
496 work with page cache in 2.6 kernel. This also fix some deadlocks
497 and remove hack for work O_SYNC with 2.6 kernel.
499 ------------------------------------------------------------------------------
501 2006-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
503 * Support for kernels:
504 2.6.16.21-0.8 (SLES10)
505 2.6.9-42.0.3EL (RHEL 4)
507 2.4.21-47.0.1.EL (RHEL 3)
508 2.6.12.6 vanilla (kernel.org)
511 * The backwards-compatible /proc/sys/portals symlink has been removed
512 in this release. Before upgrading, please ensure that you change
513 any configuration scripts or /etc/sysctl.conf files that access
514 /proc/sys/portals/* or sysctl portals.* to use the corresponding
515 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
516 in advance of the upgrade on any system running Lustre 1.4.6 or
517 newer, since /proc/sys/lnet was added in that version.
518 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
522 Frequency : MDS failover only, very rarely
524 Description: "went back in time" messages on mds failover
525 Details : The greatest transno may be lost when the current operation
526 finishes with an error (transno==0) and the client's last_rcvd
527 record is over-written. Save the greatest transno in the
528 mds_last_transno for this case.
531 Frequency : always for specific kernels and striping counts
533 Description: client may get "Matching packet too big" without ACL support
534 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
535 when trying to access files in certain configurations. The
536 clients should in fact be denied when mounting because they do
540 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
542 Description: many clients connecting with IO in progress causes connect timeouts
543 Details : Avoid synchronous journal commits to avoid delays caused by many
544 clients connecting/disconnecting when bulk IO is in progress.
545 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
546 OST_IO_PORTAL to avoid delays behind potentially many pending
550 Frequency : occasionally with multiple writers to a single file
552 Description: shared writes to file may result in wrong size reported by stat()
553 Details : Allow growing of kms when extent lock is cancelled
556 Frequency : always with random mmap IO to multi-striped file
558 Description: mmap write might be lost if we are writing to a 'hole' in stripe
559 Details : Only if the hole is at the end of OST object so that kms is too
560 small. Fix is to increase kms accordingly in ll_nopage.
563 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
565 Description: writes to a missing object would leak memory on the OST
566 Details : If there is an inconsistency between the MDS and OST filesystems,
567 such that the MDS references an object that doesn't exist, writes
568 to that object will leak memory due to incorrect cleanup in the
569 error handling path, eventually running out of memory on the OST.
574 Description: Creating too long symlink causes lustre errors
575 Details : Check symlink and name lengths before sending requests to MDS.
578 Frequency : only if flock is enabled (not on by default)
580 Description: posix locks not released on fd closure on 2.6.9+
581 Details : We failed to add posix locks to list of inode locks on 2.6.9+
582 kernels, this caused such locks not to be released on fd close and
583 then assertions on fs unmount about still used locks.
586 Frequency : MDS failover only, very rarely
588 Description: clients may get ASSERTION(granted_lock != NULL)
589 Details : When request was taking a long time, and a client was resending
590 a getattr by name lock request. The were multiple lock requests
591 with the same client lock handle and
592 mds_getattr_name->fixup_handle_for_resent_request found one of the
593 lock handles but later failed with ASSERTION(granted_lock != NULL).
598 Description: handle->h_buffer_credits > 0, assertion failure
599 Details : h_buffer_credits is zero after truncate, causing assertion
600 failure. This patch extends the transaction or creates a new
604 Frequency : NFS re-export or patchless client
605 Bugzilla : 11179, 10796
606 Description: Crash on NFS re-export node (__d_move)
607 Details : We do not want to hash the dentry if we don't have a lock.
608 But if this dentry is later used in d_move, we'd hit uninitialised
609 list head d_hash, so we just do this to init d_hash field but
610 leave dentry unhashed.
613 Frequency : NFS re-export or patchless client
615 Description: NFS exports has problem with symbolic link
616 Details : lustre client didn't properly install dentry when re-exported
617 to NFS or running patchless client.
620 Frequency : NFS re-export or patchless client
622 Description: Various nfs/patchless fixes.
623 Details : fixes reuse disconected alias for lookup process - this fixes
624 warning "find_exported_dentry: npd != pd",
625 fix permission error with open files at nfs.
626 fix apply umaks when do revalidate.
629 Frequency : occasional
631 Description: Crash on NFS re-export node
632 Details : calling clear_page() on the wrong pointer triggered oops in
633 generic_mapping_read().
636 Frequency : rarely, using O_DIRECT IO
638 Description: unaligned directio crashes client with LASSERT
639 Details : check for unaligned buffers before trying any requests.
642 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
644 Description: stale data returned from RAID cache
645 Details : If only a small amount of IO is done to the RAID device before
646 reading it again it is possible to get stale data from the RAID
647 cache instead of reading it from disk.
650 Frequency : depends on arch, kernel and compiler version, always on sles10
653 Description: recursive or deep enough symlinks cause stack overflow
654 Details : getting rid of large stack-allocated variable in
658 Frequency : depends on hardware
660 Description: lustre write performance loss in the SLES10 kernel
661 Details : the performance loss is caused by using of write barriers in the
662 ext3 code. The SLES10 kernel turns barrier support on by
663 default. The fix is to undo that change for ldiskfs.
665 ------------------------------------------------------------------------------
667 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
669 * Support for kernels:
670 2.6.9-42.0.3EL (RHEL 4)
672 2.4.21-47.0.1.EL (RHEL 3)
673 2.6.12.6 vanilla (kernel.org)
677 Frequency : quota enabled and large files being deleted
679 Description: releasing more than 4GB of quota at once hangs OST
680 Details : If a user deletes more than 4GB of files on a single OST it
681 will cause the OST to spin in an infinite loop. Release
682 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
687 Description: statfs data retrieved from /proc may be stale or zero
688 Details : When reading per-device statfs data from /proc, in the
689 {kbytes,files}_{total,free,avail} files, it may appear
690 as zero or be out of date.
693 Frequency : systems with MD RAID1 external journal devices
695 Description: lconf's call to blkid is confused by RAID1 journal devices
696 Details : Use the "blkid -l" flag to locate the MD RAID device instead
697 of returning all block devices that match the journal UUID.
700 Frequency : always, for aggregate stripe size over 4GB
702 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
703 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
704 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
707 Frequency : always if "lfs find" used on a local file/directory
709 Description: "lfs find" segfaults if used on a local file/directory
710 Details : The case where a directory component was not specified wasn't
711 handled correctly. Handle this properly.
714 Frequency : always on ppc64
716 Description: the write to an ext3 filesystem mounted with mballoc got stuck
717 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
718 perform endianness conversion.
721 Frequency : rarely (truncate to non-zero file size after write under load)
722 Bugzilla : 10730, 10687
723 Description: Files padded with zeros to next 4K multiple
724 Details : With filesystems mounted using the "extents" option (2.6 kernels)
725 it is possible that files that are truncated to a non-zero size
726 immediately after being written are filled with zero bytes beyond
727 the truncated size. No file data is lost.
729 Severity : enhancement
731 Description: Allow recovery/failover for liblustre clients.
732 Details : liblustre clients were unaware of failover configurations until
735 Severity : enhancement
737 Description: user file locks should fail when not mounting with flock option
738 Details : Set up an error-returning stub in ll_file_operations.lock field
739 to prevent incorrect behaviour when client is mounted without
740 flock option. Also, set up properly f_op->flock field for
744 Frequency : always on ia64
746 Description: "lfs df" loops on printing out MDS statfs information
747 Details : The obd_ioctl_data was not initialized and in some systems
748 this caused a failure during the ioctl that did not return
749 an error. Initialize the struct and return an error on failure.
752 Frequency : SLES 9 only
754 Description: Error of copying files with lustre special EAs as root
755 Details : Client side always return success for setxattr call for lustre
756 special xattr (currently only "trusted.lov").
759 Frequency : rarely on clusters with both ia64+i386 clients
761 Description: ia64+i686 clients doing shared IO on the same file may LBUG
762 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
763 clients are doing concurrent writes to the same file it is
764 possible that the ia64 clients may LASSERT because the OST
765 extent locks are not PAGE_SIZE aligned. Ensure that grown
766 locks are always aligned on the request boundary.
769 Frequency : specific use, occasional
771 Description: Overwriting in use executable truncates on-disk binary image
772 Details : If one node attempts to overwrite an executable in use by
773 another node, we now correctly return ETXTBSY instead of
779 Description: chmod on Lustre root is propagated to other clients
780 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
781 invalid by the follow_mount time.
784 Frequency : liblustre clients only
786 Description: Race in 'instant cancel' lock handling could lead to such locks
787 never to be granted in case of SMP MDS
788 Details : Do not destroy not yet granted but cbpending locks in
792 Frequency : replay/resend of open
794 Description: non null lock assetion failure in mds_intent_policy
795 Details : Trying to replay/resend lockless open requests resulted in
796 mds_open() returning 0 with no lock. Now it sets a flag if
797 a lock is going to be returned.
799 Severity : enhancement
801 Description: Checksum enhancements
802 Details : New checksum enhancements allow for resending RPCs that failed
805 Severity : enhancement
807 Description: Tunables on number of dirty pages in cacche
808 Details : Allow to set limit on number of dirty pages cached.
813 Description: client crash on unmount - lock still has references
814 Details : In some error handling cases it was possible to leak a lock
815 reference on a client while accessing a file. This was not
816 harmful to the client during operation, but would cause the
817 client to crash when the filesystem is unmounted.
820 Frequency : specific case, rare
822 Description: ETXTBSY on mds though file not in use by client
823 Details : ETXTBSY is no longer incorrectly returned when attempting to
824 chmod or chown a directory that the user previously tried to
825 execute or a currently-executing binary.
828 Frequency : extremely rare except on liblustre-based clients
830 Description: Lustre space not freed when files are deleted
831 Details : Clean up open-unlinked files after client eviction. Previously
832 the unlink was skipped and the files remained as orphans.
837 Description: OST failure "would be an LBUG" in waiting_locks_callback()
838 Details : In some cases it was possible to send a blocking callback to
839 a client doing a glimpse, even though that client didn't get
840 a lock granted. When the glimpse lock is cancelled on the OST
841 the freed lock is left on the waiting list and corrupted the list.
844 Frequency : all core dumps
846 Description: Broke core dumps to lustre
847 Details : Negative dentry may be unhashed if parent does not have UPDATE
848 lock, but some callers, e.g. do_coredump, expect dentry to be
849 hashed after successful create, hash it in ll_create_it.
851 ------------------------------------------------------------------------------
853 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
855 * Support for kernels:
856 2.6.9-42.0.2.EL (RHEL 4)
858 2.4.21-40.EL (RHEL 3)
859 2.6.12.6 vanilla (kernel.org)
863 Frequency : always on RHEL 3
865 Description: Number of open files grows over time
866 Details : The number of open files grows over time, whether or not
867 Lustre is started. This was due to a filp leak introduced
868 by one of our kernel patches.
870 ------------------------------------------------------------------------------
872 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
874 * Support for kernels:
877 2.4.21-40.EL (RHEL 3)
878 2.6.12.6 vanilla (kernel.org)
883 Bugzilla : 5719, 9635, 9792, 9684
884 Description: OST (or MDS) trips assertions in (re)connection under heavy load
885 Details : If a server is under heavy load and cannot reply to new
886 connection requests before the client resends the (re)connect,
887 the connection handling code can behave badly if two service
888 threads are concurrently handing separate (re)connections from
889 the same client. Add better locking to the connection handling
890 code, and ensure that only a single connection will be processed
891 for a given client UUID, even if the lock is dropped.
893 Severity : enhancement
895 Description: add TCP zero-copy support to kernel
896 Details : Add support to the kernel TCP stack to allow zero-copy bulk
897 sends if the hardware supports scatter-gather and checksumming.
898 This allows socklnd to do client-write and server-read more
899 efficiently and reduce CPU utilization from skbuf copying.
902 Frequency : only if NFS exporting from client
904 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
905 Details : If setting attributes on a file created under NFS that had
906 never been opened it would be possible to oops the client
907 if the file had no objects.
911 Bugzilla : 9326, 10402, 10897
912 Description: client crash in ptlrpcd_wake() thread when sending async RPC
913 Details : It is possible that ptlrpcd_wake() dereferences a freed async
914 RPC. In rare cases the ptlrpcd thread alread processed the RPC
915 before ptlrpcd_wake() was called and the request was freed.
918 Frequency : always for liblustre
920 Description: liblustre client does MDS+OSTs setattr RPC for each write
921 Details : When doing a write from a liblustre client, the client
922 incorrectly issued an RPC to the MDS and each OST the file was
923 striped over in order to update the timestamps. When writing
924 with small chunks and many clients this could overwhelm the MDS
925 with RPCs. In all cases it would slow down the write because
926 these RPCs are unnecessary.
928 Severity : enhancement
930 Description: allow number of MDS service threads to be changed at module load
931 Details : It is now possible to change the number of MDS service threads
932 running. Adding "options mds mds_num_threads={N}" to the MDS's
933 /etc/modprobe.conf will set the number of threads for the next
934 time Lustre is restarted (assuming the "mds" module is also
935 reloaded at that time). The default number of threads will
936 stay the same, 32 for most systems.
941 Description: OST crash if filesystem is unformatted or corrupt
942 Details : If an OST is started on a device that has never been formatted
943 or if the filesystem is corrupt and cannot even mount then the
944 error handling cleanup routines would dereference a NULL pointer.
949 Description: NULL pointer deref in llap_from_page.
950 Details : get_cache_page_nowait can return a page with NULL (or otherwise
951 incorrect) mapping if the page was truncated/reclaimed while it was
952 searched for. Check for this condition and skip such pages when
953 doing readahead. Introduce extra check to llap_from_page() to
954 verify page->mapping->host is non-NULL (so page is not anonymous).
957 Frequency : Sometimes when using sys_sendfile
959 Description: "page not covered by a lock" warnings from ll_readpage
960 Details : sendfile called ll_readpage without right page locks present.
961 Now we introduced ll_file_sendfile that does necessary locking
962 around call to generic_file_sendfile() much like we do in
966 Frequency : with certain MDS communication failures at client mount time
968 Description: NULL pointer deref after failed client mount
969 Details : a client connection request may delayed by the network layer
970 and not be sent until after the PTLRPC layer has timed out the
971 request. If the client fails the mount immediately it will try
972 to clean up before the network times out the request. Add a
973 reference from the request import to the obd device and delay
974 the cleanup until the network drops the request.
977 Frequency : occasionally during client (re)connect
979 Description: assertion failure during client (re)connect
980 Details : processing a client connection request may be delayed by the
981 client or server longer than the client connect timeout. This
982 causes the client to resend the connection request. If the
983 original connection request is replied in this interval, the
984 client may trip an assertion failure in ptlrpc_connect_interpret()
985 which thought it would be the only running connect process.
988 Frequency : only with obd_echo servers and clients that are rebooted
990 Description: kernel BUG accessing uninitialized data structure
991 Details : When running an obd_echo server it did not start the ping_evictor
992 thread, and when a client was evicted an uninitialized data
993 structure was accessed. Start the ping_evictor in the RPC
994 service startup instead of the OBD startup.
996 Severity : enhancement
997 Bugzilla : 10193 (patchless)
998 Description: Remove dependency on various unexported kernel interfaces.
999 Details : No longer need reparent_to_init, exit_mm, exit_files,
1000 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1003 Frequency : rare (only users of deprecated and unsupported LDAP config)
1005 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1006 Details : LDAP apparently contains 'lustreName' attributes instead of
1007 'name'. A simple remapping of the name is sufficient.
1010 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1012 Description: waiting_locks_callback trips kernel BUG if client is evicted
1013 Details : Running with the dump_on_timeout debug flag turned on makes
1014 it possible that the waiting_locks_callback() can try to dump
1015 the Lustre kernel debug logs from an interrupt handler. Defer
1016 this log dumping to the expired_lock_main() thread.
1018 Severity : enhancement
1020 Description: Support NFS exporting on 2.6 kernels.
1021 Details : Implement non-rawops metadata methods for NFS server to use without
1022 changing NFS server code.
1025 Frequency : very rare (synthetic metadata workload only)
1027 Description: two racing renames might cause an MDS thread to deadlock
1028 Details : Running the "racer" program may cause one MDS thread to rename
1029 a file from being the source of a rename to being the target of
1030 a rename at exactly the same time that another thread is doing
1031 so, and the second thread has already enqueued these locks after
1032 doing a lookup of the target and is trying to relock them in
1033 order. Ensure that we don't try to re-lock the same resource.
1036 Frequency : only very large systems with liblustre clients
1038 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1039 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1040 all OSTs in parallel. This allows the eviction of stale liblustre
1041 clients to proceed much faster than if they were done in series,
1042 and also offers similar improvements for other set_info RPCs.
1047 Description: excessive CPU usage during initial read phase on client
1048 Details : During the initial read phase on a client, it would agressively
1049 retry readahead on the file, consuming too much CPU and impacting
1050 performance (since 1.4.5.8). Improve the readahead algorithm
1051 to avoid this, and also improve some other common cases (read
1052 of small files in particular, where "small" is files smaller than
1053 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1058 Description: MDS crash when receiving packet with unknown intent.
1059 Details : Do not LBUG in unknown intent case, just return -EFAULT
1061 Severity : enhancement
1062 Bugzilla : 9293, 9385
1063 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1064 Details : Do not serialize getattr (non-intent version) and statfs.
1067 Frequency : occasional, when OST network is overloaded/intermittent
1069 Description: client evicted by OST after bulk IO timeout
1070 Details : If a client sends a bulk IO request (read or write) the OST
1071 may evict the client if it is unresposive to its data GET/PUT
1072 request. This is incorrect if the network is overloaded (takes
1073 too long to transfer the RPC data) or dropped the OST GET/PUT
1074 request. There is no need to evict the client at all, since
1075 the pinger and/or lock callbacks will handle this, and the
1076 client can restart the bulk request.
1079 Frequency : Always when mmapping file with no objects
1081 Description: client crashes when mmapping file with no objects
1082 Details : Check that we actually have objects in a file before doing any
1083 operations on objects in ll_vm_open, ll_vm_close and
1089 Description: Request leak when working with deleted CWD
1090 Details : Introduce advanced request refcount tracking for requests
1091 referenced from lustre intent.
1093 Severity : Enhancement
1095 Description: Cache open file handles on client.
1096 Details : MDS now will return special lock along with openhandle, if
1097 requested and client is allowed to hold openhandle, even if unused,
1098 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1099 closing files for every read/write openration.
1101 Severity : Enhancement
1103 Description: Cache open negative dentries on client when possible.
1104 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1105 negative dentries on lock revocation.
1110 Description: Remounting a client read-only wasn't possible with a zconf mount
1111 Details : It wasn't possible to remount a client read-only with llmount.
1113 Severity : enhancement
1114 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1115 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1116 MPICH is included to add Lustre-specific ADIO interfaces.
1117 This is based closely on the UFS ADIO layer and only differs
1118 in file creation, in order to allow the OST striping to be set.
1119 This is user-contributed code and not supported by CFS.
1124 Description: extended inode attributes (immutable, append-only) work improperly
1125 when 2.4 and 2.6 kernels are used on client/server or vice versa
1126 Details : Introduce kernel-independent values for these flags.
1128 Severity : enhancement
1131 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1132 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1133 granularity. Now, Fractional MB granularity is be supported,
1134 this is very useful for low memory system.
1136 Severity : enhancement
1138 Description: Getattr by fid
1139 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1140 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1141 have locks and avoids getattr rpc requests for it on every stat.
1144 Frequency : Always, for filesystems larger than 2TB
1146 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1147 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1148 filesystems larger than 2TB to be mounted with mballoc enabled.
1151 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1153 Description: filesystem corruption for non-standard kernels and very large OSTs
1154 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1155 filesystems larger than 2TB is mounted then the kernel will
1156 silently corrupt the start of the filesystem. CONFIG_LBD is
1157 enabled for all CFS-supported kernels, but the possibility of
1158 this happening with a modified kernel config exists.
1160 Severity : enhancement
1162 Description: add client O_DIRECT support for 2.6 kernels
1163 Details : It is now possible to do O_DIRECT reads and writes to files
1164 in the Lustre client mountpoint on 2.6 kernel clients.
1166 Severity : enhancement
1168 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1169 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1170 parallel, not waiting for response from every OST before sending
1171 a rpc to the next OST.
1176 Description: setattr vs write race when updating file timestamps
1177 Details : Client processes that update a file timestamp into the past
1178 right after writing to the file (e.g. tar) it is possible that
1179 the updated file modification time can be reset to the current
1180 time due to a race between processing the setattr and write RPC.
1182 Severity : enhancement
1184 Description: Bring 'lfs find' closer in line with regular Linux find.
1185 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1186 -print0 options and obtains all the needed info through the lustre
1189 Severity : enhancement
1191 Description: support up to 1024 configured devices on one node
1192 Details : change obd_dev array from statically allocated to dynamically
1193 allocated structs as they are first used to reduce memory usage
1198 Description: Flush dirty partially truncated pages during truncate
1199 Details : Immediatelly flush partially truncated pages in filter_setattr,
1200 this way we completely avoid having any pages in page cache on OST
1201 and can retire ugly workarounds during writes to flush such pages.
1206 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1207 Details : protect inode from truncation within vfs_unlink() context
1208 just take a reference before calling vfs_unlink() and release it
1209 when parent's i_sem is free.
1212 Frequency : always, if extents are used on OSTs
1214 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1215 Details : OSTs using the extents format would not zero the high 16 bits of
1216 the index physical block number. This is not a problem for any
1217 OST filesystems smaller than 16TB, and no kernels support ext3
1218 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1219 new/modified files) and can be fixed for existing filesystems
1220 with e2fsprogs-1.39-cfs1.
1225 Description: import connection selection may be incorrect if timer wraps
1226 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1227 import connections to be ignored if the 32-bit jiffies counter
1228 wraps. Use a 64-bit jiffies counter.
1231 Frequency : during server recovery
1233 Description: crash after server is denying duplicate export
1234 Details : If clients are resending connect requests to the server, the
1235 server refuses to allow a client to connect multiple times.
1236 Fixed a bug in the handling of this case.
1239 Frequency : very large clusters immediately after boot
1241 Description: LNET request buffers exhausted under heavy short-term load
1242 Details : If a large number of client requests are generated on a service
1243 that has previously never seen so many requests it is possible
1244 that the request buffer growth cannot keep up with the spike in
1245 demand. Instead of dropping incoming requests, they are held in
1246 the LND until the RPC service can accept more requests.
1249 Frequency : Sometimes during replay
1251 Description: Assertion failure in ll_local_open after replay.
1252 Details : If replay happened on an open request reply before we were able
1253 to set replay handler, reply will become not swabbed tripping the
1254 assertion in ll_local_open. Now we set the handler right after
1255 recognising of open request
1258 Frequency : very rare
1260 Description: kernel reports "badness in vsnprintf"
1261 Details : Reading from the "recovery_status" /proc file in small chunks
1262 may cause a negative length in lprocfs_obd_rd_recovery_status()
1263 call to vsnprintf() (which is otherwise harmless). Exit early
1264 if there is no more space in the output buffer.
1266 Severity : enhancement
1268 Description: clear OBD RPC statistics by writing to them
1269 Details : It is now possible to clear the OBD RPC statistics by writing
1270 to the "stats" file.
1275 Description: Client mtime is not the same on different clients after utimes
1276 Details : In some cases, the client was using the utimes() syscall on
1277 a file cached on another node. The clients now validate the
1278 ctime from the MDS + OSTs to determine which one is right.
1283 Description: Inability to activate failout mode
1284 Details : lconf script incorrectly assumed that in python string's numeric
1285 value is used in comparisons.
1288 Frequency : always with multiple stripes per file
1290 Description: Inefficient object allocation for mutli-stripe files
1291 Details : When selecting which OSTs to stripe files over, for files with
1292 a stripe count that divides evenly into the number of OSTs,
1293 the MDS is always picking the same starting OST for each file.
1294 Return the OST selection heuristic to the original design.
1299 Description: mount failures may take full timeout to return an error
1300 Details : Under some heavy load conditions it is possible that a
1301 failed mount can wait for the full obd_timeout interval,
1302 possibly several minutes, before reporting an error.
1303 Instead return an error as soon as the status is known.
1305 ------------------------------------------------------------------------------
1307 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
1309 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1310 INTEROPERATE with older versions automatically. Please read the
1311 user documentation before upgrading any part of a live system.
1312 * WARNING: Lustre networking configuration changes are required with
1313 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1316 * Support for kernels:
1317 2.6.9-22.0.2.EL (RHEL 4)
1318 2.6.5-7.244 (SLES 9)
1319 2.6.12.6 vanilla (kernel.org)
1322 Severity : enhancement
1323 Bugzilla : 7981/8208
1324 Description: Introduced Lustre Networking (LNET)
1325 Details : LNET is new networking infrastructure for Lustre, it includes
1326 a reorganized network configuration mode (see the user
1327 documentation for full details) as well as support for routing
1328 between different network fabrics. Lustre Networking Devices
1329 (LNDs) for the supported network fabrics have also been
1330 created for this new infrastructure.
1332 Severity : enhancement
1333 Description: Introduced Access control lists
1334 Details : clients can set ACLs on files and directories in order to have
1335 more fine-grained permissions than the standard Unix UGO+RWX.
1336 The MDS must be started with the "-o acl" mount option.
1338 Severity : enhancement
1339 Description: Introduced filesystem quotas
1340 Details : Administrators may now establish per-user quotas on the
1343 Severity : enhancement
1345 Description: Configuration change for the XT3
1346 The PTLLND is now used to run Lustre over Portals on the XT3
1347 The configure option(s) --with-cray-portals are no longer used.
1348 Rather --with-portals=<path-to-portals-includes> is used to
1349 enable building on the XT3. In addition to enable XT3 specific
1350 features the option --enable-cray-xt3 must be used.
1355 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1356 Details : A race between allocating a new llcd and re-getting the llcd_lock
1357 allowed another thread to grab newly-allocated llcd.
1359 Severity : enhancement
1361 Description: 2.6 OST async journal commit and locking fix to improve performance
1362 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1363 2.6 kernels are now async as they already were in 2.4 kernels so
1364 that they can commit concurrently with the network bulk transfer.
1365 For block-allocated files the filter allocation semaphore is held
1366 to avoid filesystem fragmentation during allocation. BKL lock
1367 removed for 2.6 xattr operations where it is no longer needed.
1372 Description: lconf incorrectly determined whether two IP networks could talk
1373 Details : In some more complicated routing and multiple-network
1374 configurations, lconf will avoid trying to make a network
1375 connection to a disjoint part of the IP space. It was doing the
1376 math incorrectly for one set of cases.
1381 Description: Fix for potential infinite loop processing records in an llog.
1382 Details : If an llog record is corrupted/zeroed, it is possible to loop
1383 forever in llog_process(). Validate the llog record length
1384 and skip the remainder of the block on error.
1387 Frequency : occasional (liblustre only)
1389 Description: liblustre could not open files whose last component is a symlink
1390 Details : sysio_path_walk() would incorrectly pass the open intent to
1391 intermediate path components.
1394 Frequency : rare (liblustre only with non-standard tuning)
1395 Bugzilla : 7201 (7350)
1396 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1397 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1398 after it has been released, only do this for referenced locks
1400 Severity : enhancement
1402 Description: specifying an (invalid) directory default stripe_size of -1
1403 would reset the directory default striping
1404 Details : stripe_size -1 was used internally to signal directory stripe
1405 removal, now use "all default" to signal dir stripe removal
1406 as a directory striping of "all default" is not useful
1409 Frequency : common for large clusters running liblustre clients
1411 Description: doing an ls when liblustre clients are running is slow
1412 Details : sending a glimpse AST to a liblustre client waits for every AST
1413 to time out, as liblustre clients will not respond. Since they
1414 cannot cache data we refresh the OST lock LVB from disk instead.
1416 Severity : enhancement
1418 Description: doing an ls at the same time as file IO can be slow
1419 Details : enqueue and other "small" requests can be blocked behind many
1420 large IO requests. Create a new OST IO portal for non-IO
1421 requests so they can be processed faster.
1424 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1426 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1427 created file with mode 000
1428 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1429 allow an ftruncate() as a non-root user to a file with mode 000.
1430 Lustre now respects this flag to disable mode checks when
1431 truncating a file owned by the user
1434 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1437 Description: Revoking locks from clients that went dead or catatonic might take
1439 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1440 cancellation of such locks instant on servers without waiting for
1441 any reply from clients. Clients drops these locks when cancel
1442 notification from server is received without replying.
1445 Frequency : liblustre-only, when liblustre client dies or becomes busy
1447 Description: Doing ls on Linux clients can take a long time with active
1449 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1450 granting such locks to it in the first place if possible. Locks
1451 are taken by proxy on the OST during the read or write and
1452 dropped immediately afterward. Add connect flags handling, do
1453 not grant locks to liblustre clients for glimpse ASTs.
1455 Severity : enhancement
1457 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1458 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1459 sequential reads and behaves badly with applications doing random
1460 reads. Improve it by reading ahead at least read region, and
1461 avoiding excessive large RPC for small reads.
1463 Severity : enhancement
1465 Description: Creating more than 1000 files for a single job may cause a load
1466 imbalance on the OSTs if there are also a large number of OSTs.
1467 Details : qos_prep_create() uses an OST index reseed value that is an
1468 even multiple of the number of available OSTs so that if the
1469 reseed happens in the middle of the object allocation it will
1470 still utilize the OSTs as uniformly as possible.
1475 Description: OST or MDS may oops in ping_evictor_main()
1476 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1477 but doesn't restart at beginning of obd_exports_timed list
1480 Severity : enhancement
1482 Description: improve by-nid export eviction on the MDS and OST
1483 Details : allow multiple exports with the same NID to be evicted at one
1484 time without re-searching the exports list.
1487 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1489 Description: MDS may oops in groups_free()
1490 Details : in rare race conditions a newly allocated group_info struct is
1491 freed again, and this can be NULL. The 2.4 compatibility code
1492 for groups_free() checked for a NULL pointer, but 2.6 did not.
1495 Frequency : common for liblustre clients doing little filesystem IO
1496 Bugzilla : 9352, 7313
1497 Description: server may evict liblustre clients accessing contended locks
1498 Details : if a client is granted a lock or receives a completion AST
1499 with a blocking AST already set it would not reply to the AST
1500 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1503 Frequency : lfs setstripe, only systems with more than 160 OSTs
1505 Description: unable to set striping with a starting offset beyond OST 160
1506 Details : llapi_create_file() incorrectly limited the starting stripe
1507 index to the maximum single-file stripe count.
1510 Frequency : LDAP users only
1512 Description: lconf did not handle in-kernel recovery with LDAP properly
1513 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1515 Severity : enhancement
1517 Description: bind OST threads to NUMA nodes to improve performance
1518 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1519 node and do their allocations there to localize memory access
1521 Severity : enhancement
1523 Description: llmount can determine client NID directly from Myrinet (GM)
1524 Details : the client NID code from gmnalnid was moved directly into
1525 llmount, removing the need to use this or specifying the
1526 client NID explicitly when mounting GM clients with zeroconf
1529 Frequency : if client is started with down MDS
1531 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1532 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1533 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1534 ptlrpc_import_delayed_req() because we didn't check if the
1535 request was interrupted, and we also didn't break out of the
1536 event loop if there was no timeout
1541 Description: data loss during non-page-aligned writes to a single file from
1542 both multiple nodes and multiple threads on one node at same time
1543 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1544 inconsistency led to false short-reads, that were cached and later
1545 used by ->prepare_write() to fill in partially written page,
1546 leading to data loss.
1549 Frequency : always, if lconf --abort_recovery used
1551 Description: lconf --abort_recovery fails with 'Operation not supported'
1552 Details : lconf was attempting to abort recovery on the MDT device and not
1555 Severity : enhancement
1557 Description: remove cleanup logs
1558 Details : replace lconf-generated cleanup logs with lustre internal
1559 cleanup routines. Eliminates the need for client-cleanup and
1562 Severity : enhancement
1564 Description: add support for EAs (user and system) on lustre filesystems
1565 Details : it is now possible to store extended attributes in the Lustre
1566 client filesystem, and with the user_xattr mount option it
1567 is possible to allow users to store EAs on their files also
1569 Severity : enhancement
1571 Description: Add possibility (config option) to show minimal available OST free
1573 Details : When compiled with --enable-mindf configure option, statfs(2)
1574 (and so, df) will return least minimal free space available from
1575 all OSTs as amount of free space on FS, instead of summary of
1576 free spaces of all OSTs.
1578 Severity : enhancement
1580 Description: do not expand extent locks acquired on OST-side
1581 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1582 by server: they are not cached anyway.
1585 Frequency : when mmap is used/binaries executed from Lustre
1587 Description: Unmmap pages before throwing them away from read cache.
1588 Details : llap_shrink cache now attempts to unmap pages before discarding
1589 them (if unmapping failed - do not discard). SLES9 kernel has
1590 extra checks that trigger if this unmapping is not done first.
1595 Description: lconf didn't resolve symlinks before checking to see whether a
1596 given mountpoint was already in use
1599 Frequency : when migrating failover services
1600 Bugzilla : 6395, 9514
1601 Description: When migrating a subset of services from a node (e.g. failback
1602 from a failover service node) the remaining services would
1603 time out and evict clients.
1604 Details : lconf --force (implied by --failover) sets the global obd_timeout
1605 to 5 seconds in order to quickly disconnect, but this caused
1606 other RPCs to time out too quickly. Do not change the global
1607 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1609 Severity : enhancement
1610 Frequency : if MDS is started with down OST
1611 Bugzilla : 9439,5706
1612 Description: Allow startup/shutdown of an MDS without depending on the
1613 availability of the OSTs.
1614 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1615 Add appropriate locking and lov-osc refcounts for safe
1616 cleaning. Add osc abort_inflight calls in case the
1617 synchronize never started.
1620 Frequency : occasional (Cray XT3 only)
1622 Description: root not authorized to access files in CRAY_PORTALS environment
1623 Details : The client process capabilities were not honoured on the MDS in
1624 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1625 been accessed by an authorized user then root was able to access
1626 the file on the local client also. The root user capabilities
1627 are now allowed on the MDS, as this environment has secure UID.
1630 Frequency : occasional
1632 Description: ldiskfs "too long searching" message happens too often
1633 Details : A debugging message (otherwise harmless) prints too often on
1634 the OST console. This has been reduced to only happen when
1635 there are fragmentation problems on the filesystem.
1640 Description: Division by zero in statfs when all OSCs are inactive
1641 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
1642 lov_statfs divides by value returned by lov_get_stripecnt().
1646 Bugzilla : 9489, 3273
1647 Description: First write from each client to each OST was only 4kB in size,
1648 to initialize client writeback cache, which caused sub-optimal
1649 RPCs and poor layout on disk for the first writen file.
1650 Details : Clients now request an initial cache grant at (re)connect time
1651 and so that they can start streaming writes to the cache right
1652 away and always do full-sized RPCs if there is enough data.
1653 If the OST is rebooted the client also re-establishes its grant
1654 so that client cached writes will be honoured under the grant.
1659 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
1660 Details : Now I/O RPCs go to different portal number and (presumably) fast
1661 lock requests (and glimses) and other RPCs get their own service
1662 threads pool that should be able to service those RPCs
1665 Severity : enhancement
1667 Description: Ability to exchange lustre version between client and servers and
1668 issue warnings at client side if client is too old. Also for
1669 liblustre clients there is ability to refuse connection of too old
1671 Details : New 'version' field is added to connect data structure that is
1672 filled with version info. That info is later checked by server and
1676 Frequency : rare, liblustre only.
1677 Bugzilla : 9296, 9581
1678 Description: Two simultaneous writes from liblustre at offset within same page
1679 might proceed at the same time overwriting eachother with stale
1681 Details : I/O lock withing llu_file_prwv was released too early, before data
1682 actually was hitting the wire. Extended lock-holding time until
1683 server acknowledges receiving data.
1686 Frequency : extremely rare. Never observed in practice.
1688 Description: avoid generating lustre_handle cookie of 0.
1689 Details : class_handle_hash() generates handle cookies by incrementing
1690 global counter, and can hit 0 occasionaly (this is unlikely, but
1691 not impossible, because initial value of cookie counter is
1692 selected randonly). Value of 0 is used as a sentinel meaning
1693 "unassigned handle" --- avoid it. Also coalesce two critical
1694 sections in this function into one.
1696 Severity : enhancement
1698 Description: allow liblustre clients to delegate truncate locking to OST
1699 Details : To avoid overhead of locking, liblustre client instructs OST to
1700 take extent lock in ost_punch() on client's behalf. New connection
1701 flag is added to handle backward compatibility.
1703 Severity : enhancement
1704 Bugzilla : 4928, 7341, 9758
1705 Description: allow number of OST service threads to be specified
1706 Details : a module parameter allows the number of OST service threads
1707 to be specified via "options ost ost_num_threads={N}" in the
1708 OSS's /etc/modules.conf or /etc/modprobe.conf.
1712 Bugzilla : 6146, 9635, 9895
1713 Description: servers crash with bad pointer in target_handle_connect()
1714 Details : In rare cases when a client is reconnecting it was possible that
1715 the connection request was the last reference for that export.
1716 We would temporarily drop the export reference and get a new
1717 one, but this may have been the last reference and the export
1718 was just destroyed. Get new reference before dropping old one.
1720 Severity : enhancement
1721 Frequency : if client is started with failover MDS
1723 Description: Allow multiple MDS hostnames in the mount command
1724 Details : Try to read the configuration from all specified MDS
1725 hostnames during a client mount in case the "primary"
1728 Severity : enhancement
1730 Description: Stop sending data to evicted clients as soon as possible.
1731 Details : Check if the client we are about to send or are sending data to
1732 was evicted already. (Check is done every second of waiting,
1733 for which l_wait_event interface was extended to allow checking
1734 of exit condition at specified intervals).
1737 Frequency : rare, normally only when NFS exporting is done from client
1739 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
1741 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
1744 Frequency : always, if srand() is called before liblustre initialization
1746 Description: Liblustre uses system PRNG disturbing its usage by user application
1747 Details : Introduce internal to lustre fast and high-quality PRNG for
1748 lustre usage and make liblustre and some other places in generic
1749 lustre code to use it.
1751 Severity : enhancement
1752 Bugzilla : 9477, 9557, 9870
1753 Description: Verify that the MDS configuration logs are updated when xml is
1754 Details : Check if the .xml configuration logs are newer than the config
1755 logs stored on the MDS and report an error if this is the case.
1756 Request --write-conf, or allow starting with --old_conf.
1758 Severity : enhancement
1760 Description: Handle symlinks in the path when checking if Lustre is mounted.
1761 Details : Resolve intermediate symlinks when checking if a client has
1762 mounted a filesystem to avoid duplicate client mounts.
1767 Description: lconf can hit an error exception but still return success.
1768 Details : The lconf command catches the Command error exception at the top
1769 level script context and will exit with the associated exit
1770 status, but doesn't ensure that this exit status is non-zero.
1775 Description: failure of ptlrpc thread startup can cause oops
1776 Details : Starting a ptlrpc service thread can fail if there are a large
1777 number of threads or the server memory is very fragmented.
1778 Handle this without oopsing.
1781 Frequency : always, only if liblustre and non-default acceptor port was used
1783 Description: liblustre cannot connect to servers with non-default acceptor port
1784 Details : tcpnal_set_default_params() was not called and was therefore
1785 ignoring the environment varaible TCPNAL_PORT, as well as other
1786 TCPNAL_ environment variables
1791 Description: two objects could be created on the same OST for a single file
1792 Details : If an OST is down, in some cases it was possible to create two
1793 objects on a single OST for a single file. No problems other
1794 than potential performance impact and spurious error messages.
1798 Bugzilla : 5681, 9562
1799 Description: Client may oops in ll_unhash_aliases
1800 Details : Client dcache may become inconsistent in race condition.
1801 In some cases "getcwd" can fail if the current directory is
1807 Description: Inode refcounting problems in NFS export code
1808 Details : link_raw functions used to call d_instantiate without obtaining
1809 extra inode reference first.
1813 Bugzilla : 9942, 9903
1814 Description: Referencing freed requests leading to crash, memleaks with NFS.
1815 Details : We used to require that call to ll_revalidate_it was always
1816 followed by ll_lookup_it. Also with revalidate_special() it is
1817 possible to call ll_revalidate_it() twice for the same dentry
1818 even if first occurence returned success. This fix changes semantic
1819 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
1820 reference on a request referred from the intent.
1821 ll_intent_release() then releases such a request.
1824 Frequency : rare, normally benchmark loads only
1826 Description: unlinked inodes were kept in memory on the client
1827 Details : If a client is repeatedly creating and unlinking files it
1828 can accumulate a lot of stale inodes in the inode slab cache.
1829 If there is no other client load running this can cause the
1830 client node to run out of memory. Instead flush old inodes
1831 from client cache that have the same inode number as a new inode.
1834 Frequency : SLES9 2.6.5 kernel and long filenames only
1835 Bugzilla : 9969, 10379
1836 Description: utime reports stale NFS file handle
1837 Details : SLES9 uses out-of-dentry names in some cases, which confused
1838 the lustre dentry revalidation. Change it to always use the
1842 Frequency : rare, unless heavy write-truncate concurrency is continuous
1843 Bugzilla : 4180, 6984, 7171, 9963, 9331
1844 Description: OST becomes very slow and/or deadlocked during object unlink
1845 Details : filter_destroy() was holding onto the parent directory lock
1846 while truncating+unlinking objects. For very large objects this
1847 may block other threads for a long time and slow overall OST
1848 responsiveness. It may also be possible to get a lock ordering
1849 deadlock in this case, or run out of journal credits because of
1850 the combined truncate+unlink. Solution is to do object truncate
1851 first in one transaction without parent lock, and then do the
1852 final unlink in a new transaction with the parent lock. This
1853 reduces the lock hold time dramatically.
1856 Frequency : rare, 2.4 kernels only
1858 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
1859 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
1860 before unmounting their filesystem in order to avoid deadlock.
1861 The kernel_locked() function in 2.4 kernels only checks whether
1862 the kernel lock is held, not whether it is this process that is
1863 holding it as 2.6 kernels do.
1868 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
1869 Details : The client ptlrpc code may be trying to reconnect to a down
1870 server before a previous connection attempt has timed out.
1871 Increase the reconnect interval to be longer than the connection
1872 timeout interval to avoid sending duplicate connections to
1876 Frequency : echo_client brw_test command
1878 Description: fix echo_client to work with OST preallocated code
1879 Details : OST preallocation code (5137) didn't take echo_client IO path
1880 into account: echo_client calls filter methods outside of any
1881 OST thread and, hence, there is no per-thread preallocated
1882 pages and buffers to use. Solution: hijack pga pages for IO. As
1883 a byproduct, this avoids unnecessary data copying.
1887 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
1888 Description: Client can oops in mdc_commit_close() after open replay
1889 Details : It was possible for the MDS to return an open request with no
1890 transaction number in mds_finish_transno() if the client was
1891 evicted, but without actually returning an error. Clients
1892 would later try to replay that open and may trip an assertion
1893 Simplify the client close codepath, and always return an error
1894 from the MDS in case the open is not successful.
1897 Frequency : rare, 2.6 OSTs only
1899 Description: OST may deadlock under high load on fragmented files
1900 Details : If there was a heavy load and highly-fragmented OST filesystems
1901 it was possible to have all the OST threads deadlock waiting on
1902 allocation of biovecs, because the biovecs were not released
1903 until the entire RPC IO was completed. Instead, release biovecs
1904 as soon as they are complete to ensure forward IO progress.
1906 Severity : enhancement
1908 Description: Support for specifying external journal device at mount
1909 Details : If an OST or MDS device is formatted with an external journal
1910 device, this device major/minor is stored in the ext3 superblock
1911 and may not be valid for failover. Allow detecting and
1912 specifying the external journal at mount time.
1917 Description: Mounting an MDS with pending unlinked files may cause oops
1918 Details : target_finish_recovery() calls mds_postrecov() which returned
1919 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
1920 considers this an error and immediately begins cleaning up the
1921 lov, just after starting the mds_lov process
1923 Severity : enhancement
1925 Description: Implement 'lfs df' to report actual free space on per-OST basis
1926 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
1927 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
1928 usage of objects; '-h' to report in human readable format.
1930 ------------------------------------------------------------------------------
1932 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
1939 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
1940 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
1941 at startup during orphan unlinking. Instead, skip buddy update
1942 before setup, as it will be regenerated after recovery is complete.
1947 Description: If an OST is inactive, its locks might reference stale inodes.
1948 Details : lov_change_cbdata() must iterate over all namespaces, even if
1949 they are inactive to clear inode references from the lock.
1951 Severity : enhancement
1952 Frequency : occasional, if non-standard max_dirty_mb used
1954 Description: Client will block write RPCs if not enough grant
1955 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
1956 then the client will block writes while waiting for another RPC
1957 to complete instead of consuming its dirty limit. With change
1958 we get improved performance when max_dirty_mb is small.
1960 Severity : enhancement
1961 Bugzilla : 3389, 6253
1962 Description: Add support for supplementary groups on the MDS.
1963 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
1964 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
1965 lookups for user supplementary groups into a cache.
1969 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
1970 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
1971 for liblustre O_EXCL is enforced by MDS.
1975 Description: readdir never returns NULL in liblustre.
1976 Details : Corrected llu_iop_getdirentries logic, to return offset of next
1977 dentry in struct dirent.
1981 Frequency : liblustre only, depends on application IO pattern
1982 Description: liblustre clients evicted if not contacting servers
1983 Details : Don't put liblustre clients into the ping_evictor list, so
1984 they will not be evicted by the pinger ever.
1986 Severity : enhancement
1988 Description: Add ability to evict clients by NID from MDS.
1989 Details : By echoing "nid:$NID" string into
1990 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
1991 $NID would be instantly evicted from this MDS and from all active
1992 OSTs connected to it.
1996 Description: Do not query file size twice, somewhat slowing stat(2) calls.
1997 Details : lookup_it_finish() used to query file size from OSTs that was not
2002 Description: service threads change working directory to that of init
2003 Details : Starting lustre service threads may pin the working directory
2004 of the parent thread, making that filesystem busy. Threads
2005 now change to the working directory of init to avoid this.
2009 Frequency : during shutdown only
2010 Description: shutdown with a failed MDS or OST can cause unmount to hang
2011 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2016 Frequency : 2.6 only
2017 Description: chmod/chown may include an extra supplementary group
2018 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2019 supplementary group and if none is specified this is used.
2022 Bugzilla : 5479 (6816)
2024 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2025 Details : If lookup is in progress on a file that is unlinked we might try
2026 to revalidate the inode and fail in revalidate after lookup is
2027 complete and ll_file_open() enqueues the open again but
2028 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2031 Frequency : always, if lconf --abort_recovery used
2033 Description: lconf --abort_recovery fails with 'Operation not supported'
2034 Details : lconf was attempting to abort recovery on the MDT device and not
2037 ------------------------------------------------------------------------------
2039 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2044 Frequency : rare (only unsupported configurations with a node running as an
2046 Bugzilla : 6514, 5137
2047 Description: Mounting a Lustre file system on a node running as an OST could
2049 Details : OSTs now preallocates memory needed to write out data at
2050 startup, instead of when needed, to avoid having to
2051 allocate memory in possibly low memory situations.
2052 Specifically, if the file system is mounted on on OST,
2053 memory pressure could force it to try to write out data,
2054 which it needed to allocate memory to do. Due to the low
2055 memory, it would be unable to do so and the node would
2056 become unresponsive.
2058 Severity : enhancement
2060 Description: Addition of lconf --service command line option
2061 Details : lconf now accepts a '--service <arg>' option, which is
2062 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2064 Severity : enhancement
2066 Description: Failover mode is now the default for OSTs.
2067 Details : By default, OSTs will now run in failover mode. To return to
2068 the old behaviour, add '--failout' to the lmc line for OSTs.
2070 Severity : enhancement
2072 Description: Health checks are now provided for MDS and OSTs
2073 Details : Additional detailed health check information on MSD and OSTs
2074 is now provided through the procfs health_check value.
2077 Frequency : occasional, depends on IO load
2079 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2080 after numerous create/delete cycles
2081 Details : The ext3 inode allocation policy would not allocate new inodes
2082 very well on the OSTs because there are no new directories
2083 being created. Instead we look for groups with free space if
2084 the parent directories are nearly full.
2089 Description: Network or server problems during mount may cause partially
2090 mounted clients instead of returning an error.
2091 Details : The config llog parsing code may overwrite the error return
2092 code during mount error handling, returning success instead
2098 Description: MDS can fail to allocate large reply buffers
2099 Details : After long uptimes the MDS can fail to allocate large reply
2100 buffers (e.g. zconf client mount config records) due to memory
2101 fragmentation or consumption by the buffer cache. Preallocate
2102 some large reply buffers so that these replies can be sent even
2103 under memory pressure.
2107 Frequency : rare (liblustre)
2108 Description: fsx running with liblustre complained that using truncate() to
2109 extend the file doesn't work. This patch corrects that issue.
2110 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2111 ATTR_SIZE and lsm use in llu_setattr_raw.
2115 Frequency : rare, only 2.6 kernels
2116 Description: Unusual file access patterns on the MDS may result in inode
2117 data being lost in very rare circumstances.
2118 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2119 code in the 2.6 kernel caused the inode and/or EA data not to
2120 be read from disk, causing single-file corruption.
2124 Frequency : rare, only 2.6 filesystems using extents
2125 Description: Heavy concurrent write and delete load may cause data corruption.
2126 Details : It was possible under high-load situations to have an extent
2127 metadata block in the block device cache from a just-unlinked
2128 file overwrite a newly-allocated data block. We now unmap any
2129 metadata buffers that alias just-allocated data blocks.
2133 Frequency : filesystems with default stripe_count larger than 77
2134 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2135 Details : lconf specifies an inode size of 4096 bytes when the default
2136 stripe_count is larger than 77. This conflicts with the default
2137 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2138 this case to avoid pinning too much memory for large EAs.
2140 ------------------------------------------------------------------------------
2142 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2147 Frequency : rare (extremely heavy IO load with hundreds of clients)
2149 Description: Client is evicted, gets IO error writing to file
2150 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2151 caused clients to be evicted for AST timeouts. The fixes in
2152 bug 5192 mean we no longer need to have such short AST timeouts
2153 so ldlm_timeout has been increased.
2156 Frequency : occasional during --force or --failover shutdown under load
2157 Bugzilla : 5949, 4834
2158 Description: Server oops/LBUG if stopped with --force or --failover under load
2159 Details : a collection of import/export refcount and cleanup ordering
2160 issues fixed for safer force cleanup
2163 Frequency : only filesystems larger than 120 OSTs
2164 Bugzilla : 5990, 6223
2165 Description: lfs getstripe would oops on a very large filesystem
2166 Details : lov_getconfig used kfree on vmalloc'd memory
2169 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2170 Bugzilla : 6242, 6243
2171 Description: reading from files that had been truncated to a non-zero size
2172 but never opened returned no data
2173 Details : ll_file_read() reads zeros from no-object files to EOF
2178 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2179 Details : There's little harm in aborting MDS/OSS recovery and letting it
2180 try again, so I removed the LASSERT and return an error instead.
2182 Severity : enhancement
2184 Description: New debugging infrastructure for tracking down data corruption
2185 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2186 (b) cover more of the client-side code path, and (c) try to narrow
2187 down where problems occurred
2191 Bugzilla : 3819, 4364, 4397, 6313
2192 Description: Racing close and eviction MDS could cause assertion in mds_close
2193 Details : It was possible to get multiple mfd references during close and
2194 client eviction, leading to one thread referencing a freed mfd.
2196 Severity: : enhancement
2197 Bugzilla : 3262, 6359
2198 Description: Attempts to reconnect to servers are now more aggressive.
2199 Details : This builds on the enhanced upcall-less recovery that was added
2200 in 1.4.2. When trying to reconnect to servers, clients will
2201 now try each server in the failover group every 10 seconds. By
2202 default, clients would previously try one server every 25 seconds.
2207 Description: After recovery, certain operations trigger a failed
2208 assertion on a client.
2209 Details : Failing over an mds, using lconf -d --failover, while a
2210 client was doing a readdir() call would cause the client to
2211 LBUG after recovery completed and the readdir() was resent.
2213 Severity : enhancement
2215 Description: Default groups are now added by lconf
2216 Details : You can now run lconf --group <servicename> without having to
2217 manually add groups with lmc.
2220 Frequency : occasional
2222 Description: Nodes with an elan id of 0 trigger a failed assertion
2225 Frequency : always when accessing e.g. tty/console device nodes
2227 Description: tty and some other devices nodes cannot be used on lustre
2228 Details : file's private_data field is used by device data and lustre
2229 values in there got lost. New field was added to struct file to
2230 store fs-specific private data.
2233 Frequency : when exporting Lustre via NFS
2235 Description: NFSD failed occasionally when looking up a path component
2236 Details : NFSD is looking up ".." which was broken in ext3 directories
2237 that had grown large enough to become hashed.
2240 Frequency : Clusters with multiple interfaces not on the same subnet
2242 Description: Nodes will repeatedly try to reconnect to an interface which it
2243 cannot reach and report an error to the log.
2244 Details : Extra peer list entries will be created by lconf with some peers
2245 unreachable. lconf now validates the peer before adding it.
2248 Frequency : Only if a default stripe is set on the filesystem root.
2250 Description: Setting a default stripe on the filesystem root prevented the
2251 filesystem from being remounted.
2252 Details : The client was sending extra request flags in the root getattr
2253 request and did not allocate a reply buffer for the dir EA.
2256 Frequency : occasional, higher if lots of files are accessed by one client
2257 Bugzilla : 6159, 6097
2258 Description: Client trips assertion regarding lsm mismatch/magic
2259 Details : While revalidating inodes the VFS looks up inodes with ifind()
2260 and in rare cases can find an inode that is being freed.
2261 The ll_test_inode() code will free the lsm during ifind()
2262 when it finds an existing inode and then the VFS later attaches
2263 this free lsm to a new inode.
2267 Bugzilla : 6422, 7030
2268 Description: MDS deadlock between mkdir and client eviction
2269 Details : Creating a new file via mkdir or mknod (starting a transaction
2270 and getting the ns lock) can deadlock with client eviction
2271 (gets ns lock and trying to finish a synchronous transaction).
2274 Frequency : occasional
2275 Description: While starting a server, the fsfilt_ext3 module could not be
2277 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2278 kernels. Previously, lconf would still use the ext3 name
2279 when trying to load modules. Now, it will correctly use
2280 ext3 on 2.4 and ldiskfs on 2.6.
2282 Severity : enhancement
2283 Description: The default stripe count has been changed to 1
2284 Details : The interpretation of the default stripe count (0, to lfs
2285 or lmc) has been changed to mean striping across a single
2286 OST, rather than all available. For general usage we have
2287 found a stripe count of 1 or 2 works best.
2289 Severity : enhancement
2290 Description: Add support for compiling against Cray portals.
2291 Details : Conditional compiling for some areas that are different
2295 Frequency : occasional
2296 Bugzilla : 6409, 6834
2297 Description: Creating files with an explicit stripe count may lead to
2298 a failed assertion on the MDS
2299 Details : If some OSTs are full or unavailable, creating files may
2300 trigger a failed assertion on the MDS. Now, Lustre will
2301 try to use other servers or return an error to the
2305 Frequency : occasional
2307 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2308 caused lock timeout during readahead (since 1.4.2).
2309 Details : Processes doing readahead might match a lock that hasn't been
2310 granted yet if there are overlapping and conflicting lock
2311 requests. The readahead process waits on ungranted lock
2312 (original lock is CBPENDING), while OST waits for that process
2313 to cancel CBPENDING read lock and eventually evicts client.
2315 Severity : enhancement
2317 Description: Initial enabling of flock support for clients
2318 Details : Implements fcntl advisory locking and file status functions.
2319 This feature is provided as an optional mount flag (default
2320 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2321 locking are implemented yet, and those that are are not guaranteed
2322 to be completely correct in production environments.
2323 mount -t lustre -o [flock|noflock] ...
2326 Frequency : occasional
2328 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2329 assertion in the ext3 JBD (journaling) layer.
2330 Details : The b_committed_data struct is protected by the big kernel lock
2331 in 2.4 kernels, serializing journal_commit_transaction() and
2332 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2333 access to this struct. In 2.6 kernels there is finer grained
2334 locking to improve SMP performance of the JBD layer.
2338 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2340 -----------------------------------------------------------------------------
2342 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2344 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2345 versions of Lustre. Please update all servers and clients to
2346 version 1.4.2 or later at the same time. You must also run
2347 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2348 to update the configuration logs.
2350 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2351 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2352 - swab llog records of type '0' so we get proper header size/idx (5861)
2353 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2354 - fix rename of one directory over another leaking an inode (5953)
2355 - avoid SetPageDirty on 2.6 (5981)
2356 - don't re-add just-being-destroyed locks to the waiting list (5653)
2357 - when creating new directories, inherit the parent's custom
2358 striping settings if present parent (3048)
2359 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2360 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2361 - don't hold client locks on temporary worklist from l_lru (5666)
2362 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2363 - automatically evict dead clients (5921)
2364 - Update file size properly in create+truncate+fstat case (6196)
2365 - Do not unhash mountpoint dentries, do not allow removal of
2367 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2368 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2369 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2370 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2371 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2372 - Correctly handle reads of files with no objects (6243)
2373 - lctl recover will also mark a device active if deactivate used (5933)
2375 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2376 - allow --write-conf on an MDS with different nettype than client (5619)
2377 - don't write config llogs to MDS for mounts not from that MDS (5617)
2378 - lconf should create multiple TCP connections from a client (5201)
2379 - init scripts are now turned off by default; run chkconfig --on
2380 lustre and chkconfig --on lustrefs to use them
2381 - upcalls are no longer needed for clients to recover to failover
2383 - add --abort-recovery option to lconf to abort recovery on device
2385 - add support for an arbitrary number of OSTs (3026)
2386 - Quota support protocol changes.
2387 - forward compatibility changes to wire structs (6007)
2388 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2389 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2390 - improved "lustre status" script
2391 - initialize blocksize for non-regular files (6062)
2392 - added --disable-server and --disable-client configure options (5782)
2393 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2394 - Vanilla 2.4.29 support
2395 - increase maximum number of obd devices to 520 (6242)
2396 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2397 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2398 SLES 9 2.6.5 kernels
2399 - limit stripes per file to 160 (the maximum EA size) (6093)
2401 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2404 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2405 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2406 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2407 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2408 - small ext3 extents cleanups and fixes (5733)
2409 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2410 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2411 - avoid needless client->OST connect, fix handle mismatch (5317)
2412 - fix DLM error path that led to out-of-sync client, long delays (5779)
2413 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2414 - fix several locking issues related to i_size (5492,5624,5654,5672)
2415 - don't move pending lock onto export if it is already evicted (5683)
2416 - fix kernel oops when creating .foo in unlinked directory (5548)
2417 - fix deadlock in obdfilter statistics vs. object create (5811)
2418 - use time_{before,after} to avoid timer jiffies wrap (5882)
2419 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2420 - Do not leak request if server was not able to process it (5154)
2421 - If mds_open unable to find parent dir, make that negative lookup(5154)
2422 - don't create new directories with extent-mapping (5909, 5936)
2424 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2425 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2426 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2427 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2428 - support for RHEL4 kernel on i686 (5773)
2429 - provide error messages when incompatible logs are encountered (5898)
2431 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2432 * version 1.4.0.10 (1.4.1 release candidate 1)
2434 - don't keep a lock reference when lock is not granted (4238)
2435 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2436 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2437 - drop import inflight refcount on signal_completed_replay error (5255)
2438 - unlock page after async write error during send (3677)
2439 - handle missing objects in filter_preprw_read properly (5265)
2440 - no transno return for symlink open, don't save no-trasno open (3440)
2441 - don't try to complete elan receive that already failed (4012)
2442 - free RPC server reply state on error (5406)
2443 - clean up thread from ptlrpc_start_thread() on error (5160)
2444 - readahead could read extra page into cache that wasn't ejected (5388)
2445 - prevent races in class_attach/setup/cleanup/detach (5260)
2446 - don't dereference de->d_inode after l_dput of de (5458)
2447 - use "int" for stripe value returned from lock_to_stripe (5544)
2448 - mballoc allocation and error-checking fixes in 2.6 (5504)
2449 - block device patches to fix I/O request sizes in 2.6 (5482)
2450 - look up hostnames for IB nals (5602)
2451 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2452 - don't start multiple acceptors for the same port (5277)
2453 - fix incorrect LASSERT in mds_getattr_name (5635)
2454 - export a proc file for general "ping" checking (5628)
2455 - fix "lfs check" to not block when the MDS is down (5628)
2457 - service request history (4965)
2458 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2459 - create an "evict_client" /proc entry on OSTs, like the MDS has
2460 - fix mount usage message, return errors per mount(8) (5168)
2461 - change grep [] to grep "[]" in tests so they work in more UMLs
2462 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2463 - remove ancient LOV_MAGIC_V0 EA support (5047)
2464 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2465 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2466 - allow readahead allocations to fail when low on memory (5383)
2467 - mmap locking landed again, after considerable improvement (2828)
2468 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2470 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2473 - send OST transaction number in read/write reply to free req (4966)
2474 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2475 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2476 - the watchdog thread now runs as interruptible (5246)
2477 - flock/lockf fixes (but it's still disabled, pending 5135)
2478 - don't use EXT3 constants in llite code (5094)
2479 - memory shortage at startup could cause assertion (5176)
2481 - reorganization of lov code
2482 - single portals codebase
2484 - add extents/mballoc support (5025)
2485 - direct I/O reads in the obdfilter (4048)
2486 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2488 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2491 - send OST transaction number in read/write reply to free req (4966)
2492 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2493 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2494 - the watchdog thread now runs as interruptible (5246)
2495 - handle missing objects in filter_preprw_read properly (5265)
2496 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2497 - drop import inflight refcount on signal_completed_replay error (5255)
2498 - unlock page after async write error during send (3677)
2499 - return original error code on reconstructed replies (3761)
2500 - no transno return for symlink open, don't save no-trasno open (3440)
2502 - add pid to ldlm debugging output (4922)
2503 - bump the watchdog timeouts -- we can't handle 30sec yet
2504 - extra debugging for orphan dentry/inode bug (5259)
2506 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2509 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2510 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2511 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2512 - fix NULL dereference in /proc/sys/portals/routes (4827)
2513 - allow failed mdc_close() operations to be interrupted (4561)
2514 - stop precreate on OST before MDS would time out on it (4778)
2515 - don't send partial-page writes before EOF from client (4410)
2516 - discard client grant for sub-page writes on large-page clients (4520)
2517 - don't free dentries not owned by NFS code, check generation (4806)
2518 - fix lsm leak if mds_create_objects() fails (4801)
2519 - limit debug_daemon file size, always print CERROR messages (4789)
2520 - use transno after validating reply (3892)
2521 - process timed out requests if import state changes (3754)
2522 - update mtime on OST during writes, return in glimpse (4829)
2523 - add mkfsoptions to LDAP (4679)
2524 - use ->max_readahead method instead of zapping global ra (5039)
2525 - don't interrupt __l_wait_event() during strace
2527 - add software watchdogs to catch hung threads quickly (4941)
2528 - make lustrefs init script start after nfs is mounted
2529 - fix CWARN/ERROR duplication (4930)
2530 - return async write errors to application if possible (2248)
2531 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2532 - print NAL number in %x format (4645)
2533 - update barely-supported suse-2.4.21-171 series (4842)
2534 - support for sles 9 %post scripts
2535 - support for building 2.6 kernel-source packages
2536 - support for sles km_* packages
2538 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2541 - ignore -ENOENT errors in osc_destroy (3639)
2542 - notify osc create thread that OSC is being cleaned up (4600)
2543 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2544 - reconstruct ost_handle() like mds_handle() (4657)
2545 - create a new thread to do import eviction to avoid deadlock (3969)
2546 - let lconf resolve symlinked-to devices (4629)
2547 - don't unlink "objects" from directory with default EA (4554)
2548 - hold socknal file ref over connect in case target is down (4394)
2549 - allow more than 32000 subdirectories in a single directory (3244)
2550 - fix blocks count for O_DIRECT writes (3751)
2551 - OST returns ENOSPC from object create when no space left (4539)
2552 - don't send truncate RPC if file size isn't changing (4410)
2553 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2554 - bind to privileged port in socknal and tcpnal (3689)
2556 - rate limit CERROR/CWARN console message to avoid overload (4519)
2557 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2558 - basic mmap support (3918)
2559 - kernel patch series update from b1_4 (4711)
2561 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2564 - avoid crash during MDS cleanup with OST shut down (2775)
2565 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2566 - don't use bad inodes on the MDS (3744)
2567 - dynamic object preallocation to improve recovery speed (4236)
2568 - don't hold spinlock over lock dumping or change debug flags (4401)
2569 - don't zero obd_dev when it is force cleaned (3651)
2570 - print grants to console if they go negative (4431)
2571 - "lctl deactivate" will stop automatic recovery attempts (3406)
2572 - look for existing locks in ldlm_handle_enqueue() (3764)
2573 - don't resolve lock handle twice in recovery avoiding race (4401)
2574 - revalidate should check working dir is a directory (4134)
2576 - don't always mark "slow" obdfilter messages as errors (4418)
2578 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2581 - don't close LustreDB during write_conf until it is done (3860)
2582 - fix typo in lconf for_each_profile (3821)
2583 - allow dumping logs from multiple threads at one time (3820)
2584 - don't allow multiple threads in OSC recovery (3812)
2585 - fix debug_size parameters (3864)
2586 - fix mds_postrecov to initialize import for llog ctxt (3121)
2587 - replace config semaphore with spinlock (3306)
2588 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2589 - don't allow enqueue to complete on a destroyed export (3822)
2590 - down write_lock before checking llog header bitmap (3825)
2591 - recover from lock replay timeout (3764)
2592 - up llog sem before sending rpc (3652)
2593 - reduce ns lock hold times when setting kms (3267)
2594 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2595 - fix NULL deref and obd_dev leak on setup error (3312)
2596 - replace some LBUG about llog ops with error handling (3841)
2597 - don't match INVALID dentries from d_lookup and spin (3784)
2598 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2599 - fix invalid assertion in ptlrpc_set_wait (3880)
2601 - add libwrap support for the TCP acceptor (3996)
2602 - add /proc/sys/portals/routes for non-root route listing (3994)
2603 - allow setting MDS UUID in .xml (2580)
2604 - print the stack of a process that LBUGs (4228)
2606 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2609 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2610 - make sure to unset replay flag from failed open requests (3440)
2611 - if default stripe count is 0, use OST count for inode size (3636)
2612 - update parent mtime/ctime on client for create/unlink (2611)
2613 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2614 - free recovery state on server during a forced cleanup (3571)
2615 - unregister_reply for resent reqs (3063)
2616 - loop back devices mounting and status check on 2.6 (3563)
2617 - fix resource-creation race that can provoke i_size == 0 (3513)
2618 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2619 - more debugging for page-accounting assertion (3746)
2620 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2621 - avoid deadlock after precreation failure (3758)
2622 - fix race and lock order deadlock in orphan handling (3450, 3750)
2623 - add validity checks when grabbing inodes from l_ast_data (3599)
2625 - add /proc/.../recovery_status to obdfilter (3428)
2626 - lightweight CDEBUG infrastructure, debug daemon (3668)
2627 - change default OSC RPC parameters to be better on small clusters
2628 - turn off OST read cache for files smaller than 32MB
2629 - install man pages and include them in rpms (3100)
2630 - add new init script for (un)mounting lustre filesystems (2593)
2631 - run chkconfig in %post for init scripts (3701)
2632 - drop scimac NAL (unmaintained)
2634 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
2637 - clean kiobufs before and after use (3485)
2638 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
2639 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
2640 - change init script to fail more clearly if not run as root (1528)
2641 - allow clients to reconnect during replay (1742)
2642 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
2643 - don't do DNS lookups on NIDs too small for IP addresses (3442)
2644 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
2645 - fix cond_resched (3554)
2646 - only evict unfinished clients after recovery (3515)
2647 - allow bulk resend, prevent data loss (3570)
2648 - dynamic ptlrpc request buffer allocation (2102)
2649 - don't allow unlinking open directory if it isn't empty (2904)
2650 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
2651 - remove extraneous obd dereference causing LASSERT failure (3334)
2652 - don't use get_cycles() when creating temp. files on the mds (3156)
2653 - hold i_sem when setting i_size in ll_extent_lock() (3564)
2654 - handle EEXIST for set-stripe, set proper directory name (3336)
2656 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
2657 - fix ksocknal_fmb_callback() error messages (2918)
2659 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
2662 - don't copy lvb into (possibly NULL) reply on error (2983)
2663 - don't deref dentry after dput, don't free lvb on error (2922)
2664 - use the kms to determine writeback rpc length (2947)
2665 - increment oti_logcookies when osc is inactive (2948)
2666 - update client's i_blocks count via lvb messages (2543)
2667 - handle intent open/close of special files properly (1557)
2668 - mount MDS with errors=remount-ro, like obdfilter (2009)
2669 - initialize lock handle to avoid ASSERT on error cleanup (3057)
2670 - don't use cancelling-locks' kms values (2947)
2671 - use highest lock extent for kms, not last one (2925)
2672 - don't dereference ERR_PTR() dentry in error handling path (3107)
2673 - fix thread race in portals_debug_dumplog() (3122)
2674 - create lprocfs device entries at setup instead of at attach (1519)
2675 - common AST error handler, don't evict client on completion race (3145)
2676 - zero nameidata in detach_mnt in 2.6 (3118)
2677 - verify d_inode after revalidate_special is valid in 2.6 (3116)
2678 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
2679 - initialize RPC timeout timer earlier for 2.6 (3219)
2680 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
2681 - print nal/nid for unknown nid (3258)
2682 - additional checks for oscc recovery before doing precreate (3284)
2683 - fix ll_extent_lock() error return code for 64-bit systems (3043)
2684 - don't crash in mdc_close for bad permissions on open (3285)
2685 - zero i_rdev for non-device files (3147)
2686 - clear page->private before handing to FS, better assertion (3119)
2687 - tune the read pipeline (3236)
2688 - fix incorrect decref of invalidated dentry (2350)
2689 - provide read-ahead stats and refine rpc in flight stats (3328)
2690 - don't hold journal transaction open across create RPC (3313)
2691 - update atime on MDS at close time (3265)
2692 - close LDAP connection when recovering to avoid server load (3315)
2693 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
2694 - don't leak open file on MDS after open resend (3325)
2695 - serialize filter_precreate and filter_destroy_precreated (3329)
2696 - loop device shouldn't call sync_dev() for nul device (3092)
2697 - clear page cache after eviction (2766)
2698 - resynchronize MDS->OST in background (2824)
2699 - refuse to mount the same filesystem twice on same mountpoint (3394)
2700 - allow llmount to create routes for mounting behind routers (3320)
2701 - push lock cancellation to blocking thread for glimpse ASTs (3409)
2702 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
2703 - fix rare problem with rename on htree directories (3417)
2705 - allow default OST striping configuration per directory (1414)
2706 - fix compilation for qswnal for 2.6 kernels (3125)
2707 - increase maximum number of MDS request buffers for large systems
2708 - change liblustreapi to be useful for external progs like lfsck (3098)
2709 - increase local configuration timeout for slow disks (3353)
2710 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
2712 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2715 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
2716 - don't overwrite extent policy data in reply if lock was blocked (2901)
2717 - drop filter export grants atomically with removal from device (2663)
2718 - del obd_self_export from work_list in class_disconnect_exports (2908)
2719 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
2720 - swab reply message in mdc_close, other PPC fixes (2464)
2721 - fix destroying of named logs (2325)
2722 - overwrite old logs when running lconf --write_conf (2264)
2723 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
2724 - fix race in target_handle_connect (2898)
2725 - mds_reint_create() should take same inode create lock (2926)
2726 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
2727 - don't close files for self_export to avoid uninitialized obd (2936)
2728 - allow MDS with the same name as client node (2939)
2729 - hold dentry reference for closed log files for unlink (2325)
2730 - reserve space for all logs during transactions (2059)
2731 - don't evict page beyond end of stripe extent (2925)
2732 - don't oops on a deleted current working directory (2399)
2733 - handle hard links to targets without a parent properly (2517)
2734 - don't dereference NULL lock when racing during eviction (2867)
2735 - don't grow lock extents when lots of conflicting locks (2919)
2737 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
2740 - account for cache space usage on clients to avoid data loss (974)
2741 - lfsck support in lustre kernel code (2349)
2742 - reduce journal credits needed for BRW writes (2370)
2743 - orphan handling to avoid losing space on client/server crashes
2744 - ptlrpcd can be blocked, stopping ALL progress (2477)
2745 - use lock value blocks to assist in proper KMS, faster stat (1021)
2746 - takes i_sem instead of DLM locks internally on obdfilter (2720)
2747 - recovery for initial connections (2355)
2748 - fixes for mds_cleanup_orphans (1934)
2749 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
2750 - block all file creations until orphan recovery completes (1901)
2751 - client remove rq_connection from request struct (2423)
2752 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
2753 - recovery timer race (2670)
2754 - mdc_close recovey bug (2532)
2755 - ptlrpc cleanup bug (2710)
2756 - mds timeout on local locks (2588)
2757 - namespace lock held during RPCs (2431)
2758 - handle interrupted sync write properly (2503)
2759 - don't try to handle a message that hasn't been replied to (2699)
2760 - client assert failure during cleanup after abort recovery (2701)
2761 - leak mdc device after failed mount (2712)
2762 - ptlrpc_check_set allows timedout requests to complete (2714)
2763 - wait for inflight reqs when ptlrpcd finishes (2710)
2764 - make sure unregistered services are removed from the srv_list
2765 - reset bulk XID's when resending them (caught by 1138 test)
2766 - unregister_bulk after timeout
2767 - fix lconf error (2694)
2768 - handle write after unfinished setstripe, stripe-only getstripe (2388)
2769 - readahead locks pages, leaves pending causing memory pressure (2673)
2770 - increase OST request buffers to 4096 on large machines (2729)
2771 - fix up permission of existing directories in simple_mkdir (2661)
2772 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
2773 - don't assign transno to errored transactions (2742)
2774 - don't delete objects on OST if given a bogus objid from MDS (2751)
2775 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
2776 - if rq_no_resend, then timeout request after recovery (2432)
2777 - fix MDS llog_logid record size, 64-bit array alignment (2733)
2778 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
2779 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
2780 - MDS recovery shouldn't delete objects that it has given out (2730)
2781 - if enqueue arrives after completion, don't clobber LVB (2819)
2782 - don't unlock pages twice when trigger_group_io returns error (2814)
2783 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
2784 - don't write pages to disk if there was an error (1450)
2785 - don't ping imports that have recovery disabled (2676)
2786 - take buffered bytes into account when balancing socknal conn (2817)
2787 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
2788 - reconnect unlink llog connection after MDS reconnects to OST (2816)
2789 - remove little-endian swabbing of llog records (1987)
2790 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
2791 - retry reposting request buffers if they fail (1191)
2792 - grow extent at grant time to avoid granting a revoked lock (2809)
2793 - lock revoke doesn't evict page if covered by a second lock (2765)
2794 - disable VM readahead to avoid reading outside lock extents (2805)
2796 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
2797 - updated kernel patches for hp-2.4.20 kernel (2681)
2799 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
2802 - fix truncated write corruption (2366)
2803 - fix for failed assertion in iopen_connect_dentry (1792,2517)
2805 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
2806 - ensure len(uuid) < 37 in lmc (1171)
2807 - fix ia64 OOPS in llog_test (2255)
2808 - zero end of page at obdfilter for partial page writes (2648)
2809 - don't leave stale dentries around after renames (bug 2428)
2810 - fix timeouts when evicting a client with a single lock held (2642)
2811 - set deadline for the initial HELLO message to drain (2634)
2812 - print out dotted-quad IP addresses in the socknal (2302)
2814 - additional debugging for MDS client eviction problem (2443)
2815 - fix mkfsoptions support for osts (2603, 2604)
2817 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
2820 - add series for the vanilla 2.6.0 kernel
2821 - add series for the vanilla 2.4.24 kernel
2822 - add series for a cray x86/64 UL kernel drop
2823 - fix xattr patches for the vanilla 2.4.19 series
2825 - generate true UUIDs in lmc (1171)
2826 - have portals stack dumping break in UML (2466)
2827 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
2828 - allocate with _NOFS in ldlm to avoid deadlock (1933)
2829 - wake callback waiting threads on client eviction (2460)
2830 - Add --ptldebug and --subsystem to lmc (1719)
2831 - update assertion to allow safe interrupt allocation
2832 - set rq_no_resend for cancel requests (2432)
2833 - recalculate ptlrpcd timeout after resend (2494)
2834 - call vfs_rmdir when removing pending directories (2368)
2835 - fix renaming a file to itself (2429)
2836 - lmc creates a default one-stripe lov (2454)
2837 - expand procfs space to handle large clusters (2326)
2838 - increase UML stack to avoid overflow
2839 - update lconf's list of debug and subsystem masks
2840 - fix lfs find --obd (2510)
2841 - /proc tunable for disabling filter read caching (2591)
2842 - stop rpm packages from altering slapd.conf (2301)
2843 - disable nagle in the socknal under 0conf (2578)
2844 - choose mds inode size based on stripe count (2572)
2845 - fix kernel-source rpm problems (2516)
2847 - add --disable-doc to avoid pdf generation (2421)
2848 - update documentation, tests, type-os, comments
2849 - avoid format warnings on ia64
2850 - remove the TOE NAL
2851 - tiny code cleanups by removing unused fields
2853 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
2856 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
2857 - remove the most blatant lies from BUILDING file (2371)
2858 - change default debug level to reasonable production setting
2859 - reduce client side cache size to reduce cache flush time
2860 - reduce max RPCs in flight to avoid unnecessary file fragmentation
2861 - make TCP zerocopy and pinger support enabled by default (2476)
2862 - sync writes completed after process exits caused crashes (2319)
2863 - maintain correct mount count on the MDS (2356)
2864 - backout 1557, because 2316 wasn't really fixed
2865 - better file I/O statistics gathering in /proc
2866 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
2867 - another kernel patch to fix zero-copy TCP function export
2868 - don't take duplicate lock when processing re-sent getattr (2420)
2869 - lctl uses obd_self_export instead of creating new conn (2353)
2870 - MDS/OST recovery case which requires object creation asserted (2425)
2871 - move lfs from /usr/sbin to /usr/bin in packages
2872 - fix race between mds_client_add and mds_client_free (2417)
2873 - use kmalloc instead of slabs in portals (2430)
2874 - don't create duplicate records when a failover MDS is present (2442)
2875 - remove unnecessary mount age check (2332)
2876 - don't remove directory inodes from locks prematurely (2451)
2877 - don't break if MDS service name is the same as hostname (2103)
2878 - fix races in client write RPC generation when cache full (2482)
2880 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
2883 - remove now-unused request->rq_obd (278)
2884 - if an allocation fails, print out how much memory we've used (1933)
2885 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
2886 - add the "configurable stack size" patch to most series files (1256)
2887 - ability to write large log records, for 100+ OST configs (2306)
2888 - fix NULL deref when filter_prep fails (2314)
2889 - fix operator precedence error in filter_sync
2890 - dynamic allocation of socknal TX descriptors (2315)
2891 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
2892 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
2893 - ensure that CWARN messages in Portals always get to the syslog
2894 - __init/__exit are not for prototype decls (ldlm_init/exit)
2895 - x86-64 compile warning fixes
2896 - fix gateway LMC keyword conflict (2318)
2897 - fix MDS lock inversions in getattr/reint paths (1844)
2898 - fix a rare lock re-ordering bug, which caused deadlock (2322)
2899 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
2900 - DLM race condition prevented some lock evictions (2328)
2901 - ENOMEM detection and retry on socknal sends (2230)
2902 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
2903 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
2904 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
2905 - better error messages when a client is rejected during recovery (1505)
2906 - avoid cancelling locks which were never granted, after failure (2330)
2907 - fix i_sem/journal inversion in mds_client_add (2333)
2908 - fix truncate/getattr lock cycle deadlock (2334)
2909 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
2910 - fix two rare exit paths which could leak an l_lock() ref (2321)
2911 - fencepost error in MDS/OST orphan recovery (2226)
2912 - make log record alignment 8 bytes (1988)
2913 - lstripe now fails when requested offset > ost_count (2237)
2914 - ensure that all kernel series have a complete list.h (1607)
2915 - fix crashes in special-file operations (2316)
2916 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
2918 - allow configurable automake binary, for testing new versions
2919 - small update to the lfs documentation
2921 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
2923 * fix negative export reference count in fsfilt_sync (2312)
2925 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
2926 * release candidate 0.9.1
2928 - orphans are moved into the PENDING directory for possible recovery
2929 - replayed opens now open by fid for orphan/rename safety (1042)
2930 - last close of an orphan inode generates a transno (683)
2931 - chdir() and mount() now pin the directory entry (1020)
2932 - avoid CERROR in normal ll_setattr_raw() error case (1500)
2933 - discard very old requests without processing them (1502)
2934 - remove some common, well-understood CERRORs (1505)
2935 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
2936 - clear "grant" flags in OST replies until OST grant code lands (1644)
2937 - fix read performance by not clobbering i_blksize on client (1598)
2938 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
2939 - make LRU size a /proc tunable, clears locks when reduced (707)
2940 - fix some lprocfs rot that prevented ptlbd from loading (1732)
2941 - server locks take references on exports now (1558)
2942 - build fixes for 2.4.20-rh trees (1663)
2943 - return an error from lov_create if all OSCs are inactive (1751)
2944 - fix import levels when a reconnect happens without a timeout (1597)
2945 - exit early from mds_open if we get a lookup error (1749)
2946 - partial page read at EOF wouldn't wait for disk before sending (1642)
2947 - avoid NULL deref in obdfilter when reading page past EOF (1592)
2948 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
2949 - fix LBUG in ll_it_open_error with rc = -2 (1861)
2950 - write/truncate lock inversion (1639)
2951 - Don't auto-load obdclass, portals modules during cleanup (1495)
2952 - fix timestamps from jumping to "now" (1763)
2953 - extra journal assertions (1648)
2954 - add an extra multiunlink test (1771)
2955 - fix read_record/write_record API (1776)
2956 - fix leak of offset_extent, possible incorrect i_size later (1772)
2957 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
2958 - Debugging for the kqswnal_get_idle_tx problems (1820)
2959 - Allow recovery to be attempted multiple times (1536)
2960 - Write out MDS last_rcvd file after it is first created (1600)
2961 - Fix tx_descriptor leak in failed transmit situations (1827)
2962 - ext3 journaling fixes for assertion failure after IO error (1871)
2963 - class_export_put() on freed export after completion AST error (1896)
2964 - Fix revalidate looping in VFS (1322)
2965 - Don't access a freed export during MDS_REINT timeout (1521)
2966 - Add open-unlink recovery support on the MDS (1673,1764)
2967 - Return an error if no MDS data was read from last_rcvd (1946)
2968 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
2969 - Don't setup a disk device that doesn't match exported UUID (317)
2970 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
2971 - avoid committing NULL handle in force close
2972 - local.sh is now a one-stripe LOV configuration
2973 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
2974 - fix ext3 htree duplicate directory entry corruption (1516)
2975 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
2976 - update to new LOV EA format (2097)
2977 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
2979 2003-06-15 Phil Schwan <phil@clusterfs.com>
2982 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
2983 - per-import/export recovery handling (958, 931, 959)
2984 - multiple last-rcvd slots, for serving multiple FSes (949)
2985 - connections are again shared between multiple imp/exports (963, 964)
2986 - "umount -f" would hang if any requests needed to be sent (393, 978)
2987 - avoid pinning large req buffer by copying for queued messages (989)
2988 - add "uuid" to "lctl device" command to help upcalls (991)
2989 - "open" RPCs with transnos would confuse recovery counters (1037)
2990 - do proper endian conversion of all wire messages (288, 340, 891)
2991 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
2992 - call waiting locks callback from LDLM recovery thread (1127, 1151)
2993 - fix ptlrpc_connection leak in target_handle_connect (1174)
2994 - fix import refcounting bug in OST and MDS cleanup (1134)
2995 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
2996 - fix very unlikely obd_types race condition (501)
2997 - remove osc_open hack for echo_client (1187)
2998 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
2999 - a failure in read_inode2 leads to deadlock (1139)
3000 - cancel ack-locks as soon as transaction is committed (1072)
3001 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3002 - make sure to commitrw after any preprw to avoid deadlock (1162)
3003 - failing to execute a file in a lustre FS would lock inode (1203)
3004 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3005 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3006 - obd_uuid2tgt would walk off the end of the list (1255)
3007 - on IA64 the osc would give portals incorrect bulk size (1258)
3008 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3009 - fix lock inversion caused by new llite matching code (1282)
3010 - limit the number of dirty pages on a client to 10MB (1286)
3011 - timed out locks were not being corrected cancelled (1289)
3012 - fix O_DIRECT above 4GB on IA-32 (1292)
3013 * major user-visible changes
3014 - fail out/fail over policy now controlled by the upcall (993)
3016 - add OBD_PING to check server availability and failure (954)
3017 - lustre messages are now sent in sending host order (288, 340, 891)
3018 - add eadatalen to MDS getattr reply (340)
3019 - OST read replies may contain second buffer, with per-page status (593)
3021 2003-03-11 Phil Schwan <phil@clusterfs.com>
3024 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3025 - failed open()s could cause deadlock; fixed (867, 869)
3026 - stop cancelling OST locks when files are closed (481)
3027 - overlapping XID spaces caused network corruption (851, 853)
3028 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3029 - setattr_raw added, to do single-RPC, server-side setattrs
3030 - lmc/lconf syntax change for OST UUIDs
3031 - fix crashy race condition between ptlrpc_free_req and osc_close
3032 - don't use request in mdc_enqueue if we hit a timeout (889)
3033 - don't set the inode i_size for regular files from the MDS (896)
3034 - handle out of order completion AST (842)
3035 - don't LBUG if a lock request times out after receiving AST (913)
3036 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3037 - if a bad lock AST arrives, send an error instead of dropping entirely
3038 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3039 - fix leak in bulk IO when only partially completed (899, 900, 926)
3040 - fix O_DIRECT for ia64 (55)
3041 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3042 - C-z after timeout could hang a process forever; fixed (977)
3044 - client-side I/O cache (678, 924, 929, 941, 970)
3046 - READPAGE and SETATTRs which don't take server-side locks get
3049 2003-02-11 Phil Schwan <phil@clusterfs.com>
3052 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3053 - Fix fsfilt_extN_readpage() to read a full page of directory
3054 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3055 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3056 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3057 - Most intent operations were removed, in favour of a new RPC mode
3058 that does a single RPC to the server and bypasses most of the VFS
3059 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3060 - Aggressively cancel local locks on DLM servers
3061 - mds_reint_unlink sends EA to the client if it's the last nlink.
3062 client uses that EA to unlink OST objects.
3063 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3064 - recursive symlinks were fixed (439)
3065 - fixed NULL deref in DEBUG_REQ
3066 - filter_update_lastobjid no longer calls sync, which annoyed extN
3067 - fixed multi-client small-writes to a single file problem (445)
3068 - fixed mtime updates during file writes (607)
3069 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3070 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3071 - fixed Linux OST size reporting problem (444, 656)
3072 - OST now updates object mtime with writes or setattr (607, 619)
3073 - client verifies file size before zeroing page past EOF (445)
3074 - OST now writes last allocated objid to disk with allocation (108)
3075 - LOV on echo now works (409)
3077 - mds_reint_unlink sends a new buffer, with the EA included. this
3078 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3079 set if a regular file was being unlinked, and it was the last link
3080 - use PtlGet from the target for bulk writes (315)
3081 - OST now updates object mtime with writes or setattr (607, 619)
3082 - LDLM now has a grant-time callback to revalidate locked items, if
3084 - Many MDS operations were reorganized to combat race conditions
3086 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3087 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3089 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3092 - Fully reactivate OST imports after reconnection (512, others)
3093 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3094 - More graceful error handling for truncating on dead OST (515)
3095 - Don't error out unless we're actually accessing dead stripes (474)
3096 - Fix garbage sizes when stripes are missing (410)
3097 - LRU counters were broken, causing constant lock purge (433, 432)
3098 - garbage on read from stripes with failed OSTs (441)
3099 - mark OSCs as active before reconnecting during recovery (438)
3100 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3101 - lfind did not preserve OST order in output (443)
3102 - symlinks cause hung clients, incorrect data (439)
3103 - stop dereferencing request after dropping refcount (457)
3104 - don't LASSERT(spin_is_locked) on non-SMP (455)
3105 - fixes for many rename() bugs
3106 - fstat didn't correctly synchronize attributes (399)
3107 - server must handle lock cancellation during blocking AST prep (487)
3108 - bulk descriptors were free()d too soon (511)
3109 - fix paths in lconf, which would load incorrect modules (451, 507)
3110 - fix confusing lconf 'host not found' error message (386)
3111 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3112 - fix race condition in mdc_blocking_ast() for inode access (526)
3113 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3114 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3115 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3116 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3117 - fix lstripe in several ways (kernel side) (527)
3118 - fix request leak in ldlm_cli_enqueue (262)
3119 - incorrect OSC was marked inactive after OST failure
3120 - call mds_fs_cleanup before unmounting filesystem (524)
3121 - fix races between taking ns_lock and ldlm_lock_change_resource
3122 - fix races updating LOV export open file list
3123 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3124 - fix recovery NULL deref in ldlm_cli_cancel_unused
3125 - fix some DLM races by using new hash table for lock handles (419)
3126 - permit the client to specify desired inodes, at replay
3127 - duplicate requests when we queue them for replay reintegration
3128 - fix last_rcvd offset calculation
3129 - sync after each recovered transaction, so we always make progress
3130 - never, not always, ERESTART requests without transnos
3131 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3133 - skip replay if the MDS says that the client is already connected
3134 - don't check for a recovery-enabled export to match lctl's UUID
3135 - don't INC_USE_COUNT for phantom exports
3136 - don't crash when cleaning up phantom exports (567)
3137 - don't double-finish or set replay data for errored mdc_open requests
3138 - abort requests when they time out, so we don't get old replies
3139 - send/receive replies for AST messages again
3140 - if the client says that it doesn't have the lock, cancel it on the
3142 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3143 mark it as destroyed, it will all work out when decref is called
3144 - fix module use counts (22, 581)
3146 - ASTs now expect a reply (server cancels lock on error reply)
3148 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3151 - fix many simultaneous client startup (392)
3152 - fix dentry->d_it clobbering
3153 - credentials weren't being shipped for readdir/getattr operations
3154 - remove invalid assertions triggered during some concurrent MD
3156 - proper Lustre versions added (336, 389)
3157 - fix memory leak for create error case (398)
3158 - fix LOV locking bug that would get cli/srv out of sync
3159 - fix echo client over LOV (409)
3160 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3161 - fix double-O_EXCL intent crash (424)
3162 - avoid sending multiple lock CANCELs (352)
3164 - MDS can do multi-client recovery (modulo bugs in new code)
3166 - many updates, edits, cleanups
3168 2002-11-18 Phil Schwan <phil@clusterfs.com>
3171 - fix null d_it dereference (346)
3172 - fix full OST/dbench hang (333)
3173 - fix permission problem with file removal (286)
3174 - fix removal of OSCs from LOV when they fail
3175 - fix NULL deref during bulk timeout (214)
3176 - fix problems related to multiple filesystems on one MDS (241)
3177 - fixed serious subtle metadata locking bugs
3178 - free locks on clients when inodes are removed due to memory
3180 - fix inode pointer in lock data (285)
3181 - partial support for multiple MDS on a single host (241)
3182 - data locks weren't cancelled at clear_inode time (290, 311)
3183 - intent locks could lead to unbounded lock growth (205)
3184 - added a maximum lock count, an LRU list, and a flusher
3185 - fix multiple rename (365)
3186 - properly abstracted the echo client
3187 - OSC locked 1 byte too many; fixed
3188 - rewrote brw callback code:
3189 - fixed recovery bugs related to LOVs (306)
3190 - fixed too-many-pages-in-one-write crash (191)
3191 - fixed (again) crash in sync_io_timeout (214)
3192 - probably fixed callback-related race (385)
3194 - Add capability to MDS protocol
3195 - LDLM cancellations and callbacks on different portals
3197 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3200 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3201 - defer open object destruction to close time (601981, 138)
3202 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3203 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3204 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3205 - don't call obd_setattr in ll_file_release for destroyed objects
3206 * protocol change to lustre_msg: move |version| and add |flags|
3207 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3208 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3209 * added replay of create, unlink, link and rename operations during
3210 MDS failover; recovery should be much more robust now
3211 * remove failed OSCs from LOVs (only lov_create uses this so far)
3212 * the lustre-HOWTO was brought (more) up to date (582544)
3214 2002-10-23 Phil Schwan <phil@clusterfs.com>
3217 - in-use dentries weren't being reused properly (617851)
3218 - prevent multiple LDLM setup (599178)
3219 - fix LOV size calculations for truncate (617853)
3220 - fix client handling of MDS intent errors (POSIX)
3221 - fix permission bug in lovstripe.c test (624321)
3222 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3223 - truncate past end of file could corrupt data
3224 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3225 - a race in recovery could return ETIMEDOUT to apps (623947)
3226 - building outside the source directory was fixed
3227 * the lustre-HOWTO was brought (more) up to date (582544)
3228 * major progress was made on recovery functionality
3230 2002-10-10 Phil Schwan <phil@clusterfs.com>
3233 - recovery deadlock fix
3234 - rm -rf causes LBUG fix (617817)
3235 - file open by multiple tasks fix (618962)
3236 - directory permissions bugs (602707 and 620007)
3237 - journal_stop fixed with locking (611313)
3238 - O_APPEND failures resolved (618273, perhaps 614459)
3239 - lconf PATH fix (619770)
3240 - IA64 build fix (621450)
3241 - RPC buffer sizes scale with amount of memory
3243 2002-10-01 Phil Schwan <phil@clusterfs.com>
3246 - locks would be cancelled without throwing away data pages,
3247 resulting in inconsistent data (605627)
3248 - inode attributes were not always being refreshed (605627, 612449)
3249 - lconf now continues to cleanup after lctl reports an error
3250 - MDS now enforces user permissions (602707)
3251 - lprocfs cleanup fixed, but not yet enabled (614157)
3252 - fixed infinite server hang, should a client not respond to an AST
3253 - avoid going into recovery if user calls readlink() with a buffer
3254 that's too small (613941)
3255 - AST RPCs no longer require replies (614867) -- this may be changed
3256 - don't crash server if client sends an IOV that's too big (611336)
3257 - fixed lock conversion deadlock (611892)
3258 - fixed the following of symlinks (614622)
3259 * recovery: the server can remove locks from a client that dies, other
3260 clients can make progress
3261 * more extN patch fixes
3262 * compile-time configurable ptlrpc buffer allocations
3264 - collaborative read cache document
3265 - Lustre Lite Performance CDR document-in-progress
3267 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3270 - fix typo in patch-2.4.18
3272 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3275 - clear ptlrpc request each time in handle_incoming_request()
3276 - unlink of files now destroys the object on the OST
3278 2002-09-19 Peter Braam <braam@clusterfs.com>
3280 * add hard link support
3281 * change obdfile creation method
3282 * kernel patch changed
3284 2002-09-19 Peter Braam <braam@clusterfs.com>
3287 - stack overflow bug in extN fixed
3289 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3291 * documentation updates
3292 - add man pages for config tools
3293 - update tests/README to describe testing with new config tools
3294 - finish metadata API descriptions
3295 * bug fixes and cleanups
3296 - statfs workaround for 16TB limit
3297 - LOV stripe allocation improved, can stripe on subset of OSTs
3298 - LOV file size/IO offset was wrong for files > 4GB in size
3299 - object EA data was being dropped, caused files to be unreadable
3300 - memory overflow with non-LOV OST caused memory corruption
3301 - fixed regression tests to work with new config tools, obdfilter
3302 - fixed bug when directory size became larger than 1 block
3303 - fixed bug (for single client case) when PWD was deleted
3304 - invalidate local directory pages when doing intent-based ops
3305 - avoid LDLM oops when lock callback contained bad data
3307 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3309 * documentation updates
3310 * bug fixes and cleanups
3311 - configuration tools
3314 - 64-bit compile warnings
3315 - 64-bit internal statfs data
3317 * test_brw on persistent OST devices
3319 * lprocfs (disabled)
3321 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3323 * documentation updates
3324 * bug fixes and cleanups
3325 * configuration tools
3327 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3330 * many small fixes to 0_5_4
3331 * io/network handling
3332 * thinkos in MDS operations
3334 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3337 * crucial basic fixes to 0.5.3
3338 * IOR, Iozone work over Elan
3341 2002-08-07 Phil Schwan <phil@clusterfs.com>
3342 * version 0_5_3, our first alpha
3343 * we use the new Portals iovs
3344 * documentation updates
3345 * bug fixes and cleanups
3346 * small changes in the DLM wire protocol
3348 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3349 * version 0_5_1 with some initial stability,
3350 * locking on MD and file I/O.
3351 * documentation updates
3352 * several bug fixes since 0.5.0
3353 * small changes in wire protocol
3355 2002-07-18 Phil Schwan <phil@clusterfs.com>
3357 * delivered as Lustre Light Alpha
3358 * fixed a crash after handling invalid MDS requests
3359 * fixed directory pages for architectures with non-4k pages sizes
3361 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3362 * release version v0_4_4
3363 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3365 * Fixes a number of interruption problems with OST operations.
3366 * Update documentation for portals header changes
3367 * Move all wire protocol structs/defines to lustre_idl.h
3368 * Fixes symlink length bug.
3369 * Add tcpdump to repository.
3371 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3372 * release version v0_4_3
3373 * Fixes statfs for inodes on extN.
3374 * Fixes bug in runtests which would delete /etc/hosts.
3375 * Use 64-bit object IDs wherever possible (not into VFS though)
3376 Remove ost_get_info, which is unused by lustre, and out of date.
3378 2002-07-03 Peter Braam <braam@clusterfs.com>
3379 * release version v0_4_2 Fixes a lookup error (type not passed)
3380 * move forward to head of Portals
3381 * move forward to latest Lustre kernel
3383 2002-06-25 Peter Braam <braam@clusterfs.com>
3384 * release version v0_4_1. Hopefully stable on single node use.