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.
38 --------------------------------------------------------------------------------
40 tbd Cluster File Systems, Inc. <info@clusterfs.com>
42 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
43 INTEROPERATE with older versions automatically. In many cases a
44 special upgrade step is needed. Please read the
45 user documentation before upgrading any part of a 1.4.x system.
46 * WARNING: Lustre configuration and startup changes are required with
47 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
49 * Support for kernels:
50 2.6.9-42.0.10.EL (RHEL 4)
52 2.4.21-47.0.1.EL (RHEL 3)
53 2.6.12.6 vanilla (kernel.org)
54 2.6.16.27-0.9 (SLES 10)
55 * Client support for unpatched kernels:
56 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
57 2.6.16 - 2.6.19 vanilla (kernel.org)
58 2.6.9-42.0.8EL (RHEL 4)
59 * Recommended e2fsprogs version: 1.39.cfs5
61 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
64 Severity : enhancement
66 Description: MountConf
67 Details : Lustre configuration is now managed via mkfs and mount
68 commands instead of lmc and lconf. New obd types (MGS, MGC)
69 are added for dynamic configuration management. See
70 https://mail.clusterfs.com/wikis/lustre/MountConf for
73 Severity : enhancement
75 Description: dynamic OST addition
76 Details : OSTs can now be added to a live filesystem
78 Severity : enhancement
80 Description: startup order invariance
81 Details : MDTs and OSTs can be started in any order. Clients only
82 require the MDT to complete startup.
84 Severity : enhancement
86 Description: parallel, asynchronous orphan cleanup
87 Details : orphan cleanup is now performed in separate threads for each
88 OST, allowing parallel non-blocking operation.
90 Severity : enhancement
92 Description: optimized stripe assignment
93 Details : stripe assignments are now made based on ost space available,
94 ost previous usage, and OSS previous usage, in order to try
95 to optimize storage space and networking resources.
97 Severity : enhancement
99 Description: Permanently set tunables
100 Details : All writable /proc/fs/lustre tunables can now be permanently
101 set on a per-server basis, at mkfs time or on a live
104 Severity : enhancement
106 Description: Lustre message v2
107 Details : Add lustre message format v2.
109 Severity : enhancement
111 Description: client OST exclusion list
112 Details : Clients can be started with a list of OSTs that should be
113 declared "inactive" for known non-responsive OSTs.
116 Frequency : SFS test only (otherwise harmless)
118 Description: SPEC SFS validation failure on NFS v2 over lustre.
119 Details : Changes the blocksize for regular files to be 2x RPC size,
120 and not depend on stripe size.
122 Severity : enhancement
124 Description: Multiple MD RPCs in flight.
125 Details : Further unserialise some read-only MDT RPCs - learn about intents.
126 To avoid overly-overloading MDT, introduce a limit on number of
127 MDT RPCs in flight for a single client and add /proc controls
128 to adjust this limit.
130 Severity : enhancement
132 Description: client read/write statistics
133 Details : Add client read/write call usage stats for performance
134 analysis of user processes.
135 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
136 file access. extents_stats shows chunk size distribution.
137 extents_stats_per_process show chunk size distribution per
140 Severity : enhancement
142 Description: per-client statistics on server
143 Details : Add ldlm and operations statistics for each client in
144 /proc/fs/lustre/mds|obdfilter/*/exports/
146 Severity : enhancement
148 Description: improved MDT statistics
149 Details : Add detailed MDT operations statistics in
150 /proc/fs/lustre/mds/*/stats
152 Severity : enhancement
154 Description: VFS operations stats
155 Details : Add client VFS call stats, trackable by pid, ppid, or gid
156 /proc/fs/lustre/llite/*/vfs_ops_stats
157 /proc/fs/lustre/llite/*/vfs_track_[pid|ppid|gid]
162 Description: Fix client-side osc byte counters
163 Details : The osc read/write byte counters in
164 /proc/fs/lustre/osc/*/stats are now working
167 Frequency : always as root on SLES
169 Description: Failure of copying files with lustre special EAs.
170 Details : Client side always return success for setxattr call for lustre
171 special xattr (currently only "trusted.lov").
176 Description: Refcount LNET uuids
177 Details : The global LNET uuid list grew linearly with every startup;
178 refcount repeated list entries instead of always adding to
181 Severity : enhancement
183 Description: Dynamic service threads
184 Details : Within a small range, start extra service threads
185 automatically when the request queue builds up.
188 Frequency : mixed-endian client/server environments
190 Description: mixed-endian crashes
191 Details : The new msg_v2 system had some failures in mixed-endian
194 Severity : enhancement
196 Description: Easy OST removal
197 Details : OSTs can be permanently deactivated with e.g. 'lctl
198 conf_param lustre-OST0001.osc.active=0'
200 Severity : enhancement
202 Description: MGS proc entries
203 Details : Added basic proc entries for the MGS showing what filesystems
206 Severity : enhancement
208 Description: provide MGS failover
209 Details : Added config lock reacquisition after MGS server failover.
211 Severity : enhancement
213 Description: add Linux 2.4 support
214 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
218 Description: a large application tries to do I/O to the same resource and dies
220 Details : Check the req->rq_arrival time after the call to
221 ost_brw_lock_get(), but before we do anything about
222 processing it & sending the BULK transfer request. This
223 should help move old stale pending locks off the queue as
224 quickly as obd_timeout.
227 Frequency : when an incorrect nid is specified during startup
229 Description: ptlrpc connect to non-existant node causes kernel crash
230 Details : LNET can't be re-entered from an event callback, which
231 happened when we expire a message after the export has been
232 cleaned up. Instead, hand the zombie cleanup off to another
235 Severity : enhancement
237 Description: plain/inodebits lock performance improvement
238 Details : Grouping plain/inodebits in granted list by their request modes
239 and bits policy, thus improving the performance of search through
243 Frequency : only for kernels with patches from Lustre below 1.4.3
245 Description: Remove old rdonly API
246 Details : Remove old rdonly API which unsed from at least lustre 1.4.3
249 Frequency : only for devices with external journals
251 Description: Set external device read-only also
252 Details : During a commanded failover stop, we set the disk device
253 read-only while the server shuts down. We now also set any
254 external journal device read-only at the same time.
257 Frequency : when upgrading from 1.4 while trying to change parameters
259 Description: The wrong (new) MDC name was used when setting parameters for
260 upgraded MDT's. Also allows changing of OSC (and MDC)
261 parameters if --writeconf is specified at tunefs upgrade time.
264 Frequency : when setting specific ost indicies
266 Description: QOS code breaks on skipped indicies
267 Details : Add checks for missing OST indicies in the QOS code, so OSTs
268 created with --index need not be sequential.
270 ------------------------------------------------------------------------------
272 TBD Cluster File Systems, Inc. <info@clusterfs.com>
274 * Support for kernels:
275 2.6.16.21-0.8 (SLES10)
276 2.6.9-42.0.8EL (RHEL 4)
278 2.4.21-47.0.1.EL (RHEL 3)
279 2.6.12.6 vanilla (kernel.org)
280 * Recommended e2fsprogs version: 1.39.cfs2-0
283 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
285 Description: System hang on startup
286 Details : This bug allowed the liblustre (e.g. catamount) client to
287 return to the app before handling all startup RPCs. This
288 could leave the node unresponsive to lustre network traffic
289 and manifested as a server ptllnd timeout.
291 Severity : enhancement
293 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
294 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
295 boolean to control whether to print peer debug info when a
296 client's RPC times out.
301 Description: make O_SYNC working on 2.6 kernels
302 Details : 2.6 kernels use different method for mark pages for write,
303 so need add a code to lustre for O_SYNC work.
308 Description: Failure to close file and release space on NFS
309 Details : Put inode details into lock acquired in ll_intent_file_open.
310 Use mdc_intent_lock in ll_intent_open to properly
311 detect all kind of errors unhandled by mdc_enqueue
316 Description: proc file read during shutdown sometimes raced obd removal,
318 Details : Add lock to prevent obd access after proc file removal.
321 Frequency : Only for files larger than 4GB on 32-bit clients.
323 Description: improperly doing page alignment of locks
324 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
325 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
330 Description: RPCs being resent when they shouldn't be
331 Details : Some RPCs that should not be resent are being resent. This
332 can cause inconsistencies in the RPC state machine. Do not
333 resend such requests.
336 Frequency : rare, only with NFS export
338 Description: Crash on NFS re-export node
339 Details : under very unusual load conditions an assertion is hit in
340 ll_intent_file_open()
343 Frequency : only if OST filesystem is corrupted
345 Description: client incorrectly hits assertion in ptlrpc_replay_req()
346 Details : for a short time RPCs with bulk IO are in the replay list,
347 but replay of bulk IOs is unimplemented. If the OST filesystem
348 is corrupted due to disk cache incoherency and then replay is
349 started it is possible to trip an assertion. Avoid putting
350 committed RPCs into the replay list at all to avoid this issue.
355 Description: large O_DIRECT requests fail under memory pressure/fragmentation
356 Details : Large single O_DIRECT read and write calls can fail to allocate
357 a sufficiently large buffer to process the request. In case of
358 allocation failure the allocation is retried with a smaller
359 buffer and broken into smaller requests.
364 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
365 > 32000 subdirectories
366 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
367 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
368 subdirectory count crosses 32000. This will aid e2fsck to
369 correctly handle more than 32000 subdirectories.
374 Description: versioning check is incomplete
375 Details : Checking the version difference of client vs. server, report
376 error if the gap is too big.
381 Description: add support PG_writeback bit
382 Details : add support for PG_writeback bit for Lustre, for more carefull
383 work with page cache in 2.6 kernel. This also fix some deadlocks
384 and remove hack for work O_SYNC with 2.6 kernel.
386 ------------------------------------------------------------------------------
388 2006-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
390 * Support for kernels:
391 2.6.16.21-0.8 (SLES10)
392 2.6.9-42.0.3EL (RHEL 4)
394 2.4.21-47.0.1.EL (RHEL 3)
395 2.6.12.6 vanilla (kernel.org)
398 * The backwards-compatible /proc/sys/portals symlink has been removed
399 in this release. Before upgrading, please ensure that you change
400 any configuration scripts or /etc/sysctl.conf files that access
401 /proc/sys/portals/* or sysctl portals.* to use the corresponding
402 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
403 in advance of the upgrade on any system running Lustre 1.4.6 or
404 newer, since /proc/sys/lnet was added in that version.
405 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
409 Frequency : MDS failover only, very rarely
411 Description: "went back in time" messages on mds failover
412 Details : The greatest transno may be lost when the current operation
413 finishes with an error (transno==0) and the client's last_rcvd
414 record is over-written. Save the greatest transno in the
415 mds_last_transno for this case.
418 Frequency : always for specific kernels and striping counts
420 Description: client may get "Matching packet too big" without ACL support
421 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
422 when trying to access files in certain configurations. The
423 clients should in fact be denied when mounting because they do
427 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
429 Description: many clients connecting with IO in progress causes connect timeouts
430 Details : Avoid synchronous journal commits to avoid delays caused by many
431 clients connecting/disconnecting when bulk IO is in progress.
432 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
433 OST_IO_PORTAL to avoid delays behind potentially many pending
437 Frequency : occasionally with multiple writers to a single file
439 Description: shared writes to file may result in wrong size reported by stat()
440 Details : Allow growing of kms when extent lock is cancelled
443 Frequency : always with random mmap IO to multi-striped file
445 Description: mmap write might be lost if we are writing to a 'hole' in stripe
446 Details : Only if the hole is at the end of OST object so that kms is too
447 small. Fix is to increase kms accordingly in ll_nopage.
450 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
452 Description: writes to a missing object would leak memory on the OST
453 Details : If there is an inconsistency between the MDS and OST filesystems,
454 such that the MDS references an object that doesn't exist, writes
455 to that object will leak memory due to incorrect cleanup in the
456 error handling path, eventually running out of memory on the OST.
461 Description: Creating too long symlink causes lustre errors
462 Details : Check symlink and name lengths before sending requests to MDS.
465 Frequency : only if flock is enabled (not on by default)
467 Description: posix locks not released on fd closure on 2.6.9+
468 Details : We failed to add posix locks to list of inode locks on 2.6.9+
469 kernels, this caused such locks not to be released on fd close and
470 then assertions on fs unmount about still used locks.
473 Frequency : MDS failover only, very rarely
475 Description: clients may get ASSERTION(granted_lock != NULL)
476 Details : When request was taking a long time, and a client was resending
477 a getattr by name lock request. The were multiple lock requests
478 with the same client lock handle and
479 mds_getattr_name->fixup_handle_for_resent_request found one of the
480 lock handles but later failed with ASSERTION(granted_lock != NULL).
485 Description: handle->h_buffer_credits > 0, assertion failure
486 Details : h_buffer_credits is zero after truncate, causing assertion
487 failure. This patch extends the transaction or creates a new
491 Frequency : NFS re-export or patchless client
492 Bugzilla : 11179, 10796
493 Description: Crash on NFS re-export node (__d_move)
494 Details : We do not want to hash the dentry if we don't have a lock.
495 But if this dentry is later used in d_move, we'd hit uninitialised
496 list head d_hash, so we just do this to init d_hash field but
497 leave dentry unhashed.
500 Frequency : NFS re-export or patchless client
502 Description: NFS exports has problem with symbolic link
503 Details : lustre client didn't properly install dentry when re-exported
504 to NFS or running patchless client.
507 Frequency : NFS re-export or patchless client
509 Description: Various nfs/patchless fixes.
510 Details : fixes reuse disconected alias for lookup process - this fixes
511 warning "find_exported_dentry: npd != pd",
512 fix permission error with open files at nfs.
513 fix apply umaks when do revalidate.
516 Frequency : occasional
518 Description: Crash on NFS re-export node
519 Details : calling clear_page() on the wrong pointer triggered oops in
520 generic_mapping_read().
523 Frequency : rarely, using O_DIRECT IO
525 Description: unaligned directio crashes client with LASSERT
526 Details : check for unaligned buffers before trying any requests.
529 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
531 Description: stale data returned from RAID cache
532 Details : If only a small amount of IO is done to the RAID device before
533 reading it again it is possible to get stale data from the RAID
534 cache instead of reading it from disk.
537 Frequency : depends on arch, kernel and compiler version, always on sles10
540 Description: recursive or deep enough symlinks cause stack overflow
541 Details : getting rid of large stack-allocated variable in
545 Frequency : depends on hardware
547 Description: lustre write performance loss in the SLES10 kernel
548 Details : the performance loss is caused by using of write barriers in the
549 ext3 code. The SLES10 kernel turns barrier support on by
550 default. The fix is to undo that change for ldiskfs.
552 ------------------------------------------------------------------------------
554 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
556 * Support for kernels:
557 2.6.9-42.0.3EL (RHEL 4)
559 2.4.21-47.0.1.EL (RHEL 3)
560 2.6.12.6 vanilla (kernel.org)
564 Frequency : quota enabled and large files being deleted
566 Description: releasing more than 4GB of quota at once hangs OST
567 Details : If a user deletes more than 4GB of files on a single OST it
568 will cause the OST to spin in an infinite loop. Release
569 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
574 Description: statfs data retrieved from /proc may be stale or zero
575 Details : When reading per-device statfs data from /proc, in the
576 {kbytes,files}_{total,free,avail} files, it may appear
577 as zero or be out of date.
580 Frequency : systems with MD RAID1 external journal devices
582 Description: lconf's call to blkid is confused by RAID1 journal devices
583 Details : Use the "blkid -l" flag to locate the MD RAID device instead
584 of returning all block devices that match the journal UUID.
587 Frequency : always, for aggregate stripe size over 4GB
589 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
590 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
591 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
594 Frequency : always if "lfs find" used on a local file/directory
596 Description: "lfs find" segfaults if used on a local file/directory
597 Details : The case where a directory component was not specified wasn't
598 handled correctly. Handle this properly.
601 Frequency : always on ppc64
603 Description: the write to an ext3 filesystem mounted with mballoc got stuck
604 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
605 perform endianness conversion.
608 Frequency : rarely (truncate to non-zero file size after write under load)
609 Bugzilla : 10730, 10687
610 Description: Files padded with zeros to next 4K multiple
611 Details : With filesystems mounted using the "extents" option (2.6 kernels)
612 it is possible that files that are truncated to a non-zero size
613 immediately after being written are filled with zero bytes beyond
614 the truncated size. No file data is lost.
616 Severity : enhancement
618 Description: Allow recovery/failover for liblustre clients.
619 Details : liblustre clients were unaware of failover configurations until
622 Severity : enhancement
624 Description: user file locks should fail when not mounting with flock option
625 Details : Set up an error-returning stub in ll_file_operations.lock field
626 to prevent incorrect behaviour when client is mounted without
627 flock option. Also, set up properly f_op->flock field for
631 Frequency : always on ia64
633 Description: "lfs df" loops on printing out MDS statfs information
634 Details : The obd_ioctl_data was not initialized and in some systems
635 this caused a failure during the ioctl that did not return
636 an error. Initialize the struct and return an error on failure.
639 Frequency : SLES 9 only
641 Description: Error of copying files with lustre special EAs as root
642 Details : Client side always return success for setxattr call for lustre
643 special xattr (currently only "trusted.lov").
646 Frequency : rarely on clusters with both ia64+i386 clients
648 Description: ia64+i686 clients doing shared IO on the same file may LBUG
649 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
650 clients are doing concurrent writes to the same file it is
651 possible that the ia64 clients may LASSERT because the OST
652 extent locks are not PAGE_SIZE aligned. Ensure that grown
653 locks are always aligned on the request boundary.
656 Frequency : specific use, occasional
658 Description: Overwriting in use executable truncates on-disk binary image
659 Details : If one node attempts to overwrite an executable in use by
660 another node, we now correctly return ETXTBSY instead of
666 Description: chmod on Lustre root is propagated to other clients
667 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
668 invalid by the follow_mount time.
671 Frequency : liblustre clients only
673 Description: Race in 'instant cancel' lock handling could lead to such locks
674 never to be granted in case of SMP MDS
675 Details : Do not destroy not yet granted but cbpending locks in
679 Frequency : replay/resend of open
681 Description: non null lock assetion failure in mds_intent_policy
682 Details : Trying to replay/resend lockless open requests resulted in
683 mds_open() returning 0 with no lock. Now it sets a flag if
684 a lock is going to be returned.
686 Severity : enhancement
688 Description: Checksum enhancements
689 Details : New checksum enhancements allow for resending RPCs that failed
692 Severity : enhancement
694 Description: Tunables on number of dirty pages in cacche
695 Details : Allow to set limit on number of dirty pages cached.
700 Description: client crash on unmount - lock still has references
701 Details : In some error handling cases it was possible to leak a lock
702 reference on a client while accessing a file. This was not
703 harmful to the client during operation, but would cause the
704 client to crash when the filesystem is unmounted.
707 Frequency : specific case, rare
709 Description: ETXTBSY on mds though file not in use by client
710 Details : ETXTBSY is no longer incorrectly returned when attempting to
711 chmod or chown a directory that the user previously tried to
712 execute or a currently-executing binary.
715 Frequency : extremely rare except on liblustre-based clients
717 Description: Lustre space not freed when files are deleted
718 Details : Clean up open-unlinked files after client eviction. Previously
719 the unlink was skipped and the files remained as orphans.
724 Description: OST failure "would be an LBUG" in waiting_locks_callback()
725 Details : In some cases it was possible to send a blocking callback to
726 a client doing a glimpse, even though that client didn't get
727 a lock granted. When the glimpse lock is cancelled on the OST
728 the freed lock is left on the waiting list and corrupted the list.
731 Frequency : all core dumps
733 Description: Broke core dumps to lustre
734 Details : Negative dentry may be unhashed if parent does not have UPDATE
735 lock, but some callers, e.g. do_coredump, expect dentry to be
736 hashed after successful create, hash it in ll_create_it.
738 ------------------------------------------------------------------------------
740 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
742 * Support for kernels:
743 2.6.9-42.0.2.EL (RHEL 4)
745 2.4.21-40.EL (RHEL 3)
746 2.6.12.6 vanilla (kernel.org)
750 Frequency : always on RHEL 3
752 Description: Number of open files grows over time
753 Details : The number of open files grows over time, whether or not
754 Lustre is started. This was due to a filp leak introduced
755 by one of our kernel patches.
757 ------------------------------------------------------------------------------
759 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
761 * Support for kernels:
764 2.4.21-40.EL (RHEL 3)
765 2.6.12.6 vanilla (kernel.org)
770 Bugzilla : 5719, 9635, 9792, 9684
771 Description: OST (or MDS) trips assertions in (re)connection under heavy load
772 Details : If a server is under heavy load and cannot reply to new
773 connection requests before the client resends the (re)connect,
774 the connection handling code can behave badly if two service
775 threads are concurrently handing separate (re)connections from
776 the same client. Add better locking to the connection handling
777 code, and ensure that only a single connection will be processed
778 for a given client UUID, even if the lock is dropped.
780 Severity : enhancement
782 Description: add TCP zero-copy support to kernel
783 Details : Add support to the kernel TCP stack to allow zero-copy bulk
784 sends if the hardware supports scatter-gather and checksumming.
785 This allows socklnd to do client-write and server-read more
786 efficiently and reduce CPU utilization from skbuf copying.
789 Frequency : only if NFS exporting from client
791 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
792 Details : If setting attributes on a file created under NFS that had
793 never been opened it would be possible to oops the client
794 if the file had no objects.
798 Bugzilla : 9326, 10402, 10897
799 Description: client crash in ptlrpcd_wake() thread when sending async RPC
800 Details : It is possible that ptlrpcd_wake() dereferences a freed async
801 RPC. In rare cases the ptlrpcd thread alread processed the RPC
802 before ptlrpcd_wake() was called and the request was freed.
805 Frequency : always for liblustre
807 Description: liblustre client does MDS+OSTs setattr RPC for each write
808 Details : When doing a write from a liblustre client, the client
809 incorrectly issued an RPC to the MDS and each OST the file was
810 striped over in order to update the timestamps. When writing
811 with small chunks and many clients this could overwhelm the MDS
812 with RPCs. In all cases it would slow down the write because
813 these RPCs are unnecessary.
815 Severity : enhancement
817 Description: allow number of MDS service threads to be changed at module load
818 Details : It is now possible to change the number of MDS service threads
819 running. Adding "options mds mds_num_threads={N}" to the MDS's
820 /etc/modprobe.conf will set the number of threads for the next
821 time Lustre is restarted (assuming the "mds" module is also
822 reloaded at that time). The default number of threads will
823 stay the same, 32 for most systems.
828 Description: OST crash if filesystem is unformatted or corrupt
829 Details : If an OST is started on a device that has never been formatted
830 or if the filesystem is corrupt and cannot even mount then the
831 error handling cleanup routines would dereference a NULL pointer.
836 Description: NULL pointer deref in llap_from_page.
837 Details : get_cache_page_nowait can return a page with NULL (or otherwise
838 incorrect) mapping if the page was truncated/reclaimed while it was
839 searched for. Check for this condition and skip such pages when
840 doing readahead. Introduce extra check to llap_from_page() to
841 verify page->mapping->host is non-NULL (so page is not anonymous).
844 Frequency : Sometimes when using sys_sendfile
846 Description: "page not covered by a lock" warnings from ll_readpage
847 Details : sendfile called ll_readpage without right page locks present.
848 Now we introduced ll_file_sendfile that does necessary locking
849 around call to generic_file_sendfile() much like we do in
853 Frequency : with certain MDS communication failures at client mount time
855 Description: NULL pointer deref after failed client mount
856 Details : a client connection request may delayed by the network layer
857 and not be sent until after the PTLRPC layer has timed out the
858 request. If the client fails the mount immediately it will try
859 to clean up before the network times out the request. Add a
860 reference from the request import to the obd device and delay
861 the cleanup until the network drops the request.
864 Frequency : occasionally during client (re)connect
866 Description: assertion failure during client (re)connect
867 Details : processing a client connection request may be delayed by the
868 client or server longer than the client connect timeout. This
869 causes the client to resend the connection request. If the
870 original connection request is replied in this interval, the
871 client may trip an assertion failure in ptlrpc_connect_interpret()
872 which thought it would be the only running connect process.
875 Frequency : only with obd_echo servers and clients that are rebooted
877 Description: kernel BUG accessing uninitialized data structure
878 Details : When running an obd_echo server it did not start the ping_evictor
879 thread, and when a client was evicted an uninitialized data
880 structure was accessed. Start the ping_evictor in the RPC
881 service startup instead of the OBD startup.
883 Severity : enhancement
884 Bugzilla : 10193 (patchless)
885 Description: Remove dependency on various unexported kernel interfaces.
886 Details : No longer need reparent_to_init, exit_mm, exit_files,
887 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
890 Frequency : rare (only users of deprecated and unsupported LDAP config)
892 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
893 Details : LDAP apparently contains 'lustreName' attributes instead of
894 'name'. A simple remapping of the name is sufficient.
897 Frequency : rare (only with non-default dump_on_timeout debug enabled)
899 Description: waiting_locks_callback trips kernel BUG if client is evicted
900 Details : Running with the dump_on_timeout debug flag turned on makes
901 it possible that the waiting_locks_callback() can try to dump
902 the Lustre kernel debug logs from an interrupt handler. Defer
903 this log dumping to the expired_lock_main() thread.
905 Severity : enhancement
907 Description: Support NFS exporting on 2.6 kernels.
908 Details : Implement non-rawops metadata methods for NFS server to use without
909 changing NFS server code.
912 Frequency : very rare (synthetic metadata workload only)
914 Description: two racing renames might cause an MDS thread to deadlock
915 Details : Running the "racer" program may cause one MDS thread to rename
916 a file from being the source of a rename to being the target of
917 a rename at exactly the same time that another thread is doing
918 so, and the second thread has already enqueued these locks after
919 doing a lookup of the target and is trying to relock them in
920 order. Ensure that we don't try to re-lock the same resource.
923 Frequency : only very large systems with liblustre clients
925 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
926 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
927 all OSTs in parallel. This allows the eviction of stale liblustre
928 clients to proceed much faster than if they were done in series,
929 and also offers similar improvements for other set_info RPCs.
934 Description: excessive CPU usage during initial read phase on client
935 Details : During the initial read phase on a client, it would agressively
936 retry readahead on the file, consuming too much CPU and impacting
937 performance (since 1.4.5.8). Improve the readahead algorithm
938 to avoid this, and also improve some other common cases (read
939 of small files in particular, where "small" is files smaller than
940 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
945 Description: MDS crash when receiving packet with unknown intent.
946 Details : Do not LBUG in unknown intent case, just return -EFAULT
948 Severity : enhancement
949 Bugzilla : 9293, 9385
950 Description: MDS RPCs are serialised on client. This is unnecessary for some.
951 Details : Do not serialize getattr (non-intent version) and statfs.
954 Frequency : occasional, when OST network is overloaded/intermittent
956 Description: client evicted by OST after bulk IO timeout
957 Details : If a client sends a bulk IO request (read or write) the OST
958 may evict the client if it is unresposive to its data GET/PUT
959 request. This is incorrect if the network is overloaded (takes
960 too long to transfer the RPC data) or dropped the OST GET/PUT
961 request. There is no need to evict the client at all, since
962 the pinger and/or lock callbacks will handle this, and the
963 client can restart the bulk request.
966 Frequency : Always when mmapping file with no objects
968 Description: client crashes when mmapping file with no objects
969 Details : Check that we actually have objects in a file before doing any
970 operations on objects in ll_vm_open, ll_vm_close and
976 Description: Request leak when working with deleted CWD
977 Details : Introduce advanced request refcount tracking for requests
978 referenced from lustre intent.
980 Severity : Enhancement
982 Description: Cache open file handles on client.
983 Details : MDS now will return special lock along with openhandle, if
984 requested and client is allowed to hold openhandle, even if unused,
985 until such a lock is revoked. Helps NFS a lot, since NFS is opening
986 closing files for every read/write openration.
988 Severity : Enhancement
990 Description: Cache open negative dentries on client when possible.
991 Details : Guard negative dentries with UPDATE lock on parent dir, drop
992 negative dentries on lock revocation.
997 Description: Remounting a client read-only wasn't possible with a zconf mount
998 Details : It wasn't possible to remount a client read-only with llmount.
1000 Severity : enhancement
1001 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1002 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1003 MPICH is included to add Lustre-specific ADIO interfaces.
1004 This is based closely on the UFS ADIO layer and only differs
1005 in file creation, in order to allow the OST striping to be set.
1006 This is user-contributed code and not supported by CFS.
1011 Description: extended inode attributes (immutable, append-only) work improperly
1012 when 2.4 and 2.6 kernels are used on client/server or vice versa
1013 Details : Introduce kernel-independent values for these flags.
1015 Severity : enhancement
1018 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1019 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1020 granularity. Now, Fractional MB granularity is be supported,
1021 this is very useful for low memory system.
1023 Severity : enhancement
1025 Description: Getattr by fid
1026 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1027 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1028 have locks and avoids getattr rpc requests for it on every stat.
1031 Frequency : Always, for filesystems larger than 2TB
1033 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1034 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1035 filesystems larger than 2TB to be mounted with mballoc enabled.
1038 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1040 Description: filesystem corruption for non-standard kernels and very large OSTs
1041 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1042 filesystems larger than 2TB is mounted then the kernel will
1043 silently corrupt the start of the filesystem. CONFIG_LBD is
1044 enabled for all CFS-supported kernels, but the possibility of
1045 this happening with a modified kernel config exists.
1047 Severity : enhancement
1049 Description: add client O_DIRECT support for 2.6 kernels
1050 Details : It is now possible to do O_DIRECT reads and writes to files
1051 in the Lustre client mountpoint on 2.6 kernel clients.
1053 Severity : enhancement
1055 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1056 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1057 parallel, not waiting for response from every OST before sending
1058 a rpc to the next OST.
1063 Description: setattr vs write race when updating file timestamps
1064 Details : Client processes that update a file timestamp into the past
1065 right after writing to the file (e.g. tar) it is possible that
1066 the updated file modification time can be reset to the current
1067 time due to a race between processing the setattr and write RPC.
1069 Severity : enhancement
1071 Description: Bring 'lfs find' closer in line with regular Linux find.
1072 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1073 -print0 options and obtains all the needed info through the lustre
1076 Severity : enhancement
1078 Description: support up to 1024 configured devices on one node
1079 Details : change obd_dev array from statically allocated to dynamically
1080 allocated structs as they are first used to reduce memory usage
1085 Description: Flush dirty partially truncated pages during truncate
1086 Details : Immediatelly flush partially truncated pages in filter_setattr,
1087 this way we completely avoid having any pages in page cache on OST
1088 and can retire ugly workarounds during writes to flush such pages.
1093 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1094 Details : protect inode from truncation within vfs_unlink() context
1095 just take a reference before calling vfs_unlink() and release it
1096 when parent's i_sem is free.
1099 Frequency : always, if extents are used on OSTs
1101 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1102 Details : OSTs using the extents format would not zero the high 16 bits of
1103 the index physical block number. This is not a problem for any
1104 OST filesystems smaller than 16TB, and no kernels support ext3
1105 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1106 new/modified files) and can be fixed for existing filesystems
1107 with e2fsprogs-1.39-cfs1.
1112 Description: import connection selection may be incorrect if timer wraps
1113 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1114 import connections to be ignored if the 32-bit jiffies counter
1115 wraps. Use a 64-bit jiffies counter.
1118 Frequency : during server recovery
1120 Description: crash after server is denying duplicate export
1121 Details : If clients are resending connect requests to the server, the
1122 server refuses to allow a client to connect multiple times.
1123 Fixed a bug in the handling of this case.
1126 Frequency : very large clusters immediately after boot
1128 Description: LNET request buffers exhausted under heavy short-term load
1129 Details : If a large number of client requests are generated on a service
1130 that has previously never seen so many requests it is possible
1131 that the request buffer growth cannot keep up with the spike in
1132 demand. Instead of dropping incoming requests, they are held in
1133 the LND until the RPC service can accept more requests.
1136 Frequency : Sometimes during replay
1138 Description: Assertion failure in ll_local_open after replay.
1139 Details : If replay happened on an open request reply before we were able
1140 to set replay handler, reply will become not swabbed tripping the
1141 assertion in ll_local_open. Now we set the handler right after
1142 recognising of open request
1145 Frequency : very rare
1147 Description: kernel reports "badness in vsnprintf"
1148 Details : Reading from the "recovery_status" /proc file in small chunks
1149 may cause a negative length in lprocfs_obd_rd_recovery_status()
1150 call to vsnprintf() (which is otherwise harmless). Exit early
1151 if there is no more space in the output buffer.
1153 Severity : enhancement
1155 Description: clear OBD RPC statistics by writing to them
1156 Details : It is now possible to clear the OBD RPC statistics by writing
1157 to the "stats" file.
1162 Description: Client mtime is not the same on different clients after utimes
1163 Details : In some cases, the client was using the utimes() syscall on
1164 a file cached on another node. The clients now validate the
1165 ctime from the MDS + OSTs to determine which one is right.
1170 Description: Inability to activate failout mode
1171 Details : lconf script incorrectly assumed that in python string's numeric
1172 value is used in comparisons.
1175 Frequency : always with multiple stripes per file
1177 Description: Inefficient object allocation for mutli-stripe files
1178 Details : When selecting which OSTs to stripe files over, for files with
1179 a stripe count that divides evenly into the number of OSTs,
1180 the MDS is always picking the same starting OST for each file.
1181 Return the OST selection heuristic to the original design.
1186 Description: mount failures may take full timeout to return an error
1187 Details : Under some heavy load conditions it is possible that a
1188 failed mount can wait for the full obd_timeout interval,
1189 possibly several minutes, before reporting an error.
1190 Instead return an error as soon as the status is known.
1192 ------------------------------------------------------------------------------
1194 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
1196 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1197 INTEROPERATE with older versions automatically. Please read the
1198 user documentation before upgrading any part of a live system.
1199 * WARNING: Lustre networking configuration changes are required with
1200 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1203 * Support for kernels:
1204 2.6.9-22.0.2.EL (RHEL 4)
1205 2.6.5-7.244 (SLES 9)
1206 2.6.12.6 vanilla (kernel.org)
1209 Severity : enhancement
1210 Bugzilla : 7981/8208
1211 Description: Introduced Lustre Networking (LNET)
1212 Details : LNET is new networking infrastructure for Lustre, it includes
1213 a reorganized network configuration mode (see the user
1214 documentation for full details) as well as support for routing
1215 between different network fabrics. Lustre Networking Devices
1216 (LNDs) for the supported network fabrics have also been
1217 created for this new infrastructure.
1219 Severity : enhancement
1220 Description: Introduced Access control lists
1221 Details : clients can set ACLs on files and directories in order to have
1222 more fine-grained permissions than the standard Unix UGO+RWX.
1223 The MDS must be started with the "-o acl" mount option.
1225 Severity : enhancement
1226 Description: Introduced filesystem quotas
1227 Details : Administrators may now establish per-user quotas on the
1230 Severity : enhancement
1232 Description: Configuration change for the XT3
1233 The PTLLND is now used to run Lustre over Portals on the XT3
1234 The configure option(s) --with-cray-portals are no longer used.
1235 Rather --with-portals=<path-to-portals-includes> is used to
1236 enable building on the XT3. In addition to enable XT3 specific
1237 features the option --enable-cray-xt3 must be used.
1242 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1243 Details : A race between allocating a new llcd and re-getting the llcd_lock
1244 allowed another thread to grab newly-allocated llcd.
1246 Severity : enhancement
1248 Description: 2.6 OST async journal commit and locking fix to improve performance
1249 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1250 2.6 kernels are now async as they already were in 2.4 kernels so
1251 that they can commit concurrently with the network bulk transfer.
1252 For block-allocated files the filter allocation semaphore is held
1253 to avoid filesystem fragmentation during allocation. BKL lock
1254 removed for 2.6 xattr operations where it is no longer needed.
1259 Description: lconf incorrectly determined whether two IP networks could talk
1260 Details : In some more complicated routing and multiple-network
1261 configurations, lconf will avoid trying to make a network
1262 connection to a disjoint part of the IP space. It was doing the
1263 math incorrectly for one set of cases.
1268 Description: Fix for potential infinite loop processing records in an llog.
1269 Details : If an llog record is corrupted/zeroed, it is possible to loop
1270 forever in llog_process(). Validate the llog record length
1271 and skip the remainder of the block on error.
1274 Frequency : occasional (liblustre only)
1276 Description: liblustre could not open files whose last component is a symlink
1277 Details : sysio_path_walk() would incorrectly pass the open intent to
1278 intermediate path components.
1281 Frequency : rare (liblustre only with non-standard tuning)
1282 Bugzilla : 7201 (7350)
1283 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1284 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1285 after it has been released, only do this for referenced locks
1287 Severity : enhancement
1289 Description: specifying an (invalid) directory default stripe_size of -1
1290 would reset the directory default striping
1291 Details : stripe_size -1 was used internally to signal directory stripe
1292 removal, now use "all default" to signal dir stripe removal
1293 as a directory striping of "all default" is not useful
1296 Frequency : common for large clusters running liblustre clients
1298 Description: doing an ls when liblustre clients are running is slow
1299 Details : sending a glimpse AST to a liblustre client waits for every AST
1300 to time out, as liblustre clients will not respond. Since they
1301 cannot cache data we refresh the OST lock LVB from disk instead.
1303 Severity : enhancement
1305 Description: doing an ls at the same time as file IO can be slow
1306 Details : enqueue and other "small" requests can be blocked behind many
1307 large IO requests. Create a new OST IO portal for non-IO
1308 requests so they can be processed faster.
1311 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1313 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1314 created file with mode 000
1315 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1316 allow an ftruncate() as a non-root user to a file with mode 000.
1317 Lustre now respects this flag to disable mode checks when
1318 truncating a file owned by the user
1321 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1324 Description: Revoking locks from clients that went dead or catatonic might take
1326 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1327 cancellation of such locks instant on servers without waiting for
1328 any reply from clients. Clients drops these locks when cancel
1329 notification from server is received without replying.
1332 Frequency : liblustre-only, when liblustre client dies or becomes busy
1334 Description: Doing ls on Linux clients can take a long time with active
1336 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1337 granting such locks to it in the first place if possible. Locks
1338 are taken by proxy on the OST during the read or write and
1339 dropped immediately afterward. Add connect flags handling, do
1340 not grant locks to liblustre clients for glimpse ASTs.
1342 Severity : enhancement
1344 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1345 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1346 sequential reads and behaves badly with applications doing random
1347 reads. Improve it by reading ahead at least read region, and
1348 avoiding excessive large RPC for small reads.
1350 Severity : enhancement
1352 Description: Creating more than 1000 files for a single job may cause a load
1353 imbalance on the OSTs if there are also a large number of OSTs.
1354 Details : qos_prep_create() uses an OST index reseed value that is an
1355 even multiple of the number of available OSTs so that if the
1356 reseed happens in the middle of the object allocation it will
1357 still utilize the OSTs as uniformly as possible.
1362 Description: OST or MDS may oops in ping_evictor_main()
1363 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1364 but doesn't restart at beginning of obd_exports_timed list
1367 Severity : enhancement
1369 Description: improve by-nid export eviction on the MDS and OST
1370 Details : allow multiple exports with the same NID to be evicted at one
1371 time without re-searching the exports list.
1374 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1376 Description: MDS may oops in groups_free()
1377 Details : in rare race conditions a newly allocated group_info struct is
1378 freed again, and this can be NULL. The 2.4 compatibility code
1379 for groups_free() checked for a NULL pointer, but 2.6 did not.
1382 Frequency : common for liblustre clients doing little filesystem IO
1383 Bugzilla : 9352, 7313
1384 Description: server may evict liblustre clients accessing contended locks
1385 Details : if a client is granted a lock or receives a completion AST
1386 with a blocking AST already set it would not reply to the AST
1387 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1390 Frequency : lfs setstripe, only systems with more than 160 OSTs
1392 Description: unable to set striping with a starting offset beyond OST 160
1393 Details : llapi_create_file() incorrectly limited the starting stripe
1394 index to the maximum single-file stripe count.
1397 Frequency : LDAP users only
1399 Description: lconf did not handle in-kernel recovery with LDAP properly
1400 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1402 Severity : enhancement
1404 Description: bind OST threads to NUMA nodes to improve performance
1405 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1406 node and do their allocations there to localize memory access
1408 Severity : enhancement
1410 Description: llmount can determine client NID directly from Myrinet (GM)
1411 Details : the client NID code from gmnalnid was moved directly into
1412 llmount, removing the need to use this or specifying the
1413 client NID explicitly when mounting GM clients with zeroconf
1416 Frequency : if client is started with down MDS
1418 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1419 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1420 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1421 ptlrpc_import_delayed_req() because we didn't check if the
1422 request was interrupted, and we also didn't break out of the
1423 event loop if there was no timeout
1428 Description: data loss during non-page-aligned writes to a single file from
1429 both multiple nodes and multiple threads on one node at same time
1430 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1431 inconsistency led to false short-reads, that were cached and later
1432 used by ->prepare_write() to fill in partially written page,
1433 leading to data loss.
1436 Frequency : always, if lconf --abort_recovery used
1438 Description: lconf --abort_recovery fails with 'Operation not supported'
1439 Details : lconf was attempting to abort recovery on the MDT device and not
1442 Severity : enhancement
1444 Description: remove cleanup logs
1445 Details : replace lconf-generated cleanup logs with lustre internal
1446 cleanup routines. Eliminates the need for client-cleanup and
1449 Severity : enhancement
1451 Description: add support for EAs (user and system) on lustre filesystems
1452 Details : it is now possible to store extended attributes in the Lustre
1453 client filesystem, and with the user_xattr mount option it
1454 is possible to allow users to store EAs on their files also
1456 Severity : enhancement
1458 Description: Add possibility (config option) to show minimal available OST free
1460 Details : When compiled with --enable-mindf configure option, statfs(2)
1461 (and so, df) will return least minimal free space available from
1462 all OSTs as amount of free space on FS, instead of summary of
1463 free spaces of all OSTs.
1465 Severity : enhancement
1467 Description: do not expand extent locks acquired on OST-side
1468 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1469 by server: they are not cached anyway.
1472 Frequency : when mmap is used/binaries executed from Lustre
1474 Description: Unmmap pages before throwing them away from read cache.
1475 Details : llap_shrink cache now attempts to unmap pages before discarding
1476 them (if unmapping failed - do not discard). SLES9 kernel has
1477 extra checks that trigger if this unmapping is not done first.
1482 Description: lconf didn't resolve symlinks before checking to see whether a
1483 given mountpoint was already in use
1486 Frequency : when migrating failover services
1487 Bugzilla : 6395, 9514
1488 Description: When migrating a subset of services from a node (e.g. failback
1489 from a failover service node) the remaining services would
1490 time out and evict clients.
1491 Details : lconf --force (implied by --failover) sets the global obd_timeout
1492 to 5 seconds in order to quickly disconnect, but this caused
1493 other RPCs to time out too quickly. Do not change the global
1494 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1496 Severity : enhancement
1497 Frequency : if MDS is started with down OST
1498 Bugzilla : 9439,5706
1499 Description: Allow startup/shutdown of an MDS without depending on the
1500 availability of the OSTs.
1501 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1502 Add appropriate locking and lov-osc refcounts for safe
1503 cleaning. Add osc abort_inflight calls in case the
1504 synchronize never started.
1507 Frequency : occasional (Cray XT3 only)
1509 Description: root not authorized to access files in CRAY_PORTALS environment
1510 Details : The client process capabilities were not honoured on the MDS in
1511 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1512 been accessed by an authorized user then root was able to access
1513 the file on the local client also. The root user capabilities
1514 are now allowed on the MDS, as this environment has secure UID.
1517 Frequency : occasional
1519 Description: ldiskfs "too long searching" message happens too often
1520 Details : A debugging message (otherwise harmless) prints too often on
1521 the OST console. This has been reduced to only happen when
1522 there are fragmentation problems on the filesystem.
1527 Description: Division by zero in statfs when all OSCs are inactive
1528 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
1529 lov_statfs divides by value returned by lov_get_stripecnt().
1533 Bugzilla : 9489, 3273
1534 Description: First write from each client to each OST was only 4kB in size,
1535 to initialize client writeback cache, which caused sub-optimal
1536 RPCs and poor layout on disk for the first writen file.
1537 Details : Clients now request an initial cache grant at (re)connect time
1538 and so that they can start streaming writes to the cache right
1539 away and always do full-sized RPCs if there is enough data.
1540 If the OST is rebooted the client also re-establishes its grant
1541 so that client cached writes will be honoured under the grant.
1546 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
1547 Details : Now I/O RPCs go to different portal number and (presumably) fast
1548 lock requests (and glimses) and other RPCs get their own service
1549 threads pool that should be able to service those RPCs
1552 Severity : enhancement
1554 Description: Ability to exchange lustre version between client and servers and
1555 issue warnings at client side if client is too old. Also for
1556 liblustre clients there is ability to refuse connection of too old
1558 Details : New 'version' field is added to connect data structure that is
1559 filled with version info. That info is later checked by server and
1563 Frequency : rare, liblustre only.
1564 Bugzilla : 9296, 9581
1565 Description: Two simultaneous writes from liblustre at offset within same page
1566 might proceed at the same time overwriting eachother with stale
1568 Details : I/O lock withing llu_file_prwv was released too early, before data
1569 actually was hitting the wire. Extended lock-holding time until
1570 server acknowledges receiving data.
1573 Frequency : extremely rare. Never observed in practice.
1575 Description: avoid generating lustre_handle cookie of 0.
1576 Details : class_handle_hash() generates handle cookies by incrementing
1577 global counter, and can hit 0 occasionaly (this is unlikely, but
1578 not impossible, because initial value of cookie counter is
1579 selected randonly). Value of 0 is used as a sentinel meaning
1580 "unassigned handle" --- avoid it. Also coalesce two critical
1581 sections in this function into one.
1583 Severity : enhancement
1585 Description: allow liblustre clients to delegate truncate locking to OST
1586 Details : To avoid overhead of locking, liblustre client instructs OST to
1587 take extent lock in ost_punch() on client's behalf. New connection
1588 flag is added to handle backward compatibility.
1590 Severity : enhancement
1591 Bugzilla : 4928, 7341, 9758
1592 Description: allow number of OST service threads to be specified
1593 Details : a module parameter allows the number of OST service threads
1594 to be specified via "options ost ost_num_threads={N}" in the
1595 OSS's /etc/modules.conf or /etc/modprobe.conf.
1599 Bugzilla : 6146, 9635, 9895
1600 Description: servers crash with bad pointer in target_handle_connect()
1601 Details : In rare cases when a client is reconnecting it was possible that
1602 the connection request was the last reference for that export.
1603 We would temporarily drop the export reference and get a new
1604 one, but this may have been the last reference and the export
1605 was just destroyed. Get new reference before dropping old one.
1607 Severity : enhancement
1608 Frequency : if client is started with failover MDS
1610 Description: Allow multiple MDS hostnames in the mount command
1611 Details : Try to read the configuration from all specified MDS
1612 hostnames during a client mount in case the "primary"
1615 Severity : enhancement
1617 Description: Stop sending data to evicted clients as soon as possible.
1618 Details : Check if the client we are about to send or are sending data to
1619 was evicted already. (Check is done every second of waiting,
1620 for which l_wait_event interface was extended to allow checking
1621 of exit condition at specified intervals).
1624 Frequency : rare, normally only when NFS exporting is done from client
1626 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
1628 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
1631 Frequency : always, if srand() is called before liblustre initialization
1633 Description: Liblustre uses system PRNG disturbing its usage by user application
1634 Details : Introduce internal to lustre fast and high-quality PRNG for
1635 lustre usage and make liblustre and some other places in generic
1636 lustre code to use it.
1638 Severity : enhancement
1639 Bugzilla : 9477, 9557, 9870
1640 Description: Verify that the MDS configuration logs are updated when xml is
1641 Details : Check if the .xml configuration logs are newer than the config
1642 logs stored on the MDS and report an error if this is the case.
1643 Request --write-conf, or allow starting with --old_conf.
1645 Severity : enhancement
1647 Description: Handle symlinks in the path when checking if Lustre is mounted.
1648 Details : Resolve intermediate symlinks when checking if a client has
1649 mounted a filesystem to avoid duplicate client mounts.
1654 Description: lconf can hit an error exception but still return success.
1655 Details : The lconf command catches the Command error exception at the top
1656 level script context and will exit with the associated exit
1657 status, but doesn't ensure that this exit status is non-zero.
1662 Description: failure of ptlrpc thread startup can cause oops
1663 Details : Starting a ptlrpc service thread can fail if there are a large
1664 number of threads or the server memory is very fragmented.
1665 Handle this without oopsing.
1668 Frequency : always, only if liblustre and non-default acceptor port was used
1670 Description: liblustre cannot connect to servers with non-default acceptor port
1671 Details : tcpnal_set_default_params() was not called and was therefore
1672 ignoring the environment varaible TCPNAL_PORT, as well as other
1673 TCPNAL_ environment variables
1678 Description: two objects could be created on the same OST for a single file
1679 Details : If an OST is down, in some cases it was possible to create two
1680 objects on a single OST for a single file. No problems other
1681 than potential performance impact and spurious error messages.
1685 Bugzilla : 5681, 9562
1686 Description: Client may oops in ll_unhash_aliases
1687 Details : Client dcache may become inconsistent in race condition.
1688 In some cases "getcwd" can fail if the current directory is
1694 Description: Inode refcounting problems in NFS export code
1695 Details : link_raw functions used to call d_instantiate without obtaining
1696 extra inode reference first.
1700 Bugzilla : 9942, 9903
1701 Description: Referencing freed requests leading to crash, memleaks with NFS.
1702 Details : We used to require that call to ll_revalidate_it was always
1703 followed by ll_lookup_it. Also with revalidate_special() it is
1704 possible to call ll_revalidate_it() twice for the same dentry
1705 even if first occurence returned success. This fix changes semantic
1706 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
1707 reference on a request referred from the intent.
1708 ll_intent_release() then releases such a request.
1711 Frequency : rare, normally benchmark loads only
1713 Description: unlinked inodes were kept in memory on the client
1714 Details : If a client is repeatedly creating and unlinking files it
1715 can accumulate a lot of stale inodes in the inode slab cache.
1716 If there is no other client load running this can cause the
1717 client node to run out of memory. Instead flush old inodes
1718 from client cache that have the same inode number as a new inode.
1721 Frequency : SLES9 2.6.5 kernel and long filenames only
1722 Bugzilla : 9969, 10379
1723 Description: utime reports stale NFS file handle
1724 Details : SLES9 uses out-of-dentry names in some cases, which confused
1725 the lustre dentry revalidation. Change it to always use the
1729 Frequency : rare, unless heavy write-truncate concurrency is continuous
1730 Bugzilla : 4180, 6984, 7171, 9963, 9331
1731 Description: OST becomes very slow and/or deadlocked during object unlink
1732 Details : filter_destroy() was holding onto the parent directory lock
1733 while truncating+unlinking objects. For very large objects this
1734 may block other threads for a long time and slow overall OST
1735 responsiveness. It may also be possible to get a lock ordering
1736 deadlock in this case, or run out of journal credits because of
1737 the combined truncate+unlink. Solution is to do object truncate
1738 first in one transaction without parent lock, and then do the
1739 final unlink in a new transaction with the parent lock. This
1740 reduces the lock hold time dramatically.
1743 Frequency : rare, 2.4 kernels only
1745 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
1746 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
1747 before unmounting their filesystem in order to avoid deadlock.
1748 The kernel_locked() function in 2.4 kernels only checks whether
1749 the kernel lock is held, not whether it is this process that is
1750 holding it as 2.6 kernels do.
1755 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
1756 Details : The client ptlrpc code may be trying to reconnect to a down
1757 server before a previous connection attempt has timed out.
1758 Increase the reconnect interval to be longer than the connection
1759 timeout interval to avoid sending duplicate connections to
1763 Frequency : echo_client brw_test command
1765 Description: fix echo_client to work with OST preallocated code
1766 Details : OST preallocation code (5137) didn't take echo_client IO path
1767 into account: echo_client calls filter methods outside of any
1768 OST thread and, hence, there is no per-thread preallocated
1769 pages and buffers to use. Solution: hijack pga pages for IO. As
1770 a byproduct, this avoids unnecessary data copying.
1774 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
1775 Description: Client can oops in mdc_commit_close() after open replay
1776 Details : It was possible for the MDS to return an open request with no
1777 transaction number in mds_finish_transno() if the client was
1778 evicted, but without actually returning an error. Clients
1779 would later try to replay that open and may trip an assertion
1780 Simplify the client close codepath, and always return an error
1781 from the MDS in case the open is not successful.
1784 Frequency : rare, 2.6 OSTs only
1786 Description: OST may deadlock under high load on fragmented files
1787 Details : If there was a heavy load and highly-fragmented OST filesystems
1788 it was possible to have all the OST threads deadlock waiting on
1789 allocation of biovecs, because the biovecs were not released
1790 until the entire RPC IO was completed. Instead, release biovecs
1791 as soon as they are complete to ensure forward IO progress.
1793 Severity : enhancement
1795 Description: Support for specifying external journal device at mount
1796 Details : If an OST or MDS device is formatted with an external journal
1797 device, this device major/minor is stored in the ext3 superblock
1798 and may not be valid for failover. Allow detecting and
1799 specifying the external journal at mount time.
1804 Description: Mounting an MDS with pending unlinked files may cause oops
1805 Details : target_finish_recovery() calls mds_postrecov() which returned
1806 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
1807 considers this an error and immediately begins cleaning up the
1808 lov, just after starting the mds_lov process
1810 Severity : enhancement
1812 Description: Implement 'lfs df' to report actual free space on per-OST basis
1813 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
1814 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
1815 usage of objects; '-h' to report in human readable format.
1817 ------------------------------------------------------------------------------
1819 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
1826 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
1827 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
1828 at startup during orphan unlinking. Instead, skip buddy update
1829 before setup, as it will be regenerated after recovery is complete.
1834 Description: If an OST is inactive, its locks might reference stale inodes.
1835 Details : lov_change_cbdata() must iterate over all namespaces, even if
1836 they are inactive to clear inode references from the lock.
1838 Severity : enhancement
1839 Frequency : occasional, if non-standard max_dirty_mb used
1841 Description: Client will block write RPCs if not enough grant
1842 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
1843 then the client will block writes while waiting for another RPC
1844 to complete instead of consuming its dirty limit. With change
1845 we get improved performance when max_dirty_mb is small.
1847 Severity : enhancement
1848 Bugzilla : 3389, 6253
1849 Description: Add support for supplementary groups on the MDS.
1850 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
1851 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
1852 lookups for user supplementary groups into a cache.
1856 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
1857 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
1858 for liblustre O_EXCL is enforced by MDS.
1862 Description: readdir never returns NULL in liblustre.
1863 Details : Corrected llu_iop_getdirentries logic, to return offset of next
1864 dentry in struct dirent.
1868 Frequency : liblustre only, depends on application IO pattern
1869 Description: liblustre clients evicted if not contacting servers
1870 Details : Don't put liblustre clients into the ping_evictor list, so
1871 they will not be evicted by the pinger ever.
1873 Severity : enhancement
1875 Description: Add ability to evict clients by NID from MDS.
1876 Details : By echoing "nid:$NID" string into
1877 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
1878 $NID would be instantly evicted from this MDS and from all active
1879 OSTs connected to it.
1883 Description: Do not query file size twice, somewhat slowing stat(2) calls.
1884 Details : lookup_it_finish() used to query file size from OSTs that was not
1889 Description: service threads change working directory to that of init
1890 Details : Starting lustre service threads may pin the working directory
1891 of the parent thread, making that filesystem busy. Threads
1892 now change to the working directory of init to avoid this.
1896 Frequency : during shutdown only
1897 Description: shutdown with a failed MDS or OST can cause unmount to hang
1898 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
1903 Frequency : 2.6 only
1904 Description: chmod/chown may include an extra supplementary group
1905 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
1906 supplementary group and if none is specified this is used.
1909 Bugzilla : 5479 (6816)
1911 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
1912 Details : If lookup is in progress on a file that is unlinked we might try
1913 to revalidate the inode and fail in revalidate after lookup is
1914 complete and ll_file_open() enqueues the open again but
1915 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
1918 Frequency : always, if lconf --abort_recovery used
1920 Description: lconf --abort_recovery fails with 'Operation not supported'
1921 Details : lconf was attempting to abort recovery on the MDT device and not
1924 ------------------------------------------------------------------------------
1926 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
1931 Frequency : rare (only unsupported configurations with a node running as an
1933 Bugzilla : 6514, 5137
1934 Description: Mounting a Lustre file system on a node running as an OST could
1936 Details : OSTs now preallocates memory needed to write out data at
1937 startup, instead of when needed, to avoid having to
1938 allocate memory in possibly low memory situations.
1939 Specifically, if the file system is mounted on on OST,
1940 memory pressure could force it to try to write out data,
1941 which it needed to allocate memory to do. Due to the low
1942 memory, it would be unable to do so and the node would
1943 become unresponsive.
1945 Severity : enhancement
1947 Description: Addition of lconf --service command line option
1948 Details : lconf now accepts a '--service <arg>' option, which is
1949 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
1951 Severity : enhancement
1953 Description: Failover mode is now the default for OSTs.
1954 Details : By default, OSTs will now run in failover mode. To return to
1955 the old behaviour, add '--failout' to the lmc line for OSTs.
1957 Severity : enhancement
1959 Description: Health checks are now provided for MDS and OSTs
1960 Details : Additional detailed health check information on MSD and OSTs
1961 is now provided through the procfs health_check value.
1964 Frequency : occasional, depends on IO load
1966 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
1967 after numerous create/delete cycles
1968 Details : The ext3 inode allocation policy would not allocate new inodes
1969 very well on the OSTs because there are no new directories
1970 being created. Instead we look for groups with free space if
1971 the parent directories are nearly full.
1976 Description: Network or server problems during mount may cause partially
1977 mounted clients instead of returning an error.
1978 Details : The config llog parsing code may overwrite the error return
1979 code during mount error handling, returning success instead
1985 Description: MDS can fail to allocate large reply buffers
1986 Details : After long uptimes the MDS can fail to allocate large reply
1987 buffers (e.g. zconf client mount config records) due to memory
1988 fragmentation or consumption by the buffer cache. Preallocate
1989 some large reply buffers so that these replies can be sent even
1990 under memory pressure.
1994 Frequency : rare (liblustre)
1995 Description: fsx running with liblustre complained that using truncate() to
1996 extend the file doesn't work. This patch corrects that issue.
1997 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
1998 ATTR_SIZE and lsm use in llu_setattr_raw.
2002 Frequency : rare, only 2.6 kernels
2003 Description: Unusual file access patterns on the MDS may result in inode
2004 data being lost in very rare circumstances.
2005 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2006 code in the 2.6 kernel caused the inode and/or EA data not to
2007 be read from disk, causing single-file corruption.
2011 Frequency : rare, only 2.6 filesystems using extents
2012 Description: Heavy concurrent write and delete load may cause data corruption.
2013 Details : It was possible under high-load situations to have an extent
2014 metadata block in the block device cache from a just-unlinked
2015 file overwrite a newly-allocated data block. We now unmap any
2016 metadata buffers that alias just-allocated data blocks.
2020 Frequency : filesystems with default stripe_count larger than 77
2021 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2022 Details : lconf specifies an inode size of 4096 bytes when the default
2023 stripe_count is larger than 77. This conflicts with the default
2024 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2025 this case to avoid pinning too much memory for large EAs.
2027 ------------------------------------------------------------------------------
2029 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2034 Frequency : rare (extremely heavy IO load with hundreds of clients)
2036 Description: Client is evicted, gets IO error writing to file
2037 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2038 caused clients to be evicted for AST timeouts. The fixes in
2039 bug 5192 mean we no longer need to have such short AST timeouts
2040 so ldlm_timeout has been increased.
2043 Frequency : occasional during --force or --failover shutdown under load
2044 Bugzilla : 5949, 4834
2045 Description: Server oops/LBUG if stopped with --force or --failover under load
2046 Details : a collection of import/export refcount and cleanup ordering
2047 issues fixed for safer force cleanup
2050 Frequency : only filesystems larger than 120 OSTs
2051 Bugzilla : 5990, 6223
2052 Description: lfs getstripe would oops on a very large filesystem
2053 Details : lov_getconfig used kfree on vmalloc'd memory
2056 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2057 Bugzilla : 6242, 6243
2058 Description: reading from files that had been truncated to a non-zero size
2059 but never opened returned no data
2060 Details : ll_file_read() reads zeros from no-object files to EOF
2065 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2066 Details : There's little harm in aborting MDS/OSS recovery and letting it
2067 try again, so I removed the LASSERT and return an error instead.
2069 Severity : enhancement
2071 Description: New debugging infrastructure for tracking down data corruption
2072 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2073 (b) cover more of the client-side code path, and (c) try to narrow
2074 down where problems occurred
2078 Bugzilla : 3819, 4364, 4397, 6313
2079 Description: Racing close and eviction MDS could cause assertion in mds_close
2080 Details : It was possible to get multiple mfd references during close and
2081 client eviction, leading to one thread referencing a freed mfd.
2083 Severity: : enhancement
2084 Bugzilla : 3262, 6359
2085 Description: Attempts to reconnect to servers are now more aggressive.
2086 Details : This builds on the enhanced upcall-less recovery that was added
2087 in 1.4.2. When trying to reconnect to servers, clients will
2088 now try each server in the failover group every 10 seconds. By
2089 default, clients would previously try one server every 25 seconds.
2094 Description: After recovery, certain operations trigger a failed
2095 assertion on a client.
2096 Details : Failing over an mds, using lconf -d --failover, while a
2097 client was doing a readdir() call would cause the client to
2098 LBUG after recovery completed and the readdir() was resent.
2100 Severity : enhancement
2102 Description: Default groups are now added by lconf
2103 Details : You can now run lconf --group <servicename> without having to
2104 manually add groups with lmc.
2107 Frequency : occasional
2109 Description: Nodes with an elan id of 0 trigger a failed assertion
2112 Frequency : always when accessing e.g. tty/console device nodes
2114 Description: tty and some other devices nodes cannot be used on lustre
2115 Details : file's private_data field is used by device data and lustre
2116 values in there got lost. New field was added to struct file to
2117 store fs-specific private data.
2120 Frequency : when exporting Lustre via NFS
2122 Description: NFSD failed occasionally when looking up a path component
2123 Details : NFSD is looking up ".." which was broken in ext3 directories
2124 that had grown large enough to become hashed.
2127 Frequency : Clusters with multiple interfaces not on the same subnet
2129 Description: Nodes will repeatedly try to reconnect to an interface which it
2130 cannot reach and report an error to the log.
2131 Details : Extra peer list entries will be created by lconf with some peers
2132 unreachable. lconf now validates the peer before adding it.
2135 Frequency : Only if a default stripe is set on the filesystem root.
2137 Description: Setting a default stripe on the filesystem root prevented the
2138 filesystem from being remounted.
2139 Details : The client was sending extra request flags in the root getattr
2140 request and did not allocate a reply buffer for the dir EA.
2143 Frequency : occasional, higher if lots of files are accessed by one client
2144 Bugzilla : 6159, 6097
2145 Description: Client trips assertion regarding lsm mismatch/magic
2146 Details : While revalidating inodes the VFS looks up inodes with ifind()
2147 and in rare cases can find an inode that is being freed.
2148 The ll_test_inode() code will free the lsm during ifind()
2149 when it finds an existing inode and then the VFS later attaches
2150 this free lsm to a new inode.
2154 Bugzilla : 6422, 7030
2155 Description: MDS deadlock between mkdir and client eviction
2156 Details : Creating a new file via mkdir or mknod (starting a transaction
2157 and getting the ns lock) can deadlock with client eviction
2158 (gets ns lock and trying to finish a synchronous transaction).
2161 Frequency : occasional
2162 Description: While starting a server, the fsfilt_ext3 module could not be
2164 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2165 kernels. Previously, lconf would still use the ext3 name
2166 when trying to load modules. Now, it will correctly use
2167 ext3 on 2.4 and ldiskfs on 2.6.
2169 Severity : enhancement
2170 Description: The default stripe count has been changed to 1
2171 Details : The interpretation of the default stripe count (0, to lfs
2172 or lmc) has been changed to mean striping across a single
2173 OST, rather than all available. For general usage we have
2174 found a stripe count of 1 or 2 works best.
2176 Severity : enhancement
2177 Description: Add support for compiling against Cray portals.
2178 Details : Conditional compiling for some areas that are different
2182 Frequency : occasional
2183 Bugzilla : 6409, 6834
2184 Description: Creating files with an explicit stripe count may lead to
2185 a failed assertion on the MDS
2186 Details : If some OSTs are full or unavailable, creating files may
2187 trigger a failed assertion on the MDS. Now, Lustre will
2188 try to use other servers or return an error to the
2192 Frequency : occasional
2194 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2195 caused lock timeout during readahead (since 1.4.2).
2196 Details : Processes doing readahead might match a lock that hasn't been
2197 granted yet if there are overlapping and conflicting lock
2198 requests. The readahead process waits on ungranted lock
2199 (original lock is CBPENDING), while OST waits for that process
2200 to cancel CBPENDING read lock and eventually evicts client.
2202 Severity : enhancement
2204 Description: Initial enabling of flock support for clients
2205 Details : Implements fcntl advisory locking and file status functions.
2206 This feature is provided as an optional mount flag (default
2207 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2208 locking are implemented yet, and those that are are not guaranteed
2209 to be completely correct in production environments.
2210 mount -t lustre -o [flock|noflock] ...
2213 Frequency : occasional
2215 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2216 assertion in the ext3 JBD (journaling) layer.
2217 Details : The b_committed_data struct is protected by the big kernel lock
2218 in 2.4 kernels, serializing journal_commit_transaction() and
2219 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2220 access to this struct. In 2.6 kernels there is finer grained
2221 locking to improve SMP performance of the JBD layer.
2225 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2227 -----------------------------------------------------------------------------
2229 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2231 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2232 versions of Lustre. Please update all servers and clients to
2233 version 1.4.2 or later at the same time. You must also run
2234 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2235 to update the configuration logs.
2237 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2238 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2239 - swab llog records of type '0' so we get proper header size/idx (5861)
2240 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2241 - fix rename of one directory over another leaking an inode (5953)
2242 - avoid SetPageDirty on 2.6 (5981)
2243 - don't re-add just-being-destroyed locks to the waiting list (5653)
2244 - when creating new directories, inherit the parent's custom
2245 striping settings if present parent (3048)
2246 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2247 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2248 - don't hold client locks on temporary worklist from l_lru (5666)
2249 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2250 - automatically evict dead clients (5921)
2251 - Update file size properly in create+truncate+fstat case (6196)
2252 - Do not unhash mountpoint dentries, do not allow removal of
2254 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2255 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2256 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2257 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2258 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2259 - Correctly handle reads of files with no objects (6243)
2260 - lctl recover will also mark a device active if deactivate used (5933)
2262 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2263 - allow --write-conf on an MDS with different nettype than client (5619)
2264 - don't write config llogs to MDS for mounts not from that MDS (5617)
2265 - lconf should create multiple TCP connections from a client (5201)
2266 - init scripts are now turned off by default; run chkconfig --on
2267 lustre and chkconfig --on lustrefs to use them
2268 - upcalls are no longer needed for clients to recover to failover
2270 - add --abort-recovery option to lconf to abort recovery on device
2272 - add support for an arbitrary number of OSTs (3026)
2273 - Quota support protocol changes.
2274 - forward compatibility changes to wire structs (6007)
2275 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2276 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2277 - improved "lustre status" script
2278 - initialize blocksize for non-regular files (6062)
2279 - added --disable-server and --disable-client configure options (5782)
2280 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2281 - Vanilla 2.4.29 support
2282 - increase maximum number of obd devices to 520 (6242)
2283 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2284 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2285 SLES 9 2.6.5 kernels
2286 - limit stripes per file to 160 (the maximum EA size) (6093)
2288 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2291 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2292 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2293 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2294 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2295 - small ext3 extents cleanups and fixes (5733)
2296 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2297 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2298 - avoid needless client->OST connect, fix handle mismatch (5317)
2299 - fix DLM error path that led to out-of-sync client, long delays (5779)
2300 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2301 - fix several locking issues related to i_size (5492,5624,5654,5672)
2302 - don't move pending lock onto export if it is already evicted (5683)
2303 - fix kernel oops when creating .foo in unlinked directory (5548)
2304 - fix deadlock in obdfilter statistics vs. object create (5811)
2305 - use time_{before,after} to avoid timer jiffies wrap (5882)
2306 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2307 - Do not leak request if server was not able to process it (5154)
2308 - If mds_open unable to find parent dir, make that negative lookup(5154)
2309 - don't create new directories with extent-mapping (5909, 5936)
2311 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2312 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2313 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2314 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2315 - support for RHEL4 kernel on i686 (5773)
2316 - provide error messages when incompatible logs are encountered (5898)
2318 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2319 * version 1.4.0.10 (1.4.1 release candidate 1)
2321 - don't keep a lock reference when lock is not granted (4238)
2322 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2323 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2324 - drop import inflight refcount on signal_completed_replay error (5255)
2325 - unlock page after async write error during send (3677)
2326 - handle missing objects in filter_preprw_read properly (5265)
2327 - no transno return for symlink open, don't save no-trasno open (3440)
2328 - don't try to complete elan receive that already failed (4012)
2329 - free RPC server reply state on error (5406)
2330 - clean up thread from ptlrpc_start_thread() on error (5160)
2331 - readahead could read extra page into cache that wasn't ejected (5388)
2332 - prevent races in class_attach/setup/cleanup/detach (5260)
2333 - don't dereference de->d_inode after l_dput of de (5458)
2334 - use "int" for stripe value returned from lock_to_stripe (5544)
2335 - mballoc allocation and error-checking fixes in 2.6 (5504)
2336 - block device patches to fix I/O request sizes in 2.6 (5482)
2337 - look up hostnames for IB nals (5602)
2338 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2339 - don't start multiple acceptors for the same port (5277)
2340 - fix incorrect LASSERT in mds_getattr_name (5635)
2341 - export a proc file for general "ping" checking (5628)
2342 - fix "lfs check" to not block when the MDS is down (5628)
2344 - service request history (4965)
2345 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2346 - create an "evict_client" /proc entry on OSTs, like the MDS has
2347 - fix mount usage message, return errors per mount(8) (5168)
2348 - change grep [] to grep "[]" in tests so they work in more UMLs
2349 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2350 - remove ancient LOV_MAGIC_V0 EA support (5047)
2351 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2352 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2353 - allow readahead allocations to fail when low on memory (5383)
2354 - mmap locking landed again, after considerable improvement (2828)
2355 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2357 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2360 - send OST transaction number in read/write reply to free req (4966)
2361 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2362 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2363 - the watchdog thread now runs as interruptible (5246)
2364 - flock/lockf fixes (but it's still disabled, pending 5135)
2365 - don't use EXT3 constants in llite code (5094)
2366 - memory shortage at startup could cause assertion (5176)
2368 - reorganization of lov code
2369 - single portals codebase
2371 - add extents/mballoc support (5025)
2372 - direct I/O reads in the obdfilter (4048)
2373 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2375 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2378 - send OST transaction number in read/write reply to free req (4966)
2379 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2380 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2381 - the watchdog thread now runs as interruptible (5246)
2382 - handle missing objects in filter_preprw_read properly (5265)
2383 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2384 - drop import inflight refcount on signal_completed_replay error (5255)
2385 - unlock page after async write error during send (3677)
2386 - return original error code on reconstructed replies (3761)
2387 - no transno return for symlink open, don't save no-trasno open (3440)
2389 - add pid to ldlm debugging output (4922)
2390 - bump the watchdog timeouts -- we can't handle 30sec yet
2391 - extra debugging for orphan dentry/inode bug (5259)
2393 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2396 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2397 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2398 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2399 - fix NULL dereference in /proc/sys/portals/routes (4827)
2400 - allow failed mdc_close() operations to be interrupted (4561)
2401 - stop precreate on OST before MDS would time out on it (4778)
2402 - don't send partial-page writes before EOF from client (4410)
2403 - discard client grant for sub-page writes on large-page clients (4520)
2404 - don't free dentries not owned by NFS code, check generation (4806)
2405 - fix lsm leak if mds_create_objects() fails (4801)
2406 - limit debug_daemon file size, always print CERROR messages (4789)
2407 - use transno after validating reply (3892)
2408 - process timed out requests if import state changes (3754)
2409 - update mtime on OST during writes, return in glimpse (4829)
2410 - add mkfsoptions to LDAP (4679)
2411 - use ->max_readahead method instead of zapping global ra (5039)
2412 - don't interrupt __l_wait_event() during strace
2414 - add software watchdogs to catch hung threads quickly (4941)
2415 - make lustrefs init script start after nfs is mounted
2416 - fix CWARN/ERROR duplication (4930)
2417 - return async write errors to application if possible (2248)
2418 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2419 - print NAL number in %x format (4645)
2420 - update barely-supported suse-2.4.21-171 series (4842)
2421 - support for sles 9 %post scripts
2422 - support for building 2.6 kernel-source packages
2423 - support for sles km_* packages
2425 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2428 - ignore -ENOENT errors in osc_destroy (3639)
2429 - notify osc create thread that OSC is being cleaned up (4600)
2430 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2431 - reconstruct ost_handle() like mds_handle() (4657)
2432 - create a new thread to do import eviction to avoid deadlock (3969)
2433 - let lconf resolve symlinked-to devices (4629)
2434 - don't unlink "objects" from directory with default EA (4554)
2435 - hold socknal file ref over connect in case target is down (4394)
2436 - allow more than 32000 subdirectories in a single directory (3244)
2437 - fix blocks count for O_DIRECT writes (3751)
2438 - OST returns ENOSPC from object create when no space left (4539)
2439 - don't send truncate RPC if file size isn't changing (4410)
2440 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2441 - bind to privileged port in socknal and tcpnal (3689)
2443 - rate limit CERROR/CWARN console message to avoid overload (4519)
2444 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2445 - basic mmap support (3918)
2446 - kernel patch series update from b1_4 (4711)
2448 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2451 - avoid crash during MDS cleanup with OST shut down (2775)
2452 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2453 - don't use bad inodes on the MDS (3744)
2454 - dynamic object preallocation to improve recovery speed (4236)
2455 - don't hold spinlock over lock dumping or change debug flags (4401)
2456 - don't zero obd_dev when it is force cleaned (3651)
2457 - print grants to console if they go negative (4431)
2458 - "lctl deactivate" will stop automatic recovery attempts (3406)
2459 - look for existing locks in ldlm_handle_enqueue() (3764)
2460 - don't resolve lock handle twice in recovery avoiding race (4401)
2461 - revalidate should check working dir is a directory (4134)
2463 - don't always mark "slow" obdfilter messages as errors (4418)
2465 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2468 - don't close LustreDB during write_conf until it is done (3860)
2469 - fix typo in lconf for_each_profile (3821)
2470 - allow dumping logs from multiple threads at one time (3820)
2471 - don't allow multiple threads in OSC recovery (3812)
2472 - fix debug_size parameters (3864)
2473 - fix mds_postrecov to initialize import for llog ctxt (3121)
2474 - replace config semaphore with spinlock (3306)
2475 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2476 - don't allow enqueue to complete on a destroyed export (3822)
2477 - down write_lock before checking llog header bitmap (3825)
2478 - recover from lock replay timeout (3764)
2479 - up llog sem before sending rpc (3652)
2480 - reduce ns lock hold times when setting kms (3267)
2481 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2482 - fix NULL deref and obd_dev leak on setup error (3312)
2483 - replace some LBUG about llog ops with error handling (3841)
2484 - don't match INVALID dentries from d_lookup and spin (3784)
2485 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2486 - fix invalid assertion in ptlrpc_set_wait (3880)
2488 - add libwrap support for the TCP acceptor (3996)
2489 - add /proc/sys/portals/routes for non-root route listing (3994)
2490 - allow setting MDS UUID in .xml (2580)
2491 - print the stack of a process that LBUGs (4228)
2493 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2496 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2497 - make sure to unset replay flag from failed open requests (3440)
2498 - if default stripe count is 0, use OST count for inode size (3636)
2499 - update parent mtime/ctime on client for create/unlink (2611)
2500 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2501 - free recovery state on server during a forced cleanup (3571)
2502 - unregister_reply for resent reqs (3063)
2503 - loop back devices mounting and status check on 2.6 (3563)
2504 - fix resource-creation race that can provoke i_size == 0 (3513)
2505 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2506 - more debugging for page-accounting assertion (3746)
2507 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2508 - avoid deadlock after precreation failure (3758)
2509 - fix race and lock order deadlock in orphan handling (3450, 3750)
2510 - add validity checks when grabbing inodes from l_ast_data (3599)
2512 - add /proc/.../recovery_status to obdfilter (3428)
2513 - lightweight CDEBUG infrastructure, debug daemon (3668)
2514 - change default OSC RPC parameters to be better on small clusters
2515 - turn off OST read cache for files smaller than 32MB
2516 - install man pages and include them in rpms (3100)
2517 - add new init script for (un)mounting lustre filesystems (2593)
2518 - run chkconfig in %post for init scripts (3701)
2519 - drop scimac NAL (unmaintained)
2521 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
2524 - clean kiobufs before and after use (3485)
2525 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
2526 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
2527 - change init script to fail more clearly if not run as root (1528)
2528 - allow clients to reconnect during replay (1742)
2529 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
2530 - don't do DNS lookups on NIDs too small for IP addresses (3442)
2531 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
2532 - fix cond_resched (3554)
2533 - only evict unfinished clients after recovery (3515)
2534 - allow bulk resend, prevent data loss (3570)
2535 - dynamic ptlrpc request buffer allocation (2102)
2536 - don't allow unlinking open directory if it isn't empty (2904)
2537 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
2538 - remove extraneous obd dereference causing LASSERT failure (3334)
2539 - don't use get_cycles() when creating temp. files on the mds (3156)
2540 - hold i_sem when setting i_size in ll_extent_lock() (3564)
2541 - handle EEXIST for set-stripe, set proper directory name (3336)
2543 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
2544 - fix ksocknal_fmb_callback() error messages (2918)
2546 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
2549 - don't copy lvb into (possibly NULL) reply on error (2983)
2550 - don't deref dentry after dput, don't free lvb on error (2922)
2551 - use the kms to determine writeback rpc length (2947)
2552 - increment oti_logcookies when osc is inactive (2948)
2553 - update client's i_blocks count via lvb messages (2543)
2554 - handle intent open/close of special files properly (1557)
2555 - mount MDS with errors=remount-ro, like obdfilter (2009)
2556 - initialize lock handle to avoid ASSERT on error cleanup (3057)
2557 - don't use cancelling-locks' kms values (2947)
2558 - use highest lock extent for kms, not last one (2925)
2559 - don't dereference ERR_PTR() dentry in error handling path (3107)
2560 - fix thread race in portals_debug_dumplog() (3122)
2561 - create lprocfs device entries at setup instead of at attach (1519)
2562 - common AST error handler, don't evict client on completion race (3145)
2563 - zero nameidata in detach_mnt in 2.6 (3118)
2564 - verify d_inode after revalidate_special is valid in 2.6 (3116)
2565 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
2566 - initialize RPC timeout timer earlier for 2.6 (3219)
2567 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
2568 - print nal/nid for unknown nid (3258)
2569 - additional checks for oscc recovery before doing precreate (3284)
2570 - fix ll_extent_lock() error return code for 64-bit systems (3043)
2571 - don't crash in mdc_close for bad permissions on open (3285)
2572 - zero i_rdev for non-device files (3147)
2573 - clear page->private before handing to FS, better assertion (3119)
2574 - tune the read pipeline (3236)
2575 - fix incorrect decref of invalidated dentry (2350)
2576 - provide read-ahead stats and refine rpc in flight stats (3328)
2577 - don't hold journal transaction open across create RPC (3313)
2578 - update atime on MDS at close time (3265)
2579 - close LDAP connection when recovering to avoid server load (3315)
2580 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
2581 - don't leak open file on MDS after open resend (3325)
2582 - serialize filter_precreate and filter_destroy_precreated (3329)
2583 - loop device shouldn't call sync_dev() for nul device (3092)
2584 - clear page cache after eviction (2766)
2585 - resynchronize MDS->OST in background (2824)
2586 - refuse to mount the same filesystem twice on same mountpoint (3394)
2587 - allow llmount to create routes for mounting behind routers (3320)
2588 - push lock cancellation to blocking thread for glimpse ASTs (3409)
2589 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
2590 - fix rare problem with rename on htree directories (3417)
2592 - allow default OST striping configuration per directory (1414)
2593 - fix compilation for qswnal for 2.6 kernels (3125)
2594 - increase maximum number of MDS request buffers for large systems
2595 - change liblustreapi to be useful for external progs like lfsck (3098)
2596 - increase local configuration timeout for slow disks (3353)
2597 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
2599 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2602 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
2603 - don't overwrite extent policy data in reply if lock was blocked (2901)
2604 - drop filter export grants atomically with removal from device (2663)
2605 - del obd_self_export from work_list in class_disconnect_exports (2908)
2606 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
2607 - swab reply message in mdc_close, other PPC fixes (2464)
2608 - fix destroying of named logs (2325)
2609 - overwrite old logs when running lconf --write_conf (2264)
2610 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
2611 - fix race in target_handle_connect (2898)
2612 - mds_reint_create() should take same inode create lock (2926)
2613 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
2614 - don't close files for self_export to avoid uninitialized obd (2936)
2615 - allow MDS with the same name as client node (2939)
2616 - hold dentry reference for closed log files for unlink (2325)
2617 - reserve space for all logs during transactions (2059)
2618 - don't evict page beyond end of stripe extent (2925)
2619 - don't oops on a deleted current working directory (2399)
2620 - handle hard links to targets without a parent properly (2517)
2621 - don't dereference NULL lock when racing during eviction (2867)
2622 - don't grow lock extents when lots of conflicting locks (2919)
2624 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
2627 - account for cache space usage on clients to avoid data loss (974)
2628 - lfsck support in lustre kernel code (2349)
2629 - reduce journal credits needed for BRW writes (2370)
2630 - orphan handling to avoid losing space on client/server crashes
2631 - ptlrpcd can be blocked, stopping ALL progress (2477)
2632 - use lock value blocks to assist in proper KMS, faster stat (1021)
2633 - takes i_sem instead of DLM locks internally on obdfilter (2720)
2634 - recovery for initial connections (2355)
2635 - fixes for mds_cleanup_orphans (1934)
2636 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
2637 - block all file creations until orphan recovery completes (1901)
2638 - client remove rq_connection from request struct (2423)
2639 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
2640 - recovery timer race (2670)
2641 - mdc_close recovey bug (2532)
2642 - ptlrpc cleanup bug (2710)
2643 - mds timeout on local locks (2588)
2644 - namespace lock held during RPCs (2431)
2645 - handle interrupted sync write properly (2503)
2646 - don't try to handle a message that hasn't been replied to (2699)
2647 - client assert failure during cleanup after abort recovery (2701)
2648 - leak mdc device after failed mount (2712)
2649 - ptlrpc_check_set allows timedout requests to complete (2714)
2650 - wait for inflight reqs when ptlrpcd finishes (2710)
2651 - make sure unregistered services are removed from the srv_list
2652 - reset bulk XID's when resending them (caught by 1138 test)
2653 - unregister_bulk after timeout
2654 - fix lconf error (2694)
2655 - handle write after unfinished setstripe, stripe-only getstripe (2388)
2656 - readahead locks pages, leaves pending causing memory pressure (2673)
2657 - increase OST request buffers to 4096 on large machines (2729)
2658 - fix up permission of existing directories in simple_mkdir (2661)
2659 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
2660 - don't assign transno to errored transactions (2742)
2661 - don't delete objects on OST if given a bogus objid from MDS (2751)
2662 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
2663 - if rq_no_resend, then timeout request after recovery (2432)
2664 - fix MDS llog_logid record size, 64-bit array alignment (2733)
2665 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
2666 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
2667 - MDS recovery shouldn't delete objects that it has given out (2730)
2668 - if enqueue arrives after completion, don't clobber LVB (2819)
2669 - don't unlock pages twice when trigger_group_io returns error (2814)
2670 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
2671 - don't write pages to disk if there was an error (1450)
2672 - don't ping imports that have recovery disabled (2676)
2673 - take buffered bytes into account when balancing socknal conn (2817)
2674 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
2675 - reconnect unlink llog connection after MDS reconnects to OST (2816)
2676 - remove little-endian swabbing of llog records (1987)
2677 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
2678 - retry reposting request buffers if they fail (1191)
2679 - grow extent at grant time to avoid granting a revoked lock (2809)
2680 - lock revoke doesn't evict page if covered by a second lock (2765)
2681 - disable VM readahead to avoid reading outside lock extents (2805)
2683 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
2684 - updated kernel patches for hp-2.4.20 kernel (2681)
2686 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
2689 - fix truncated write corruption (2366)
2690 - fix for failed assertion in iopen_connect_dentry (1792,2517)
2692 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
2693 - ensure len(uuid) < 37 in lmc (1171)
2694 - fix ia64 OOPS in llog_test (2255)
2695 - zero end of page at obdfilter for partial page writes (2648)
2696 - don't leave stale dentries around after renames (bug 2428)
2697 - fix timeouts when evicting a client with a single lock held (2642)
2698 - set deadline for the initial HELLO message to drain (2634)
2699 - print out dotted-quad IP addresses in the socknal (2302)
2701 - additional debugging for MDS client eviction problem (2443)
2702 - fix mkfsoptions support for osts (2603, 2604)
2704 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
2707 - add series for the vanilla 2.6.0 kernel
2708 - add series for the vanilla 2.4.24 kernel
2709 - add series for a cray x86/64 UL kernel drop
2710 - fix xattr patches for the vanilla 2.4.19 series
2712 - generate true UUIDs in lmc (1171)
2713 - have portals stack dumping break in UML (2466)
2714 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
2715 - allocate with _NOFS in ldlm to avoid deadlock (1933)
2716 - wake callback waiting threads on client eviction (2460)
2717 - Add --ptldebug and --subsystem to lmc (1719)
2718 - update assertion to allow safe interrupt allocation
2719 - set rq_no_resend for cancel requests (2432)
2720 - recalculate ptlrpcd timeout after resend (2494)
2721 - call vfs_rmdir when removing pending directories (2368)
2722 - fix renaming a file to itself (2429)
2723 - lmc creates a default one-stripe lov (2454)
2724 - expand procfs space to handle large clusters (2326)
2725 - increase UML stack to avoid overflow
2726 - update lconf's list of debug and subsystem masks
2727 - fix lfs find --obd (2510)
2728 - /proc tunable for disabling filter read caching (2591)
2729 - stop rpm packages from altering slapd.conf (2301)
2730 - disable nagle in the socknal under 0conf (2578)
2731 - choose mds inode size based on stripe count (2572)
2732 - fix kernel-source rpm problems (2516)
2734 - add --disable-doc to avoid pdf generation (2421)
2735 - update documentation, tests, type-os, comments
2736 - avoid format warnings on ia64
2737 - remove the TOE NAL
2738 - tiny code cleanups by removing unused fields
2740 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
2743 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
2744 - remove the most blatant lies from BUILDING file (2371)
2745 - change default debug level to reasonable production setting
2746 - reduce client side cache size to reduce cache flush time
2747 - reduce max RPCs in flight to avoid unnecessary file fragmentation
2748 - make TCP zerocopy and pinger support enabled by default (2476)
2749 - sync writes completed after process exits caused crashes (2319)
2750 - maintain correct mount count on the MDS (2356)
2751 - backout 1557, because 2316 wasn't really fixed
2752 - better file I/O statistics gathering in /proc
2753 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
2754 - another kernel patch to fix zero-copy TCP function export
2755 - don't take duplicate lock when processing re-sent getattr (2420)
2756 - lctl uses obd_self_export instead of creating new conn (2353)
2757 - MDS/OST recovery case which requires object creation asserted (2425)
2758 - move lfs from /usr/sbin to /usr/bin in packages
2759 - fix race between mds_client_add and mds_client_free (2417)
2760 - use kmalloc instead of slabs in portals (2430)
2761 - don't create duplicate records when a failover MDS is present (2442)
2762 - remove unnecessary mount age check (2332)
2763 - don't remove directory inodes from locks prematurely (2451)
2764 - don't break if MDS service name is the same as hostname (2103)
2765 - fix races in client write RPC generation when cache full (2482)
2767 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
2770 - remove now-unused request->rq_obd (278)
2771 - if an allocation fails, print out how much memory we've used (1933)
2772 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
2773 - add the "configurable stack size" patch to most series files (1256)
2774 - ability to write large log records, for 100+ OST configs (2306)
2775 - fix NULL deref when filter_prep fails (2314)
2776 - fix operator precedence error in filter_sync
2777 - dynamic allocation of socknal TX descriptors (2315)
2778 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
2779 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
2780 - ensure that CWARN messages in Portals always get to the syslog
2781 - __init/__exit are not for prototype decls (ldlm_init/exit)
2782 - x86-64 compile warning fixes
2783 - fix gateway LMC keyword conflict (2318)
2784 - fix MDS lock inversions in getattr/reint paths (1844)
2785 - fix a rare lock re-ordering bug, which caused deadlock (2322)
2786 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
2787 - DLM race condition prevented some lock evictions (2328)
2788 - ENOMEM detection and retry on socknal sends (2230)
2789 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
2790 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
2791 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
2792 - better error messages when a client is rejected during recovery (1505)
2793 - avoid cancelling locks which were never granted, after failure (2330)
2794 - fix i_sem/journal inversion in mds_client_add (2333)
2795 - fix truncate/getattr lock cycle deadlock (2334)
2796 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
2797 - fix two rare exit paths which could leak an l_lock() ref (2321)
2798 - fencepost error in MDS/OST orphan recovery (2226)
2799 - make log record alignment 8 bytes (1988)
2800 - lstripe now fails when requested offset > ost_count (2237)
2801 - ensure that all kernel series have a complete list.h (1607)
2802 - fix crashes in special-file operations (2316)
2803 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
2805 - allow configurable automake binary, for testing new versions
2806 - small update to the lfs documentation
2808 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
2810 * fix negative export reference count in fsfilt_sync (2312)
2812 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
2813 * release candidate 0.9.1
2815 - orphans are moved into the PENDING directory for possible recovery
2816 - replayed opens now open by fid for orphan/rename safety (1042)
2817 - last close of an orphan inode generates a transno (683)
2818 - chdir() and mount() now pin the directory entry (1020)
2819 - avoid CERROR in normal ll_setattr_raw() error case (1500)
2820 - discard very old requests without processing them (1502)
2821 - remove some common, well-understood CERRORs (1505)
2822 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
2823 - clear "grant" flags in OST replies until OST grant code lands (1644)
2824 - fix read performance by not clobbering i_blksize on client (1598)
2825 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
2826 - make LRU size a /proc tunable, clears locks when reduced (707)
2827 - fix some lprocfs rot that prevented ptlbd from loading (1732)
2828 - server locks take references on exports now (1558)
2829 - build fixes for 2.4.20-rh trees (1663)
2830 - return an error from lov_create if all OSCs are inactive (1751)
2831 - fix import levels when a reconnect happens without a timeout (1597)
2832 - exit early from mds_open if we get a lookup error (1749)
2833 - partial page read at EOF wouldn't wait for disk before sending (1642)
2834 - avoid NULL deref in obdfilter when reading page past EOF (1592)
2835 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
2836 - fix LBUG in ll_it_open_error with rc = -2 (1861)
2837 - write/truncate lock inversion (1639)
2838 - Don't auto-load obdclass, portals modules during cleanup (1495)
2839 - fix timestamps from jumping to "now" (1763)
2840 - extra journal assertions (1648)
2841 - add an extra multiunlink test (1771)
2842 - fix read_record/write_record API (1776)
2843 - fix leak of offset_extent, possible incorrect i_size later (1772)
2844 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
2845 - Debugging for the kqswnal_get_idle_tx problems (1820)
2846 - Allow recovery to be attempted multiple times (1536)
2847 - Write out MDS last_rcvd file after it is first created (1600)
2848 - Fix tx_descriptor leak in failed transmit situations (1827)
2849 - ext3 journaling fixes for assertion failure after IO error (1871)
2850 - class_export_put() on freed export after completion AST error (1896)
2851 - Fix revalidate looping in VFS (1322)
2852 - Don't access a freed export during MDS_REINT timeout (1521)
2853 - Add open-unlink recovery support on the MDS (1673,1764)
2854 - Return an error if no MDS data was read from last_rcvd (1946)
2855 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
2856 - Don't setup a disk device that doesn't match exported UUID (317)
2857 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
2858 - avoid committing NULL handle in force close
2859 - local.sh is now a one-stripe LOV configuration
2860 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
2861 - fix ext3 htree duplicate directory entry corruption (1516)
2862 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
2863 - update to new LOV EA format (2097)
2864 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
2866 2003-06-15 Phil Schwan <phil@clusterfs.com>
2869 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
2870 - per-import/export recovery handling (958, 931, 959)
2871 - multiple last-rcvd slots, for serving multiple FSes (949)
2872 - connections are again shared between multiple imp/exports (963, 964)
2873 - "umount -f" would hang if any requests needed to be sent (393, 978)
2874 - avoid pinning large req buffer by copying for queued messages (989)
2875 - add "uuid" to "lctl device" command to help upcalls (991)
2876 - "open" RPCs with transnos would confuse recovery counters (1037)
2877 - do proper endian conversion of all wire messages (288, 340, 891)
2878 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
2879 - call waiting locks callback from LDLM recovery thread (1127, 1151)
2880 - fix ptlrpc_connection leak in target_handle_connect (1174)
2881 - fix import refcounting bug in OST and MDS cleanup (1134)
2882 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
2883 - fix very unlikely obd_types race condition (501)
2884 - remove osc_open hack for echo_client (1187)
2885 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
2886 - a failure in read_inode2 leads to deadlock (1139)
2887 - cancel ack-locks as soon as transaction is committed (1072)
2888 - fix major leaks and crashes in the bulk I/O path (937, 1057)
2889 - make sure to commitrw after any preprw to avoid deadlock (1162)
2890 - failing to execute a file in a lustre FS would lock inode (1203)
2891 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
2892 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
2893 - obd_uuid2tgt would walk off the end of the list (1255)
2894 - on IA64 the osc would give portals incorrect bulk size (1258)
2895 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
2896 - fix lock inversion caused by new llite matching code (1282)
2897 - limit the number of dirty pages on a client to 10MB (1286)
2898 - timed out locks were not being corrected cancelled (1289)
2899 - fix O_DIRECT above 4GB on IA-32 (1292)
2900 * major user-visible changes
2901 - fail out/fail over policy now controlled by the upcall (993)
2903 - add OBD_PING to check server availability and failure (954)
2904 - lustre messages are now sent in sending host order (288, 340, 891)
2905 - add eadatalen to MDS getattr reply (340)
2906 - OST read replies may contain second buffer, with per-page status (593)
2908 2003-03-11 Phil Schwan <phil@clusterfs.com>
2911 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
2912 - failed open()s could cause deadlock; fixed (867, 869)
2913 - stop cancelling OST locks when files are closed (481)
2914 - overlapping XID spaces caused network corruption (851, 853)
2915 - fix unsafe fsfilt counter arithmetic; change to atomic_t
2916 - setattr_raw added, to do single-RPC, server-side setattrs
2917 - lmc/lconf syntax change for OST UUIDs
2918 - fix crashy race condition between ptlrpc_free_req and osc_close
2919 - don't use request in mdc_enqueue if we hit a timeout (889)
2920 - don't set the inode i_size for regular files from the MDS (896)
2921 - handle out of order completion AST (842)
2922 - don't LBUG if a lock request times out after receiving AST (913)
2923 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
2924 - if a bad lock AST arrives, send an error instead of dropping entirely
2925 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
2926 - fix leak in bulk IO when only partially completed (899, 900, 926)
2927 - fix O_DIRECT for ia64 (55)
2928 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
2929 - C-z after timeout could hang a process forever; fixed (977)
2931 - client-side I/O cache (678, 924, 929, 941, 970)
2933 - READPAGE and SETATTRs which don't take server-side locks get
2936 2003-02-11 Phil Schwan <phil@clusterfs.com>
2939 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
2940 - Fix fsfilt_extN_readpage() to read a full page of directory
2941 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
2942 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
2943 - Fix imbalanced LOV object allocation and out-of-bound access (469)
2944 - Most intent operations were removed, in favour of a new RPC mode
2945 that does a single RPC to the server and bypasses most of the VFS
2946 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
2947 - Aggressively cancel local locks on DLM servers
2948 - mds_reint_unlink sends EA to the client if it's the last nlink.
2949 client uses that EA to unlink OST objects.
2950 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
2951 - recursive symlinks were fixed (439)
2952 - fixed NULL deref in DEBUG_REQ
2953 - filter_update_lastobjid no longer calls sync, which annoyed extN
2954 - fixed multi-client small-writes to a single file problem (445)
2955 - fixed mtime updates during file writes (607)
2956 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
2957 - fixed bug in obd_brw_read/write() (under guise of testing 367)
2958 - fixed Linux OST size reporting problem (444, 656)
2959 - OST now updates object mtime with writes or setattr (607, 619)
2960 - client verifies file size before zeroing page past EOF (445)
2961 - OST now writes last allocated objid to disk with allocation (108)
2962 - LOV on echo now works (409)
2964 - mds_reint_unlink sends a new buffer, with the EA included. this
2965 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
2966 set if a regular file was being unlinked, and it was the last link
2967 - use PtlGet from the target for bulk writes (315)
2968 - OST now updates object mtime with writes or setattr (607, 619)
2969 - LDLM now has a grant-time callback to revalidate locked items, if
2971 - Many MDS operations were reorganized to combat race conditions
2973 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
2974 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
2976 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
2979 - Fully reactivate OST imports after reconnection (512, others)
2980 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
2981 - More graceful error handling for truncating on dead OST (515)
2982 - Don't error out unless we're actually accessing dead stripes (474)
2983 - Fix garbage sizes when stripes are missing (410)
2984 - LRU counters were broken, causing constant lock purge (433, 432)
2985 - garbage on read from stripes with failed OSTs (441)
2986 - mark OSCs as active before reconnecting during recovery (438)
2987 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
2988 - lfind did not preserve OST order in output (443)
2989 - symlinks cause hung clients, incorrect data (439)
2990 - stop dereferencing request after dropping refcount (457)
2991 - don't LASSERT(spin_is_locked) on non-SMP (455)
2992 - fixes for many rename() bugs
2993 - fstat didn't correctly synchronize attributes (399)
2994 - server must handle lock cancellation during blocking AST prep (487)
2995 - bulk descriptors were free()d too soon (511)
2996 - fix paths in lconf, which would load incorrect modules (451, 507)
2997 - fix confusing lconf 'host not found' error message (386)
2998 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
2999 - fix race condition in mdc_blocking_ast() for inode access (526)
3000 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3001 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3002 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3003 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3004 - fix lstripe in several ways (kernel side) (527)
3005 - fix request leak in ldlm_cli_enqueue (262)
3006 - incorrect OSC was marked inactive after OST failure
3007 - call mds_fs_cleanup before unmounting filesystem (524)
3008 - fix races between taking ns_lock and ldlm_lock_change_resource
3009 - fix races updating LOV export open file list
3010 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3011 - fix recovery NULL deref in ldlm_cli_cancel_unused
3012 - fix some DLM races by using new hash table for lock handles (419)
3013 - permit the client to specify desired inodes, at replay
3014 - duplicate requests when we queue them for replay reintegration
3015 - fix last_rcvd offset calculation
3016 - sync after each recovered transaction, so we always make progress
3017 - never, not always, ERESTART requests without transnos
3018 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3020 - skip replay if the MDS says that the client is already connected
3021 - don't check for a recovery-enabled export to match lctl's UUID
3022 - don't INC_USE_COUNT for phantom exports
3023 - don't crash when cleaning up phantom exports (567)
3024 - don't double-finish or set replay data for errored mdc_open requests
3025 - abort requests when they time out, so we don't get old replies
3026 - send/receive replies for AST messages again
3027 - if the client says that it doesn't have the lock, cancel it on the
3029 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3030 mark it as destroyed, it will all work out when decref is called
3031 - fix module use counts (22, 581)
3033 - ASTs now expect a reply (server cancels lock on error reply)
3035 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3038 - fix many simultaneous client startup (392)
3039 - fix dentry->d_it clobbering
3040 - credentials weren't being shipped for readdir/getattr operations
3041 - remove invalid assertions triggered during some concurrent MD
3043 - proper Lustre versions added (336, 389)
3044 - fix memory leak for create error case (398)
3045 - fix LOV locking bug that would get cli/srv out of sync
3046 - fix echo client over LOV (409)
3047 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3048 - fix double-O_EXCL intent crash (424)
3049 - avoid sending multiple lock CANCELs (352)
3051 - MDS can do multi-client recovery (modulo bugs in new code)
3053 - many updates, edits, cleanups
3055 2002-11-18 Phil Schwan <phil@clusterfs.com>
3058 - fix null d_it dereference (346)
3059 - fix full OST/dbench hang (333)
3060 - fix permission problem with file removal (286)
3061 - fix removal of OSCs from LOV when they fail
3062 - fix NULL deref during bulk timeout (214)
3063 - fix problems related to multiple filesystems on one MDS (241)
3064 - fixed serious subtle metadata locking bugs
3065 - free locks on clients when inodes are removed due to memory
3067 - fix inode pointer in lock data (285)
3068 - partial support for multiple MDS on a single host (241)
3069 - data locks weren't cancelled at clear_inode time (290, 311)
3070 - intent locks could lead to unbounded lock growth (205)
3071 - added a maximum lock count, an LRU list, and a flusher
3072 - fix multiple rename (365)
3073 - properly abstracted the echo client
3074 - OSC locked 1 byte too many; fixed
3075 - rewrote brw callback code:
3076 - fixed recovery bugs related to LOVs (306)
3077 - fixed too-many-pages-in-one-write crash (191)
3078 - fixed (again) crash in sync_io_timeout (214)
3079 - probably fixed callback-related race (385)
3081 - Add capability to MDS protocol
3082 - LDLM cancellations and callbacks on different portals
3084 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3087 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3088 - defer open object destruction to close time (601981, 138)
3089 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3090 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3091 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3092 - don't call obd_setattr in ll_file_release for destroyed objects
3093 * protocol change to lustre_msg: move |version| and add |flags|
3094 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3095 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3096 * added replay of create, unlink, link and rename operations during
3097 MDS failover; recovery should be much more robust now
3098 * remove failed OSCs from LOVs (only lov_create uses this so far)
3099 * the lustre-HOWTO was brought (more) up to date (582544)
3101 2002-10-23 Phil Schwan <phil@clusterfs.com>
3104 - in-use dentries weren't being reused properly (617851)
3105 - prevent multiple LDLM setup (599178)
3106 - fix LOV size calculations for truncate (617853)
3107 - fix client handling of MDS intent errors (POSIX)
3108 - fix permission bug in lovstripe.c test (624321)
3109 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3110 - truncate past end of file could corrupt data
3111 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3112 - a race in recovery could return ETIMEDOUT to apps (623947)
3113 - building outside the source directory was fixed
3114 * the lustre-HOWTO was brought (more) up to date (582544)
3115 * major progress was made on recovery functionality
3117 2002-10-10 Phil Schwan <phil@clusterfs.com>
3120 - recovery deadlock fix
3121 - rm -rf causes LBUG fix (617817)
3122 - file open by multiple tasks fix (618962)
3123 - directory permissions bugs (602707 and 620007)
3124 - journal_stop fixed with locking (611313)
3125 - O_APPEND failures resolved (618273, perhaps 614459)
3126 - lconf PATH fix (619770)
3127 - IA64 build fix (621450)
3128 - RPC buffer sizes scale with amount of memory
3130 2002-10-01 Phil Schwan <phil@clusterfs.com>
3133 - locks would be cancelled without throwing away data pages,
3134 resulting in inconsistent data (605627)
3135 - inode attributes were not always being refreshed (605627, 612449)
3136 - lconf now continues to cleanup after lctl reports an error
3137 - MDS now enforces user permissions (602707)
3138 - lprocfs cleanup fixed, but not yet enabled (614157)
3139 - fixed infinite server hang, should a client not respond to an AST
3140 - avoid going into recovery if user calls readlink() with a buffer
3141 that's too small (613941)
3142 - AST RPCs no longer require replies (614867) -- this may be changed
3143 - don't crash server if client sends an IOV that's too big (611336)
3144 - fixed lock conversion deadlock (611892)
3145 - fixed the following of symlinks (614622)
3146 * recovery: the server can remove locks from a client that dies, other
3147 clients can make progress
3148 * more extN patch fixes
3149 * compile-time configurable ptlrpc buffer allocations
3151 - collaborative read cache document
3152 - Lustre Lite Performance CDR document-in-progress
3154 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3157 - fix typo in patch-2.4.18
3159 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3162 - clear ptlrpc request each time in handle_incoming_request()
3163 - unlink of files now destroys the object on the OST
3165 2002-09-19 Peter Braam <braam@clusterfs.com>
3167 * add hard link support
3168 * change obdfile creation method
3169 * kernel patch changed
3171 2002-09-19 Peter Braam <braam@clusterfs.com>
3174 - stack overflow bug in extN fixed
3176 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3178 * documentation updates
3179 - add man pages for config tools
3180 - update tests/README to describe testing with new config tools
3181 - finish metadata API descriptions
3182 * bug fixes and cleanups
3183 - statfs workaround for 16TB limit
3184 - LOV stripe allocation improved, can stripe on subset of OSTs
3185 - LOV file size/IO offset was wrong for files > 4GB in size
3186 - object EA data was being dropped, caused files to be unreadable
3187 - memory overflow with non-LOV OST caused memory corruption
3188 - fixed regression tests to work with new config tools, obdfilter
3189 - fixed bug when directory size became larger than 1 block
3190 - fixed bug (for single client case) when PWD was deleted
3191 - invalidate local directory pages when doing intent-based ops
3192 - avoid LDLM oops when lock callback contained bad data
3194 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3196 * documentation updates
3197 * bug fixes and cleanups
3198 - configuration tools
3201 - 64-bit compile warnings
3202 - 64-bit internal statfs data
3204 * test_brw on persistent OST devices
3206 * lprocfs (disabled)
3208 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3210 * documentation updates
3211 * bug fixes and cleanups
3212 * configuration tools
3214 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3217 * many small fixes to 0_5_4
3218 * io/network handling
3219 * thinkos in MDS operations
3221 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3224 * crucial basic fixes to 0.5.3
3225 * IOR, Iozone work over Elan
3228 2002-08-07 Phil Schwan <phil@clusterfs.com>
3229 * version 0_5_3, our first alpha
3230 * we use the new Portals iovs
3231 * documentation updates
3232 * bug fixes and cleanups
3233 * small changes in the DLM wire protocol
3235 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3236 * version 0_5_1 with some initial stability,
3237 * locking on MD and file I/O.
3238 * documentation updates
3239 * several bug fixes since 0.5.0
3240 * small changes in wire protocol
3242 2002-07-18 Phil Schwan <phil@clusterfs.com>
3244 * delivered as Lustre Light Alpha
3245 * fixed a crash after handling invalid MDS requests
3246 * fixed directory pages for architectures with non-4k pages sizes
3248 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3249 * release version v0_4_4
3250 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3252 * Fixes a number of interruption problems with OST operations.
3253 * Update documentation for portals header changes
3254 * Move all wire protocol structs/defines to lustre_idl.h
3255 * Fixes symlink length bug.
3256 * Add tcpdump to repository.
3258 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3259 * release version v0_4_3
3260 * Fixes statfs for inodes on extN.
3261 * Fixes bug in runtests which would delete /etc/hosts.
3262 * Use 64-bit object IDs wherever possible (not into VFS though)
3263 Remove ost_get_info, which is unused by lustre, and out of date.
3265 2002-07-03 Peter Braam <braam@clusterfs.com>
3266 * release version v0_4_2 Fixes a lookup error (type not passed)
3267 * move forward to head of Portals
3268 * move forward to latest Lustre kernel
3270 2002-06-25 Peter Braam <braam@clusterfs.com>
3271 * release version v0_4_1. Hopefully stable on single node use.