1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
4 INTEROPERATE with 1.4.x versions automatically. In many cases a
5 special upgrade step is needed. Please read the user documentation
6 before upgrading any part of a 1.4.x system.
7 * WARNING: Lustre configuration and startup changes are required with
8 1.6.x releases. See https://mail.clusterfs.com/wikis/lustre/MountConf
10 * Support for kernels:
11 2.6.9-42.0.10.EL (RHEL 4)
13 2.4.21-47.0.1.EL (RHEL 3)
14 2.6.12.6 vanilla (kernel.org)
15 2.6.16.27-0.9 (SLES 10)
16 * Client support for unpatched kernels:
17 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
18 2.6.16 - 2.6.19 vanilla (kernel.org)
19 2.6.9-42.0.8EL (RHEL 4)
20 * Recommended e2fsprogs version: 1.39.cfs6
22 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
25 Severity : enhancement
27 Description: lfs setstripe use optional parameters instead of postional
30 Severity : enhancement
32 Description: Nanosecond timestamp support.
34 Severity : enhancement
36 Description: Move random uuid functions to prng.c.
39 Frequency : during server recovery
41 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
42 Details : request with rq_no_resend flag not awake l_wait_event if they get a
46 --------------------------------------------------------------------------------
48 tbd Cluster File Systems, Inc. <info@clusterfs.com>
50 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
51 INTEROPERATE with older versions automatically. In many cases a
52 special upgrade step is needed. Please read the
53 user documentation before upgrading any part of a 1.4.x system.
54 * WARNING: Lustre configuration and startup changes are required with
55 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
57 * Support for kernels:
58 2.6.9-42.0.10.EL (RHEL 4)
60 2.4.21-47.0.1.EL (RHEL 3)
61 2.6.12.6 vanilla (kernel.org)
62 2.6.16.27-0.9 (SLES 10)
63 * Client support for unpatched kernels:
64 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
65 2.6.16 - 2.6.19 vanilla (kernel.org)
66 2.6.9-42.0.8EL (RHEL 4)
67 * Recommended e2fsprogs version: 1.39.cfs5
69 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
72 Severity : enhancement
74 Description: MountConf
75 Details : Lustre configuration is now managed via mkfs and mount
76 commands instead of lmc and lconf. New obd types (MGS, MGC)
77 are added for dynamic configuration management. See
78 https://mail.clusterfs.com/wikis/lustre/MountConf for
81 Severity : enhancement
83 Description: dynamic OST addition
84 Details : OSTs can now be added to a live filesystem
86 Severity : enhancement
88 Description: startup order invariance
89 Details : MDTs and OSTs can be started in any order. Clients only
90 require the MDT to complete startup.
92 Severity : enhancement
94 Description: parallel, asynchronous orphan cleanup
95 Details : orphan cleanup is now performed in separate threads for each
96 OST, allowing parallel non-blocking operation.
98 Severity : enhancement
100 Description: optimized stripe assignment
101 Details : stripe assignments are now made based on ost space available,
102 ost previous usage, and OSS previous usage, in order to try
103 to optimize storage space and networking resources.
105 Severity : enhancement
107 Description: Permanently set tunables
108 Details : All writable /proc/fs/lustre tunables can now be permanently
109 set on a per-server basis, at mkfs time or on a live
112 Severity : enhancement
114 Description: Lustre message v2
115 Details : Add lustre message format v2.
117 Severity : enhancement
119 Description: client OST exclusion list
120 Details : Clients can be started with a list of OSTs that should be
121 declared "inactive" for known non-responsive OSTs.
124 Frequency : SFS test only (otherwise harmless)
126 Description: SPEC SFS validation failure on NFS v2 over lustre.
127 Details : Changes the blocksize for regular files to be 2x RPC size,
128 and not depend on stripe size.
130 Severity : enhancement
132 Description: Multiple MD RPCs in flight.
133 Details : Further unserialise some read-only MDT RPCs - learn about intents.
134 To avoid overly-overloading MDT, introduce a limit on number of
135 MDT RPCs in flight for a single client and add /proc controls
136 to adjust this limit.
138 Severity : enhancement
140 Description: client read/write statistics
141 Details : Add client read/write call usage stats for performance
142 analysis of user processes.
143 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
144 file access. extents_stats shows chunk size distribution.
145 extents_stats_per_process show chunk size distribution per
148 Severity : enhancement
150 Description: per-client statistics on server
151 Details : Add ldlm and operations statistics for each client in
152 /proc/fs/lustre/mds|obdfilter/*/exports/
154 Severity : enhancement
156 Description: improved MDT statistics
157 Details : Add detailed MDT operations statistics in
158 /proc/fs/lustre/mds/*/stats
160 Severity : enhancement
162 Description: VFS operations stats
163 Details : Add client VFS call stats, trackable by pid, ppid, or gid
164 /proc/fs/lustre/llite/*/vfs_ops_stats
165 /proc/fs/lustre/llite/*/vfs_track_[pid|ppid|gid]
170 Description: Fix client-side osc byte counters
171 Details : The osc read/write byte counters in
172 /proc/fs/lustre/osc/*/stats are now working
175 Frequency : always as root on SLES
177 Description: Failure of copying files with lustre special EAs.
178 Details : Client side always return success for setxattr call for lustre
179 special xattr (currently only "trusted.lov").
184 Description: Refcount LNET uuids
185 Details : The global LNET uuid list grew linearly with every startup;
186 refcount repeated list entries instead of always adding to
189 Severity : enhancement
191 Description: Dynamic service threads
192 Details : Within a small range, start extra service threads
193 automatically when the request queue builds up.
196 Frequency : mixed-endian client/server environments
198 Description: mixed-endian crashes
199 Details : The new msg_v2 system had some failures in mixed-endian
202 Severity : enhancement
204 Description: Easy OST removal
205 Details : OSTs can be permanently deactivated with e.g. 'lctl
206 conf_param lustre-OST0001.osc.active=0'
208 Severity : enhancement
210 Description: MGS proc entries
211 Details : Added basic proc entries for the MGS showing what filesystems
214 Severity : enhancement
216 Description: provide MGS failover
217 Details : Added config lock reacquisition after MGS server failover.
219 Severity : enhancement
221 Description: add Linux 2.4 support
222 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
226 Description: a large application tries to do I/O to the same resource and dies
228 Details : Check the req->rq_arrival time after the call to
229 ost_brw_lock_get(), but before we do anything about
230 processing it & sending the BULK transfer request. This
231 should help move old stale pending locks off the queue as
232 quickly as obd_timeout.
235 Frequency : when an incorrect nid is specified during startup
237 Description: ptlrpc connect to non-existant node causes kernel crash
238 Details : LNET can't be re-entered from an event callback, which
239 happened when we expire a message after the export has been
240 cleaned up. Instead, hand the zombie cleanup off to another
243 Severity : enhancement
245 Description: plain/inodebits lock performance improvement
246 Details : Grouping plain/inodebits in granted list by their request modes
247 and bits policy, thus improving the performance of search through
251 Frequency : only for kernels with patches from Lustre below 1.4.3
253 Description: Remove old rdonly API
254 Details : Remove old rdonly API which unsed from at least lustre 1.4.3
257 Frequency : only for devices with external journals
259 Description: Set external device read-only also
260 Details : During a commanded failover stop, we set the disk device
261 read-only while the server shuts down. We now also set any
262 external journal device read-only at the same time.
265 Frequency : when upgrading from 1.4 while trying to change parameters
267 Description: The wrong (new) MDC name was used when setting parameters for
268 upgraded MDT's. Also allows changing of OSC (and MDC)
269 parameters if --writeconf is specified at tunefs upgrade time.
272 Frequency : when setting specific ost indicies
274 Description: QOS code breaks on skipped indicies
275 Details : Add checks for missing OST indicies in the QOS code, so OSTs
276 created with --index need not be sequential.
278 ------------------------------------------------------------------------------
280 TBD Cluster File Systems, Inc. <info@clusterfs.com>
282 * Support for kernels:
283 2.6.16.21-0.8 (SLES10)
284 2.6.9-42.0.8EL (RHEL 4)
286 2.4.21-47.0.1.EL (RHEL 3)
287 2.6.12.6 vanilla (kernel.org)
288 * Recommended e2fsprogs version: 1.39.cfs2-0
291 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
293 Description: System hang on startup
294 Details : This bug allowed the liblustre (e.g. catamount) client to
295 return to the app before handling all startup RPCs. This
296 could leave the node unresponsive to lustre network traffic
297 and manifested as a server ptllnd timeout.
299 Severity : enhancement
301 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
302 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
303 boolean to control whether to print peer debug info when a
304 client's RPC times out.
309 Description: make O_SYNC working on 2.6 kernels
310 Details : 2.6 kernels use different method for mark pages for write,
311 so need add a code to lustre for O_SYNC work.
316 Description: Failure to close file and release space on NFS
317 Details : Put inode details into lock acquired in ll_intent_file_open.
318 Use mdc_intent_lock in ll_intent_open to properly
319 detect all kind of errors unhandled by mdc_enqueue
324 Description: proc file read during shutdown sometimes raced obd removal,
326 Details : Add lock to prevent obd access after proc file removal.
329 Frequency : Only for files larger than 4GB on 32-bit clients.
331 Description: improperly doing page alignment of locks
332 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
333 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
338 Description: RPCs being resent when they shouldn't be
339 Details : Some RPCs that should not be resent are being resent. This
340 can cause inconsistencies in the RPC state machine. Do not
341 resend such requests.
344 Frequency : rare, only with NFS export
346 Description: Crash on NFS re-export node
347 Details : under very unusual load conditions an assertion is hit in
348 ll_intent_file_open()
351 Frequency : only if OST filesystem is corrupted
353 Description: client incorrectly hits assertion in ptlrpc_replay_req()
354 Details : for a short time RPCs with bulk IO are in the replay list,
355 but replay of bulk IOs is unimplemented. If the OST filesystem
356 is corrupted due to disk cache incoherency and then replay is
357 started it is possible to trip an assertion. Avoid putting
358 committed RPCs into the replay list at all to avoid this issue.
363 Description: large O_DIRECT requests fail under memory pressure/fragmentation
364 Details : Large single O_DIRECT read and write calls can fail to allocate
365 a sufficiently large buffer to process the request. In case of
366 allocation failure the allocation is retried with a smaller
367 buffer and broken into smaller requests.
372 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
373 > 32000 subdirectories
374 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
375 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
376 subdirectory count crosses 32000. This will aid e2fsck to
377 correctly handle more than 32000 subdirectories.
382 Description: versioning check is incomplete
383 Details : Checking the version difference of client vs. server, report
384 error if the gap is too big.
389 Description: add support PG_writeback bit
390 Details : add support for PG_writeback bit for Lustre, for more carefull
391 work with page cache in 2.6 kernel. This also fix some deadlocks
392 and remove hack for work O_SYNC with 2.6 kernel.
394 ------------------------------------------------------------------------------
396 2006-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
398 * Support for kernels:
399 2.6.16.21-0.8 (SLES10)
400 2.6.9-42.0.3EL (RHEL 4)
402 2.4.21-47.0.1.EL (RHEL 3)
403 2.6.12.6 vanilla (kernel.org)
406 * The backwards-compatible /proc/sys/portals symlink has been removed
407 in this release. Before upgrading, please ensure that you change
408 any configuration scripts or /etc/sysctl.conf files that access
409 /proc/sys/portals/* or sysctl portals.* to use the corresponding
410 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
411 in advance of the upgrade on any system running Lustre 1.4.6 or
412 newer, since /proc/sys/lnet was added in that version.
413 * Note that reiserfs quotas are temporarily disabled on SLES 10 in this
417 Frequency : MDS failover only, very rarely
419 Description: "went back in time" messages on mds failover
420 Details : The greatest transno may be lost when the current operation
421 finishes with an error (transno==0) and the client's last_rcvd
422 record is over-written. Save the greatest transno in the
423 mds_last_transno for this case.
426 Frequency : always for specific kernels and striping counts
428 Description: client may get "Matching packet too big" without ACL support
429 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
430 when trying to access files in certain configurations. The
431 clients should in fact be denied when mounting because they do
435 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
437 Description: many clients connecting with IO in progress causes connect timeouts
438 Details : Avoid synchronous journal commits to avoid delays caused by many
439 clients connecting/disconnecting when bulk IO is in progress.
440 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
441 OST_IO_PORTAL to avoid delays behind potentially many pending
445 Frequency : occasionally with multiple writers to a single file
447 Description: shared writes to file may result in wrong size reported by stat()
448 Details : Allow growing of kms when extent lock is cancelled
451 Frequency : always with random mmap IO to multi-striped file
453 Description: mmap write might be lost if we are writing to a 'hole' in stripe
454 Details : Only if the hole is at the end of OST object so that kms is too
455 small. Fix is to increase kms accordingly in ll_nopage.
458 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
460 Description: writes to a missing object would leak memory on the OST
461 Details : If there is an inconsistency between the MDS and OST filesystems,
462 such that the MDS references an object that doesn't exist, writes
463 to that object will leak memory due to incorrect cleanup in the
464 error handling path, eventually running out of memory on the OST.
469 Description: Creating too long symlink causes lustre errors
470 Details : Check symlink and name lengths before sending requests to MDS.
473 Frequency : only if flock is enabled (not on by default)
475 Description: posix locks not released on fd closure on 2.6.9+
476 Details : We failed to add posix locks to list of inode locks on 2.6.9+
477 kernels, this caused such locks not to be released on fd close and
478 then assertions on fs unmount about still used locks.
481 Frequency : MDS failover only, very rarely
483 Description: clients may get ASSERTION(granted_lock != NULL)
484 Details : When request was taking a long time, and a client was resending
485 a getattr by name lock request. The were multiple lock requests
486 with the same client lock handle and
487 mds_getattr_name->fixup_handle_for_resent_request found one of the
488 lock handles but later failed with ASSERTION(granted_lock != NULL).
493 Description: handle->h_buffer_credits > 0, assertion failure
494 Details : h_buffer_credits is zero after truncate, causing assertion
495 failure. This patch extends the transaction or creates a new
499 Frequency : NFS re-export or patchless client
500 Bugzilla : 11179, 10796
501 Description: Crash on NFS re-export node (__d_move)
502 Details : We do not want to hash the dentry if we don't have a lock.
503 But if this dentry is later used in d_move, we'd hit uninitialised
504 list head d_hash, so we just do this to init d_hash field but
505 leave dentry unhashed.
508 Frequency : NFS re-export or patchless client
510 Description: NFS exports has problem with symbolic link
511 Details : lustre client didn't properly install dentry when re-exported
512 to NFS or running patchless client.
515 Frequency : NFS re-export or patchless client
517 Description: Various nfs/patchless fixes.
518 Details : fixes reuse disconected alias for lookup process - this fixes
519 warning "find_exported_dentry: npd != pd",
520 fix permission error with open files at nfs.
521 fix apply umaks when do revalidate.
524 Frequency : occasional
526 Description: Crash on NFS re-export node
527 Details : calling clear_page() on the wrong pointer triggered oops in
528 generic_mapping_read().
531 Frequency : rarely, using O_DIRECT IO
533 Description: unaligned directio crashes client with LASSERT
534 Details : check for unaligned buffers before trying any requests.
537 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
539 Description: stale data returned from RAID cache
540 Details : If only a small amount of IO is done to the RAID device before
541 reading it again it is possible to get stale data from the RAID
542 cache instead of reading it from disk.
545 Frequency : depends on arch, kernel and compiler version, always on sles10
548 Description: recursive or deep enough symlinks cause stack overflow
549 Details : getting rid of large stack-allocated variable in
553 Frequency : depends on hardware
555 Description: lustre write performance loss in the SLES10 kernel
556 Details : the performance loss is caused by using of write barriers in the
557 ext3 code. The SLES10 kernel turns barrier support on by
558 default. The fix is to undo that change for ldiskfs.
560 ------------------------------------------------------------------------------
562 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
564 * Support for kernels:
565 2.6.9-42.0.3EL (RHEL 4)
567 2.4.21-47.0.1.EL (RHEL 3)
568 2.6.12.6 vanilla (kernel.org)
572 Frequency : quota enabled and large files being deleted
574 Description: releasing more than 4GB of quota at once hangs OST
575 Details : If a user deletes more than 4GB of files on a single OST it
576 will cause the OST to spin in an infinite loop. Release
577 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
582 Description: statfs data retrieved from /proc may be stale or zero
583 Details : When reading per-device statfs data from /proc, in the
584 {kbytes,files}_{total,free,avail} files, it may appear
585 as zero or be out of date.
588 Frequency : systems with MD RAID1 external journal devices
590 Description: lconf's call to blkid is confused by RAID1 journal devices
591 Details : Use the "blkid -l" flag to locate the MD RAID device instead
592 of returning all block devices that match the journal UUID.
595 Frequency : always, for aggregate stripe size over 4GB
597 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
598 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
599 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
602 Frequency : always if "lfs find" used on a local file/directory
604 Description: "lfs find" segfaults if used on a local file/directory
605 Details : The case where a directory component was not specified wasn't
606 handled correctly. Handle this properly.
609 Frequency : always on ppc64
611 Description: the write to an ext3 filesystem mounted with mballoc got stuck
612 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
613 perform endianness conversion.
616 Frequency : rarely (truncate to non-zero file size after write under load)
617 Bugzilla : 10730, 10687
618 Description: Files padded with zeros to next 4K multiple
619 Details : With filesystems mounted using the "extents" option (2.6 kernels)
620 it is possible that files that are truncated to a non-zero size
621 immediately after being written are filled with zero bytes beyond
622 the truncated size. No file data is lost.
624 Severity : enhancement
626 Description: Allow recovery/failover for liblustre clients.
627 Details : liblustre clients were unaware of failover configurations until
630 Severity : enhancement
632 Description: user file locks should fail when not mounting with flock option
633 Details : Set up an error-returning stub in ll_file_operations.lock field
634 to prevent incorrect behaviour when client is mounted without
635 flock option. Also, set up properly f_op->flock field for
639 Frequency : always on ia64
641 Description: "lfs df" loops on printing out MDS statfs information
642 Details : The obd_ioctl_data was not initialized and in some systems
643 this caused a failure during the ioctl that did not return
644 an error. Initialize the struct and return an error on failure.
647 Frequency : SLES 9 only
649 Description: Error of copying files with lustre special EAs as root
650 Details : Client side always return success for setxattr call for lustre
651 special xattr (currently only "trusted.lov").
654 Frequency : rarely on clusters with both ia64+i386 clients
656 Description: ia64+i686 clients doing shared IO on the same file may LBUG
657 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
658 clients are doing concurrent writes to the same file it is
659 possible that the ia64 clients may LASSERT because the OST
660 extent locks are not PAGE_SIZE aligned. Ensure that grown
661 locks are always aligned on the request boundary.
664 Frequency : specific use, occasional
666 Description: Overwriting in use executable truncates on-disk binary image
667 Details : If one node attempts to overwrite an executable in use by
668 another node, we now correctly return ETXTBSY instead of
674 Description: chmod on Lustre root is propagated to other clients
675 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
676 invalid by the follow_mount time.
679 Frequency : liblustre clients only
681 Description: Race in 'instant cancel' lock handling could lead to such locks
682 never to be granted in case of SMP MDS
683 Details : Do not destroy not yet granted but cbpending locks in
687 Frequency : replay/resend of open
689 Description: non null lock assetion failure in mds_intent_policy
690 Details : Trying to replay/resend lockless open requests resulted in
691 mds_open() returning 0 with no lock. Now it sets a flag if
692 a lock is going to be returned.
694 Severity : enhancement
696 Description: Checksum enhancements
697 Details : New checksum enhancements allow for resending RPCs that failed
700 Severity : enhancement
702 Description: Tunables on number of dirty pages in cacche
703 Details : Allow to set limit on number of dirty pages cached.
708 Description: client crash on unmount - lock still has references
709 Details : In some error handling cases it was possible to leak a lock
710 reference on a client while accessing a file. This was not
711 harmful to the client during operation, but would cause the
712 client to crash when the filesystem is unmounted.
715 Frequency : specific case, rare
717 Description: ETXTBSY on mds though file not in use by client
718 Details : ETXTBSY is no longer incorrectly returned when attempting to
719 chmod or chown a directory that the user previously tried to
720 execute or a currently-executing binary.
723 Frequency : extremely rare except on liblustre-based clients
725 Description: Lustre space not freed when files are deleted
726 Details : Clean up open-unlinked files after client eviction. Previously
727 the unlink was skipped and the files remained as orphans.
732 Description: OST failure "would be an LBUG" in waiting_locks_callback()
733 Details : In some cases it was possible to send a blocking callback to
734 a client doing a glimpse, even though that client didn't get
735 a lock granted. When the glimpse lock is cancelled on the OST
736 the freed lock is left on the waiting list and corrupted the list.
739 Frequency : all core dumps
741 Description: Broke core dumps to lustre
742 Details : Negative dentry may be unhashed if parent does not have UPDATE
743 lock, but some callers, e.g. do_coredump, expect dentry to be
744 hashed after successful create, hash it in ll_create_it.
746 ------------------------------------------------------------------------------
748 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
750 * Support for kernels:
751 2.6.9-42.0.2.EL (RHEL 4)
753 2.4.21-40.EL (RHEL 3)
754 2.6.12.6 vanilla (kernel.org)
758 Frequency : always on RHEL 3
760 Description: Number of open files grows over time
761 Details : The number of open files grows over time, whether or not
762 Lustre is started. This was due to a filp leak introduced
763 by one of our kernel patches.
765 ------------------------------------------------------------------------------
767 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
769 * Support for kernels:
772 2.4.21-40.EL (RHEL 3)
773 2.6.12.6 vanilla (kernel.org)
778 Bugzilla : 5719, 9635, 9792, 9684
779 Description: OST (or MDS) trips assertions in (re)connection under heavy load
780 Details : If a server is under heavy load and cannot reply to new
781 connection requests before the client resends the (re)connect,
782 the connection handling code can behave badly if two service
783 threads are concurrently handing separate (re)connections from
784 the same client. Add better locking to the connection handling
785 code, and ensure that only a single connection will be processed
786 for a given client UUID, even if the lock is dropped.
788 Severity : enhancement
790 Description: add TCP zero-copy support to kernel
791 Details : Add support to the kernel TCP stack to allow zero-copy bulk
792 sends if the hardware supports scatter-gather and checksumming.
793 This allows socklnd to do client-write and server-read more
794 efficiently and reduce CPU utilization from skbuf copying.
797 Frequency : only if NFS exporting from client
799 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
800 Details : If setting attributes on a file created under NFS that had
801 never been opened it would be possible to oops the client
802 if the file had no objects.
806 Bugzilla : 9326, 10402, 10897
807 Description: client crash in ptlrpcd_wake() thread when sending async RPC
808 Details : It is possible that ptlrpcd_wake() dereferences a freed async
809 RPC. In rare cases the ptlrpcd thread alread processed the RPC
810 before ptlrpcd_wake() was called and the request was freed.
813 Frequency : always for liblustre
815 Description: liblustre client does MDS+OSTs setattr RPC for each write
816 Details : When doing a write from a liblustre client, the client
817 incorrectly issued an RPC to the MDS and each OST the file was
818 striped over in order to update the timestamps. When writing
819 with small chunks and many clients this could overwhelm the MDS
820 with RPCs. In all cases it would slow down the write because
821 these RPCs are unnecessary.
823 Severity : enhancement
825 Description: allow number of MDS service threads to be changed at module load
826 Details : It is now possible to change the number of MDS service threads
827 running. Adding "options mds mds_num_threads={N}" to the MDS's
828 /etc/modprobe.conf will set the number of threads for the next
829 time Lustre is restarted (assuming the "mds" module is also
830 reloaded at that time). The default number of threads will
831 stay the same, 32 for most systems.
836 Description: OST crash if filesystem is unformatted or corrupt
837 Details : If an OST is started on a device that has never been formatted
838 or if the filesystem is corrupt and cannot even mount then the
839 error handling cleanup routines would dereference a NULL pointer.
844 Description: NULL pointer deref in llap_from_page.
845 Details : get_cache_page_nowait can return a page with NULL (or otherwise
846 incorrect) mapping if the page was truncated/reclaimed while it was
847 searched for. Check for this condition and skip such pages when
848 doing readahead. Introduce extra check to llap_from_page() to
849 verify page->mapping->host is non-NULL (so page is not anonymous).
852 Frequency : Sometimes when using sys_sendfile
854 Description: "page not covered by a lock" warnings from ll_readpage
855 Details : sendfile called ll_readpage without right page locks present.
856 Now we introduced ll_file_sendfile that does necessary locking
857 around call to generic_file_sendfile() much like we do in
861 Frequency : with certain MDS communication failures at client mount time
863 Description: NULL pointer deref after failed client mount
864 Details : a client connection request may delayed by the network layer
865 and not be sent until after the PTLRPC layer has timed out the
866 request. If the client fails the mount immediately it will try
867 to clean up before the network times out the request. Add a
868 reference from the request import to the obd device and delay
869 the cleanup until the network drops the request.
872 Frequency : occasionally during client (re)connect
874 Description: assertion failure during client (re)connect
875 Details : processing a client connection request may be delayed by the
876 client or server longer than the client connect timeout. This
877 causes the client to resend the connection request. If the
878 original connection request is replied in this interval, the
879 client may trip an assertion failure in ptlrpc_connect_interpret()
880 which thought it would be the only running connect process.
883 Frequency : only with obd_echo servers and clients that are rebooted
885 Description: kernel BUG accessing uninitialized data structure
886 Details : When running an obd_echo server it did not start the ping_evictor
887 thread, and when a client was evicted an uninitialized data
888 structure was accessed. Start the ping_evictor in the RPC
889 service startup instead of the OBD startup.
891 Severity : enhancement
892 Bugzilla : 10193 (patchless)
893 Description: Remove dependency on various unexported kernel interfaces.
894 Details : No longer need reparent_to_init, exit_mm, exit_files,
895 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
898 Frequency : rare (only users of deprecated and unsupported LDAP config)
900 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
901 Details : LDAP apparently contains 'lustreName' attributes instead of
902 'name'. A simple remapping of the name is sufficient.
905 Frequency : rare (only with non-default dump_on_timeout debug enabled)
907 Description: waiting_locks_callback trips kernel BUG if client is evicted
908 Details : Running with the dump_on_timeout debug flag turned on makes
909 it possible that the waiting_locks_callback() can try to dump
910 the Lustre kernel debug logs from an interrupt handler. Defer
911 this log dumping to the expired_lock_main() thread.
913 Severity : enhancement
915 Description: Support NFS exporting on 2.6 kernels.
916 Details : Implement non-rawops metadata methods for NFS server to use without
917 changing NFS server code.
920 Frequency : very rare (synthetic metadata workload only)
922 Description: two racing renames might cause an MDS thread to deadlock
923 Details : Running the "racer" program may cause one MDS thread to rename
924 a file from being the source of a rename to being the target of
925 a rename at exactly the same time that another thread is doing
926 so, and the second thread has already enqueued these locks after
927 doing a lookup of the target and is trying to relock them in
928 order. Ensure that we don't try to re-lock the same resource.
931 Frequency : only very large systems with liblustre clients
933 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
934 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
935 all OSTs in parallel. This allows the eviction of stale liblustre
936 clients to proceed much faster than if they were done in series,
937 and also offers similar improvements for other set_info RPCs.
942 Description: excessive CPU usage during initial read phase on client
943 Details : During the initial read phase on a client, it would agressively
944 retry readahead on the file, consuming too much CPU and impacting
945 performance (since 1.4.5.8). Improve the readahead algorithm
946 to avoid this, and also improve some other common cases (read
947 of small files in particular, where "small" is files smaller than
948 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
953 Description: MDS crash when receiving packet with unknown intent.
954 Details : Do not LBUG in unknown intent case, just return -EFAULT
956 Severity : enhancement
957 Bugzilla : 9293, 9385
958 Description: MDS RPCs are serialised on client. This is unnecessary for some.
959 Details : Do not serialize getattr (non-intent version) and statfs.
962 Frequency : occasional, when OST network is overloaded/intermittent
964 Description: client evicted by OST after bulk IO timeout
965 Details : If a client sends a bulk IO request (read or write) the OST
966 may evict the client if it is unresposive to its data GET/PUT
967 request. This is incorrect if the network is overloaded (takes
968 too long to transfer the RPC data) or dropped the OST GET/PUT
969 request. There is no need to evict the client at all, since
970 the pinger and/or lock callbacks will handle this, and the
971 client can restart the bulk request.
974 Frequency : Always when mmapping file with no objects
976 Description: client crashes when mmapping file with no objects
977 Details : Check that we actually have objects in a file before doing any
978 operations on objects in ll_vm_open, ll_vm_close and
984 Description: Request leak when working with deleted CWD
985 Details : Introduce advanced request refcount tracking for requests
986 referenced from lustre intent.
988 Severity : Enhancement
990 Description: Cache open file handles on client.
991 Details : MDS now will return special lock along with openhandle, if
992 requested and client is allowed to hold openhandle, even if unused,
993 until such a lock is revoked. Helps NFS a lot, since NFS is opening
994 closing files for every read/write openration.
996 Severity : Enhancement
998 Description: Cache open negative dentries on client when possible.
999 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1000 negative dentries on lock revocation.
1005 Description: Remounting a client read-only wasn't possible with a zconf mount
1006 Details : It wasn't possible to remount a client read-only with llmount.
1008 Severity : enhancement
1009 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1010 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1011 MPICH is included to add Lustre-specific ADIO interfaces.
1012 This is based closely on the UFS ADIO layer and only differs
1013 in file creation, in order to allow the OST striping to be set.
1014 This is user-contributed code and not supported by CFS.
1019 Description: extended inode attributes (immutable, append-only) work improperly
1020 when 2.4 and 2.6 kernels are used on client/server or vice versa
1021 Details : Introduce kernel-independent values for these flags.
1023 Severity : enhancement
1026 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1027 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1028 granularity. Now, Fractional MB granularity is be supported,
1029 this is very useful for low memory system.
1031 Severity : enhancement
1033 Description: Getattr by fid
1034 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1035 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1036 have locks and avoids getattr rpc requests for it on every stat.
1039 Frequency : Always, for filesystems larger than 2TB
1041 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1042 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1043 filesystems larger than 2TB to be mounted with mballoc enabled.
1046 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1048 Description: filesystem corruption for non-standard kernels and very large OSTs
1049 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1050 filesystems larger than 2TB is mounted then the kernel will
1051 silently corrupt the start of the filesystem. CONFIG_LBD is
1052 enabled for all CFS-supported kernels, but the possibility of
1053 this happening with a modified kernel config exists.
1055 Severity : enhancement
1057 Description: add client O_DIRECT support for 2.6 kernels
1058 Details : It is now possible to do O_DIRECT reads and writes to files
1059 in the Lustre client mountpoint on 2.6 kernel clients.
1061 Severity : enhancement
1063 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1064 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1065 parallel, not waiting for response from every OST before sending
1066 a rpc to the next OST.
1071 Description: setattr vs write race when updating file timestamps
1072 Details : Client processes that update a file timestamp into the past
1073 right after writing to the file (e.g. tar) it is possible that
1074 the updated file modification time can be reset to the current
1075 time due to a race between processing the setattr and write RPC.
1077 Severity : enhancement
1079 Description: Bring 'lfs find' closer in line with regular Linux find.
1080 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1081 -print0 options and obtains all the needed info through the lustre
1084 Severity : enhancement
1086 Description: support up to 1024 configured devices on one node
1087 Details : change obd_dev array from statically allocated to dynamically
1088 allocated structs as they are first used to reduce memory usage
1093 Description: Flush dirty partially truncated pages during truncate
1094 Details : Immediatelly flush partially truncated pages in filter_setattr,
1095 this way we completely avoid having any pages in page cache on OST
1096 and can retire ugly workarounds during writes to flush such pages.
1101 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1102 Details : protect inode from truncation within vfs_unlink() context
1103 just take a reference before calling vfs_unlink() and release it
1104 when parent's i_sem is free.
1107 Frequency : always, if extents are used on OSTs
1109 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1110 Details : OSTs using the extents format would not zero the high 16 bits of
1111 the index physical block number. This is not a problem for any
1112 OST filesystems smaller than 16TB, and no kernels support ext3
1113 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1114 new/modified files) and can be fixed for existing filesystems
1115 with e2fsprogs-1.39-cfs1.
1120 Description: import connection selection may be incorrect if timer wraps
1121 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1122 import connections to be ignored if the 32-bit jiffies counter
1123 wraps. Use a 64-bit jiffies counter.
1126 Frequency : during server recovery
1128 Description: crash after server is denying duplicate export
1129 Details : If clients are resending connect requests to the server, the
1130 server refuses to allow a client to connect multiple times.
1131 Fixed a bug in the handling of this case.
1134 Frequency : very large clusters immediately after boot
1136 Description: LNET request buffers exhausted under heavy short-term load
1137 Details : If a large number of client requests are generated on a service
1138 that has previously never seen so many requests it is possible
1139 that the request buffer growth cannot keep up with the spike in
1140 demand. Instead of dropping incoming requests, they are held in
1141 the LND until the RPC service can accept more requests.
1144 Frequency : Sometimes during replay
1146 Description: Assertion failure in ll_local_open after replay.
1147 Details : If replay happened on an open request reply before we were able
1148 to set replay handler, reply will become not swabbed tripping the
1149 assertion in ll_local_open. Now we set the handler right after
1150 recognising of open request
1153 Frequency : very rare
1155 Description: kernel reports "badness in vsnprintf"
1156 Details : Reading from the "recovery_status" /proc file in small chunks
1157 may cause a negative length in lprocfs_obd_rd_recovery_status()
1158 call to vsnprintf() (which is otherwise harmless). Exit early
1159 if there is no more space in the output buffer.
1161 Severity : enhancement
1163 Description: clear OBD RPC statistics by writing to them
1164 Details : It is now possible to clear the OBD RPC statistics by writing
1165 to the "stats" file.
1170 Description: Client mtime is not the same on different clients after utimes
1171 Details : In some cases, the client was using the utimes() syscall on
1172 a file cached on another node. The clients now validate the
1173 ctime from the MDS + OSTs to determine which one is right.
1178 Description: Inability to activate failout mode
1179 Details : lconf script incorrectly assumed that in python string's numeric
1180 value is used in comparisons.
1183 Frequency : always with multiple stripes per file
1185 Description: Inefficient object allocation for mutli-stripe files
1186 Details : When selecting which OSTs to stripe files over, for files with
1187 a stripe count that divides evenly into the number of OSTs,
1188 the MDS is always picking the same starting OST for each file.
1189 Return the OST selection heuristic to the original design.
1194 Description: mount failures may take full timeout to return an error
1195 Details : Under some heavy load conditions it is possible that a
1196 failed mount can wait for the full obd_timeout interval,
1197 possibly several minutes, before reporting an error.
1198 Instead return an error as soon as the status is known.
1200 ------------------------------------------------------------------------------
1202 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
1204 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1205 INTEROPERATE with older versions automatically. Please read the
1206 user documentation before upgrading any part of a live system.
1207 * WARNING: Lustre networking configuration changes are required with
1208 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1211 * Support for kernels:
1212 2.6.9-22.0.2.EL (RHEL 4)
1213 2.6.5-7.244 (SLES 9)
1214 2.6.12.6 vanilla (kernel.org)
1217 Severity : enhancement
1218 Bugzilla : 7981/8208
1219 Description: Introduced Lustre Networking (LNET)
1220 Details : LNET is new networking infrastructure for Lustre, it includes
1221 a reorganized network configuration mode (see the user
1222 documentation for full details) as well as support for routing
1223 between different network fabrics. Lustre Networking Devices
1224 (LNDs) for the supported network fabrics have also been
1225 created for this new infrastructure.
1227 Severity : enhancement
1228 Description: Introduced Access control lists
1229 Details : clients can set ACLs on files and directories in order to have
1230 more fine-grained permissions than the standard Unix UGO+RWX.
1231 The MDS must be started with the "-o acl" mount option.
1233 Severity : enhancement
1234 Description: Introduced filesystem quotas
1235 Details : Administrators may now establish per-user quotas on the
1238 Severity : enhancement
1240 Description: Configuration change for the XT3
1241 The PTLLND is now used to run Lustre over Portals on the XT3
1242 The configure option(s) --with-cray-portals are no longer used.
1243 Rather --with-portals=<path-to-portals-includes> is used to
1244 enable building on the XT3. In addition to enable XT3 specific
1245 features the option --enable-cray-xt3 must be used.
1250 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1251 Details : A race between allocating a new llcd and re-getting the llcd_lock
1252 allowed another thread to grab newly-allocated llcd.
1254 Severity : enhancement
1256 Description: 2.6 OST async journal commit and locking fix to improve performance
1257 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1258 2.6 kernels are now async as they already were in 2.4 kernels so
1259 that they can commit concurrently with the network bulk transfer.
1260 For block-allocated files the filter allocation semaphore is held
1261 to avoid filesystem fragmentation during allocation. BKL lock
1262 removed for 2.6 xattr operations where it is no longer needed.
1267 Description: lconf incorrectly determined whether two IP networks could talk
1268 Details : In some more complicated routing and multiple-network
1269 configurations, lconf will avoid trying to make a network
1270 connection to a disjoint part of the IP space. It was doing the
1271 math incorrectly for one set of cases.
1276 Description: Fix for potential infinite loop processing records in an llog.
1277 Details : If an llog record is corrupted/zeroed, it is possible to loop
1278 forever in llog_process(). Validate the llog record length
1279 and skip the remainder of the block on error.
1282 Frequency : occasional (liblustre only)
1284 Description: liblustre could not open files whose last component is a symlink
1285 Details : sysio_path_walk() would incorrectly pass the open intent to
1286 intermediate path components.
1289 Frequency : rare (liblustre only with non-standard tuning)
1290 Bugzilla : 7201 (7350)
1291 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1292 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1293 after it has been released, only do this for referenced locks
1295 Severity : enhancement
1297 Description: specifying an (invalid) directory default stripe_size of -1
1298 would reset the directory default striping
1299 Details : stripe_size -1 was used internally to signal directory stripe
1300 removal, now use "all default" to signal dir stripe removal
1301 as a directory striping of "all default" is not useful
1304 Frequency : common for large clusters running liblustre clients
1306 Description: doing an ls when liblustre clients are running is slow
1307 Details : sending a glimpse AST to a liblustre client waits for every AST
1308 to time out, as liblustre clients will not respond. Since they
1309 cannot cache data we refresh the OST lock LVB from disk instead.
1311 Severity : enhancement
1313 Description: doing an ls at the same time as file IO can be slow
1314 Details : enqueue and other "small" requests can be blocked behind many
1315 large IO requests. Create a new OST IO portal for non-IO
1316 requests so they can be processed faster.
1319 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1321 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1322 created file with mode 000
1323 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1324 allow an ftruncate() as a non-root user to a file with mode 000.
1325 Lustre now respects this flag to disable mode checks when
1326 truncating a file owned by the user
1329 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1332 Description: Revoking locks from clients that went dead or catatonic might take
1334 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1335 cancellation of such locks instant on servers without waiting for
1336 any reply from clients. Clients drops these locks when cancel
1337 notification from server is received without replying.
1340 Frequency : liblustre-only, when liblustre client dies or becomes busy
1342 Description: Doing ls on Linux clients can take a long time with active
1344 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1345 granting such locks to it in the first place if possible. Locks
1346 are taken by proxy on the OST during the read or write and
1347 dropped immediately afterward. Add connect flags handling, do
1348 not grant locks to liblustre clients for glimpse ASTs.
1350 Severity : enhancement
1352 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1353 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1354 sequential reads and behaves badly with applications doing random
1355 reads. Improve it by reading ahead at least read region, and
1356 avoiding excessive large RPC for small reads.
1358 Severity : enhancement
1360 Description: Creating more than 1000 files for a single job may cause a load
1361 imbalance on the OSTs if there are also a large number of OSTs.
1362 Details : qos_prep_create() uses an OST index reseed value that is an
1363 even multiple of the number of available OSTs so that if the
1364 reseed happens in the middle of the object allocation it will
1365 still utilize the OSTs as uniformly as possible.
1370 Description: OST or MDS may oops in ping_evictor_main()
1371 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1372 but doesn't restart at beginning of obd_exports_timed list
1375 Severity : enhancement
1377 Description: improve by-nid export eviction on the MDS and OST
1378 Details : allow multiple exports with the same NID to be evicted at one
1379 time without re-searching the exports list.
1382 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1384 Description: MDS may oops in groups_free()
1385 Details : in rare race conditions a newly allocated group_info struct is
1386 freed again, and this can be NULL. The 2.4 compatibility code
1387 for groups_free() checked for a NULL pointer, but 2.6 did not.
1390 Frequency : common for liblustre clients doing little filesystem IO
1391 Bugzilla : 9352, 7313
1392 Description: server may evict liblustre clients accessing contended locks
1393 Details : if a client is granted a lock or receives a completion AST
1394 with a blocking AST already set it would not reply to the AST
1395 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1398 Frequency : lfs setstripe, only systems with more than 160 OSTs
1400 Description: unable to set striping with a starting offset beyond OST 160
1401 Details : llapi_create_file() incorrectly limited the starting stripe
1402 index to the maximum single-file stripe count.
1405 Frequency : LDAP users only
1407 Description: lconf did not handle in-kernel recovery with LDAP properly
1408 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1410 Severity : enhancement
1412 Description: bind OST threads to NUMA nodes to improve performance
1413 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1414 node and do their allocations there to localize memory access
1416 Severity : enhancement
1418 Description: llmount can determine client NID directly from Myrinet (GM)
1419 Details : the client NID code from gmnalnid was moved directly into
1420 llmount, removing the need to use this or specifying the
1421 client NID explicitly when mounting GM clients with zeroconf
1424 Frequency : if client is started with down MDS
1426 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1427 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1428 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1429 ptlrpc_import_delayed_req() because we didn't check if the
1430 request was interrupted, and we also didn't break out of the
1431 event loop if there was no timeout
1436 Description: data loss during non-page-aligned writes to a single file from
1437 both multiple nodes and multiple threads on one node at same time
1438 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1439 inconsistency led to false short-reads, that were cached and later
1440 used by ->prepare_write() to fill in partially written page,
1441 leading to data loss.
1444 Frequency : always, if lconf --abort_recovery used
1446 Description: lconf --abort_recovery fails with 'Operation not supported'
1447 Details : lconf was attempting to abort recovery on the MDT device and not
1450 Severity : enhancement
1452 Description: remove cleanup logs
1453 Details : replace lconf-generated cleanup logs with lustre internal
1454 cleanup routines. Eliminates the need for client-cleanup and
1457 Severity : enhancement
1459 Description: add support for EAs (user and system) on lustre filesystems
1460 Details : it is now possible to store extended attributes in the Lustre
1461 client filesystem, and with the user_xattr mount option it
1462 is possible to allow users to store EAs on their files also
1464 Severity : enhancement
1466 Description: Add possibility (config option) to show minimal available OST free
1468 Details : When compiled with --enable-mindf configure option, statfs(2)
1469 (and so, df) will return least minimal free space available from
1470 all OSTs as amount of free space on FS, instead of summary of
1471 free spaces of all OSTs.
1473 Severity : enhancement
1475 Description: do not expand extent locks acquired on OST-side
1476 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1477 by server: they are not cached anyway.
1480 Frequency : when mmap is used/binaries executed from Lustre
1482 Description: Unmmap pages before throwing them away from read cache.
1483 Details : llap_shrink cache now attempts to unmap pages before discarding
1484 them (if unmapping failed - do not discard). SLES9 kernel has
1485 extra checks that trigger if this unmapping is not done first.
1490 Description: lconf didn't resolve symlinks before checking to see whether a
1491 given mountpoint was already in use
1494 Frequency : when migrating failover services
1495 Bugzilla : 6395, 9514
1496 Description: When migrating a subset of services from a node (e.g. failback
1497 from a failover service node) the remaining services would
1498 time out and evict clients.
1499 Details : lconf --force (implied by --failover) sets the global obd_timeout
1500 to 5 seconds in order to quickly disconnect, but this caused
1501 other RPCs to time out too quickly. Do not change the global
1502 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1504 Severity : enhancement
1505 Frequency : if MDS is started with down OST
1506 Bugzilla : 9439,5706
1507 Description: Allow startup/shutdown of an MDS without depending on the
1508 availability of the OSTs.
1509 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1510 Add appropriate locking and lov-osc refcounts for safe
1511 cleaning. Add osc abort_inflight calls in case the
1512 synchronize never started.
1515 Frequency : occasional (Cray XT3 only)
1517 Description: root not authorized to access files in CRAY_PORTALS environment
1518 Details : The client process capabilities were not honoured on the MDS in
1519 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1520 been accessed by an authorized user then root was able to access
1521 the file on the local client also. The root user capabilities
1522 are now allowed on the MDS, as this environment has secure UID.
1525 Frequency : occasional
1527 Description: ldiskfs "too long searching" message happens too often
1528 Details : A debugging message (otherwise harmless) prints too often on
1529 the OST console. This has been reduced to only happen when
1530 there are fragmentation problems on the filesystem.
1535 Description: Division by zero in statfs when all OSCs are inactive
1536 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
1537 lov_statfs divides by value returned by lov_get_stripecnt().
1541 Bugzilla : 9489, 3273
1542 Description: First write from each client to each OST was only 4kB in size,
1543 to initialize client writeback cache, which caused sub-optimal
1544 RPCs and poor layout on disk for the first writen file.
1545 Details : Clients now request an initial cache grant at (re)connect time
1546 and so that they can start streaming writes to the cache right
1547 away and always do full-sized RPCs if there is enough data.
1548 If the OST is rebooted the client also re-establishes its grant
1549 so that client cached writes will be honoured under the grant.
1554 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
1555 Details : Now I/O RPCs go to different portal number and (presumably) fast
1556 lock requests (and glimses) and other RPCs get their own service
1557 threads pool that should be able to service those RPCs
1560 Severity : enhancement
1562 Description: Ability to exchange lustre version between client and servers and
1563 issue warnings at client side if client is too old. Also for
1564 liblustre clients there is ability to refuse connection of too old
1566 Details : New 'version' field is added to connect data structure that is
1567 filled with version info. That info is later checked by server and
1571 Frequency : rare, liblustre only.
1572 Bugzilla : 9296, 9581
1573 Description: Two simultaneous writes from liblustre at offset within same page
1574 might proceed at the same time overwriting eachother with stale
1576 Details : I/O lock withing llu_file_prwv was released too early, before data
1577 actually was hitting the wire. Extended lock-holding time until
1578 server acknowledges receiving data.
1581 Frequency : extremely rare. Never observed in practice.
1583 Description: avoid generating lustre_handle cookie of 0.
1584 Details : class_handle_hash() generates handle cookies by incrementing
1585 global counter, and can hit 0 occasionaly (this is unlikely, but
1586 not impossible, because initial value of cookie counter is
1587 selected randonly). Value of 0 is used as a sentinel meaning
1588 "unassigned handle" --- avoid it. Also coalesce two critical
1589 sections in this function into one.
1591 Severity : enhancement
1593 Description: allow liblustre clients to delegate truncate locking to OST
1594 Details : To avoid overhead of locking, liblustre client instructs OST to
1595 take extent lock in ost_punch() on client's behalf. New connection
1596 flag is added to handle backward compatibility.
1598 Severity : enhancement
1599 Bugzilla : 4928, 7341, 9758
1600 Description: allow number of OST service threads to be specified
1601 Details : a module parameter allows the number of OST service threads
1602 to be specified via "options ost ost_num_threads={N}" in the
1603 OSS's /etc/modules.conf or /etc/modprobe.conf.
1607 Bugzilla : 6146, 9635, 9895
1608 Description: servers crash with bad pointer in target_handle_connect()
1609 Details : In rare cases when a client is reconnecting it was possible that
1610 the connection request was the last reference for that export.
1611 We would temporarily drop the export reference and get a new
1612 one, but this may have been the last reference and the export
1613 was just destroyed. Get new reference before dropping old one.
1615 Severity : enhancement
1616 Frequency : if client is started with failover MDS
1618 Description: Allow multiple MDS hostnames in the mount command
1619 Details : Try to read the configuration from all specified MDS
1620 hostnames during a client mount in case the "primary"
1623 Severity : enhancement
1625 Description: Stop sending data to evicted clients as soon as possible.
1626 Details : Check if the client we are about to send or are sending data to
1627 was evicted already. (Check is done every second of waiting,
1628 for which l_wait_event interface was extended to allow checking
1629 of exit condition at specified intervals).
1632 Frequency : rare, normally only when NFS exporting is done from client
1634 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
1636 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
1639 Frequency : always, if srand() is called before liblustre initialization
1641 Description: Liblustre uses system PRNG disturbing its usage by user application
1642 Details : Introduce internal to lustre fast and high-quality PRNG for
1643 lustre usage and make liblustre and some other places in generic
1644 lustre code to use it.
1646 Severity : enhancement
1647 Bugzilla : 9477, 9557, 9870
1648 Description: Verify that the MDS configuration logs are updated when xml is
1649 Details : Check if the .xml configuration logs are newer than the config
1650 logs stored on the MDS and report an error if this is the case.
1651 Request --write-conf, or allow starting with --old_conf.
1653 Severity : enhancement
1655 Description: Handle symlinks in the path when checking if Lustre is mounted.
1656 Details : Resolve intermediate symlinks when checking if a client has
1657 mounted a filesystem to avoid duplicate client mounts.
1662 Description: lconf can hit an error exception but still return success.
1663 Details : The lconf command catches the Command error exception at the top
1664 level script context and will exit with the associated exit
1665 status, but doesn't ensure that this exit status is non-zero.
1670 Description: failure of ptlrpc thread startup can cause oops
1671 Details : Starting a ptlrpc service thread can fail if there are a large
1672 number of threads or the server memory is very fragmented.
1673 Handle this without oopsing.
1676 Frequency : always, only if liblustre and non-default acceptor port was used
1678 Description: liblustre cannot connect to servers with non-default acceptor port
1679 Details : tcpnal_set_default_params() was not called and was therefore
1680 ignoring the environment varaible TCPNAL_PORT, as well as other
1681 TCPNAL_ environment variables
1686 Description: two objects could be created on the same OST for a single file
1687 Details : If an OST is down, in some cases it was possible to create two
1688 objects on a single OST for a single file. No problems other
1689 than potential performance impact and spurious error messages.
1693 Bugzilla : 5681, 9562
1694 Description: Client may oops in ll_unhash_aliases
1695 Details : Client dcache may become inconsistent in race condition.
1696 In some cases "getcwd" can fail if the current directory is
1702 Description: Inode refcounting problems in NFS export code
1703 Details : link_raw functions used to call d_instantiate without obtaining
1704 extra inode reference first.
1708 Bugzilla : 9942, 9903
1709 Description: Referencing freed requests leading to crash, memleaks with NFS.
1710 Details : We used to require that call to ll_revalidate_it was always
1711 followed by ll_lookup_it. Also with revalidate_special() it is
1712 possible to call ll_revalidate_it() twice for the same dentry
1713 even if first occurence returned success. This fix changes semantic
1714 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
1715 reference on a request referred from the intent.
1716 ll_intent_release() then releases such a request.
1719 Frequency : rare, normally benchmark loads only
1721 Description: unlinked inodes were kept in memory on the client
1722 Details : If a client is repeatedly creating and unlinking files it
1723 can accumulate a lot of stale inodes in the inode slab cache.
1724 If there is no other client load running this can cause the
1725 client node to run out of memory. Instead flush old inodes
1726 from client cache that have the same inode number as a new inode.
1729 Frequency : SLES9 2.6.5 kernel and long filenames only
1730 Bugzilla : 9969, 10379
1731 Description: utime reports stale NFS file handle
1732 Details : SLES9 uses out-of-dentry names in some cases, which confused
1733 the lustre dentry revalidation. Change it to always use the
1737 Frequency : rare, unless heavy write-truncate concurrency is continuous
1738 Bugzilla : 4180, 6984, 7171, 9963, 9331
1739 Description: OST becomes very slow and/or deadlocked during object unlink
1740 Details : filter_destroy() was holding onto the parent directory lock
1741 while truncating+unlinking objects. For very large objects this
1742 may block other threads for a long time and slow overall OST
1743 responsiveness. It may also be possible to get a lock ordering
1744 deadlock in this case, or run out of journal credits because of
1745 the combined truncate+unlink. Solution is to do object truncate
1746 first in one transaction without parent lock, and then do the
1747 final unlink in a new transaction with the parent lock. This
1748 reduces the lock hold time dramatically.
1751 Frequency : rare, 2.4 kernels only
1753 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
1754 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
1755 before unmounting their filesystem in order to avoid deadlock.
1756 The kernel_locked() function in 2.4 kernels only checks whether
1757 the kernel lock is held, not whether it is this process that is
1758 holding it as 2.6 kernels do.
1763 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
1764 Details : The client ptlrpc code may be trying to reconnect to a down
1765 server before a previous connection attempt has timed out.
1766 Increase the reconnect interval to be longer than the connection
1767 timeout interval to avoid sending duplicate connections to
1771 Frequency : echo_client brw_test command
1773 Description: fix echo_client to work with OST preallocated code
1774 Details : OST preallocation code (5137) didn't take echo_client IO path
1775 into account: echo_client calls filter methods outside of any
1776 OST thread and, hence, there is no per-thread preallocated
1777 pages and buffers to use. Solution: hijack pga pages for IO. As
1778 a byproduct, this avoids unnecessary data copying.
1782 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
1783 Description: Client can oops in mdc_commit_close() after open replay
1784 Details : It was possible for the MDS to return an open request with no
1785 transaction number in mds_finish_transno() if the client was
1786 evicted, but without actually returning an error. Clients
1787 would later try to replay that open and may trip an assertion
1788 Simplify the client close codepath, and always return an error
1789 from the MDS in case the open is not successful.
1792 Frequency : rare, 2.6 OSTs only
1794 Description: OST may deadlock under high load on fragmented files
1795 Details : If there was a heavy load and highly-fragmented OST filesystems
1796 it was possible to have all the OST threads deadlock waiting on
1797 allocation of biovecs, because the biovecs were not released
1798 until the entire RPC IO was completed. Instead, release biovecs
1799 as soon as they are complete to ensure forward IO progress.
1801 Severity : enhancement
1803 Description: Support for specifying external journal device at mount
1804 Details : If an OST or MDS device is formatted with an external journal
1805 device, this device major/minor is stored in the ext3 superblock
1806 and may not be valid for failover. Allow detecting and
1807 specifying the external journal at mount time.
1812 Description: Mounting an MDS with pending unlinked files may cause oops
1813 Details : target_finish_recovery() calls mds_postrecov() which returned
1814 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
1815 considers this an error and immediately begins cleaning up the
1816 lov, just after starting the mds_lov process
1818 Severity : enhancement
1820 Description: Implement 'lfs df' to report actual free space on per-OST basis
1821 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
1822 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
1823 usage of objects; '-h' to report in human readable format.
1825 ------------------------------------------------------------------------------
1827 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
1834 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
1835 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
1836 at startup during orphan unlinking. Instead, skip buddy update
1837 before setup, as it will be regenerated after recovery is complete.
1842 Description: If an OST is inactive, its locks might reference stale inodes.
1843 Details : lov_change_cbdata() must iterate over all namespaces, even if
1844 they are inactive to clear inode references from the lock.
1846 Severity : enhancement
1847 Frequency : occasional, if non-standard max_dirty_mb used
1849 Description: Client will block write RPCs if not enough grant
1850 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
1851 then the client will block writes while waiting for another RPC
1852 to complete instead of consuming its dirty limit. With change
1853 we get improved performance when max_dirty_mb is small.
1855 Severity : enhancement
1856 Bugzilla : 3389, 6253
1857 Description: Add support for supplementary groups on the MDS.
1858 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
1859 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
1860 lookups for user supplementary groups into a cache.
1864 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
1865 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
1866 for liblustre O_EXCL is enforced by MDS.
1870 Description: readdir never returns NULL in liblustre.
1871 Details : Corrected llu_iop_getdirentries logic, to return offset of next
1872 dentry in struct dirent.
1876 Frequency : liblustre only, depends on application IO pattern
1877 Description: liblustre clients evicted if not contacting servers
1878 Details : Don't put liblustre clients into the ping_evictor list, so
1879 they will not be evicted by the pinger ever.
1881 Severity : enhancement
1883 Description: Add ability to evict clients by NID from MDS.
1884 Details : By echoing "nid:$NID" string into
1885 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
1886 $NID would be instantly evicted from this MDS and from all active
1887 OSTs connected to it.
1891 Description: Do not query file size twice, somewhat slowing stat(2) calls.
1892 Details : lookup_it_finish() used to query file size from OSTs that was not
1897 Description: service threads change working directory to that of init
1898 Details : Starting lustre service threads may pin the working directory
1899 of the parent thread, making that filesystem busy. Threads
1900 now change to the working directory of init to avoid this.
1904 Frequency : during shutdown only
1905 Description: shutdown with a failed MDS or OST can cause unmount to hang
1906 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
1911 Frequency : 2.6 only
1912 Description: chmod/chown may include an extra supplementary group
1913 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
1914 supplementary group and if none is specified this is used.
1917 Bugzilla : 5479 (6816)
1919 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
1920 Details : If lookup is in progress on a file that is unlinked we might try
1921 to revalidate the inode and fail in revalidate after lookup is
1922 complete and ll_file_open() enqueues the open again but
1923 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
1926 Frequency : always, if lconf --abort_recovery used
1928 Description: lconf --abort_recovery fails with 'Operation not supported'
1929 Details : lconf was attempting to abort recovery on the MDT device and not
1932 ------------------------------------------------------------------------------
1934 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
1939 Frequency : rare (only unsupported configurations with a node running as an
1941 Bugzilla : 6514, 5137
1942 Description: Mounting a Lustre file system on a node running as an OST could
1944 Details : OSTs now preallocates memory needed to write out data at
1945 startup, instead of when needed, to avoid having to
1946 allocate memory in possibly low memory situations.
1947 Specifically, if the file system is mounted on on OST,
1948 memory pressure could force it to try to write out data,
1949 which it needed to allocate memory to do. Due to the low
1950 memory, it would be unable to do so and the node would
1951 become unresponsive.
1953 Severity : enhancement
1955 Description: Addition of lconf --service command line option
1956 Details : lconf now accepts a '--service <arg>' option, which is
1957 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
1959 Severity : enhancement
1961 Description: Failover mode is now the default for OSTs.
1962 Details : By default, OSTs will now run in failover mode. To return to
1963 the old behaviour, add '--failout' to the lmc line for OSTs.
1965 Severity : enhancement
1967 Description: Health checks are now provided for MDS and OSTs
1968 Details : Additional detailed health check information on MSD and OSTs
1969 is now provided through the procfs health_check value.
1972 Frequency : occasional, depends on IO load
1974 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
1975 after numerous create/delete cycles
1976 Details : The ext3 inode allocation policy would not allocate new inodes
1977 very well on the OSTs because there are no new directories
1978 being created. Instead we look for groups with free space if
1979 the parent directories are nearly full.
1984 Description: Network or server problems during mount may cause partially
1985 mounted clients instead of returning an error.
1986 Details : The config llog parsing code may overwrite the error return
1987 code during mount error handling, returning success instead
1993 Description: MDS can fail to allocate large reply buffers
1994 Details : After long uptimes the MDS can fail to allocate large reply
1995 buffers (e.g. zconf client mount config records) due to memory
1996 fragmentation or consumption by the buffer cache. Preallocate
1997 some large reply buffers so that these replies can be sent even
1998 under memory pressure.
2002 Frequency : rare (liblustre)
2003 Description: fsx running with liblustre complained that using truncate() to
2004 extend the file doesn't work. This patch corrects that issue.
2005 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2006 ATTR_SIZE and lsm use in llu_setattr_raw.
2010 Frequency : rare, only 2.6 kernels
2011 Description: Unusual file access patterns on the MDS may result in inode
2012 data being lost in very rare circumstances.
2013 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2014 code in the 2.6 kernel caused the inode and/or EA data not to
2015 be read from disk, causing single-file corruption.
2019 Frequency : rare, only 2.6 filesystems using extents
2020 Description: Heavy concurrent write and delete load may cause data corruption.
2021 Details : It was possible under high-load situations to have an extent
2022 metadata block in the block device cache from a just-unlinked
2023 file overwrite a newly-allocated data block. We now unmap any
2024 metadata buffers that alias just-allocated data blocks.
2028 Frequency : filesystems with default stripe_count larger than 77
2029 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2030 Details : lconf specifies an inode size of 4096 bytes when the default
2031 stripe_count is larger than 77. This conflicts with the default
2032 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2033 this case to avoid pinning too much memory for large EAs.
2035 ------------------------------------------------------------------------------
2037 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2042 Frequency : rare (extremely heavy IO load with hundreds of clients)
2044 Description: Client is evicted, gets IO error writing to file
2045 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2046 caused clients to be evicted for AST timeouts. The fixes in
2047 bug 5192 mean we no longer need to have such short AST timeouts
2048 so ldlm_timeout has been increased.
2051 Frequency : occasional during --force or --failover shutdown under load
2052 Bugzilla : 5949, 4834
2053 Description: Server oops/LBUG if stopped with --force or --failover under load
2054 Details : a collection of import/export refcount and cleanup ordering
2055 issues fixed for safer force cleanup
2058 Frequency : only filesystems larger than 120 OSTs
2059 Bugzilla : 5990, 6223
2060 Description: lfs getstripe would oops on a very large filesystem
2061 Details : lov_getconfig used kfree on vmalloc'd memory
2064 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2065 Bugzilla : 6242, 6243
2066 Description: reading from files that had been truncated to a non-zero size
2067 but never opened returned no data
2068 Details : ll_file_read() reads zeros from no-object files to EOF
2073 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2074 Details : There's little harm in aborting MDS/OSS recovery and letting it
2075 try again, so I removed the LASSERT and return an error instead.
2077 Severity : enhancement
2079 Description: New debugging infrastructure for tracking down data corruption
2080 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2081 (b) cover more of the client-side code path, and (c) try to narrow
2082 down where problems occurred
2086 Bugzilla : 3819, 4364, 4397, 6313
2087 Description: Racing close and eviction MDS could cause assertion in mds_close
2088 Details : It was possible to get multiple mfd references during close and
2089 client eviction, leading to one thread referencing a freed mfd.
2091 Severity: : enhancement
2092 Bugzilla : 3262, 6359
2093 Description: Attempts to reconnect to servers are now more aggressive.
2094 Details : This builds on the enhanced upcall-less recovery that was added
2095 in 1.4.2. When trying to reconnect to servers, clients will
2096 now try each server in the failover group every 10 seconds. By
2097 default, clients would previously try one server every 25 seconds.
2102 Description: After recovery, certain operations trigger a failed
2103 assertion on a client.
2104 Details : Failing over an mds, using lconf -d --failover, while a
2105 client was doing a readdir() call would cause the client to
2106 LBUG after recovery completed and the readdir() was resent.
2108 Severity : enhancement
2110 Description: Default groups are now added by lconf
2111 Details : You can now run lconf --group <servicename> without having to
2112 manually add groups with lmc.
2115 Frequency : occasional
2117 Description: Nodes with an elan id of 0 trigger a failed assertion
2120 Frequency : always when accessing e.g. tty/console device nodes
2122 Description: tty and some other devices nodes cannot be used on lustre
2123 Details : file's private_data field is used by device data and lustre
2124 values in there got lost. New field was added to struct file to
2125 store fs-specific private data.
2128 Frequency : when exporting Lustre via NFS
2130 Description: NFSD failed occasionally when looking up a path component
2131 Details : NFSD is looking up ".." which was broken in ext3 directories
2132 that had grown large enough to become hashed.
2135 Frequency : Clusters with multiple interfaces not on the same subnet
2137 Description: Nodes will repeatedly try to reconnect to an interface which it
2138 cannot reach and report an error to the log.
2139 Details : Extra peer list entries will be created by lconf with some peers
2140 unreachable. lconf now validates the peer before adding it.
2143 Frequency : Only if a default stripe is set on the filesystem root.
2145 Description: Setting a default stripe on the filesystem root prevented the
2146 filesystem from being remounted.
2147 Details : The client was sending extra request flags in the root getattr
2148 request and did not allocate a reply buffer for the dir EA.
2151 Frequency : occasional, higher if lots of files are accessed by one client
2152 Bugzilla : 6159, 6097
2153 Description: Client trips assertion regarding lsm mismatch/magic
2154 Details : While revalidating inodes the VFS looks up inodes with ifind()
2155 and in rare cases can find an inode that is being freed.
2156 The ll_test_inode() code will free the lsm during ifind()
2157 when it finds an existing inode and then the VFS later attaches
2158 this free lsm to a new inode.
2162 Bugzilla : 6422, 7030
2163 Description: MDS deadlock between mkdir and client eviction
2164 Details : Creating a new file via mkdir or mknod (starting a transaction
2165 and getting the ns lock) can deadlock with client eviction
2166 (gets ns lock and trying to finish a synchronous transaction).
2169 Frequency : occasional
2170 Description: While starting a server, the fsfilt_ext3 module could not be
2172 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2173 kernels. Previously, lconf would still use the ext3 name
2174 when trying to load modules. Now, it will correctly use
2175 ext3 on 2.4 and ldiskfs on 2.6.
2177 Severity : enhancement
2178 Description: The default stripe count has been changed to 1
2179 Details : The interpretation of the default stripe count (0, to lfs
2180 or lmc) has been changed to mean striping across a single
2181 OST, rather than all available. For general usage we have
2182 found a stripe count of 1 or 2 works best.
2184 Severity : enhancement
2185 Description: Add support for compiling against Cray portals.
2186 Details : Conditional compiling for some areas that are different
2190 Frequency : occasional
2191 Bugzilla : 6409, 6834
2192 Description: Creating files with an explicit stripe count may lead to
2193 a failed assertion on the MDS
2194 Details : If some OSTs are full or unavailable, creating files may
2195 trigger a failed assertion on the MDS. Now, Lustre will
2196 try to use other servers or return an error to the
2200 Frequency : occasional
2202 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2203 caused lock timeout during readahead (since 1.4.2).
2204 Details : Processes doing readahead might match a lock that hasn't been
2205 granted yet if there are overlapping and conflicting lock
2206 requests. The readahead process waits on ungranted lock
2207 (original lock is CBPENDING), while OST waits for that process
2208 to cancel CBPENDING read lock and eventually evicts client.
2210 Severity : enhancement
2212 Description: Initial enabling of flock support for clients
2213 Details : Implements fcntl advisory locking and file status functions.
2214 This feature is provided as an optional mount flag (default
2215 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2216 locking are implemented yet, and those that are are not guaranteed
2217 to be completely correct in production environments.
2218 mount -t lustre -o [flock|noflock] ...
2221 Frequency : occasional
2223 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2224 assertion in the ext3 JBD (journaling) layer.
2225 Details : The b_committed_data struct is protected by the big kernel lock
2226 in 2.4 kernels, serializing journal_commit_transaction() and
2227 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2228 access to this struct. In 2.6 kernels there is finer grained
2229 locking to improve SMP performance of the JBD layer.
2233 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2235 -----------------------------------------------------------------------------
2237 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2239 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2240 versions of Lustre. Please update all servers and clients to
2241 version 1.4.2 or later at the same time. You must also run
2242 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2243 to update the configuration logs.
2245 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2246 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2247 - swab llog records of type '0' so we get proper header size/idx (5861)
2248 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2249 - fix rename of one directory over another leaking an inode (5953)
2250 - avoid SetPageDirty on 2.6 (5981)
2251 - don't re-add just-being-destroyed locks to the waiting list (5653)
2252 - when creating new directories, inherit the parent's custom
2253 striping settings if present parent (3048)
2254 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2255 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2256 - don't hold client locks on temporary worklist from l_lru (5666)
2257 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2258 - automatically evict dead clients (5921)
2259 - Update file size properly in create+truncate+fstat case (6196)
2260 - Do not unhash mountpoint dentries, do not allow removal of
2262 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2263 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2264 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2265 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2266 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2267 - Correctly handle reads of files with no objects (6243)
2268 - lctl recover will also mark a device active if deactivate used (5933)
2270 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2271 - allow --write-conf on an MDS with different nettype than client (5619)
2272 - don't write config llogs to MDS for mounts not from that MDS (5617)
2273 - lconf should create multiple TCP connections from a client (5201)
2274 - init scripts are now turned off by default; run chkconfig --on
2275 lustre and chkconfig --on lustrefs to use them
2276 - upcalls are no longer needed for clients to recover to failover
2278 - add --abort-recovery option to lconf to abort recovery on device
2280 - add support for an arbitrary number of OSTs (3026)
2281 - Quota support protocol changes.
2282 - forward compatibility changes to wire structs (6007)
2283 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2284 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2285 - improved "lustre status" script
2286 - initialize blocksize for non-regular files (6062)
2287 - added --disable-server and --disable-client configure options (5782)
2288 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2289 - Vanilla 2.4.29 support
2290 - increase maximum number of obd devices to 520 (6242)
2291 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2292 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2293 SLES 9 2.6.5 kernels
2294 - limit stripes per file to 160 (the maximum EA size) (6093)
2296 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2299 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2300 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2301 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2302 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2303 - small ext3 extents cleanups and fixes (5733)
2304 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2305 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2306 - avoid needless client->OST connect, fix handle mismatch (5317)
2307 - fix DLM error path that led to out-of-sync client, long delays (5779)
2308 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2309 - fix several locking issues related to i_size (5492,5624,5654,5672)
2310 - don't move pending lock onto export if it is already evicted (5683)
2311 - fix kernel oops when creating .foo in unlinked directory (5548)
2312 - fix deadlock in obdfilter statistics vs. object create (5811)
2313 - use time_{before,after} to avoid timer jiffies wrap (5882)
2314 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2315 - Do not leak request if server was not able to process it (5154)
2316 - If mds_open unable to find parent dir, make that negative lookup(5154)
2317 - don't create new directories with extent-mapping (5909, 5936)
2319 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2320 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2321 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2322 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2323 - support for RHEL4 kernel on i686 (5773)
2324 - provide error messages when incompatible logs are encountered (5898)
2326 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2327 * version 1.4.0.10 (1.4.1 release candidate 1)
2329 - don't keep a lock reference when lock is not granted (4238)
2330 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2331 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2332 - drop import inflight refcount on signal_completed_replay error (5255)
2333 - unlock page after async write error during send (3677)
2334 - handle missing objects in filter_preprw_read properly (5265)
2335 - no transno return for symlink open, don't save no-trasno open (3440)
2336 - don't try to complete elan receive that already failed (4012)
2337 - free RPC server reply state on error (5406)
2338 - clean up thread from ptlrpc_start_thread() on error (5160)
2339 - readahead could read extra page into cache that wasn't ejected (5388)
2340 - prevent races in class_attach/setup/cleanup/detach (5260)
2341 - don't dereference de->d_inode after l_dput of de (5458)
2342 - use "int" for stripe value returned from lock_to_stripe (5544)
2343 - mballoc allocation and error-checking fixes in 2.6 (5504)
2344 - block device patches to fix I/O request sizes in 2.6 (5482)
2345 - look up hostnames for IB nals (5602)
2346 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2347 - don't start multiple acceptors for the same port (5277)
2348 - fix incorrect LASSERT in mds_getattr_name (5635)
2349 - export a proc file for general "ping" checking (5628)
2350 - fix "lfs check" to not block when the MDS is down (5628)
2352 - service request history (4965)
2353 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2354 - create an "evict_client" /proc entry on OSTs, like the MDS has
2355 - fix mount usage message, return errors per mount(8) (5168)
2356 - change grep [] to grep "[]" in tests so they work in more UMLs
2357 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2358 - remove ancient LOV_MAGIC_V0 EA support (5047)
2359 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2360 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2361 - allow readahead allocations to fail when low on memory (5383)
2362 - mmap locking landed again, after considerable improvement (2828)
2363 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2365 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2368 - send OST transaction number in read/write reply to free req (4966)
2369 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2370 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2371 - the watchdog thread now runs as interruptible (5246)
2372 - flock/lockf fixes (but it's still disabled, pending 5135)
2373 - don't use EXT3 constants in llite code (5094)
2374 - memory shortage at startup could cause assertion (5176)
2376 - reorganization of lov code
2377 - single portals codebase
2379 - add extents/mballoc support (5025)
2380 - direct I/O reads in the obdfilter (4048)
2381 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2383 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2386 - send OST transaction number in read/write reply to free req (4966)
2387 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2388 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2389 - the watchdog thread now runs as interruptible (5246)
2390 - handle missing objects in filter_preprw_read properly (5265)
2391 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2392 - drop import inflight refcount on signal_completed_replay error (5255)
2393 - unlock page after async write error during send (3677)
2394 - return original error code on reconstructed replies (3761)
2395 - no transno return for symlink open, don't save no-trasno open (3440)
2397 - add pid to ldlm debugging output (4922)
2398 - bump the watchdog timeouts -- we can't handle 30sec yet
2399 - extra debugging for orphan dentry/inode bug (5259)
2401 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2404 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2405 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2406 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2407 - fix NULL dereference in /proc/sys/portals/routes (4827)
2408 - allow failed mdc_close() operations to be interrupted (4561)
2409 - stop precreate on OST before MDS would time out on it (4778)
2410 - don't send partial-page writes before EOF from client (4410)
2411 - discard client grant for sub-page writes on large-page clients (4520)
2412 - don't free dentries not owned by NFS code, check generation (4806)
2413 - fix lsm leak if mds_create_objects() fails (4801)
2414 - limit debug_daemon file size, always print CERROR messages (4789)
2415 - use transno after validating reply (3892)
2416 - process timed out requests if import state changes (3754)
2417 - update mtime on OST during writes, return in glimpse (4829)
2418 - add mkfsoptions to LDAP (4679)
2419 - use ->max_readahead method instead of zapping global ra (5039)
2420 - don't interrupt __l_wait_event() during strace
2422 - add software watchdogs to catch hung threads quickly (4941)
2423 - make lustrefs init script start after nfs is mounted
2424 - fix CWARN/ERROR duplication (4930)
2425 - return async write errors to application if possible (2248)
2426 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2427 - print NAL number in %x format (4645)
2428 - update barely-supported suse-2.4.21-171 series (4842)
2429 - support for sles 9 %post scripts
2430 - support for building 2.6 kernel-source packages
2431 - support for sles km_* packages
2433 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2436 - ignore -ENOENT errors in osc_destroy (3639)
2437 - notify osc create thread that OSC is being cleaned up (4600)
2438 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2439 - reconstruct ost_handle() like mds_handle() (4657)
2440 - create a new thread to do import eviction to avoid deadlock (3969)
2441 - let lconf resolve symlinked-to devices (4629)
2442 - don't unlink "objects" from directory with default EA (4554)
2443 - hold socknal file ref over connect in case target is down (4394)
2444 - allow more than 32000 subdirectories in a single directory (3244)
2445 - fix blocks count for O_DIRECT writes (3751)
2446 - OST returns ENOSPC from object create when no space left (4539)
2447 - don't send truncate RPC if file size isn't changing (4410)
2448 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2449 - bind to privileged port in socknal and tcpnal (3689)
2451 - rate limit CERROR/CWARN console message to avoid overload (4519)
2452 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2453 - basic mmap support (3918)
2454 - kernel patch series update from b1_4 (4711)
2456 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2459 - avoid crash during MDS cleanup with OST shut down (2775)
2460 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2461 - don't use bad inodes on the MDS (3744)
2462 - dynamic object preallocation to improve recovery speed (4236)
2463 - don't hold spinlock over lock dumping or change debug flags (4401)
2464 - don't zero obd_dev when it is force cleaned (3651)
2465 - print grants to console if they go negative (4431)
2466 - "lctl deactivate" will stop automatic recovery attempts (3406)
2467 - look for existing locks in ldlm_handle_enqueue() (3764)
2468 - don't resolve lock handle twice in recovery avoiding race (4401)
2469 - revalidate should check working dir is a directory (4134)
2471 - don't always mark "slow" obdfilter messages as errors (4418)
2473 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2476 - don't close LustreDB during write_conf until it is done (3860)
2477 - fix typo in lconf for_each_profile (3821)
2478 - allow dumping logs from multiple threads at one time (3820)
2479 - don't allow multiple threads in OSC recovery (3812)
2480 - fix debug_size parameters (3864)
2481 - fix mds_postrecov to initialize import for llog ctxt (3121)
2482 - replace config semaphore with spinlock (3306)
2483 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2484 - don't allow enqueue to complete on a destroyed export (3822)
2485 - down write_lock before checking llog header bitmap (3825)
2486 - recover from lock replay timeout (3764)
2487 - up llog sem before sending rpc (3652)
2488 - reduce ns lock hold times when setting kms (3267)
2489 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2490 - fix NULL deref and obd_dev leak on setup error (3312)
2491 - replace some LBUG about llog ops with error handling (3841)
2492 - don't match INVALID dentries from d_lookup and spin (3784)
2493 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2494 - fix invalid assertion in ptlrpc_set_wait (3880)
2496 - add libwrap support for the TCP acceptor (3996)
2497 - add /proc/sys/portals/routes for non-root route listing (3994)
2498 - allow setting MDS UUID in .xml (2580)
2499 - print the stack of a process that LBUGs (4228)
2501 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2504 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2505 - make sure to unset replay flag from failed open requests (3440)
2506 - if default stripe count is 0, use OST count for inode size (3636)
2507 - update parent mtime/ctime on client for create/unlink (2611)
2508 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2509 - free recovery state on server during a forced cleanup (3571)
2510 - unregister_reply for resent reqs (3063)
2511 - loop back devices mounting and status check on 2.6 (3563)
2512 - fix resource-creation race that can provoke i_size == 0 (3513)
2513 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2514 - more debugging for page-accounting assertion (3746)
2515 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2516 - avoid deadlock after precreation failure (3758)
2517 - fix race and lock order deadlock in orphan handling (3450, 3750)
2518 - add validity checks when grabbing inodes from l_ast_data (3599)
2520 - add /proc/.../recovery_status to obdfilter (3428)
2521 - lightweight CDEBUG infrastructure, debug daemon (3668)
2522 - change default OSC RPC parameters to be better on small clusters
2523 - turn off OST read cache for files smaller than 32MB
2524 - install man pages and include them in rpms (3100)
2525 - add new init script for (un)mounting lustre filesystems (2593)
2526 - run chkconfig in %post for init scripts (3701)
2527 - drop scimac NAL (unmaintained)
2529 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
2532 - clean kiobufs before and after use (3485)
2533 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
2534 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
2535 - change init script to fail more clearly if not run as root (1528)
2536 - allow clients to reconnect during replay (1742)
2537 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
2538 - don't do DNS lookups on NIDs too small for IP addresses (3442)
2539 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
2540 - fix cond_resched (3554)
2541 - only evict unfinished clients after recovery (3515)
2542 - allow bulk resend, prevent data loss (3570)
2543 - dynamic ptlrpc request buffer allocation (2102)
2544 - don't allow unlinking open directory if it isn't empty (2904)
2545 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
2546 - remove extraneous obd dereference causing LASSERT failure (3334)
2547 - don't use get_cycles() when creating temp. files on the mds (3156)
2548 - hold i_sem when setting i_size in ll_extent_lock() (3564)
2549 - handle EEXIST for set-stripe, set proper directory name (3336)
2551 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
2552 - fix ksocknal_fmb_callback() error messages (2918)
2554 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
2557 - don't copy lvb into (possibly NULL) reply on error (2983)
2558 - don't deref dentry after dput, don't free lvb on error (2922)
2559 - use the kms to determine writeback rpc length (2947)
2560 - increment oti_logcookies when osc is inactive (2948)
2561 - update client's i_blocks count via lvb messages (2543)
2562 - handle intent open/close of special files properly (1557)
2563 - mount MDS with errors=remount-ro, like obdfilter (2009)
2564 - initialize lock handle to avoid ASSERT on error cleanup (3057)
2565 - don't use cancelling-locks' kms values (2947)
2566 - use highest lock extent for kms, not last one (2925)
2567 - don't dereference ERR_PTR() dentry in error handling path (3107)
2568 - fix thread race in portals_debug_dumplog() (3122)
2569 - create lprocfs device entries at setup instead of at attach (1519)
2570 - common AST error handler, don't evict client on completion race (3145)
2571 - zero nameidata in detach_mnt in 2.6 (3118)
2572 - verify d_inode after revalidate_special is valid in 2.6 (3116)
2573 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
2574 - initialize RPC timeout timer earlier for 2.6 (3219)
2575 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
2576 - print nal/nid for unknown nid (3258)
2577 - additional checks for oscc recovery before doing precreate (3284)
2578 - fix ll_extent_lock() error return code for 64-bit systems (3043)
2579 - don't crash in mdc_close for bad permissions on open (3285)
2580 - zero i_rdev for non-device files (3147)
2581 - clear page->private before handing to FS, better assertion (3119)
2582 - tune the read pipeline (3236)
2583 - fix incorrect decref of invalidated dentry (2350)
2584 - provide read-ahead stats and refine rpc in flight stats (3328)
2585 - don't hold journal transaction open across create RPC (3313)
2586 - update atime on MDS at close time (3265)
2587 - close LDAP connection when recovering to avoid server load (3315)
2588 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
2589 - don't leak open file on MDS after open resend (3325)
2590 - serialize filter_precreate and filter_destroy_precreated (3329)
2591 - loop device shouldn't call sync_dev() for nul device (3092)
2592 - clear page cache after eviction (2766)
2593 - resynchronize MDS->OST in background (2824)
2594 - refuse to mount the same filesystem twice on same mountpoint (3394)
2595 - allow llmount to create routes for mounting behind routers (3320)
2596 - push lock cancellation to blocking thread for glimpse ASTs (3409)
2597 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
2598 - fix rare problem with rename on htree directories (3417)
2600 - allow default OST striping configuration per directory (1414)
2601 - fix compilation for qswnal for 2.6 kernels (3125)
2602 - increase maximum number of MDS request buffers for large systems
2603 - change liblustreapi to be useful for external progs like lfsck (3098)
2604 - increase local configuration timeout for slow disks (3353)
2605 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
2607 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2610 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
2611 - don't overwrite extent policy data in reply if lock was blocked (2901)
2612 - drop filter export grants atomically with removal from device (2663)
2613 - del obd_self_export from work_list in class_disconnect_exports (2908)
2614 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
2615 - swab reply message in mdc_close, other PPC fixes (2464)
2616 - fix destroying of named logs (2325)
2617 - overwrite old logs when running lconf --write_conf (2264)
2618 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
2619 - fix race in target_handle_connect (2898)
2620 - mds_reint_create() should take same inode create lock (2926)
2621 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
2622 - don't close files for self_export to avoid uninitialized obd (2936)
2623 - allow MDS with the same name as client node (2939)
2624 - hold dentry reference for closed log files for unlink (2325)
2625 - reserve space for all logs during transactions (2059)
2626 - don't evict page beyond end of stripe extent (2925)
2627 - don't oops on a deleted current working directory (2399)
2628 - handle hard links to targets without a parent properly (2517)
2629 - don't dereference NULL lock when racing during eviction (2867)
2630 - don't grow lock extents when lots of conflicting locks (2919)
2632 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
2635 - account for cache space usage on clients to avoid data loss (974)
2636 - lfsck support in lustre kernel code (2349)
2637 - reduce journal credits needed for BRW writes (2370)
2638 - orphan handling to avoid losing space on client/server crashes
2639 - ptlrpcd can be blocked, stopping ALL progress (2477)
2640 - use lock value blocks to assist in proper KMS, faster stat (1021)
2641 - takes i_sem instead of DLM locks internally on obdfilter (2720)
2642 - recovery for initial connections (2355)
2643 - fixes for mds_cleanup_orphans (1934)
2644 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
2645 - block all file creations until orphan recovery completes (1901)
2646 - client remove rq_connection from request struct (2423)
2647 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
2648 - recovery timer race (2670)
2649 - mdc_close recovey bug (2532)
2650 - ptlrpc cleanup bug (2710)
2651 - mds timeout on local locks (2588)
2652 - namespace lock held during RPCs (2431)
2653 - handle interrupted sync write properly (2503)
2654 - don't try to handle a message that hasn't been replied to (2699)
2655 - client assert failure during cleanup after abort recovery (2701)
2656 - leak mdc device after failed mount (2712)
2657 - ptlrpc_check_set allows timedout requests to complete (2714)
2658 - wait for inflight reqs when ptlrpcd finishes (2710)
2659 - make sure unregistered services are removed from the srv_list
2660 - reset bulk XID's when resending them (caught by 1138 test)
2661 - unregister_bulk after timeout
2662 - fix lconf error (2694)
2663 - handle write after unfinished setstripe, stripe-only getstripe (2388)
2664 - readahead locks pages, leaves pending causing memory pressure (2673)
2665 - increase OST request buffers to 4096 on large machines (2729)
2666 - fix up permission of existing directories in simple_mkdir (2661)
2667 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
2668 - don't assign transno to errored transactions (2742)
2669 - don't delete objects on OST if given a bogus objid from MDS (2751)
2670 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
2671 - if rq_no_resend, then timeout request after recovery (2432)
2672 - fix MDS llog_logid record size, 64-bit array alignment (2733)
2673 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
2674 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
2675 - MDS recovery shouldn't delete objects that it has given out (2730)
2676 - if enqueue arrives after completion, don't clobber LVB (2819)
2677 - don't unlock pages twice when trigger_group_io returns error (2814)
2678 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
2679 - don't write pages to disk if there was an error (1450)
2680 - don't ping imports that have recovery disabled (2676)
2681 - take buffered bytes into account when balancing socknal conn (2817)
2682 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
2683 - reconnect unlink llog connection after MDS reconnects to OST (2816)
2684 - remove little-endian swabbing of llog records (1987)
2685 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
2686 - retry reposting request buffers if they fail (1191)
2687 - grow extent at grant time to avoid granting a revoked lock (2809)
2688 - lock revoke doesn't evict page if covered by a second lock (2765)
2689 - disable VM readahead to avoid reading outside lock extents (2805)
2691 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
2692 - updated kernel patches for hp-2.4.20 kernel (2681)
2694 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
2697 - fix truncated write corruption (2366)
2698 - fix for failed assertion in iopen_connect_dentry (1792,2517)
2700 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
2701 - ensure len(uuid) < 37 in lmc (1171)
2702 - fix ia64 OOPS in llog_test (2255)
2703 - zero end of page at obdfilter for partial page writes (2648)
2704 - don't leave stale dentries around after renames (bug 2428)
2705 - fix timeouts when evicting a client with a single lock held (2642)
2706 - set deadline for the initial HELLO message to drain (2634)
2707 - print out dotted-quad IP addresses in the socknal (2302)
2709 - additional debugging for MDS client eviction problem (2443)
2710 - fix mkfsoptions support for osts (2603, 2604)
2712 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
2715 - add series for the vanilla 2.6.0 kernel
2716 - add series for the vanilla 2.4.24 kernel
2717 - add series for a cray x86/64 UL kernel drop
2718 - fix xattr patches for the vanilla 2.4.19 series
2720 - generate true UUIDs in lmc (1171)
2721 - have portals stack dumping break in UML (2466)
2722 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
2723 - allocate with _NOFS in ldlm to avoid deadlock (1933)
2724 - wake callback waiting threads on client eviction (2460)
2725 - Add --ptldebug and --subsystem to lmc (1719)
2726 - update assertion to allow safe interrupt allocation
2727 - set rq_no_resend for cancel requests (2432)
2728 - recalculate ptlrpcd timeout after resend (2494)
2729 - call vfs_rmdir when removing pending directories (2368)
2730 - fix renaming a file to itself (2429)
2731 - lmc creates a default one-stripe lov (2454)
2732 - expand procfs space to handle large clusters (2326)
2733 - increase UML stack to avoid overflow
2734 - update lconf's list of debug and subsystem masks
2735 - fix lfs find --obd (2510)
2736 - /proc tunable for disabling filter read caching (2591)
2737 - stop rpm packages from altering slapd.conf (2301)
2738 - disable nagle in the socknal under 0conf (2578)
2739 - choose mds inode size based on stripe count (2572)
2740 - fix kernel-source rpm problems (2516)
2742 - add --disable-doc to avoid pdf generation (2421)
2743 - update documentation, tests, type-os, comments
2744 - avoid format warnings on ia64
2745 - remove the TOE NAL
2746 - tiny code cleanups by removing unused fields
2748 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
2751 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
2752 - remove the most blatant lies from BUILDING file (2371)
2753 - change default debug level to reasonable production setting
2754 - reduce client side cache size to reduce cache flush time
2755 - reduce max RPCs in flight to avoid unnecessary file fragmentation
2756 - make TCP zerocopy and pinger support enabled by default (2476)
2757 - sync writes completed after process exits caused crashes (2319)
2758 - maintain correct mount count on the MDS (2356)
2759 - backout 1557, because 2316 wasn't really fixed
2760 - better file I/O statistics gathering in /proc
2761 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
2762 - another kernel patch to fix zero-copy TCP function export
2763 - don't take duplicate lock when processing re-sent getattr (2420)
2764 - lctl uses obd_self_export instead of creating new conn (2353)
2765 - MDS/OST recovery case which requires object creation asserted (2425)
2766 - move lfs from /usr/sbin to /usr/bin in packages
2767 - fix race between mds_client_add and mds_client_free (2417)
2768 - use kmalloc instead of slabs in portals (2430)
2769 - don't create duplicate records when a failover MDS is present (2442)
2770 - remove unnecessary mount age check (2332)
2771 - don't remove directory inodes from locks prematurely (2451)
2772 - don't break if MDS service name is the same as hostname (2103)
2773 - fix races in client write RPC generation when cache full (2482)
2775 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
2778 - remove now-unused request->rq_obd (278)
2779 - if an allocation fails, print out how much memory we've used (1933)
2780 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
2781 - add the "configurable stack size" patch to most series files (1256)
2782 - ability to write large log records, for 100+ OST configs (2306)
2783 - fix NULL deref when filter_prep fails (2314)
2784 - fix operator precedence error in filter_sync
2785 - dynamic allocation of socknal TX descriptors (2315)
2786 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
2787 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
2788 - ensure that CWARN messages in Portals always get to the syslog
2789 - __init/__exit are not for prototype decls (ldlm_init/exit)
2790 - x86-64 compile warning fixes
2791 - fix gateway LMC keyword conflict (2318)
2792 - fix MDS lock inversions in getattr/reint paths (1844)
2793 - fix a rare lock re-ordering bug, which caused deadlock (2322)
2794 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
2795 - DLM race condition prevented some lock evictions (2328)
2796 - ENOMEM detection and retry on socknal sends (2230)
2797 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
2798 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
2799 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
2800 - better error messages when a client is rejected during recovery (1505)
2801 - avoid cancelling locks which were never granted, after failure (2330)
2802 - fix i_sem/journal inversion in mds_client_add (2333)
2803 - fix truncate/getattr lock cycle deadlock (2334)
2804 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
2805 - fix two rare exit paths which could leak an l_lock() ref (2321)
2806 - fencepost error in MDS/OST orphan recovery (2226)
2807 - make log record alignment 8 bytes (1988)
2808 - lstripe now fails when requested offset > ost_count (2237)
2809 - ensure that all kernel series have a complete list.h (1607)
2810 - fix crashes in special-file operations (2316)
2811 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
2813 - allow configurable automake binary, for testing new versions
2814 - small update to the lfs documentation
2816 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
2818 * fix negative export reference count in fsfilt_sync (2312)
2820 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
2821 * release candidate 0.9.1
2823 - orphans are moved into the PENDING directory for possible recovery
2824 - replayed opens now open by fid for orphan/rename safety (1042)
2825 - last close of an orphan inode generates a transno (683)
2826 - chdir() and mount() now pin the directory entry (1020)
2827 - avoid CERROR in normal ll_setattr_raw() error case (1500)
2828 - discard very old requests without processing them (1502)
2829 - remove some common, well-understood CERRORs (1505)
2830 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
2831 - clear "grant" flags in OST replies until OST grant code lands (1644)
2832 - fix read performance by not clobbering i_blksize on client (1598)
2833 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
2834 - make LRU size a /proc tunable, clears locks when reduced (707)
2835 - fix some lprocfs rot that prevented ptlbd from loading (1732)
2836 - server locks take references on exports now (1558)
2837 - build fixes for 2.4.20-rh trees (1663)
2838 - return an error from lov_create if all OSCs are inactive (1751)
2839 - fix import levels when a reconnect happens without a timeout (1597)
2840 - exit early from mds_open if we get a lookup error (1749)
2841 - partial page read at EOF wouldn't wait for disk before sending (1642)
2842 - avoid NULL deref in obdfilter when reading page past EOF (1592)
2843 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
2844 - fix LBUG in ll_it_open_error with rc = -2 (1861)
2845 - write/truncate lock inversion (1639)
2846 - Don't auto-load obdclass, portals modules during cleanup (1495)
2847 - fix timestamps from jumping to "now" (1763)
2848 - extra journal assertions (1648)
2849 - add an extra multiunlink test (1771)
2850 - fix read_record/write_record API (1776)
2851 - fix leak of offset_extent, possible incorrect i_size later (1772)
2852 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
2853 - Debugging for the kqswnal_get_idle_tx problems (1820)
2854 - Allow recovery to be attempted multiple times (1536)
2855 - Write out MDS last_rcvd file after it is first created (1600)
2856 - Fix tx_descriptor leak in failed transmit situations (1827)
2857 - ext3 journaling fixes for assertion failure after IO error (1871)
2858 - class_export_put() on freed export after completion AST error (1896)
2859 - Fix revalidate looping in VFS (1322)
2860 - Don't access a freed export during MDS_REINT timeout (1521)
2861 - Add open-unlink recovery support on the MDS (1673,1764)
2862 - Return an error if no MDS data was read from last_rcvd (1946)
2863 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
2864 - Don't setup a disk device that doesn't match exported UUID (317)
2865 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
2866 - avoid committing NULL handle in force close
2867 - local.sh is now a one-stripe LOV configuration
2868 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
2869 - fix ext3 htree duplicate directory entry corruption (1516)
2870 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
2871 - update to new LOV EA format (2097)
2872 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
2874 2003-06-15 Phil Schwan <phil@clusterfs.com>
2877 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
2878 - per-import/export recovery handling (958, 931, 959)
2879 - multiple last-rcvd slots, for serving multiple FSes (949)
2880 - connections are again shared between multiple imp/exports (963, 964)
2881 - "umount -f" would hang if any requests needed to be sent (393, 978)
2882 - avoid pinning large req buffer by copying for queued messages (989)
2883 - add "uuid" to "lctl device" command to help upcalls (991)
2884 - "open" RPCs with transnos would confuse recovery counters (1037)
2885 - do proper endian conversion of all wire messages (288, 340, 891)
2886 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
2887 - call waiting locks callback from LDLM recovery thread (1127, 1151)
2888 - fix ptlrpc_connection leak in target_handle_connect (1174)
2889 - fix import refcounting bug in OST and MDS cleanup (1134)
2890 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
2891 - fix very unlikely obd_types race condition (501)
2892 - remove osc_open hack for echo_client (1187)
2893 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
2894 - a failure in read_inode2 leads to deadlock (1139)
2895 - cancel ack-locks as soon as transaction is committed (1072)
2896 - fix major leaks and crashes in the bulk I/O path (937, 1057)
2897 - make sure to commitrw after any preprw to avoid deadlock (1162)
2898 - failing to execute a file in a lustre FS would lock inode (1203)
2899 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
2900 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
2901 - obd_uuid2tgt would walk off the end of the list (1255)
2902 - on IA64 the osc would give portals incorrect bulk size (1258)
2903 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
2904 - fix lock inversion caused by new llite matching code (1282)
2905 - limit the number of dirty pages on a client to 10MB (1286)
2906 - timed out locks were not being corrected cancelled (1289)
2907 - fix O_DIRECT above 4GB on IA-32 (1292)
2908 * major user-visible changes
2909 - fail out/fail over policy now controlled by the upcall (993)
2911 - add OBD_PING to check server availability and failure (954)
2912 - lustre messages are now sent in sending host order (288, 340, 891)
2913 - add eadatalen to MDS getattr reply (340)
2914 - OST read replies may contain second buffer, with per-page status (593)
2916 2003-03-11 Phil Schwan <phil@clusterfs.com>
2919 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
2920 - failed open()s could cause deadlock; fixed (867, 869)
2921 - stop cancelling OST locks when files are closed (481)
2922 - overlapping XID spaces caused network corruption (851, 853)
2923 - fix unsafe fsfilt counter arithmetic; change to atomic_t
2924 - setattr_raw added, to do single-RPC, server-side setattrs
2925 - lmc/lconf syntax change for OST UUIDs
2926 - fix crashy race condition between ptlrpc_free_req and osc_close
2927 - don't use request in mdc_enqueue if we hit a timeout (889)
2928 - don't set the inode i_size for regular files from the MDS (896)
2929 - handle out of order completion AST (842)
2930 - don't LBUG if a lock request times out after receiving AST (913)
2931 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
2932 - if a bad lock AST arrives, send an error instead of dropping entirely
2933 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
2934 - fix leak in bulk IO when only partially completed (899, 900, 926)
2935 - fix O_DIRECT for ia64 (55)
2936 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
2937 - C-z after timeout could hang a process forever; fixed (977)
2939 - client-side I/O cache (678, 924, 929, 941, 970)
2941 - READPAGE and SETATTRs which don't take server-side locks get
2944 2003-02-11 Phil Schwan <phil@clusterfs.com>
2947 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
2948 - Fix fsfilt_extN_readpage() to read a full page of directory
2949 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
2950 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
2951 - Fix imbalanced LOV object allocation and out-of-bound access (469)
2952 - Most intent operations were removed, in favour of a new RPC mode
2953 that does a single RPC to the server and bypasses most of the VFS
2954 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
2955 - Aggressively cancel local locks on DLM servers
2956 - mds_reint_unlink sends EA to the client if it's the last nlink.
2957 client uses that EA to unlink OST objects.
2958 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
2959 - recursive symlinks were fixed (439)
2960 - fixed NULL deref in DEBUG_REQ
2961 - filter_update_lastobjid no longer calls sync, which annoyed extN
2962 - fixed multi-client small-writes to a single file problem (445)
2963 - fixed mtime updates during file writes (607)
2964 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
2965 - fixed bug in obd_brw_read/write() (under guise of testing 367)
2966 - fixed Linux OST size reporting problem (444, 656)
2967 - OST now updates object mtime with writes or setattr (607, 619)
2968 - client verifies file size before zeroing page past EOF (445)
2969 - OST now writes last allocated objid to disk with allocation (108)
2970 - LOV on echo now works (409)
2972 - mds_reint_unlink sends a new buffer, with the EA included. this
2973 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
2974 set if a regular file was being unlinked, and it was the last link
2975 - use PtlGet from the target for bulk writes (315)
2976 - OST now updates object mtime with writes or setattr (607, 619)
2977 - LDLM now has a grant-time callback to revalidate locked items, if
2979 - Many MDS operations were reorganized to combat race conditions
2981 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
2982 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
2984 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
2987 - Fully reactivate OST imports after reconnection (512, others)
2988 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
2989 - More graceful error handling for truncating on dead OST (515)
2990 - Don't error out unless we're actually accessing dead stripes (474)
2991 - Fix garbage sizes when stripes are missing (410)
2992 - LRU counters were broken, causing constant lock purge (433, 432)
2993 - garbage on read from stripes with failed OSTs (441)
2994 - mark OSCs as active before reconnecting during recovery (438)
2995 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
2996 - lfind did not preserve OST order in output (443)
2997 - symlinks cause hung clients, incorrect data (439)
2998 - stop dereferencing request after dropping refcount (457)
2999 - don't LASSERT(spin_is_locked) on non-SMP (455)
3000 - fixes for many rename() bugs
3001 - fstat didn't correctly synchronize attributes (399)
3002 - server must handle lock cancellation during blocking AST prep (487)
3003 - bulk descriptors were free()d too soon (511)
3004 - fix paths in lconf, which would load incorrect modules (451, 507)
3005 - fix confusing lconf 'host not found' error message (386)
3006 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3007 - fix race condition in mdc_blocking_ast() for inode access (526)
3008 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3009 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3010 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3011 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3012 - fix lstripe in several ways (kernel side) (527)
3013 - fix request leak in ldlm_cli_enqueue (262)
3014 - incorrect OSC was marked inactive after OST failure
3015 - call mds_fs_cleanup before unmounting filesystem (524)
3016 - fix races between taking ns_lock and ldlm_lock_change_resource
3017 - fix races updating LOV export open file list
3018 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3019 - fix recovery NULL deref in ldlm_cli_cancel_unused
3020 - fix some DLM races by using new hash table for lock handles (419)
3021 - permit the client to specify desired inodes, at replay
3022 - duplicate requests when we queue them for replay reintegration
3023 - fix last_rcvd offset calculation
3024 - sync after each recovered transaction, so we always make progress
3025 - never, not always, ERESTART requests without transnos
3026 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3028 - skip replay if the MDS says that the client is already connected
3029 - don't check for a recovery-enabled export to match lctl's UUID
3030 - don't INC_USE_COUNT for phantom exports
3031 - don't crash when cleaning up phantom exports (567)
3032 - don't double-finish or set replay data for errored mdc_open requests
3033 - abort requests when they time out, so we don't get old replies
3034 - send/receive replies for AST messages again
3035 - if the client says that it doesn't have the lock, cancel it on the
3037 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3038 mark it as destroyed, it will all work out when decref is called
3039 - fix module use counts (22, 581)
3041 - ASTs now expect a reply (server cancels lock on error reply)
3043 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3046 - fix many simultaneous client startup (392)
3047 - fix dentry->d_it clobbering
3048 - credentials weren't being shipped for readdir/getattr operations
3049 - remove invalid assertions triggered during some concurrent MD
3051 - proper Lustre versions added (336, 389)
3052 - fix memory leak for create error case (398)
3053 - fix LOV locking bug that would get cli/srv out of sync
3054 - fix echo client over LOV (409)
3055 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3056 - fix double-O_EXCL intent crash (424)
3057 - avoid sending multiple lock CANCELs (352)
3059 - MDS can do multi-client recovery (modulo bugs in new code)
3061 - many updates, edits, cleanups
3063 2002-11-18 Phil Schwan <phil@clusterfs.com>
3066 - fix null d_it dereference (346)
3067 - fix full OST/dbench hang (333)
3068 - fix permission problem with file removal (286)
3069 - fix removal of OSCs from LOV when they fail
3070 - fix NULL deref during bulk timeout (214)
3071 - fix problems related to multiple filesystems on one MDS (241)
3072 - fixed serious subtle metadata locking bugs
3073 - free locks on clients when inodes are removed due to memory
3075 - fix inode pointer in lock data (285)
3076 - partial support for multiple MDS on a single host (241)
3077 - data locks weren't cancelled at clear_inode time (290, 311)
3078 - intent locks could lead to unbounded lock growth (205)
3079 - added a maximum lock count, an LRU list, and a flusher
3080 - fix multiple rename (365)
3081 - properly abstracted the echo client
3082 - OSC locked 1 byte too many; fixed
3083 - rewrote brw callback code:
3084 - fixed recovery bugs related to LOVs (306)
3085 - fixed too-many-pages-in-one-write crash (191)
3086 - fixed (again) crash in sync_io_timeout (214)
3087 - probably fixed callback-related race (385)
3089 - Add capability to MDS protocol
3090 - LDLM cancellations and callbacks on different portals
3092 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3095 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3096 - defer open object destruction to close time (601981, 138)
3097 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3098 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3099 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3100 - don't call obd_setattr in ll_file_release for destroyed objects
3101 * protocol change to lustre_msg: move |version| and add |flags|
3102 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3103 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3104 * added replay of create, unlink, link and rename operations during
3105 MDS failover; recovery should be much more robust now
3106 * remove failed OSCs from LOVs (only lov_create uses this so far)
3107 * the lustre-HOWTO was brought (more) up to date (582544)
3109 2002-10-23 Phil Schwan <phil@clusterfs.com>
3112 - in-use dentries weren't being reused properly (617851)
3113 - prevent multiple LDLM setup (599178)
3114 - fix LOV size calculations for truncate (617853)
3115 - fix client handling of MDS intent errors (POSIX)
3116 - fix permission bug in lovstripe.c test (624321)
3117 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3118 - truncate past end of file could corrupt data
3119 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3120 - a race in recovery could return ETIMEDOUT to apps (623947)
3121 - building outside the source directory was fixed
3122 * the lustre-HOWTO was brought (more) up to date (582544)
3123 * major progress was made on recovery functionality
3125 2002-10-10 Phil Schwan <phil@clusterfs.com>
3128 - recovery deadlock fix
3129 - rm -rf causes LBUG fix (617817)
3130 - file open by multiple tasks fix (618962)
3131 - directory permissions bugs (602707 and 620007)
3132 - journal_stop fixed with locking (611313)
3133 - O_APPEND failures resolved (618273, perhaps 614459)
3134 - lconf PATH fix (619770)
3135 - IA64 build fix (621450)
3136 - RPC buffer sizes scale with amount of memory
3138 2002-10-01 Phil Schwan <phil@clusterfs.com>
3141 - locks would be cancelled without throwing away data pages,
3142 resulting in inconsistent data (605627)
3143 - inode attributes were not always being refreshed (605627, 612449)
3144 - lconf now continues to cleanup after lctl reports an error
3145 - MDS now enforces user permissions (602707)
3146 - lprocfs cleanup fixed, but not yet enabled (614157)
3147 - fixed infinite server hang, should a client not respond to an AST
3148 - avoid going into recovery if user calls readlink() with a buffer
3149 that's too small (613941)
3150 - AST RPCs no longer require replies (614867) -- this may be changed
3151 - don't crash server if client sends an IOV that's too big (611336)
3152 - fixed lock conversion deadlock (611892)
3153 - fixed the following of symlinks (614622)
3154 * recovery: the server can remove locks from a client that dies, other
3155 clients can make progress
3156 * more extN patch fixes
3157 * compile-time configurable ptlrpc buffer allocations
3159 - collaborative read cache document
3160 - Lustre Lite Performance CDR document-in-progress
3162 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3165 - fix typo in patch-2.4.18
3167 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3170 - clear ptlrpc request each time in handle_incoming_request()
3171 - unlink of files now destroys the object on the OST
3173 2002-09-19 Peter Braam <braam@clusterfs.com>
3175 * add hard link support
3176 * change obdfile creation method
3177 * kernel patch changed
3179 2002-09-19 Peter Braam <braam@clusterfs.com>
3182 - stack overflow bug in extN fixed
3184 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3186 * documentation updates
3187 - add man pages for config tools
3188 - update tests/README to describe testing with new config tools
3189 - finish metadata API descriptions
3190 * bug fixes and cleanups
3191 - statfs workaround for 16TB limit
3192 - LOV stripe allocation improved, can stripe on subset of OSTs
3193 - LOV file size/IO offset was wrong for files > 4GB in size
3194 - object EA data was being dropped, caused files to be unreadable
3195 - memory overflow with non-LOV OST caused memory corruption
3196 - fixed regression tests to work with new config tools, obdfilter
3197 - fixed bug when directory size became larger than 1 block
3198 - fixed bug (for single client case) when PWD was deleted
3199 - invalidate local directory pages when doing intent-based ops
3200 - avoid LDLM oops when lock callback contained bad data
3202 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3204 * documentation updates
3205 * bug fixes and cleanups
3206 - configuration tools
3209 - 64-bit compile warnings
3210 - 64-bit internal statfs data
3212 * test_brw on persistent OST devices
3214 * lprocfs (disabled)
3216 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3218 * documentation updates
3219 * bug fixes and cleanups
3220 * configuration tools
3222 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3225 * many small fixes to 0_5_4
3226 * io/network handling
3227 * thinkos in MDS operations
3229 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3232 * crucial basic fixes to 0.5.3
3233 * IOR, Iozone work over Elan
3236 2002-08-07 Phil Schwan <phil@clusterfs.com>
3237 * version 0_5_3, our first alpha
3238 * we use the new Portals iovs
3239 * documentation updates
3240 * bug fixes and cleanups
3241 * small changes in the DLM wire protocol
3243 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3244 * version 0_5_1 with some initial stability,
3245 * locking on MD and file I/O.
3246 * documentation updates
3247 * several bug fixes since 0.5.0
3248 * small changes in wire protocol
3250 2002-07-18 Phil Schwan <phil@clusterfs.com>
3252 * delivered as Lustre Light Alpha
3253 * fixed a crash after handling invalid MDS requests
3254 * fixed directory pages for architectures with non-4k pages sizes
3256 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3257 * release version v0_4_4
3258 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3260 * Fixes a number of interruption problems with OST operations.
3261 * Update documentation for portals header changes
3262 * Move all wire protocol structs/defines to lustre_idl.h
3263 * Fixes symlink length bug.
3264 * Add tcpdump to repository.
3266 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3267 * release version v0_4_3
3268 * Fixes statfs for inodes on extN.
3269 * Fixes bug in runtests which would delete /etc/hosts.
3270 * Use 64-bit object IDs wherever possible (not into VFS though)
3271 Remove ost_get_info, which is unused by lustre, and out of date.
3273 2002-07-03 Peter Braam <braam@clusterfs.com>
3274 * release version v0_4_2 Fixes a lookup error (type not passed)
3275 * move forward to head of Portals
3276 * move forward to latest Lustre kernel
3278 2002-06-25 Peter Braam <braam@clusterfs.com>
3279 * release version v0_4_1. Hopefully stable on single node use.