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.
34 Severity : enhancement
36 Description: Move random uuid functions to prng.c.
39 Frequency : during server recovery
41 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
42 Details : request with rq_no_resend flag not awake l_wait_event if they get a
46 Frequency : nfs export on patchless client
48 Description: connectathon hang when test nfs export over patchless client
49 Details : Disconnected dentry cannot be found with lookup, so we do not need
50 to unhash it or make it invalid
52 --------------------------------------------------------------------------------
54 tbd Cluster File Systems, Inc. <info@clusterfs.com>
56 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
57 INTEROPERATE with older versions automatically. In many cases a
58 special upgrade step is needed. Please read the
59 user documentation before upgrading any part of a 1.4.x system.
60 * WARNING: Lustre configuration and startup changes are required with
61 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
63 * Support for kernels:
64 2.6.9-42.0.10.EL (RHEL 4)
66 2.4.21-47.0.1.EL (RHEL 3)
67 2.6.12.6 vanilla (kernel.org)
68 2.6.16.27-0.9 (SLES 10)
69 * Client support for unpatched kernels:
70 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
71 2.6.16 - 2.6.19 vanilla (kernel.org)
72 2.6.9-42.0.8EL (RHEL 4)
73 * Recommended e2fsprogs version: 1.39.cfs5
75 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
78 Severity : enhancement
80 Description: MountConf
81 Details : Lustre configuration is now managed via mkfs and mount
82 commands instead of lmc and lconf. New obd types (MGS, MGC)
83 are added for dynamic configuration management. See
84 https://mail.clusterfs.com/wikis/lustre/MountConf for
87 Severity : enhancement
89 Description: dynamic OST addition
90 Details : OSTs can now be added to a live filesystem
92 Severity : enhancement
94 Description: startup order invariance
95 Details : MDTs and OSTs can be started in any order. Clients only
96 require the MDT to complete startup.
98 Severity : enhancement
100 Description: parallel, asynchronous orphan cleanup
101 Details : orphan cleanup is now performed in separate threads for each
102 OST, allowing parallel non-blocking operation.
104 Severity : enhancement
106 Description: optimized stripe assignment
107 Details : stripe assignments are now made based on ost space available,
108 ost previous usage, and OSS previous usage, in order to try
109 to optimize storage space and networking resources.
111 Severity : enhancement
113 Description: Permanently set tunables
114 Details : All writable /proc/fs/lustre tunables can now be permanently
115 set on a per-server basis, at mkfs time or on a live
118 Severity : enhancement
120 Description: Lustre message v2
121 Details : Add lustre message format v2.
123 Severity : enhancement
125 Description: client OST exclusion list
126 Details : Clients can be started with a list of OSTs that should be
127 declared "inactive" for known non-responsive OSTs.
130 Frequency : SFS test only (otherwise harmless)
132 Description: SPEC SFS validation failure on NFS v2 over lustre.
133 Details : Changes the blocksize for regular files to be 2x RPC size,
134 and not depend on stripe size.
136 Severity : enhancement
138 Description: Multiple MD RPCs in flight.
139 Details : Further unserialise some read-only MDT RPCs - learn about intents.
140 To avoid overly-overloading MDT, introduce a limit on number of
141 MDT RPCs in flight for a single client and add /proc controls
142 to adjust this limit.
144 Severity : enhancement
146 Description: client read/write statistics
147 Details : Add client read/write call usage stats for performance
148 analysis of user processes.
149 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
150 file access. extents_stats shows chunk size distribution.
151 extents_stats_per_process show chunk size distribution per
154 Severity : enhancement
156 Description: per-client statistics on server
157 Details : Add ldlm and operations statistics for each client in
158 /proc/fs/lustre/mds|obdfilter/*/exports/
160 Severity : enhancement
162 Description: improved MDT statistics
163 Details : Add detailed MDT operations statistics in
164 /proc/fs/lustre/mds/*/stats
166 Severity : enhancement
168 Description: VFS operations stats
169 Details : Add client VFS call stats, trackable by pid, ppid, or gid
170 /proc/fs/lustre/llite/*/vfs_ops_stats
171 /proc/fs/lustre/llite/*/vfs_track_[pid|ppid|gid]
176 Description: Fix client-side osc byte counters
177 Details : The osc read/write byte counters in
178 /proc/fs/lustre/osc/*/stats are now working
181 Frequency : always as root on SLES
183 Description: Failure of copying files with lustre special EAs.
184 Details : Client side always return success for setxattr call for lustre
185 special xattr (currently only "trusted.lov").
190 Description: Refcount LNET uuids
191 Details : The global LNET uuid list grew linearly with every startup;
192 refcount repeated list entries instead of always adding to
195 Severity : enhancement
197 Description: Dynamic service threads
198 Details : Within a small range, start extra service threads
199 automatically when the request queue builds up.
202 Frequency : mixed-endian client/server environments
204 Description: mixed-endian crashes
205 Details : The new msg_v2 system had some failures in mixed-endian
208 Severity : enhancement
210 Description: Easy OST removal
211 Details : OSTs can be permanently deactivated with e.g. 'lctl
212 conf_param lustre-OST0001.osc.active=0'
214 Severity : enhancement
216 Description: MGS proc entries
217 Details : Added basic proc entries for the MGS showing what filesystems
220 Severity : enhancement
222 Description: provide MGS failover
223 Details : Added config lock reacquisition after MGS server failover.
225 Severity : enhancement
227 Description: add Linux 2.4 support
228 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
232 Description: a large application tries to do I/O to the same resource and dies
234 Details : Check the req->rq_arrival time after the call to
235 ost_brw_lock_get(), but before we do anything about
236 processing it & sending the BULK transfer request. This
237 should help move old stale pending locks off the queue as
238 quickly as obd_timeout.
241 Frequency : when an incorrect nid is specified during startup
243 Description: ptlrpc connect to non-existant node causes kernel crash
244 Details : LNET can't be re-entered from an event callback, which
245 happened when we expire a message after the export has been
246 cleaned up. Instead, hand the zombie cleanup off to another
249 Severity : enhancement
251 Description: plain/inodebits lock performance improvement
252 Details : Grouping plain/inodebits in granted list by their request modes
253 and bits policy, thus improving the performance of search through
257 Frequency : only for kernels with patches from Lustre below 1.4.3
259 Description: Remove old rdonly API
260 Details : Remove old rdonly API which unsed from at least lustre 1.4.3
263 Frequency : only for devices with external journals
265 Description: Set external device read-only also
266 Details : During a commanded failover stop, we set the disk device
267 read-only while the server shuts down. We now also set any
268 external journal device read-only at the same time.
271 Frequency : when upgrading from 1.4 while trying to change parameters
273 Description: The wrong (new) MDC name was used when setting parameters for
274 upgraded MDT's. Also allows changing of OSC (and MDC)
275 parameters if --writeconf is specified at tunefs upgrade time.
278 Frequency : when setting specific ost indicies
280 Description: QOS code breaks on skipped indicies
281 Details : Add checks for missing OST indicies in the QOS code, so OSTs
282 created with --index need not be sequential.
284 ------------------------------------------------------------------------------
286 TBD Cluster File Systems, Inc. <info@clusterfs.com>
288 * Support for kernels:
289 2.6.16.21-0.8 (SLES10)
290 2.6.9-42.0.8EL (RHEL 4)
292 2.4.21-47.0.1.EL (RHEL 3)
293 2.6.12.6 vanilla (kernel.org)
294 * Recommended e2fsprogs version: 1.39.cfs2-0
297 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
299 Description: System hang on startup
300 Details : This bug allowed the liblustre (e.g. catamount) client to
301 return to the app before handling all startup RPCs. This
302 could leave the node unresponsive to lustre network traffic
303 and manifested as a server ptllnd timeout.
305 Severity : enhancement
307 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
308 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
309 boolean to control whether to print peer debug info when a
310 client's RPC times out.
315 Description: make O_SYNC working on 2.6 kernels
316 Details : 2.6 kernels use different method for mark pages for write,
317 so need add a code to lustre for O_SYNC work.
322 Description: Failure to close file and release space on NFS
323 Details : Put inode details into lock acquired in ll_intent_file_open.
324 Use mdc_intent_lock in ll_intent_open to properly
325 detect all kind of errors unhandled by mdc_enqueue
330 Description: proc file read during shutdown sometimes raced obd removal,
332 Details : Add lock to prevent obd access after proc file removal.
335 Frequency : Only for files larger than 4GB on 32-bit clients.
337 Description: improperly doing page alignment of locks
338 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
339 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
344 Description: RPCs being resent when they shouldn't be
345 Details : Some RPCs that should not be resent are being resent. This
346 can cause inconsistencies in the RPC state machine. Do not
347 resend such requests.
350 Frequency : rare, only with NFS export
352 Description: Crash on NFS re-export node
353 Details : under very unusual load conditions an assertion is hit in
354 ll_intent_file_open()
357 Frequency : only if OST filesystem is corrupted
359 Description: client incorrectly hits assertion in ptlrpc_replay_req()
360 Details : for a short time RPCs with bulk IO are in the replay list,
361 but replay of bulk IOs is unimplemented. If the OST filesystem
362 is corrupted due to disk cache incoherency and then replay is
363 started it is possible to trip an assertion. Avoid putting
364 committed RPCs into the replay list at all to avoid this issue.
369 Description: large O_DIRECT requests fail under memory pressure/fragmentation
370 Details : Large single O_DIRECT read and write calls can fail to allocate
371 a sufficiently large buffer to process the request. In case of
372 allocation failure the allocation is retried with a smaller
373 buffer and broken into smaller requests.
378 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
379 > 32000 subdirectories
380 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
381 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
382 subdirectory count crosses 32000. This will aid e2fsck to
383 correctly handle more than 32000 subdirectories.
388 Description: versioning check is incomplete
389 Details : Checking the version difference of client vs. server, report
390 error if the gap is too big.
395 Description: add support PG_writeback bit
396 Details : add support for PG_writeback bit for Lustre, for more carefull
397 work with page cache in 2.6 kernel. This also fix some deadlocks
398 and remove hack for work O_SYNC with 2.6 kernel.
400 ------------------------------------------------------------------------------
402 2006-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
404 * Support for kernels:
405 2.6.16.21-0.8 (SLES10)
406 2.6.9-42.0.3EL (RHEL 4)
408 2.4.21-47.0.1.EL (RHEL 3)
409 2.6.12.6 vanilla (kernel.org)
412 * The backwards-compatible /proc/sys/portals symlink has been removed
413 in this release. Before upgrading, please ensure that you change
414 any configuration scripts or /etc/sysctl.conf files that access
415 /proc/sys/portals/* or sysctl portals.* to use the corresponding
416 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
417 in advance of the upgrade on any system running Lustre 1.4.6 or
418 newer, since /proc/sys/lnet was added in that version.
419 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
423 Frequency : MDS failover only, very rarely
425 Description: "went back in time" messages on mds failover
426 Details : The greatest transno may be lost when the current operation
427 finishes with an error (transno==0) and the client's last_rcvd
428 record is over-written. Save the greatest transno in the
429 mds_last_transno for this case.
432 Frequency : always for specific kernels and striping counts
434 Description: client may get "Matching packet too big" without ACL support
435 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
436 when trying to access files in certain configurations. The
437 clients should in fact be denied when mounting because they do
441 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
443 Description: many clients connecting with IO in progress causes connect timeouts
444 Details : Avoid synchronous journal commits to avoid delays caused by many
445 clients connecting/disconnecting when bulk IO is in progress.
446 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
447 OST_IO_PORTAL to avoid delays behind potentially many pending
451 Frequency : occasionally with multiple writers to a single file
453 Description: shared writes to file may result in wrong size reported by stat()
454 Details : Allow growing of kms when extent lock is cancelled
457 Frequency : always with random mmap IO to multi-striped file
459 Description: mmap write might be lost if we are writing to a 'hole' in stripe
460 Details : Only if the hole is at the end of OST object so that kms is too
461 small. Fix is to increase kms accordingly in ll_nopage.
464 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
466 Description: writes to a missing object would leak memory on the OST
467 Details : If there is an inconsistency between the MDS and OST filesystems,
468 such that the MDS references an object that doesn't exist, writes
469 to that object will leak memory due to incorrect cleanup in the
470 error handling path, eventually running out of memory on the OST.
475 Description: Creating too long symlink causes lustre errors
476 Details : Check symlink and name lengths before sending requests to MDS.
479 Frequency : only if flock is enabled (not on by default)
481 Description: posix locks not released on fd closure on 2.6.9+
482 Details : We failed to add posix locks to list of inode locks on 2.6.9+
483 kernels, this caused such locks not to be released on fd close and
484 then assertions on fs unmount about still used locks.
487 Frequency : MDS failover only, very rarely
489 Description: clients may get ASSERTION(granted_lock != NULL)
490 Details : When request was taking a long time, and a client was resending
491 a getattr by name lock request. The were multiple lock requests
492 with the same client lock handle and
493 mds_getattr_name->fixup_handle_for_resent_request found one of the
494 lock handles but later failed with ASSERTION(granted_lock != NULL).
499 Description: handle->h_buffer_credits > 0, assertion failure
500 Details : h_buffer_credits is zero after truncate, causing assertion
501 failure. This patch extends the transaction or creates a new
505 Frequency : NFS re-export or patchless client
506 Bugzilla : 11179, 10796
507 Description: Crash on NFS re-export node (__d_move)
508 Details : We do not want to hash the dentry if we don't have a lock.
509 But if this dentry is later used in d_move, we'd hit uninitialised
510 list head d_hash, so we just do this to init d_hash field but
511 leave dentry unhashed.
514 Frequency : NFS re-export or patchless client
516 Description: NFS exports has problem with symbolic link
517 Details : lustre client didn't properly install dentry when re-exported
518 to NFS or running patchless client.
521 Frequency : NFS re-export or patchless client
523 Description: Various nfs/patchless fixes.
524 Details : fixes reuse disconected alias for lookup process - this fixes
525 warning "find_exported_dentry: npd != pd",
526 fix permission error with open files at nfs.
527 fix apply umaks when do revalidate.
530 Frequency : occasional
532 Description: Crash on NFS re-export node
533 Details : calling clear_page() on the wrong pointer triggered oops in
534 generic_mapping_read().
537 Frequency : rarely, using O_DIRECT IO
539 Description: unaligned directio crashes client with LASSERT
540 Details : check for unaligned buffers before trying any requests.
543 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
545 Description: stale data returned from RAID cache
546 Details : If only a small amount of IO is done to the RAID device before
547 reading it again it is possible to get stale data from the RAID
548 cache instead of reading it from disk.
551 Frequency : depends on arch, kernel and compiler version, always on sles10
554 Description: recursive or deep enough symlinks cause stack overflow
555 Details : getting rid of large stack-allocated variable in
559 Frequency : depends on hardware
561 Description: lustre write performance loss in the SLES10 kernel
562 Details : the performance loss is caused by using of write barriers in the
563 ext3 code. The SLES10 kernel turns barrier support on by
564 default. The fix is to undo that change for ldiskfs.
566 ------------------------------------------------------------------------------
568 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
570 * Support for kernels:
571 2.6.9-42.0.3EL (RHEL 4)
573 2.4.21-47.0.1.EL (RHEL 3)
574 2.6.12.6 vanilla (kernel.org)
578 Frequency : quota enabled and large files being deleted
580 Description: releasing more than 4GB of quota at once hangs OST
581 Details : If a user deletes more than 4GB of files on a single OST it
582 will cause the OST to spin in an infinite loop. Release
583 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
588 Description: statfs data retrieved from /proc may be stale or zero
589 Details : When reading per-device statfs data from /proc, in the
590 {kbytes,files}_{total,free,avail} files, it may appear
591 as zero or be out of date.
594 Frequency : systems with MD RAID1 external journal devices
596 Description: lconf's call to blkid is confused by RAID1 journal devices
597 Details : Use the "blkid -l" flag to locate the MD RAID device instead
598 of returning all block devices that match the journal UUID.
601 Frequency : always, for aggregate stripe size over 4GB
603 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
604 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
605 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
608 Frequency : always if "lfs find" used on a local file/directory
610 Description: "lfs find" segfaults if used on a local file/directory
611 Details : The case where a directory component was not specified wasn't
612 handled correctly. Handle this properly.
615 Frequency : always on ppc64
617 Description: the write to an ext3 filesystem mounted with mballoc got stuck
618 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
619 perform endianness conversion.
622 Frequency : rarely (truncate to non-zero file size after write under load)
623 Bugzilla : 10730, 10687
624 Description: Files padded with zeros to next 4K multiple
625 Details : With filesystems mounted using the "extents" option (2.6 kernels)
626 it is possible that files that are truncated to a non-zero size
627 immediately after being written are filled with zero bytes beyond
628 the truncated size. No file data is lost.
630 Severity : enhancement
632 Description: Allow recovery/failover for liblustre clients.
633 Details : liblustre clients were unaware of failover configurations until
636 Severity : enhancement
638 Description: user file locks should fail when not mounting with flock option
639 Details : Set up an error-returning stub in ll_file_operations.lock field
640 to prevent incorrect behaviour when client is mounted without
641 flock option. Also, set up properly f_op->flock field for
645 Frequency : always on ia64
647 Description: "lfs df" loops on printing out MDS statfs information
648 Details : The obd_ioctl_data was not initialized and in some systems
649 this caused a failure during the ioctl that did not return
650 an error. Initialize the struct and return an error on failure.
653 Frequency : SLES 9 only
655 Description: Error of copying files with lustre special EAs as root
656 Details : Client side always return success for setxattr call for lustre
657 special xattr (currently only "trusted.lov").
660 Frequency : rarely on clusters with both ia64+i386 clients
662 Description: ia64+i686 clients doing shared IO on the same file may LBUG
663 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
664 clients are doing concurrent writes to the same file it is
665 possible that the ia64 clients may LASSERT because the OST
666 extent locks are not PAGE_SIZE aligned. Ensure that grown
667 locks are always aligned on the request boundary.
670 Frequency : specific use, occasional
672 Description: Overwriting in use executable truncates on-disk binary image
673 Details : If one node attempts to overwrite an executable in use by
674 another node, we now correctly return ETXTBSY instead of
680 Description: chmod on Lustre root is propagated to other clients
681 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
682 invalid by the follow_mount time.
685 Frequency : liblustre clients only
687 Description: Race in 'instant cancel' lock handling could lead to such locks
688 never to be granted in case of SMP MDS
689 Details : Do not destroy not yet granted but cbpending locks in
693 Frequency : replay/resend of open
695 Description: non null lock assetion failure in mds_intent_policy
696 Details : Trying to replay/resend lockless open requests resulted in
697 mds_open() returning 0 with no lock. Now it sets a flag if
698 a lock is going to be returned.
700 Severity : enhancement
702 Description: Checksum enhancements
703 Details : New checksum enhancements allow for resending RPCs that failed
706 Severity : enhancement
708 Description: Tunables on number of dirty pages in cacche
709 Details : Allow to set limit on number of dirty pages cached.
714 Description: client crash on unmount - lock still has references
715 Details : In some error handling cases it was possible to leak a lock
716 reference on a client while accessing a file. This was not
717 harmful to the client during operation, but would cause the
718 client to crash when the filesystem is unmounted.
721 Frequency : specific case, rare
723 Description: ETXTBSY on mds though file not in use by client
724 Details : ETXTBSY is no longer incorrectly returned when attempting to
725 chmod or chown a directory that the user previously tried to
726 execute or a currently-executing binary.
729 Frequency : extremely rare except on liblustre-based clients
731 Description: Lustre space not freed when files are deleted
732 Details : Clean up open-unlinked files after client eviction. Previously
733 the unlink was skipped and the files remained as orphans.
738 Description: OST failure "would be an LBUG" in waiting_locks_callback()
739 Details : In some cases it was possible to send a blocking callback to
740 a client doing a glimpse, even though that client didn't get
741 a lock granted. When the glimpse lock is cancelled on the OST
742 the freed lock is left on the waiting list and corrupted the list.
745 Frequency : all core dumps
747 Description: Broke core dumps to lustre
748 Details : Negative dentry may be unhashed if parent does not have UPDATE
749 lock, but some callers, e.g. do_coredump, expect dentry to be
750 hashed after successful create, hash it in ll_create_it.
752 ------------------------------------------------------------------------------
754 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
756 * Support for kernels:
757 2.6.9-42.0.2.EL (RHEL 4)
759 2.4.21-40.EL (RHEL 3)
760 2.6.12.6 vanilla (kernel.org)
764 Frequency : always on RHEL 3
766 Description: Number of open files grows over time
767 Details : The number of open files grows over time, whether or not
768 Lustre is started. This was due to a filp leak introduced
769 by one of our kernel patches.
771 ------------------------------------------------------------------------------
773 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
775 * Support for kernels:
778 2.4.21-40.EL (RHEL 3)
779 2.6.12.6 vanilla (kernel.org)
784 Bugzilla : 5719, 9635, 9792, 9684
785 Description: OST (or MDS) trips assertions in (re)connection under heavy load
786 Details : If a server is under heavy load and cannot reply to new
787 connection requests before the client resends the (re)connect,
788 the connection handling code can behave badly if two service
789 threads are concurrently handing separate (re)connections from
790 the same client. Add better locking to the connection handling
791 code, and ensure that only a single connection will be processed
792 for a given client UUID, even if the lock is dropped.
794 Severity : enhancement
796 Description: add TCP zero-copy support to kernel
797 Details : Add support to the kernel TCP stack to allow zero-copy bulk
798 sends if the hardware supports scatter-gather and checksumming.
799 This allows socklnd to do client-write and server-read more
800 efficiently and reduce CPU utilization from skbuf copying.
803 Frequency : only if NFS exporting from client
805 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
806 Details : If setting attributes on a file created under NFS that had
807 never been opened it would be possible to oops the client
808 if the file had no objects.
812 Bugzilla : 9326, 10402, 10897
813 Description: client crash in ptlrpcd_wake() thread when sending async RPC
814 Details : It is possible that ptlrpcd_wake() dereferences a freed async
815 RPC. In rare cases the ptlrpcd thread alread processed the RPC
816 before ptlrpcd_wake() was called and the request was freed.
819 Frequency : always for liblustre
821 Description: liblustre client does MDS+OSTs setattr RPC for each write
822 Details : When doing a write from a liblustre client, the client
823 incorrectly issued an RPC to the MDS and each OST the file was
824 striped over in order to update the timestamps. When writing
825 with small chunks and many clients this could overwhelm the MDS
826 with RPCs. In all cases it would slow down the write because
827 these RPCs are unnecessary.
829 Severity : enhancement
831 Description: allow number of MDS service threads to be changed at module load
832 Details : It is now possible to change the number of MDS service threads
833 running. Adding "options mds mds_num_threads={N}" to the MDS's
834 /etc/modprobe.conf will set the number of threads for the next
835 time Lustre is restarted (assuming the "mds" module is also
836 reloaded at that time). The default number of threads will
837 stay the same, 32 for most systems.
842 Description: OST crash if filesystem is unformatted or corrupt
843 Details : If an OST is started on a device that has never been formatted
844 or if the filesystem is corrupt and cannot even mount then the
845 error handling cleanup routines would dereference a NULL pointer.
850 Description: NULL pointer deref in llap_from_page.
851 Details : get_cache_page_nowait can return a page with NULL (or otherwise
852 incorrect) mapping if the page was truncated/reclaimed while it was
853 searched for. Check for this condition and skip such pages when
854 doing readahead. Introduce extra check to llap_from_page() to
855 verify page->mapping->host is non-NULL (so page is not anonymous).
858 Frequency : Sometimes when using sys_sendfile
860 Description: "page not covered by a lock" warnings from ll_readpage
861 Details : sendfile called ll_readpage without right page locks present.
862 Now we introduced ll_file_sendfile that does necessary locking
863 around call to generic_file_sendfile() much like we do in
867 Frequency : with certain MDS communication failures at client mount time
869 Description: NULL pointer deref after failed client mount
870 Details : a client connection request may delayed by the network layer
871 and not be sent until after the PTLRPC layer has timed out the
872 request. If the client fails the mount immediately it will try
873 to clean up before the network times out the request. Add a
874 reference from the request import to the obd device and delay
875 the cleanup until the network drops the request.
878 Frequency : occasionally during client (re)connect
880 Description: assertion failure during client (re)connect
881 Details : processing a client connection request may be delayed by the
882 client or server longer than the client connect timeout. This
883 causes the client to resend the connection request. If the
884 original connection request is replied in this interval, the
885 client may trip an assertion failure in ptlrpc_connect_interpret()
886 which thought it would be the only running connect process.
889 Frequency : only with obd_echo servers and clients that are rebooted
891 Description: kernel BUG accessing uninitialized data structure
892 Details : When running an obd_echo server it did not start the ping_evictor
893 thread, and when a client was evicted an uninitialized data
894 structure was accessed. Start the ping_evictor in the RPC
895 service startup instead of the OBD startup.
897 Severity : enhancement
898 Bugzilla : 10193 (patchless)
899 Description: Remove dependency on various unexported kernel interfaces.
900 Details : No longer need reparent_to_init, exit_mm, exit_files,
901 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
904 Frequency : rare (only users of deprecated and unsupported LDAP config)
906 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
907 Details : LDAP apparently contains 'lustreName' attributes instead of
908 'name'. A simple remapping of the name is sufficient.
911 Frequency : rare (only with non-default dump_on_timeout debug enabled)
913 Description: waiting_locks_callback trips kernel BUG if client is evicted
914 Details : Running with the dump_on_timeout debug flag turned on makes
915 it possible that the waiting_locks_callback() can try to dump
916 the Lustre kernel debug logs from an interrupt handler. Defer
917 this log dumping to the expired_lock_main() thread.
919 Severity : enhancement
921 Description: Support NFS exporting on 2.6 kernels.
922 Details : Implement non-rawops metadata methods for NFS server to use without
923 changing NFS server code.
926 Frequency : very rare (synthetic metadata workload only)
928 Description: two racing renames might cause an MDS thread to deadlock
929 Details : Running the "racer" program may cause one MDS thread to rename
930 a file from being the source of a rename to being the target of
931 a rename at exactly the same time that another thread is doing
932 so, and the second thread has already enqueued these locks after
933 doing a lookup of the target and is trying to relock them in
934 order. Ensure that we don't try to re-lock the same resource.
937 Frequency : only very large systems with liblustre clients
939 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
940 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
941 all OSTs in parallel. This allows the eviction of stale liblustre
942 clients to proceed much faster than if they were done in series,
943 and also offers similar improvements for other set_info RPCs.
948 Description: excessive CPU usage during initial read phase on client
949 Details : During the initial read phase on a client, it would agressively
950 retry readahead on the file, consuming too much CPU and impacting
951 performance (since 1.4.5.8). Improve the readahead algorithm
952 to avoid this, and also improve some other common cases (read
953 of small files in particular, where "small" is files smaller than
954 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
959 Description: MDS crash when receiving packet with unknown intent.
960 Details : Do not LBUG in unknown intent case, just return -EFAULT
962 Severity : enhancement
963 Bugzilla : 9293, 9385
964 Description: MDS RPCs are serialised on client. This is unnecessary for some.
965 Details : Do not serialize getattr (non-intent version) and statfs.
968 Frequency : occasional, when OST network is overloaded/intermittent
970 Description: client evicted by OST after bulk IO timeout
971 Details : If a client sends a bulk IO request (read or write) the OST
972 may evict the client if it is unresposive to its data GET/PUT
973 request. This is incorrect if the network is overloaded (takes
974 too long to transfer the RPC data) or dropped the OST GET/PUT
975 request. There is no need to evict the client at all, since
976 the pinger and/or lock callbacks will handle this, and the
977 client can restart the bulk request.
980 Frequency : Always when mmapping file with no objects
982 Description: client crashes when mmapping file with no objects
983 Details : Check that we actually have objects in a file before doing any
984 operations on objects in ll_vm_open, ll_vm_close and
990 Description: Request leak when working with deleted CWD
991 Details : Introduce advanced request refcount tracking for requests
992 referenced from lustre intent.
994 Severity : Enhancement
996 Description: Cache open file handles on client.
997 Details : MDS now will return special lock along with openhandle, if
998 requested and client is allowed to hold openhandle, even if unused,
999 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1000 closing files for every read/write openration.
1002 Severity : Enhancement
1004 Description: Cache open negative dentries on client when possible.
1005 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1006 negative dentries on lock revocation.
1011 Description: Remounting a client read-only wasn't possible with a zconf mount
1012 Details : It wasn't possible to remount a client read-only with llmount.
1014 Severity : enhancement
1015 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1016 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1017 MPICH is included to add Lustre-specific ADIO interfaces.
1018 This is based closely on the UFS ADIO layer and only differs
1019 in file creation, in order to allow the OST striping to be set.
1020 This is user-contributed code and not supported by CFS.
1025 Description: extended inode attributes (immutable, append-only) work improperly
1026 when 2.4 and 2.6 kernels are used on client/server or vice versa
1027 Details : Introduce kernel-independent values for these flags.
1029 Severity : enhancement
1032 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1033 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1034 granularity. Now, Fractional MB granularity is be supported,
1035 this is very useful for low memory system.
1037 Severity : enhancement
1039 Description: Getattr by fid
1040 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1041 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1042 have locks and avoids getattr rpc requests for it on every stat.
1045 Frequency : Always, for filesystems larger than 2TB
1047 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1048 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1049 filesystems larger than 2TB to be mounted with mballoc enabled.
1052 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1054 Description: filesystem corruption for non-standard kernels and very large OSTs
1055 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1056 filesystems larger than 2TB is mounted then the kernel will
1057 silently corrupt the start of the filesystem. CONFIG_LBD is
1058 enabled for all CFS-supported kernels, but the possibility of
1059 this happening with a modified kernel config exists.
1061 Severity : enhancement
1063 Description: add client O_DIRECT support for 2.6 kernels
1064 Details : It is now possible to do O_DIRECT reads and writes to files
1065 in the Lustre client mountpoint on 2.6 kernel clients.
1067 Severity : enhancement
1069 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1070 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1071 parallel, not waiting for response from every OST before sending
1072 a rpc to the next OST.
1077 Description: setattr vs write race when updating file timestamps
1078 Details : Client processes that update a file timestamp into the past
1079 right after writing to the file (e.g. tar) it is possible that
1080 the updated file modification time can be reset to the current
1081 time due to a race between processing the setattr and write RPC.
1083 Severity : enhancement
1085 Description: Bring 'lfs find' closer in line with regular Linux find.
1086 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1087 -print0 options and obtains all the needed info through the lustre
1090 Severity : enhancement
1092 Description: support up to 1024 configured devices on one node
1093 Details : change obd_dev array from statically allocated to dynamically
1094 allocated structs as they are first used to reduce memory usage
1099 Description: Flush dirty partially truncated pages during truncate
1100 Details : Immediatelly flush partially truncated pages in filter_setattr,
1101 this way we completely avoid having any pages in page cache on OST
1102 and can retire ugly workarounds during writes to flush such pages.
1107 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1108 Details : protect inode from truncation within vfs_unlink() context
1109 just take a reference before calling vfs_unlink() and release it
1110 when parent's i_sem is free.
1113 Frequency : always, if extents are used on OSTs
1115 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1116 Details : OSTs using the extents format would not zero the high 16 bits of
1117 the index physical block number. This is not a problem for any
1118 OST filesystems smaller than 16TB, and no kernels support ext3
1119 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1120 new/modified files) and can be fixed for existing filesystems
1121 with e2fsprogs-1.39-cfs1.
1126 Description: import connection selection may be incorrect if timer wraps
1127 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1128 import connections to be ignored if the 32-bit jiffies counter
1129 wraps. Use a 64-bit jiffies counter.
1132 Frequency : during server recovery
1134 Description: crash after server is denying duplicate export
1135 Details : If clients are resending connect requests to the server, the
1136 server refuses to allow a client to connect multiple times.
1137 Fixed a bug in the handling of this case.
1140 Frequency : very large clusters immediately after boot
1142 Description: LNET request buffers exhausted under heavy short-term load
1143 Details : If a large number of client requests are generated on a service
1144 that has previously never seen so many requests it is possible
1145 that the request buffer growth cannot keep up with the spike in
1146 demand. Instead of dropping incoming requests, they are held in
1147 the LND until the RPC service can accept more requests.
1150 Frequency : Sometimes during replay
1152 Description: Assertion failure in ll_local_open after replay.
1153 Details : If replay happened on an open request reply before we were able
1154 to set replay handler, reply will become not swabbed tripping the
1155 assertion in ll_local_open. Now we set the handler right after
1156 recognising of open request
1159 Frequency : very rare
1161 Description: kernel reports "badness in vsnprintf"
1162 Details : Reading from the "recovery_status" /proc file in small chunks
1163 may cause a negative length in lprocfs_obd_rd_recovery_status()
1164 call to vsnprintf() (which is otherwise harmless). Exit early
1165 if there is no more space in the output buffer.
1167 Severity : enhancement
1169 Description: clear OBD RPC statistics by writing to them
1170 Details : It is now possible to clear the OBD RPC statistics by writing
1171 to the "stats" file.
1176 Description: Client mtime is not the same on different clients after utimes
1177 Details : In some cases, the client was using the utimes() syscall on
1178 a file cached on another node. The clients now validate the
1179 ctime from the MDS + OSTs to determine which one is right.
1184 Description: Inability to activate failout mode
1185 Details : lconf script incorrectly assumed that in python string's numeric
1186 value is used in comparisons.
1189 Frequency : always with multiple stripes per file
1191 Description: Inefficient object allocation for mutli-stripe files
1192 Details : When selecting which OSTs to stripe files over, for files with
1193 a stripe count that divides evenly into the number of OSTs,
1194 the MDS is always picking the same starting OST for each file.
1195 Return the OST selection heuristic to the original design.
1200 Description: mount failures may take full timeout to return an error
1201 Details : Under some heavy load conditions it is possible that a
1202 failed mount can wait for the full obd_timeout interval,
1203 possibly several minutes, before reporting an error.
1204 Instead return an error as soon as the status is known.
1206 ------------------------------------------------------------------------------
1208 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
1210 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1211 INTEROPERATE with older versions automatically. Please read the
1212 user documentation before upgrading any part of a live system.
1213 * WARNING: Lustre networking configuration changes are required with
1214 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1217 * Support for kernels:
1218 2.6.9-22.0.2.EL (RHEL 4)
1219 2.6.5-7.244 (SLES 9)
1220 2.6.12.6 vanilla (kernel.org)
1223 Severity : enhancement
1224 Bugzilla : 7981/8208
1225 Description: Introduced Lustre Networking (LNET)
1226 Details : LNET is new networking infrastructure for Lustre, it includes
1227 a reorganized network configuration mode (see the user
1228 documentation for full details) as well as support for routing
1229 between different network fabrics. Lustre Networking Devices
1230 (LNDs) for the supported network fabrics have also been
1231 created for this new infrastructure.
1233 Severity : enhancement
1234 Description: Introduced Access control lists
1235 Details : clients can set ACLs on files and directories in order to have
1236 more fine-grained permissions than the standard Unix UGO+RWX.
1237 The MDS must be started with the "-o acl" mount option.
1239 Severity : enhancement
1240 Description: Introduced filesystem quotas
1241 Details : Administrators may now establish per-user quotas on the
1244 Severity : enhancement
1246 Description: Configuration change for the XT3
1247 The PTLLND is now used to run Lustre over Portals on the XT3
1248 The configure option(s) --with-cray-portals are no longer used.
1249 Rather --with-portals=<path-to-portals-includes> is used to
1250 enable building on the XT3. In addition to enable XT3 specific
1251 features the option --enable-cray-xt3 must be used.
1256 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1257 Details : A race between allocating a new llcd and re-getting the llcd_lock
1258 allowed another thread to grab newly-allocated llcd.
1260 Severity : enhancement
1262 Description: 2.6 OST async journal commit and locking fix to improve performance
1263 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1264 2.6 kernels are now async as they already were in 2.4 kernels so
1265 that they can commit concurrently with the network bulk transfer.
1266 For block-allocated files the filter allocation semaphore is held
1267 to avoid filesystem fragmentation during allocation. BKL lock
1268 removed for 2.6 xattr operations where it is no longer needed.
1273 Description: lconf incorrectly determined whether two IP networks could talk
1274 Details : In some more complicated routing and multiple-network
1275 configurations, lconf will avoid trying to make a network
1276 connection to a disjoint part of the IP space. It was doing the
1277 math incorrectly for one set of cases.
1282 Description: Fix for potential infinite loop processing records in an llog.
1283 Details : If an llog record is corrupted/zeroed, it is possible to loop
1284 forever in llog_process(). Validate the llog record length
1285 and skip the remainder of the block on error.
1288 Frequency : occasional (liblustre only)
1290 Description: liblustre could not open files whose last component is a symlink
1291 Details : sysio_path_walk() would incorrectly pass the open intent to
1292 intermediate path components.
1295 Frequency : rare (liblustre only with non-standard tuning)
1296 Bugzilla : 7201 (7350)
1297 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1298 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1299 after it has been released, only do this for referenced locks
1301 Severity : enhancement
1303 Description: specifying an (invalid) directory default stripe_size of -1
1304 would reset the directory default striping
1305 Details : stripe_size -1 was used internally to signal directory stripe
1306 removal, now use "all default" to signal dir stripe removal
1307 as a directory striping of "all default" is not useful
1310 Frequency : common for large clusters running liblustre clients
1312 Description: doing an ls when liblustre clients are running is slow
1313 Details : sending a glimpse AST to a liblustre client waits for every AST
1314 to time out, as liblustre clients will not respond. Since they
1315 cannot cache data we refresh the OST lock LVB from disk instead.
1317 Severity : enhancement
1319 Description: doing an ls at the same time as file IO can be slow
1320 Details : enqueue and other "small" requests can be blocked behind many
1321 large IO requests. Create a new OST IO portal for non-IO
1322 requests so they can be processed faster.
1325 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1327 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1328 created file with mode 000
1329 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1330 allow an ftruncate() as a non-root user to a file with mode 000.
1331 Lustre now respects this flag to disable mode checks when
1332 truncating a file owned by the user
1335 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1338 Description: Revoking locks from clients that went dead or catatonic might take
1340 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1341 cancellation of such locks instant on servers without waiting for
1342 any reply from clients. Clients drops these locks when cancel
1343 notification from server is received without replying.
1346 Frequency : liblustre-only, when liblustre client dies or becomes busy
1348 Description: Doing ls on Linux clients can take a long time with active
1350 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1351 granting such locks to it in the first place if possible. Locks
1352 are taken by proxy on the OST during the read or write and
1353 dropped immediately afterward. Add connect flags handling, do
1354 not grant locks to liblustre clients for glimpse ASTs.
1356 Severity : enhancement
1358 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1359 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1360 sequential reads and behaves badly with applications doing random
1361 reads. Improve it by reading ahead at least read region, and
1362 avoiding excessive large RPC for small reads.
1364 Severity : enhancement
1366 Description: Creating more than 1000 files for a single job may cause a load
1367 imbalance on the OSTs if there are also a large number of OSTs.
1368 Details : qos_prep_create() uses an OST index reseed value that is an
1369 even multiple of the number of available OSTs so that if the
1370 reseed happens in the middle of the object allocation it will
1371 still utilize the OSTs as uniformly as possible.
1376 Description: OST or MDS may oops in ping_evictor_main()
1377 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1378 but doesn't restart at beginning of obd_exports_timed list
1381 Severity : enhancement
1383 Description: improve by-nid export eviction on the MDS and OST
1384 Details : allow multiple exports with the same NID to be evicted at one
1385 time without re-searching the exports list.
1388 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1390 Description: MDS may oops in groups_free()
1391 Details : in rare race conditions a newly allocated group_info struct is
1392 freed again, and this can be NULL. The 2.4 compatibility code
1393 for groups_free() checked for a NULL pointer, but 2.6 did not.
1396 Frequency : common for liblustre clients doing little filesystem IO
1397 Bugzilla : 9352, 7313
1398 Description: server may evict liblustre clients accessing contended locks
1399 Details : if a client is granted a lock or receives a completion AST
1400 with a blocking AST already set it would not reply to the AST
1401 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1404 Frequency : lfs setstripe, only systems with more than 160 OSTs
1406 Description: unable to set striping with a starting offset beyond OST 160
1407 Details : llapi_create_file() incorrectly limited the starting stripe
1408 index to the maximum single-file stripe count.
1411 Frequency : LDAP users only
1413 Description: lconf did not handle in-kernel recovery with LDAP properly
1414 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1416 Severity : enhancement
1418 Description: bind OST threads to NUMA nodes to improve performance
1419 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1420 node and do their allocations there to localize memory access
1422 Severity : enhancement
1424 Description: llmount can determine client NID directly from Myrinet (GM)
1425 Details : the client NID code from gmnalnid was moved directly into
1426 llmount, removing the need to use this or specifying the
1427 client NID explicitly when mounting GM clients with zeroconf
1430 Frequency : if client is started with down MDS
1432 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1433 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1434 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1435 ptlrpc_import_delayed_req() because we didn't check if the
1436 request was interrupted, and we also didn't break out of the
1437 event loop if there was no timeout
1442 Description: data loss during non-page-aligned writes to a single file from
1443 both multiple nodes and multiple threads on one node at same time
1444 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1445 inconsistency led to false short-reads, that were cached and later
1446 used by ->prepare_write() to fill in partially written page,
1447 leading to data loss.
1450 Frequency : always, if lconf --abort_recovery used
1452 Description: lconf --abort_recovery fails with 'Operation not supported'
1453 Details : lconf was attempting to abort recovery on the MDT device and not
1456 Severity : enhancement
1458 Description: remove cleanup logs
1459 Details : replace lconf-generated cleanup logs with lustre internal
1460 cleanup routines. Eliminates the need for client-cleanup and
1463 Severity : enhancement
1465 Description: add support for EAs (user and system) on lustre filesystems
1466 Details : it is now possible to store extended attributes in the Lustre
1467 client filesystem, and with the user_xattr mount option it
1468 is possible to allow users to store EAs on their files also
1470 Severity : enhancement
1472 Description: Add possibility (config option) to show minimal available OST free
1474 Details : When compiled with --enable-mindf configure option, statfs(2)
1475 (and so, df) will return least minimal free space available from
1476 all OSTs as amount of free space on FS, instead of summary of
1477 free spaces of all OSTs.
1479 Severity : enhancement
1481 Description: do not expand extent locks acquired on OST-side
1482 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1483 by server: they are not cached anyway.
1486 Frequency : when mmap is used/binaries executed from Lustre
1488 Description: Unmmap pages before throwing them away from read cache.
1489 Details : llap_shrink cache now attempts to unmap pages before discarding
1490 them (if unmapping failed - do not discard). SLES9 kernel has
1491 extra checks that trigger if this unmapping is not done first.
1496 Description: lconf didn't resolve symlinks before checking to see whether a
1497 given mountpoint was already in use
1500 Frequency : when migrating failover services
1501 Bugzilla : 6395, 9514
1502 Description: When migrating a subset of services from a node (e.g. failback
1503 from a failover service node) the remaining services would
1504 time out and evict clients.
1505 Details : lconf --force (implied by --failover) sets the global obd_timeout
1506 to 5 seconds in order to quickly disconnect, but this caused
1507 other RPCs to time out too quickly. Do not change the global
1508 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1510 Severity : enhancement
1511 Frequency : if MDS is started with down OST
1512 Bugzilla : 9439,5706
1513 Description: Allow startup/shutdown of an MDS without depending on the
1514 availability of the OSTs.
1515 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1516 Add appropriate locking and lov-osc refcounts for safe
1517 cleaning. Add osc abort_inflight calls in case the
1518 synchronize never started.
1521 Frequency : occasional (Cray XT3 only)
1523 Description: root not authorized to access files in CRAY_PORTALS environment
1524 Details : The client process capabilities were not honoured on the MDS in
1525 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1526 been accessed by an authorized user then root was able to access
1527 the file on the local client also. The root user capabilities
1528 are now allowed on the MDS, as this environment has secure UID.
1531 Frequency : occasional
1533 Description: ldiskfs "too long searching" message happens too often
1534 Details : A debugging message (otherwise harmless) prints too often on
1535 the OST console. This has been reduced to only happen when
1536 there are fragmentation problems on the filesystem.
1541 Description: Division by zero in statfs when all OSCs are inactive
1542 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
1543 lov_statfs divides by value returned by lov_get_stripecnt().
1547 Bugzilla : 9489, 3273
1548 Description: First write from each client to each OST was only 4kB in size,
1549 to initialize client writeback cache, which caused sub-optimal
1550 RPCs and poor layout on disk for the first writen file.
1551 Details : Clients now request an initial cache grant at (re)connect time
1552 and so that they can start streaming writes to the cache right
1553 away and always do full-sized RPCs if there is enough data.
1554 If the OST is rebooted the client also re-establishes its grant
1555 so that client cached writes will be honoured under the grant.
1560 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
1561 Details : Now I/O RPCs go to different portal number and (presumably) fast
1562 lock requests (and glimses) and other RPCs get their own service
1563 threads pool that should be able to service those RPCs
1566 Severity : enhancement
1568 Description: Ability to exchange lustre version between client and servers and
1569 issue warnings at client side if client is too old. Also for
1570 liblustre clients there is ability to refuse connection of too old
1572 Details : New 'version' field is added to connect data structure that is
1573 filled with version info. That info is later checked by server and
1577 Frequency : rare, liblustre only.
1578 Bugzilla : 9296, 9581
1579 Description: Two simultaneous writes from liblustre at offset within same page
1580 might proceed at the same time overwriting eachother with stale
1582 Details : I/O lock withing llu_file_prwv was released too early, before data
1583 actually was hitting the wire. Extended lock-holding time until
1584 server acknowledges receiving data.
1587 Frequency : extremely rare. Never observed in practice.
1589 Description: avoid generating lustre_handle cookie of 0.
1590 Details : class_handle_hash() generates handle cookies by incrementing
1591 global counter, and can hit 0 occasionaly (this is unlikely, but
1592 not impossible, because initial value of cookie counter is
1593 selected randonly). Value of 0 is used as a sentinel meaning
1594 "unassigned handle" --- avoid it. Also coalesce two critical
1595 sections in this function into one.
1597 Severity : enhancement
1599 Description: allow liblustre clients to delegate truncate locking to OST
1600 Details : To avoid overhead of locking, liblustre client instructs OST to
1601 take extent lock in ost_punch() on client's behalf. New connection
1602 flag is added to handle backward compatibility.
1604 Severity : enhancement
1605 Bugzilla : 4928, 7341, 9758
1606 Description: allow number of OST service threads to be specified
1607 Details : a module parameter allows the number of OST service threads
1608 to be specified via "options ost ost_num_threads={N}" in the
1609 OSS's /etc/modules.conf or /etc/modprobe.conf.
1613 Bugzilla : 6146, 9635, 9895
1614 Description: servers crash with bad pointer in target_handle_connect()
1615 Details : In rare cases when a client is reconnecting it was possible that
1616 the connection request was the last reference for that export.
1617 We would temporarily drop the export reference and get a new
1618 one, but this may have been the last reference and the export
1619 was just destroyed. Get new reference before dropping old one.
1621 Severity : enhancement
1622 Frequency : if client is started with failover MDS
1624 Description: Allow multiple MDS hostnames in the mount command
1625 Details : Try to read the configuration from all specified MDS
1626 hostnames during a client mount in case the "primary"
1629 Severity : enhancement
1631 Description: Stop sending data to evicted clients as soon as possible.
1632 Details : Check if the client we are about to send or are sending data to
1633 was evicted already. (Check is done every second of waiting,
1634 for which l_wait_event interface was extended to allow checking
1635 of exit condition at specified intervals).
1638 Frequency : rare, normally only when NFS exporting is done from client
1640 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
1642 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
1645 Frequency : always, if srand() is called before liblustre initialization
1647 Description: Liblustre uses system PRNG disturbing its usage by user application
1648 Details : Introduce internal to lustre fast and high-quality PRNG for
1649 lustre usage and make liblustre and some other places in generic
1650 lustre code to use it.
1652 Severity : enhancement
1653 Bugzilla : 9477, 9557, 9870
1654 Description: Verify that the MDS configuration logs are updated when xml is
1655 Details : Check if the .xml configuration logs are newer than the config
1656 logs stored on the MDS and report an error if this is the case.
1657 Request --write-conf, or allow starting with --old_conf.
1659 Severity : enhancement
1661 Description: Handle symlinks in the path when checking if Lustre is mounted.
1662 Details : Resolve intermediate symlinks when checking if a client has
1663 mounted a filesystem to avoid duplicate client mounts.
1668 Description: lconf can hit an error exception but still return success.
1669 Details : The lconf command catches the Command error exception at the top
1670 level script context and will exit with the associated exit
1671 status, but doesn't ensure that this exit status is non-zero.
1676 Description: failure of ptlrpc thread startup can cause oops
1677 Details : Starting a ptlrpc service thread can fail if there are a large
1678 number of threads or the server memory is very fragmented.
1679 Handle this without oopsing.
1682 Frequency : always, only if liblustre and non-default acceptor port was used
1684 Description: liblustre cannot connect to servers with non-default acceptor port
1685 Details : tcpnal_set_default_params() was not called and was therefore
1686 ignoring the environment varaible TCPNAL_PORT, as well as other
1687 TCPNAL_ environment variables
1692 Description: two objects could be created on the same OST for a single file
1693 Details : If an OST is down, in some cases it was possible to create two
1694 objects on a single OST for a single file. No problems other
1695 than potential performance impact and spurious error messages.
1699 Bugzilla : 5681, 9562
1700 Description: Client may oops in ll_unhash_aliases
1701 Details : Client dcache may become inconsistent in race condition.
1702 In some cases "getcwd" can fail if the current directory is
1708 Description: Inode refcounting problems in NFS export code
1709 Details : link_raw functions used to call d_instantiate without obtaining
1710 extra inode reference first.
1714 Bugzilla : 9942, 9903
1715 Description: Referencing freed requests leading to crash, memleaks with NFS.
1716 Details : We used to require that call to ll_revalidate_it was always
1717 followed by ll_lookup_it. Also with revalidate_special() it is
1718 possible to call ll_revalidate_it() twice for the same dentry
1719 even if first occurence returned success. This fix changes semantic
1720 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
1721 reference on a request referred from the intent.
1722 ll_intent_release() then releases such a request.
1725 Frequency : rare, normally benchmark loads only
1727 Description: unlinked inodes were kept in memory on the client
1728 Details : If a client is repeatedly creating and unlinking files it
1729 can accumulate a lot of stale inodes in the inode slab cache.
1730 If there is no other client load running this can cause the
1731 client node to run out of memory. Instead flush old inodes
1732 from client cache that have the same inode number as a new inode.
1735 Frequency : SLES9 2.6.5 kernel and long filenames only
1736 Bugzilla : 9969, 10379
1737 Description: utime reports stale NFS file handle
1738 Details : SLES9 uses out-of-dentry names in some cases, which confused
1739 the lustre dentry revalidation. Change it to always use the
1743 Frequency : rare, unless heavy write-truncate concurrency is continuous
1744 Bugzilla : 4180, 6984, 7171, 9963, 9331
1745 Description: OST becomes very slow and/or deadlocked during object unlink
1746 Details : filter_destroy() was holding onto the parent directory lock
1747 while truncating+unlinking objects. For very large objects this
1748 may block other threads for a long time and slow overall OST
1749 responsiveness. It may also be possible to get a lock ordering
1750 deadlock in this case, or run out of journal credits because of
1751 the combined truncate+unlink. Solution is to do object truncate
1752 first in one transaction without parent lock, and then do the
1753 final unlink in a new transaction with the parent lock. This
1754 reduces the lock hold time dramatically.
1757 Frequency : rare, 2.4 kernels only
1759 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
1760 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
1761 before unmounting their filesystem in order to avoid deadlock.
1762 The kernel_locked() function in 2.4 kernels only checks whether
1763 the kernel lock is held, not whether it is this process that is
1764 holding it as 2.6 kernels do.
1769 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
1770 Details : The client ptlrpc code may be trying to reconnect to a down
1771 server before a previous connection attempt has timed out.
1772 Increase the reconnect interval to be longer than the connection
1773 timeout interval to avoid sending duplicate connections to
1777 Frequency : echo_client brw_test command
1779 Description: fix echo_client to work with OST preallocated code
1780 Details : OST preallocation code (5137) didn't take echo_client IO path
1781 into account: echo_client calls filter methods outside of any
1782 OST thread and, hence, there is no per-thread preallocated
1783 pages and buffers to use. Solution: hijack pga pages for IO. As
1784 a byproduct, this avoids unnecessary data copying.
1788 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
1789 Description: Client can oops in mdc_commit_close() after open replay
1790 Details : It was possible for the MDS to return an open request with no
1791 transaction number in mds_finish_transno() if the client was
1792 evicted, but without actually returning an error. Clients
1793 would later try to replay that open and may trip an assertion
1794 Simplify the client close codepath, and always return an error
1795 from the MDS in case the open is not successful.
1798 Frequency : rare, 2.6 OSTs only
1800 Description: OST may deadlock under high load on fragmented files
1801 Details : If there was a heavy load and highly-fragmented OST filesystems
1802 it was possible to have all the OST threads deadlock waiting on
1803 allocation of biovecs, because the biovecs were not released
1804 until the entire RPC IO was completed. Instead, release biovecs
1805 as soon as they are complete to ensure forward IO progress.
1807 Severity : enhancement
1809 Description: Support for specifying external journal device at mount
1810 Details : If an OST or MDS device is formatted with an external journal
1811 device, this device major/minor is stored in the ext3 superblock
1812 and may not be valid for failover. Allow detecting and
1813 specifying the external journal at mount time.
1818 Description: Mounting an MDS with pending unlinked files may cause oops
1819 Details : target_finish_recovery() calls mds_postrecov() which returned
1820 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
1821 considers this an error and immediately begins cleaning up the
1822 lov, just after starting the mds_lov process
1824 Severity : enhancement
1826 Description: Implement 'lfs df' to report actual free space on per-OST basis
1827 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
1828 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
1829 usage of objects; '-h' to report in human readable format.
1831 ------------------------------------------------------------------------------
1833 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
1840 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
1841 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
1842 at startup during orphan unlinking. Instead, skip buddy update
1843 before setup, as it will be regenerated after recovery is complete.
1848 Description: If an OST is inactive, its locks might reference stale inodes.
1849 Details : lov_change_cbdata() must iterate over all namespaces, even if
1850 they are inactive to clear inode references from the lock.
1852 Severity : enhancement
1853 Frequency : occasional, if non-standard max_dirty_mb used
1855 Description: Client will block write RPCs if not enough grant
1856 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
1857 then the client will block writes while waiting for another RPC
1858 to complete instead of consuming its dirty limit. With change
1859 we get improved performance when max_dirty_mb is small.
1861 Severity : enhancement
1862 Bugzilla : 3389, 6253
1863 Description: Add support for supplementary groups on the MDS.
1864 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
1865 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
1866 lookups for user supplementary groups into a cache.
1870 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
1871 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
1872 for liblustre O_EXCL is enforced by MDS.
1876 Description: readdir never returns NULL in liblustre.
1877 Details : Corrected llu_iop_getdirentries logic, to return offset of next
1878 dentry in struct dirent.
1882 Frequency : liblustre only, depends on application IO pattern
1883 Description: liblustre clients evicted if not contacting servers
1884 Details : Don't put liblustre clients into the ping_evictor list, so
1885 they will not be evicted by the pinger ever.
1887 Severity : enhancement
1889 Description: Add ability to evict clients by NID from MDS.
1890 Details : By echoing "nid:$NID" string into
1891 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
1892 $NID would be instantly evicted from this MDS and from all active
1893 OSTs connected to it.
1897 Description: Do not query file size twice, somewhat slowing stat(2) calls.
1898 Details : lookup_it_finish() used to query file size from OSTs that was not
1903 Description: service threads change working directory to that of init
1904 Details : Starting lustre service threads may pin the working directory
1905 of the parent thread, making that filesystem busy. Threads
1906 now change to the working directory of init to avoid this.
1910 Frequency : during shutdown only
1911 Description: shutdown with a failed MDS or OST can cause unmount to hang
1912 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
1917 Frequency : 2.6 only
1918 Description: chmod/chown may include an extra supplementary group
1919 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
1920 supplementary group and if none is specified this is used.
1923 Bugzilla : 5479 (6816)
1925 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
1926 Details : If lookup is in progress on a file that is unlinked we might try
1927 to revalidate the inode and fail in revalidate after lookup is
1928 complete and ll_file_open() enqueues the open again but
1929 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
1932 Frequency : always, if lconf --abort_recovery used
1934 Description: lconf --abort_recovery fails with 'Operation not supported'
1935 Details : lconf was attempting to abort recovery on the MDT device and not
1938 ------------------------------------------------------------------------------
1940 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
1945 Frequency : rare (only unsupported configurations with a node running as an
1947 Bugzilla : 6514, 5137
1948 Description: Mounting a Lustre file system on a node running as an OST could
1950 Details : OSTs now preallocates memory needed to write out data at
1951 startup, instead of when needed, to avoid having to
1952 allocate memory in possibly low memory situations.
1953 Specifically, if the file system is mounted on on OST,
1954 memory pressure could force it to try to write out data,
1955 which it needed to allocate memory to do. Due to the low
1956 memory, it would be unable to do so and the node would
1957 become unresponsive.
1959 Severity : enhancement
1961 Description: Addition of lconf --service command line option
1962 Details : lconf now accepts a '--service <arg>' option, which is
1963 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
1965 Severity : enhancement
1967 Description: Failover mode is now the default for OSTs.
1968 Details : By default, OSTs will now run in failover mode. To return to
1969 the old behaviour, add '--failout' to the lmc line for OSTs.
1971 Severity : enhancement
1973 Description: Health checks are now provided for MDS and OSTs
1974 Details : Additional detailed health check information on MSD and OSTs
1975 is now provided through the procfs health_check value.
1978 Frequency : occasional, depends on IO load
1980 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
1981 after numerous create/delete cycles
1982 Details : The ext3 inode allocation policy would not allocate new inodes
1983 very well on the OSTs because there are no new directories
1984 being created. Instead we look for groups with free space if
1985 the parent directories are nearly full.
1990 Description: Network or server problems during mount may cause partially
1991 mounted clients instead of returning an error.
1992 Details : The config llog parsing code may overwrite the error return
1993 code during mount error handling, returning success instead
1999 Description: MDS can fail to allocate large reply buffers
2000 Details : After long uptimes the MDS can fail to allocate large reply
2001 buffers (e.g. zconf client mount config records) due to memory
2002 fragmentation or consumption by the buffer cache. Preallocate
2003 some large reply buffers so that these replies can be sent even
2004 under memory pressure.
2008 Frequency : rare (liblustre)
2009 Description: fsx running with liblustre complained that using truncate() to
2010 extend the file doesn't work. This patch corrects that issue.
2011 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2012 ATTR_SIZE and lsm use in llu_setattr_raw.
2016 Frequency : rare, only 2.6 kernels
2017 Description: Unusual file access patterns on the MDS may result in inode
2018 data being lost in very rare circumstances.
2019 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2020 code in the 2.6 kernel caused the inode and/or EA data not to
2021 be read from disk, causing single-file corruption.
2025 Frequency : rare, only 2.6 filesystems using extents
2026 Description: Heavy concurrent write and delete load may cause data corruption.
2027 Details : It was possible under high-load situations to have an extent
2028 metadata block in the block device cache from a just-unlinked
2029 file overwrite a newly-allocated data block. We now unmap any
2030 metadata buffers that alias just-allocated data blocks.
2034 Frequency : filesystems with default stripe_count larger than 77
2035 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2036 Details : lconf specifies an inode size of 4096 bytes when the default
2037 stripe_count is larger than 77. This conflicts with the default
2038 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2039 this case to avoid pinning too much memory for large EAs.
2041 ------------------------------------------------------------------------------
2043 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2048 Frequency : rare (extremely heavy IO load with hundreds of clients)
2050 Description: Client is evicted, gets IO error writing to file
2051 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2052 caused clients to be evicted for AST timeouts. The fixes in
2053 bug 5192 mean we no longer need to have such short AST timeouts
2054 so ldlm_timeout has been increased.
2057 Frequency : occasional during --force or --failover shutdown under load
2058 Bugzilla : 5949, 4834
2059 Description: Server oops/LBUG if stopped with --force or --failover under load
2060 Details : a collection of import/export refcount and cleanup ordering
2061 issues fixed for safer force cleanup
2064 Frequency : only filesystems larger than 120 OSTs
2065 Bugzilla : 5990, 6223
2066 Description: lfs getstripe would oops on a very large filesystem
2067 Details : lov_getconfig used kfree on vmalloc'd memory
2070 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2071 Bugzilla : 6242, 6243
2072 Description: reading from files that had been truncated to a non-zero size
2073 but never opened returned no data
2074 Details : ll_file_read() reads zeros from no-object files to EOF
2079 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2080 Details : There's little harm in aborting MDS/OSS recovery and letting it
2081 try again, so I removed the LASSERT and return an error instead.
2083 Severity : enhancement
2085 Description: New debugging infrastructure for tracking down data corruption
2086 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2087 (b) cover more of the client-side code path, and (c) try to narrow
2088 down where problems occurred
2092 Bugzilla : 3819, 4364, 4397, 6313
2093 Description: Racing close and eviction MDS could cause assertion in mds_close
2094 Details : It was possible to get multiple mfd references during close and
2095 client eviction, leading to one thread referencing a freed mfd.
2097 Severity: : enhancement
2098 Bugzilla : 3262, 6359
2099 Description: Attempts to reconnect to servers are now more aggressive.
2100 Details : This builds on the enhanced upcall-less recovery that was added
2101 in 1.4.2. When trying to reconnect to servers, clients will
2102 now try each server in the failover group every 10 seconds. By
2103 default, clients would previously try one server every 25 seconds.
2108 Description: After recovery, certain operations trigger a failed
2109 assertion on a client.
2110 Details : Failing over an mds, using lconf -d --failover, while a
2111 client was doing a readdir() call would cause the client to
2112 LBUG after recovery completed and the readdir() was resent.
2114 Severity : enhancement
2116 Description: Default groups are now added by lconf
2117 Details : You can now run lconf --group <servicename> without having to
2118 manually add groups with lmc.
2121 Frequency : occasional
2123 Description: Nodes with an elan id of 0 trigger a failed assertion
2126 Frequency : always when accessing e.g. tty/console device nodes
2128 Description: tty and some other devices nodes cannot be used on lustre
2129 Details : file's private_data field is used by device data and lustre
2130 values in there got lost. New field was added to struct file to
2131 store fs-specific private data.
2134 Frequency : when exporting Lustre via NFS
2136 Description: NFSD failed occasionally when looking up a path component
2137 Details : NFSD is looking up ".." which was broken in ext3 directories
2138 that had grown large enough to become hashed.
2141 Frequency : Clusters with multiple interfaces not on the same subnet
2143 Description: Nodes will repeatedly try to reconnect to an interface which it
2144 cannot reach and report an error to the log.
2145 Details : Extra peer list entries will be created by lconf with some peers
2146 unreachable. lconf now validates the peer before adding it.
2149 Frequency : Only if a default stripe is set on the filesystem root.
2151 Description: Setting a default stripe on the filesystem root prevented the
2152 filesystem from being remounted.
2153 Details : The client was sending extra request flags in the root getattr
2154 request and did not allocate a reply buffer for the dir EA.
2157 Frequency : occasional, higher if lots of files are accessed by one client
2158 Bugzilla : 6159, 6097
2159 Description: Client trips assertion regarding lsm mismatch/magic
2160 Details : While revalidating inodes the VFS looks up inodes with ifind()
2161 and in rare cases can find an inode that is being freed.
2162 The ll_test_inode() code will free the lsm during ifind()
2163 when it finds an existing inode and then the VFS later attaches
2164 this free lsm to a new inode.
2168 Bugzilla : 6422, 7030
2169 Description: MDS deadlock between mkdir and client eviction
2170 Details : Creating a new file via mkdir or mknod (starting a transaction
2171 and getting the ns lock) can deadlock with client eviction
2172 (gets ns lock and trying to finish a synchronous transaction).
2175 Frequency : occasional
2176 Description: While starting a server, the fsfilt_ext3 module could not be
2178 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2179 kernels. Previously, lconf would still use the ext3 name
2180 when trying to load modules. Now, it will correctly use
2181 ext3 on 2.4 and ldiskfs on 2.6.
2183 Severity : enhancement
2184 Description: The default stripe count has been changed to 1
2185 Details : The interpretation of the default stripe count (0, to lfs
2186 or lmc) has been changed to mean striping across a single
2187 OST, rather than all available. For general usage we have
2188 found a stripe count of 1 or 2 works best.
2190 Severity : enhancement
2191 Description: Add support for compiling against Cray portals.
2192 Details : Conditional compiling for some areas that are different
2196 Frequency : occasional
2197 Bugzilla : 6409, 6834
2198 Description: Creating files with an explicit stripe count may lead to
2199 a failed assertion on the MDS
2200 Details : If some OSTs are full or unavailable, creating files may
2201 trigger a failed assertion on the MDS. Now, Lustre will
2202 try to use other servers or return an error to the
2206 Frequency : occasional
2208 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2209 caused lock timeout during readahead (since 1.4.2).
2210 Details : Processes doing readahead might match a lock that hasn't been
2211 granted yet if there are overlapping and conflicting lock
2212 requests. The readahead process waits on ungranted lock
2213 (original lock is CBPENDING), while OST waits for that process
2214 to cancel CBPENDING read lock and eventually evicts client.
2216 Severity : enhancement
2218 Description: Initial enabling of flock support for clients
2219 Details : Implements fcntl advisory locking and file status functions.
2220 This feature is provided as an optional mount flag (default
2221 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2222 locking are implemented yet, and those that are are not guaranteed
2223 to be completely correct in production environments.
2224 mount -t lustre -o [flock|noflock] ...
2227 Frequency : occasional
2229 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2230 assertion in the ext3 JBD (journaling) layer.
2231 Details : The b_committed_data struct is protected by the big kernel lock
2232 in 2.4 kernels, serializing journal_commit_transaction() and
2233 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2234 access to this struct. In 2.6 kernels there is finer grained
2235 locking to improve SMP performance of the JBD layer.
2239 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2241 -----------------------------------------------------------------------------
2243 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2245 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2246 versions of Lustre. Please update all servers and clients to
2247 version 1.4.2 or later at the same time. You must also run
2248 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2249 to update the configuration logs.
2251 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2252 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2253 - swab llog records of type '0' so we get proper header size/idx (5861)
2254 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2255 - fix rename of one directory over another leaking an inode (5953)
2256 - avoid SetPageDirty on 2.6 (5981)
2257 - don't re-add just-being-destroyed locks to the waiting list (5653)
2258 - when creating new directories, inherit the parent's custom
2259 striping settings if present parent (3048)
2260 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2261 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2262 - don't hold client locks on temporary worklist from l_lru (5666)
2263 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2264 - automatically evict dead clients (5921)
2265 - Update file size properly in create+truncate+fstat case (6196)
2266 - Do not unhash mountpoint dentries, do not allow removal of
2268 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2269 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2270 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2271 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2272 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2273 - Correctly handle reads of files with no objects (6243)
2274 - lctl recover will also mark a device active if deactivate used (5933)
2276 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2277 - allow --write-conf on an MDS with different nettype than client (5619)
2278 - don't write config llogs to MDS for mounts not from that MDS (5617)
2279 - lconf should create multiple TCP connections from a client (5201)
2280 - init scripts are now turned off by default; run chkconfig --on
2281 lustre and chkconfig --on lustrefs to use them
2282 - upcalls are no longer needed for clients to recover to failover
2284 - add --abort-recovery option to lconf to abort recovery on device
2286 - add support for an arbitrary number of OSTs (3026)
2287 - Quota support protocol changes.
2288 - forward compatibility changes to wire structs (6007)
2289 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2290 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2291 - improved "lustre status" script
2292 - initialize blocksize for non-regular files (6062)
2293 - added --disable-server and --disable-client configure options (5782)
2294 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2295 - Vanilla 2.4.29 support
2296 - increase maximum number of obd devices to 520 (6242)
2297 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2298 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2299 SLES 9 2.6.5 kernels
2300 - limit stripes per file to 160 (the maximum EA size) (6093)
2302 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2305 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2306 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2307 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2308 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2309 - small ext3 extents cleanups and fixes (5733)
2310 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2311 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2312 - avoid needless client->OST connect, fix handle mismatch (5317)
2313 - fix DLM error path that led to out-of-sync client, long delays (5779)
2314 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2315 - fix several locking issues related to i_size (5492,5624,5654,5672)
2316 - don't move pending lock onto export if it is already evicted (5683)
2317 - fix kernel oops when creating .foo in unlinked directory (5548)
2318 - fix deadlock in obdfilter statistics vs. object create (5811)
2319 - use time_{before,after} to avoid timer jiffies wrap (5882)
2320 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2321 - Do not leak request if server was not able to process it (5154)
2322 - If mds_open unable to find parent dir, make that negative lookup(5154)
2323 - don't create new directories with extent-mapping (5909, 5936)
2325 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2326 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2327 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2328 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2329 - support for RHEL4 kernel on i686 (5773)
2330 - provide error messages when incompatible logs are encountered (5898)
2332 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2333 * version 1.4.0.10 (1.4.1 release candidate 1)
2335 - don't keep a lock reference when lock is not granted (4238)
2336 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2337 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2338 - drop import inflight refcount on signal_completed_replay error (5255)
2339 - unlock page after async write error during send (3677)
2340 - handle missing objects in filter_preprw_read properly (5265)
2341 - no transno return for symlink open, don't save no-trasno open (3440)
2342 - don't try to complete elan receive that already failed (4012)
2343 - free RPC server reply state on error (5406)
2344 - clean up thread from ptlrpc_start_thread() on error (5160)
2345 - readahead could read extra page into cache that wasn't ejected (5388)
2346 - prevent races in class_attach/setup/cleanup/detach (5260)
2347 - don't dereference de->d_inode after l_dput of de (5458)
2348 - use "int" for stripe value returned from lock_to_stripe (5544)
2349 - mballoc allocation and error-checking fixes in 2.6 (5504)
2350 - block device patches to fix I/O request sizes in 2.6 (5482)
2351 - look up hostnames for IB nals (5602)
2352 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2353 - don't start multiple acceptors for the same port (5277)
2354 - fix incorrect LASSERT in mds_getattr_name (5635)
2355 - export a proc file for general "ping" checking (5628)
2356 - fix "lfs check" to not block when the MDS is down (5628)
2358 - service request history (4965)
2359 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2360 - create an "evict_client" /proc entry on OSTs, like the MDS has
2361 - fix mount usage message, return errors per mount(8) (5168)
2362 - change grep [] to grep "[]" in tests so they work in more UMLs
2363 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2364 - remove ancient LOV_MAGIC_V0 EA support (5047)
2365 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2366 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2367 - allow readahead allocations to fail when low on memory (5383)
2368 - mmap locking landed again, after considerable improvement (2828)
2369 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2371 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2374 - send OST transaction number in read/write reply to free req (4966)
2375 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2376 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2377 - the watchdog thread now runs as interruptible (5246)
2378 - flock/lockf fixes (but it's still disabled, pending 5135)
2379 - don't use EXT3 constants in llite code (5094)
2380 - memory shortage at startup could cause assertion (5176)
2382 - reorganization of lov code
2383 - single portals codebase
2385 - add extents/mballoc support (5025)
2386 - direct I/O reads in the obdfilter (4048)
2387 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2389 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2392 - send OST transaction number in read/write reply to free req (4966)
2393 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2394 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2395 - the watchdog thread now runs as interruptible (5246)
2396 - handle missing objects in filter_preprw_read properly (5265)
2397 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2398 - drop import inflight refcount on signal_completed_replay error (5255)
2399 - unlock page after async write error during send (3677)
2400 - return original error code on reconstructed replies (3761)
2401 - no transno return for symlink open, don't save no-trasno open (3440)
2403 - add pid to ldlm debugging output (4922)
2404 - bump the watchdog timeouts -- we can't handle 30sec yet
2405 - extra debugging for orphan dentry/inode bug (5259)
2407 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2410 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2411 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2412 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2413 - fix NULL dereference in /proc/sys/portals/routes (4827)
2414 - allow failed mdc_close() operations to be interrupted (4561)
2415 - stop precreate on OST before MDS would time out on it (4778)
2416 - don't send partial-page writes before EOF from client (4410)
2417 - discard client grant for sub-page writes on large-page clients (4520)
2418 - don't free dentries not owned by NFS code, check generation (4806)
2419 - fix lsm leak if mds_create_objects() fails (4801)
2420 - limit debug_daemon file size, always print CERROR messages (4789)
2421 - use transno after validating reply (3892)
2422 - process timed out requests if import state changes (3754)
2423 - update mtime on OST during writes, return in glimpse (4829)
2424 - add mkfsoptions to LDAP (4679)
2425 - use ->max_readahead method instead of zapping global ra (5039)
2426 - don't interrupt __l_wait_event() during strace
2428 - add software watchdogs to catch hung threads quickly (4941)
2429 - make lustrefs init script start after nfs is mounted
2430 - fix CWARN/ERROR duplication (4930)
2431 - return async write errors to application if possible (2248)
2432 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2433 - print NAL number in %x format (4645)
2434 - update barely-supported suse-2.4.21-171 series (4842)
2435 - support for sles 9 %post scripts
2436 - support for building 2.6 kernel-source packages
2437 - support for sles km_* packages
2439 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2442 - ignore -ENOENT errors in osc_destroy (3639)
2443 - notify osc create thread that OSC is being cleaned up (4600)
2444 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2445 - reconstruct ost_handle() like mds_handle() (4657)
2446 - create a new thread to do import eviction to avoid deadlock (3969)
2447 - let lconf resolve symlinked-to devices (4629)
2448 - don't unlink "objects" from directory with default EA (4554)
2449 - hold socknal file ref over connect in case target is down (4394)
2450 - allow more than 32000 subdirectories in a single directory (3244)
2451 - fix blocks count for O_DIRECT writes (3751)
2452 - OST returns ENOSPC from object create when no space left (4539)
2453 - don't send truncate RPC if file size isn't changing (4410)
2454 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2455 - bind to privileged port in socknal and tcpnal (3689)
2457 - rate limit CERROR/CWARN console message to avoid overload (4519)
2458 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2459 - basic mmap support (3918)
2460 - kernel patch series update from b1_4 (4711)
2462 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2465 - avoid crash during MDS cleanup with OST shut down (2775)
2466 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2467 - don't use bad inodes on the MDS (3744)
2468 - dynamic object preallocation to improve recovery speed (4236)
2469 - don't hold spinlock over lock dumping or change debug flags (4401)
2470 - don't zero obd_dev when it is force cleaned (3651)
2471 - print grants to console if they go negative (4431)
2472 - "lctl deactivate" will stop automatic recovery attempts (3406)
2473 - look for existing locks in ldlm_handle_enqueue() (3764)
2474 - don't resolve lock handle twice in recovery avoiding race (4401)
2475 - revalidate should check working dir is a directory (4134)
2477 - don't always mark "slow" obdfilter messages as errors (4418)
2479 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2482 - don't close LustreDB during write_conf until it is done (3860)
2483 - fix typo in lconf for_each_profile (3821)
2484 - allow dumping logs from multiple threads at one time (3820)
2485 - don't allow multiple threads in OSC recovery (3812)
2486 - fix debug_size parameters (3864)
2487 - fix mds_postrecov to initialize import for llog ctxt (3121)
2488 - replace config semaphore with spinlock (3306)
2489 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2490 - don't allow enqueue to complete on a destroyed export (3822)
2491 - down write_lock before checking llog header bitmap (3825)
2492 - recover from lock replay timeout (3764)
2493 - up llog sem before sending rpc (3652)
2494 - reduce ns lock hold times when setting kms (3267)
2495 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2496 - fix NULL deref and obd_dev leak on setup error (3312)
2497 - replace some LBUG about llog ops with error handling (3841)
2498 - don't match INVALID dentries from d_lookup and spin (3784)
2499 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2500 - fix invalid assertion in ptlrpc_set_wait (3880)
2502 - add libwrap support for the TCP acceptor (3996)
2503 - add /proc/sys/portals/routes for non-root route listing (3994)
2504 - allow setting MDS UUID in .xml (2580)
2505 - print the stack of a process that LBUGs (4228)
2507 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2510 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2511 - make sure to unset replay flag from failed open requests (3440)
2512 - if default stripe count is 0, use OST count for inode size (3636)
2513 - update parent mtime/ctime on client for create/unlink (2611)
2514 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2515 - free recovery state on server during a forced cleanup (3571)
2516 - unregister_reply for resent reqs (3063)
2517 - loop back devices mounting and status check on 2.6 (3563)
2518 - fix resource-creation race that can provoke i_size == 0 (3513)
2519 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2520 - more debugging for page-accounting assertion (3746)
2521 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2522 - avoid deadlock after precreation failure (3758)
2523 - fix race and lock order deadlock in orphan handling (3450, 3750)
2524 - add validity checks when grabbing inodes from l_ast_data (3599)
2526 - add /proc/.../recovery_status to obdfilter (3428)
2527 - lightweight CDEBUG infrastructure, debug daemon (3668)
2528 - change default OSC RPC parameters to be better on small clusters
2529 - turn off OST read cache for files smaller than 32MB
2530 - install man pages and include them in rpms (3100)
2531 - add new init script for (un)mounting lustre filesystems (2593)
2532 - run chkconfig in %post for init scripts (3701)
2533 - drop scimac NAL (unmaintained)
2535 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
2538 - clean kiobufs before and after use (3485)
2539 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
2540 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
2541 - change init script to fail more clearly if not run as root (1528)
2542 - allow clients to reconnect during replay (1742)
2543 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
2544 - don't do DNS lookups on NIDs too small for IP addresses (3442)
2545 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
2546 - fix cond_resched (3554)
2547 - only evict unfinished clients after recovery (3515)
2548 - allow bulk resend, prevent data loss (3570)
2549 - dynamic ptlrpc request buffer allocation (2102)
2550 - don't allow unlinking open directory if it isn't empty (2904)
2551 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
2552 - remove extraneous obd dereference causing LASSERT failure (3334)
2553 - don't use get_cycles() when creating temp. files on the mds (3156)
2554 - hold i_sem when setting i_size in ll_extent_lock() (3564)
2555 - handle EEXIST for set-stripe, set proper directory name (3336)
2557 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
2558 - fix ksocknal_fmb_callback() error messages (2918)
2560 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
2563 - don't copy lvb into (possibly NULL) reply on error (2983)
2564 - don't deref dentry after dput, don't free lvb on error (2922)
2565 - use the kms to determine writeback rpc length (2947)
2566 - increment oti_logcookies when osc is inactive (2948)
2567 - update client's i_blocks count via lvb messages (2543)
2568 - handle intent open/close of special files properly (1557)
2569 - mount MDS with errors=remount-ro, like obdfilter (2009)
2570 - initialize lock handle to avoid ASSERT on error cleanup (3057)
2571 - don't use cancelling-locks' kms values (2947)
2572 - use highest lock extent for kms, not last one (2925)
2573 - don't dereference ERR_PTR() dentry in error handling path (3107)
2574 - fix thread race in portals_debug_dumplog() (3122)
2575 - create lprocfs device entries at setup instead of at attach (1519)
2576 - common AST error handler, don't evict client on completion race (3145)
2577 - zero nameidata in detach_mnt in 2.6 (3118)
2578 - verify d_inode after revalidate_special is valid in 2.6 (3116)
2579 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
2580 - initialize RPC timeout timer earlier for 2.6 (3219)
2581 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
2582 - print nal/nid for unknown nid (3258)
2583 - additional checks for oscc recovery before doing precreate (3284)
2584 - fix ll_extent_lock() error return code for 64-bit systems (3043)
2585 - don't crash in mdc_close for bad permissions on open (3285)
2586 - zero i_rdev for non-device files (3147)
2587 - clear page->private before handing to FS, better assertion (3119)
2588 - tune the read pipeline (3236)
2589 - fix incorrect decref of invalidated dentry (2350)
2590 - provide read-ahead stats and refine rpc in flight stats (3328)
2591 - don't hold journal transaction open across create RPC (3313)
2592 - update atime on MDS at close time (3265)
2593 - close LDAP connection when recovering to avoid server load (3315)
2594 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
2595 - don't leak open file on MDS after open resend (3325)
2596 - serialize filter_precreate and filter_destroy_precreated (3329)
2597 - loop device shouldn't call sync_dev() for nul device (3092)
2598 - clear page cache after eviction (2766)
2599 - resynchronize MDS->OST in background (2824)
2600 - refuse to mount the same filesystem twice on same mountpoint (3394)
2601 - allow llmount to create routes for mounting behind routers (3320)
2602 - push lock cancellation to blocking thread for glimpse ASTs (3409)
2603 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
2604 - fix rare problem with rename on htree directories (3417)
2606 - allow default OST striping configuration per directory (1414)
2607 - fix compilation for qswnal for 2.6 kernels (3125)
2608 - increase maximum number of MDS request buffers for large systems
2609 - change liblustreapi to be useful for external progs like lfsck (3098)
2610 - increase local configuration timeout for slow disks (3353)
2611 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
2613 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2616 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
2617 - don't overwrite extent policy data in reply if lock was blocked (2901)
2618 - drop filter export grants atomically with removal from device (2663)
2619 - del obd_self_export from work_list in class_disconnect_exports (2908)
2620 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
2621 - swab reply message in mdc_close, other PPC fixes (2464)
2622 - fix destroying of named logs (2325)
2623 - overwrite old logs when running lconf --write_conf (2264)
2624 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
2625 - fix race in target_handle_connect (2898)
2626 - mds_reint_create() should take same inode create lock (2926)
2627 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
2628 - don't close files for self_export to avoid uninitialized obd (2936)
2629 - allow MDS with the same name as client node (2939)
2630 - hold dentry reference for closed log files for unlink (2325)
2631 - reserve space for all logs during transactions (2059)
2632 - don't evict page beyond end of stripe extent (2925)
2633 - don't oops on a deleted current working directory (2399)
2634 - handle hard links to targets without a parent properly (2517)
2635 - don't dereference NULL lock when racing during eviction (2867)
2636 - don't grow lock extents when lots of conflicting locks (2919)
2638 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
2641 - account for cache space usage on clients to avoid data loss (974)
2642 - lfsck support in lustre kernel code (2349)
2643 - reduce journal credits needed for BRW writes (2370)
2644 - orphan handling to avoid losing space on client/server crashes
2645 - ptlrpcd can be blocked, stopping ALL progress (2477)
2646 - use lock value blocks to assist in proper KMS, faster stat (1021)
2647 - takes i_sem instead of DLM locks internally on obdfilter (2720)
2648 - recovery for initial connections (2355)
2649 - fixes for mds_cleanup_orphans (1934)
2650 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
2651 - block all file creations until orphan recovery completes (1901)
2652 - client remove rq_connection from request struct (2423)
2653 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
2654 - recovery timer race (2670)
2655 - mdc_close recovey bug (2532)
2656 - ptlrpc cleanup bug (2710)
2657 - mds timeout on local locks (2588)
2658 - namespace lock held during RPCs (2431)
2659 - handle interrupted sync write properly (2503)
2660 - don't try to handle a message that hasn't been replied to (2699)
2661 - client assert failure during cleanup after abort recovery (2701)
2662 - leak mdc device after failed mount (2712)
2663 - ptlrpc_check_set allows timedout requests to complete (2714)
2664 - wait for inflight reqs when ptlrpcd finishes (2710)
2665 - make sure unregistered services are removed from the srv_list
2666 - reset bulk XID's when resending them (caught by 1138 test)
2667 - unregister_bulk after timeout
2668 - fix lconf error (2694)
2669 - handle write after unfinished setstripe, stripe-only getstripe (2388)
2670 - readahead locks pages, leaves pending causing memory pressure (2673)
2671 - increase OST request buffers to 4096 on large machines (2729)
2672 - fix up permission of existing directories in simple_mkdir (2661)
2673 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
2674 - don't assign transno to errored transactions (2742)
2675 - don't delete objects on OST if given a bogus objid from MDS (2751)
2676 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
2677 - if rq_no_resend, then timeout request after recovery (2432)
2678 - fix MDS llog_logid record size, 64-bit array alignment (2733)
2679 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
2680 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
2681 - MDS recovery shouldn't delete objects that it has given out (2730)
2682 - if enqueue arrives after completion, don't clobber LVB (2819)
2683 - don't unlock pages twice when trigger_group_io returns error (2814)
2684 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
2685 - don't write pages to disk if there was an error (1450)
2686 - don't ping imports that have recovery disabled (2676)
2687 - take buffered bytes into account when balancing socknal conn (2817)
2688 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
2689 - reconnect unlink llog connection after MDS reconnects to OST (2816)
2690 - remove little-endian swabbing of llog records (1987)
2691 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
2692 - retry reposting request buffers if they fail (1191)
2693 - grow extent at grant time to avoid granting a revoked lock (2809)
2694 - lock revoke doesn't evict page if covered by a second lock (2765)
2695 - disable VM readahead to avoid reading outside lock extents (2805)
2697 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
2698 - updated kernel patches for hp-2.4.20 kernel (2681)
2700 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
2703 - fix truncated write corruption (2366)
2704 - fix for failed assertion in iopen_connect_dentry (1792,2517)
2706 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
2707 - ensure len(uuid) < 37 in lmc (1171)
2708 - fix ia64 OOPS in llog_test (2255)
2709 - zero end of page at obdfilter for partial page writes (2648)
2710 - don't leave stale dentries around after renames (bug 2428)
2711 - fix timeouts when evicting a client with a single lock held (2642)
2712 - set deadline for the initial HELLO message to drain (2634)
2713 - print out dotted-quad IP addresses in the socknal (2302)
2715 - additional debugging for MDS client eviction problem (2443)
2716 - fix mkfsoptions support for osts (2603, 2604)
2718 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
2721 - add series for the vanilla 2.6.0 kernel
2722 - add series for the vanilla 2.4.24 kernel
2723 - add series for a cray x86/64 UL kernel drop
2724 - fix xattr patches for the vanilla 2.4.19 series
2726 - generate true UUIDs in lmc (1171)
2727 - have portals stack dumping break in UML (2466)
2728 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
2729 - allocate with _NOFS in ldlm to avoid deadlock (1933)
2730 - wake callback waiting threads on client eviction (2460)
2731 - Add --ptldebug and --subsystem to lmc (1719)
2732 - update assertion to allow safe interrupt allocation
2733 - set rq_no_resend for cancel requests (2432)
2734 - recalculate ptlrpcd timeout after resend (2494)
2735 - call vfs_rmdir when removing pending directories (2368)
2736 - fix renaming a file to itself (2429)
2737 - lmc creates a default one-stripe lov (2454)
2738 - expand procfs space to handle large clusters (2326)
2739 - increase UML stack to avoid overflow
2740 - update lconf's list of debug and subsystem masks
2741 - fix lfs find --obd (2510)
2742 - /proc tunable for disabling filter read caching (2591)
2743 - stop rpm packages from altering slapd.conf (2301)
2744 - disable nagle in the socknal under 0conf (2578)
2745 - choose mds inode size based on stripe count (2572)
2746 - fix kernel-source rpm problems (2516)
2748 - add --disable-doc to avoid pdf generation (2421)
2749 - update documentation, tests, type-os, comments
2750 - avoid format warnings on ia64
2751 - remove the TOE NAL
2752 - tiny code cleanups by removing unused fields
2754 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
2757 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
2758 - remove the most blatant lies from BUILDING file (2371)
2759 - change default debug level to reasonable production setting
2760 - reduce client side cache size to reduce cache flush time
2761 - reduce max RPCs in flight to avoid unnecessary file fragmentation
2762 - make TCP zerocopy and pinger support enabled by default (2476)
2763 - sync writes completed after process exits caused crashes (2319)
2764 - maintain correct mount count on the MDS (2356)
2765 - backout 1557, because 2316 wasn't really fixed
2766 - better file I/O statistics gathering in /proc
2767 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
2768 - another kernel patch to fix zero-copy TCP function export
2769 - don't take duplicate lock when processing re-sent getattr (2420)
2770 - lctl uses obd_self_export instead of creating new conn (2353)
2771 - MDS/OST recovery case which requires object creation asserted (2425)
2772 - move lfs from /usr/sbin to /usr/bin in packages
2773 - fix race between mds_client_add and mds_client_free (2417)
2774 - use kmalloc instead of slabs in portals (2430)
2775 - don't create duplicate records when a failover MDS is present (2442)
2776 - remove unnecessary mount age check (2332)
2777 - don't remove directory inodes from locks prematurely (2451)
2778 - don't break if MDS service name is the same as hostname (2103)
2779 - fix races in client write RPC generation when cache full (2482)
2781 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
2784 - remove now-unused request->rq_obd (278)
2785 - if an allocation fails, print out how much memory we've used (1933)
2786 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
2787 - add the "configurable stack size" patch to most series files (1256)
2788 - ability to write large log records, for 100+ OST configs (2306)
2789 - fix NULL deref when filter_prep fails (2314)
2790 - fix operator precedence error in filter_sync
2791 - dynamic allocation of socknal TX descriptors (2315)
2792 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
2793 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
2794 - ensure that CWARN messages in Portals always get to the syslog
2795 - __init/__exit are not for prototype decls (ldlm_init/exit)
2796 - x86-64 compile warning fixes
2797 - fix gateway LMC keyword conflict (2318)
2798 - fix MDS lock inversions in getattr/reint paths (1844)
2799 - fix a rare lock re-ordering bug, which caused deadlock (2322)
2800 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
2801 - DLM race condition prevented some lock evictions (2328)
2802 - ENOMEM detection and retry on socknal sends (2230)
2803 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
2804 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
2805 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
2806 - better error messages when a client is rejected during recovery (1505)
2807 - avoid cancelling locks which were never granted, after failure (2330)
2808 - fix i_sem/journal inversion in mds_client_add (2333)
2809 - fix truncate/getattr lock cycle deadlock (2334)
2810 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
2811 - fix two rare exit paths which could leak an l_lock() ref (2321)
2812 - fencepost error in MDS/OST orphan recovery (2226)
2813 - make log record alignment 8 bytes (1988)
2814 - lstripe now fails when requested offset > ost_count (2237)
2815 - ensure that all kernel series have a complete list.h (1607)
2816 - fix crashes in special-file operations (2316)
2817 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
2819 - allow configurable automake binary, for testing new versions
2820 - small update to the lfs documentation
2822 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
2824 * fix negative export reference count in fsfilt_sync (2312)
2826 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
2827 * release candidate 0.9.1
2829 - orphans are moved into the PENDING directory for possible recovery
2830 - replayed opens now open by fid for orphan/rename safety (1042)
2831 - last close of an orphan inode generates a transno (683)
2832 - chdir() and mount() now pin the directory entry (1020)
2833 - avoid CERROR in normal ll_setattr_raw() error case (1500)
2834 - discard very old requests without processing them (1502)
2835 - remove some common, well-understood CERRORs (1505)
2836 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
2837 - clear "grant" flags in OST replies until OST grant code lands (1644)
2838 - fix read performance by not clobbering i_blksize on client (1598)
2839 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
2840 - make LRU size a /proc tunable, clears locks when reduced (707)
2841 - fix some lprocfs rot that prevented ptlbd from loading (1732)
2842 - server locks take references on exports now (1558)
2843 - build fixes for 2.4.20-rh trees (1663)
2844 - return an error from lov_create if all OSCs are inactive (1751)
2845 - fix import levels when a reconnect happens without a timeout (1597)
2846 - exit early from mds_open if we get a lookup error (1749)
2847 - partial page read at EOF wouldn't wait for disk before sending (1642)
2848 - avoid NULL deref in obdfilter when reading page past EOF (1592)
2849 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
2850 - fix LBUG in ll_it_open_error with rc = -2 (1861)
2851 - write/truncate lock inversion (1639)
2852 - Don't auto-load obdclass, portals modules during cleanup (1495)
2853 - fix timestamps from jumping to "now" (1763)
2854 - extra journal assertions (1648)
2855 - add an extra multiunlink test (1771)
2856 - fix read_record/write_record API (1776)
2857 - fix leak of offset_extent, possible incorrect i_size later (1772)
2858 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
2859 - Debugging for the kqswnal_get_idle_tx problems (1820)
2860 - Allow recovery to be attempted multiple times (1536)
2861 - Write out MDS last_rcvd file after it is first created (1600)
2862 - Fix tx_descriptor leak in failed transmit situations (1827)
2863 - ext3 journaling fixes for assertion failure after IO error (1871)
2864 - class_export_put() on freed export after completion AST error (1896)
2865 - Fix revalidate looping in VFS (1322)
2866 - Don't access a freed export during MDS_REINT timeout (1521)
2867 - Add open-unlink recovery support on the MDS (1673,1764)
2868 - Return an error if no MDS data was read from last_rcvd (1946)
2869 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
2870 - Don't setup a disk device that doesn't match exported UUID (317)
2871 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
2872 - avoid committing NULL handle in force close
2873 - local.sh is now a one-stripe LOV configuration
2874 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
2875 - fix ext3 htree duplicate directory entry corruption (1516)
2876 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
2877 - update to new LOV EA format (2097)
2878 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
2880 2003-06-15 Phil Schwan <phil@clusterfs.com>
2883 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
2884 - per-import/export recovery handling (958, 931, 959)
2885 - multiple last-rcvd slots, for serving multiple FSes (949)
2886 - connections are again shared between multiple imp/exports (963, 964)
2887 - "umount -f" would hang if any requests needed to be sent (393, 978)
2888 - avoid pinning large req buffer by copying for queued messages (989)
2889 - add "uuid" to "lctl device" command to help upcalls (991)
2890 - "open" RPCs with transnos would confuse recovery counters (1037)
2891 - do proper endian conversion of all wire messages (288, 340, 891)
2892 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
2893 - call waiting locks callback from LDLM recovery thread (1127, 1151)
2894 - fix ptlrpc_connection leak in target_handle_connect (1174)
2895 - fix import refcounting bug in OST and MDS cleanup (1134)
2896 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
2897 - fix very unlikely obd_types race condition (501)
2898 - remove osc_open hack for echo_client (1187)
2899 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
2900 - a failure in read_inode2 leads to deadlock (1139)
2901 - cancel ack-locks as soon as transaction is committed (1072)
2902 - fix major leaks and crashes in the bulk I/O path (937, 1057)
2903 - make sure to commitrw after any preprw to avoid deadlock (1162)
2904 - failing to execute a file in a lustre FS would lock inode (1203)
2905 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
2906 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
2907 - obd_uuid2tgt would walk off the end of the list (1255)
2908 - on IA64 the osc would give portals incorrect bulk size (1258)
2909 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
2910 - fix lock inversion caused by new llite matching code (1282)
2911 - limit the number of dirty pages on a client to 10MB (1286)
2912 - timed out locks were not being corrected cancelled (1289)
2913 - fix O_DIRECT above 4GB on IA-32 (1292)
2914 * major user-visible changes
2915 - fail out/fail over policy now controlled by the upcall (993)
2917 - add OBD_PING to check server availability and failure (954)
2918 - lustre messages are now sent in sending host order (288, 340, 891)
2919 - add eadatalen to MDS getattr reply (340)
2920 - OST read replies may contain second buffer, with per-page status (593)
2922 2003-03-11 Phil Schwan <phil@clusterfs.com>
2925 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
2926 - failed open()s could cause deadlock; fixed (867, 869)
2927 - stop cancelling OST locks when files are closed (481)
2928 - overlapping XID spaces caused network corruption (851, 853)
2929 - fix unsafe fsfilt counter arithmetic; change to atomic_t
2930 - setattr_raw added, to do single-RPC, server-side setattrs
2931 - lmc/lconf syntax change for OST UUIDs
2932 - fix crashy race condition between ptlrpc_free_req and osc_close
2933 - don't use request in mdc_enqueue if we hit a timeout (889)
2934 - don't set the inode i_size for regular files from the MDS (896)
2935 - handle out of order completion AST (842)
2936 - don't LBUG if a lock request times out after receiving AST (913)
2937 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
2938 - if a bad lock AST arrives, send an error instead of dropping entirely
2939 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
2940 - fix leak in bulk IO when only partially completed (899, 900, 926)
2941 - fix O_DIRECT for ia64 (55)
2942 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
2943 - C-z after timeout could hang a process forever; fixed (977)
2945 - client-side I/O cache (678, 924, 929, 941, 970)
2947 - READPAGE and SETATTRs which don't take server-side locks get
2950 2003-02-11 Phil Schwan <phil@clusterfs.com>
2953 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
2954 - Fix fsfilt_extN_readpage() to read a full page of directory
2955 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
2956 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
2957 - Fix imbalanced LOV object allocation and out-of-bound access (469)
2958 - Most intent operations were removed, in favour of a new RPC mode
2959 that does a single RPC to the server and bypasses most of the VFS
2960 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
2961 - Aggressively cancel local locks on DLM servers
2962 - mds_reint_unlink sends EA to the client if it's the last nlink.
2963 client uses that EA to unlink OST objects.
2964 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
2965 - recursive symlinks were fixed (439)
2966 - fixed NULL deref in DEBUG_REQ
2967 - filter_update_lastobjid no longer calls sync, which annoyed extN
2968 - fixed multi-client small-writes to a single file problem (445)
2969 - fixed mtime updates during file writes (607)
2970 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
2971 - fixed bug in obd_brw_read/write() (under guise of testing 367)
2972 - fixed Linux OST size reporting problem (444, 656)
2973 - OST now updates object mtime with writes or setattr (607, 619)
2974 - client verifies file size before zeroing page past EOF (445)
2975 - OST now writes last allocated objid to disk with allocation (108)
2976 - LOV on echo now works (409)
2978 - mds_reint_unlink sends a new buffer, with the EA included. this
2979 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
2980 set if a regular file was being unlinked, and it was the last link
2981 - use PtlGet from the target for bulk writes (315)
2982 - OST now updates object mtime with writes or setattr (607, 619)
2983 - LDLM now has a grant-time callback to revalidate locked items, if
2985 - Many MDS operations were reorganized to combat race conditions
2987 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
2988 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
2990 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
2993 - Fully reactivate OST imports after reconnection (512, others)
2994 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
2995 - More graceful error handling for truncating on dead OST (515)
2996 - Don't error out unless we're actually accessing dead stripes (474)
2997 - Fix garbage sizes when stripes are missing (410)
2998 - LRU counters were broken, causing constant lock purge (433, 432)
2999 - garbage on read from stripes with failed OSTs (441)
3000 - mark OSCs as active before reconnecting during recovery (438)
3001 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3002 - lfind did not preserve OST order in output (443)
3003 - symlinks cause hung clients, incorrect data (439)
3004 - stop dereferencing request after dropping refcount (457)
3005 - don't LASSERT(spin_is_locked) on non-SMP (455)
3006 - fixes for many rename() bugs
3007 - fstat didn't correctly synchronize attributes (399)
3008 - server must handle lock cancellation during blocking AST prep (487)
3009 - bulk descriptors were free()d too soon (511)
3010 - fix paths in lconf, which would load incorrect modules (451, 507)
3011 - fix confusing lconf 'host not found' error message (386)
3012 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3013 - fix race condition in mdc_blocking_ast() for inode access (526)
3014 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3015 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3016 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3017 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3018 - fix lstripe in several ways (kernel side) (527)
3019 - fix request leak in ldlm_cli_enqueue (262)
3020 - incorrect OSC was marked inactive after OST failure
3021 - call mds_fs_cleanup before unmounting filesystem (524)
3022 - fix races between taking ns_lock and ldlm_lock_change_resource
3023 - fix races updating LOV export open file list
3024 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3025 - fix recovery NULL deref in ldlm_cli_cancel_unused
3026 - fix some DLM races by using new hash table for lock handles (419)
3027 - permit the client to specify desired inodes, at replay
3028 - duplicate requests when we queue them for replay reintegration
3029 - fix last_rcvd offset calculation
3030 - sync after each recovered transaction, so we always make progress
3031 - never, not always, ERESTART requests without transnos
3032 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3034 - skip replay if the MDS says that the client is already connected
3035 - don't check for a recovery-enabled export to match lctl's UUID
3036 - don't INC_USE_COUNT for phantom exports
3037 - don't crash when cleaning up phantom exports (567)
3038 - don't double-finish or set replay data for errored mdc_open requests
3039 - abort requests when they time out, so we don't get old replies
3040 - send/receive replies for AST messages again
3041 - if the client says that it doesn't have the lock, cancel it on the
3043 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3044 mark it as destroyed, it will all work out when decref is called
3045 - fix module use counts (22, 581)
3047 - ASTs now expect a reply (server cancels lock on error reply)
3049 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3052 - fix many simultaneous client startup (392)
3053 - fix dentry->d_it clobbering
3054 - credentials weren't being shipped for readdir/getattr operations
3055 - remove invalid assertions triggered during some concurrent MD
3057 - proper Lustre versions added (336, 389)
3058 - fix memory leak for create error case (398)
3059 - fix LOV locking bug that would get cli/srv out of sync
3060 - fix echo client over LOV (409)
3061 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3062 - fix double-O_EXCL intent crash (424)
3063 - avoid sending multiple lock CANCELs (352)
3065 - MDS can do multi-client recovery (modulo bugs in new code)
3067 - many updates, edits, cleanups
3069 2002-11-18 Phil Schwan <phil@clusterfs.com>
3072 - fix null d_it dereference (346)
3073 - fix full OST/dbench hang (333)
3074 - fix permission problem with file removal (286)
3075 - fix removal of OSCs from LOV when they fail
3076 - fix NULL deref during bulk timeout (214)
3077 - fix problems related to multiple filesystems on one MDS (241)
3078 - fixed serious subtle metadata locking bugs
3079 - free locks on clients when inodes are removed due to memory
3081 - fix inode pointer in lock data (285)
3082 - partial support for multiple MDS on a single host (241)
3083 - data locks weren't cancelled at clear_inode time (290, 311)
3084 - intent locks could lead to unbounded lock growth (205)
3085 - added a maximum lock count, an LRU list, and a flusher
3086 - fix multiple rename (365)
3087 - properly abstracted the echo client
3088 - OSC locked 1 byte too many; fixed
3089 - rewrote brw callback code:
3090 - fixed recovery bugs related to LOVs (306)
3091 - fixed too-many-pages-in-one-write crash (191)
3092 - fixed (again) crash in sync_io_timeout (214)
3093 - probably fixed callback-related race (385)
3095 - Add capability to MDS protocol
3096 - LDLM cancellations and callbacks on different portals
3098 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3101 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3102 - defer open object destruction to close time (601981, 138)
3103 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3104 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3105 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3106 - don't call obd_setattr in ll_file_release for destroyed objects
3107 * protocol change to lustre_msg: move |version| and add |flags|
3108 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3109 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3110 * added replay of create, unlink, link and rename operations during
3111 MDS failover; recovery should be much more robust now
3112 * remove failed OSCs from LOVs (only lov_create uses this so far)
3113 * the lustre-HOWTO was brought (more) up to date (582544)
3115 2002-10-23 Phil Schwan <phil@clusterfs.com>
3118 - in-use dentries weren't being reused properly (617851)
3119 - prevent multiple LDLM setup (599178)
3120 - fix LOV size calculations for truncate (617853)
3121 - fix client handling of MDS intent errors (POSIX)
3122 - fix permission bug in lovstripe.c test (624321)
3123 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3124 - truncate past end of file could corrupt data
3125 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3126 - a race in recovery could return ETIMEDOUT to apps (623947)
3127 - building outside the source directory was fixed
3128 * the lustre-HOWTO was brought (more) up to date (582544)
3129 * major progress was made on recovery functionality
3131 2002-10-10 Phil Schwan <phil@clusterfs.com>
3134 - recovery deadlock fix
3135 - rm -rf causes LBUG fix (617817)
3136 - file open by multiple tasks fix (618962)
3137 - directory permissions bugs (602707 and 620007)
3138 - journal_stop fixed with locking (611313)
3139 - O_APPEND failures resolved (618273, perhaps 614459)
3140 - lconf PATH fix (619770)
3141 - IA64 build fix (621450)
3142 - RPC buffer sizes scale with amount of memory
3144 2002-10-01 Phil Schwan <phil@clusterfs.com>
3147 - locks would be cancelled without throwing away data pages,
3148 resulting in inconsistent data (605627)
3149 - inode attributes were not always being refreshed (605627, 612449)
3150 - lconf now continues to cleanup after lctl reports an error
3151 - MDS now enforces user permissions (602707)
3152 - lprocfs cleanup fixed, but not yet enabled (614157)
3153 - fixed infinite server hang, should a client not respond to an AST
3154 - avoid going into recovery if user calls readlink() with a buffer
3155 that's too small (613941)
3156 - AST RPCs no longer require replies (614867) -- this may be changed
3157 - don't crash server if client sends an IOV that's too big (611336)
3158 - fixed lock conversion deadlock (611892)
3159 - fixed the following of symlinks (614622)
3160 * recovery: the server can remove locks from a client that dies, other
3161 clients can make progress
3162 * more extN patch fixes
3163 * compile-time configurable ptlrpc buffer allocations
3165 - collaborative read cache document
3166 - Lustre Lite Performance CDR document-in-progress
3168 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3171 - fix typo in patch-2.4.18
3173 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3176 - clear ptlrpc request each time in handle_incoming_request()
3177 - unlink of files now destroys the object on the OST
3179 2002-09-19 Peter Braam <braam@clusterfs.com>
3181 * add hard link support
3182 * change obdfile creation method
3183 * kernel patch changed
3185 2002-09-19 Peter Braam <braam@clusterfs.com>
3188 - stack overflow bug in extN fixed
3190 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3192 * documentation updates
3193 - add man pages for config tools
3194 - update tests/README to describe testing with new config tools
3195 - finish metadata API descriptions
3196 * bug fixes and cleanups
3197 - statfs workaround for 16TB limit
3198 - LOV stripe allocation improved, can stripe on subset of OSTs
3199 - LOV file size/IO offset was wrong for files > 4GB in size
3200 - object EA data was being dropped, caused files to be unreadable
3201 - memory overflow with non-LOV OST caused memory corruption
3202 - fixed regression tests to work with new config tools, obdfilter
3203 - fixed bug when directory size became larger than 1 block
3204 - fixed bug (for single client case) when PWD was deleted
3205 - invalidate local directory pages when doing intent-based ops
3206 - avoid LDLM oops when lock callback contained bad data
3208 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3210 * documentation updates
3211 * bug fixes and cleanups
3212 - configuration tools
3215 - 64-bit compile warnings
3216 - 64-bit internal statfs data
3218 * test_brw on persistent OST devices
3220 * lprocfs (disabled)
3222 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3224 * documentation updates
3225 * bug fixes and cleanups
3226 * configuration tools
3228 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3231 * many small fixes to 0_5_4
3232 * io/network handling
3233 * thinkos in MDS operations
3235 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3238 * crucial basic fixes to 0.5.3
3239 * IOR, Iozone work over Elan
3242 2002-08-07 Phil Schwan <phil@clusterfs.com>
3243 * version 0_5_3, our first alpha
3244 * we use the new Portals iovs
3245 * documentation updates
3246 * bug fixes and cleanups
3247 * small changes in the DLM wire protocol
3249 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3250 * version 0_5_1 with some initial stability,
3251 * locking on MD and file I/O.
3252 * documentation updates
3253 * several bug fixes since 0.5.0
3254 * small changes in wire protocol
3256 2002-07-18 Phil Schwan <phil@clusterfs.com>
3258 * delivered as Lustre Light Alpha
3259 * fixed a crash after handling invalid MDS requests
3260 * fixed directory pages for architectures with non-4k pages sizes
3262 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3263 * release version v0_4_4
3264 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3266 * Fixes a number of interruption problems with OST operations.
3267 * Update documentation for portals header changes
3268 * Move all wire protocol structs/defines to lustre_idl.h
3269 * Fixes symlink length bug.
3270 * Add tcpdump to repository.
3272 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3273 * release version v0_4_3
3274 * Fixes statfs for inodes on extN.
3275 * Fixes bug in runtests which would delete /etc/hosts.
3276 * Use 64-bit object IDs wherever possible (not into VFS though)
3277 Remove ost_get_info, which is unused by lustre, and out of date.
3279 2002-07-03 Peter Braam <braam@clusterfs.com>
3280 * release version v0_4_2 Fixes a lookup error (type not passed)
3281 * move forward to head of Portals
3282 * move forward to latest Lustre kernel
3284 2002-06-25 Peter Braam <braam@clusterfs.com>
3285 * release version v0_4_1. Hopefully stable on single node use.