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: Add LNET router traceability for debug purposes
28 Details : If a checksum failure occurs with a router as part of the
29 IO path, the NID of the last router that forwarded the bulk data
30 is printed so it can be identified.
35 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
36 Details : Due to a race condition, liblustre clients were occasionally
39 Severity : enhancement
41 Description: lfs setstripe use optional parameters instead of postional
44 Severity : enhancement
46 Description: Nanosecond timestamp support for ldiskfs
47 Details : The on-disk ldiskfs filesystem has added support for nanosecond
48 resolution timestamps. There is not yet support for this at
49 the Lustre filesystem level.
52 Frequency : during server recovery
54 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
55 Details : request with rq_no_resend flag not awake l_wait_event if they get a
59 Frequency : nfs export on patchless client
61 Description: connectathon hang when test nfs export over patchless client
62 Details : Disconnected dentry cannot be found with lookup, so we do not need
63 to unhash it or make it invalid
66 Description: ENOENT returned for valid filehandle during dbench.
67 Details: : Check if a directory has children when invalidating dentries
68 associated with an inode during lock cancellation. This fixes
69 an incorrect ENOENT sometimes seen for valid filehandles during
72 --------------------------------------------------------------------------------
74 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
76 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
77 INTEROPERATE with older versions automatically. In many cases a
78 special upgrade step is needed. Please read the
79 user documentation before upgrading any part of a 1.4.x system.
80 * WARNING: Lustre configuration and startup changes are required with
81 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
83 * Support for kernels:
84 2.4.21-47.0.1.EL (RHEL 3)
86 2.6.9-42.0.10.EL (RHEL 4)
87 2.6.12.6 vanilla (kernel.org)
88 2.6.16.27-0.9 (SLES10)
89 * Client support for unpatched kernels:
90 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
91 2.6.16 - 2.6.19 vanilla (kernel.org)
92 2.6.9-42.0.8EL (RHEL 4)
93 * Recommended e2fsprogs version: 1.39.cfs6
94 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
97 Severity : enhancement
99 Description: MountConf
100 Details : Lustre configuration is now managed via mkfs and mount
101 commands instead of lmc and lconf. New obd types (MGS, MGC)
102 are added for dynamic configuration management. See
103 https://mail.clusterfs.com/wikis/lustre/MountConf for
106 Severity : enhancement
108 Description: dynamic OST addition
109 Details : OSTs can now be added to a live filesystem
111 Severity : enhancement
113 Description: startup order invariance
114 Details : MDTs and OSTs can be started in any order. Clients only
115 require the MDT to complete startup.
117 Severity : enhancement
119 Description: parallel, asynchronous orphan cleanup
120 Details : orphan cleanup is now performed in separate threads for each
121 OST, allowing parallel non-blocking operation.
123 Severity : enhancement
125 Description: optimized stripe assignment
126 Details : stripe assignments are now made based on ost space available,
127 ost previous usage, and OSS previous usage, in order to try
128 to optimize storage space and networking resources.
130 Severity : enhancement
132 Description: Permanently set tunables
133 Details : All writable /proc/fs/lustre tunables can now be permanently
134 set on a per-server basis, at mkfs time or on a live
137 Severity : enhancement
139 Description: Lustre message v2
140 Details : Add lustre message format v2.
142 Severity : enhancement
144 Description: client OST exclusion list
145 Details : Clients can be started with a list of OSTs that should be
146 declared "inactive" for known non-responsive OSTs.
150 Description: ENOENT returned for valid filehandle during dbench.
151 Details : Check if a directory has children when invalidating dentries
152 associated with an inode during lock cancellation. This fixes
153 an incorrect ENOENT sometimes seen for valid filehandles during
157 Frequency : SFS test only (otherwise harmless)
159 Description: SPEC SFS validation failure on NFS v2 over lustre.
160 Details : Changes the blocksize for regular files to be 2x RPC size,
161 and not depend on stripe size.
163 Severity : enhancement
165 Description: fine-grained SMP locking inside DLM
166 Details : Improve DLM performance on SMP systems by removing the single
167 per-namespace lock and replace it with per-resource locks.
169 Severity : enhancement
171 Description: don't hold multiple extent locks at one time
172 Details : To avoid client eviction during large writes, locks are not
173 held on multiple stripes at one time or for very large writes.
174 Otherwise, clients can block waiting for a lock on a failed OST
175 while holding locks on other OSTs and be evicted.
177 Severity : enhancement
179 Description: Multiple MD RPCs in flight.
180 Details : Further unserialise some read-only MDT RPCs - learn about intents.
181 To avoid overly-overloading MDT, introduce a limit on number of
182 MDT RPCs in flight for a single client and add /proc controls
183 to adjust this limit.
185 Severity : enhancement
187 Description: client read/write statistics
188 Details : Add client read/write call usage stats for performance
189 analysis of user processes.
190 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
191 file access. extents_stats shows chunk size distribution.
192 extents_stats_per_process show chunk size distribution per
195 Severity : enhancement
197 Description: per-client statistics on server
198 Details : Add ldlm and operations statistics for each client in
199 /proc/fs/lustre/mds|obdfilter/*/exports/
201 Severity : enhancement
203 Description: improved MDT statistics
204 Details : Add detailed MDT operations statistics in
205 /proc/fs/lustre/mds/*/stats
207 Severity : enhancement
209 Description: VFS operations stats
210 Details : Add client VFS call stats, trackable by pid, ppid, or gid
211 /proc/fs/lustre/llite/*/vfs_ops_stats
212 /proc/fs/lustre/llite/*/vfs_track_[pid|ppid|gid]
217 Description: Fix client-side osc byte counters
218 Details : The osc read/write byte counters in
219 /proc/fs/lustre/osc/*/stats are now working
222 Frequency : always as root on SLES
224 Description: Failure of copying files with lustre special EAs.
225 Details : Client side always return success for setxattr call for lustre
226 special xattr (currently only "trusted.lov").
231 Description: Refcount LNET uuids
232 Details : The global LNET uuid list grew linearly with every startup;
233 refcount repeated list entries instead of always adding to
236 Severity : enhancement
238 Description: Dynamic service threads
239 Details : Within a small range, start extra service threads
240 automatically when the request queue builds up.
243 Frequency : mixed-endian client/server environments
245 Description: mixed-endian crashes
246 Details : The new msg_v2 system had some failures in mixed-endian
249 Severity : enhancement
251 Description: Easy OST removal
252 Details : OSTs can be permanently deactivated with e.g. 'lctl
253 conf_param lustre-OST0001.osc.active=0'
255 Severity : enhancement
257 Description: MGS proc entries
258 Details : Added basic proc entries for the MGS showing what filesystems
261 Severity : enhancement
263 Description: provide MGS failover
264 Details : Added config lock reacquisition after MGS server failover.
266 Severity : enhancement
268 Description: add Linux 2.4 support
269 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
273 Description: a large application tries to do I/O to the same resource and dies
275 Details : Check the req->rq_arrival time after the call to
276 ost_brw_lock_get(), but before we do anything about
277 processing it & sending the BULK transfer request. This
278 should help move old stale pending locks off the queue as
279 quickly as obd_timeout.
282 Frequency : when an incorrect nid is specified during startup
284 Description: ptlrpc connect to non-existant node causes kernel crash
285 Details : LNET can't be re-entered from an event callback, which
286 happened when we expire a message after the export has been
287 cleaned up. Instead, hand the zombie cleanup off to another
290 Severity : enhancement
292 Description: plain/inodebits lock performance improvement
293 Details : Grouping plain/inodebits in granted list by their request modes
294 and bits policy, thus improving the performance of search through
298 Frequency : only if OST filesystem is corrupted
300 Description: client incorrectly hits assertion in ptlrpc_replay_req()
301 Details : for a short time RPCs with bulk IO are in the replay list,
302 but replay of bulk IOs is unimplemented. If the OST filesystem
303 is corrupted due to disk cache incoherency and then replay is
304 started it is possible to trip an assertion. Avoid putting
305 committed RPCs into the replay list at all to avoid this issue.
308 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
310 Description: System hang on startup
311 Details : This bug allowed the liblustre (e.g. catamount) client to
312 return to the app before handling all startup RPCs. This
313 could leave the node unresponsive to lustre network traffic
314 and manifested as a server ptllnd timeout.
316 Severity : enhancement
318 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
319 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
320 boolean to control whether to print peer debug info when a
321 client's RPC times out.
324 Frequency : only for kernels with patches from Lustre below 1.4.3
326 Description: Remove old rdonly API
327 Details : Remove old rdonly API which unused from at least lustre 1.4.3
330 Frequency : only for devices with external journals
332 Description: Set external device read-only also
333 Details : During a commanded failover stop, we set the disk device
334 read-only while the server shuts down. We now also set any
335 external journal device read-only at the same time.
338 Frequency : when upgrading from 1.4 while trying to change parameters
340 Description: The wrong (new) MDC name was used when setting parameters for
341 upgraded MDT's. Also allows changing of OSC (and MDC)
342 parameters if --writeconf is specified at tunefs upgrade time.
345 Frequency : when setting specific ost indicies
347 Description: QOS code breaks on skipped indicies
348 Details : Add checks for missing OST indicies in the QOS code, so OSTs
349 created with --index need not be sequential.
351 Severity : enhancement
353 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
354 Details : The uninit_groups feature works in conjunction with the kernel
355 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
356 up the pass1 processing of e2fsck. This is a read-only feature
357 in ldiskfs2 only, so older kernels and current ldiskfs cannot
358 mount filesystems that have had this feature enabled.
360 Severity : enhancement
362 Description: Improve multi-block allocation algorithm to avoid fragmentation
363 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
364 allocation locality and avoid filesystem fragmentation.
366 ------------------------------------------------------------------------------
368 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
370 * Support for kernels:
371 2.4.21-47.0.1.EL (RHEL 3)
373 2.6.9-42.0.10.EL (RHEL 4)
374 2.6.12.6 vanilla (kernel.org)
375 2.6.16.27-0.9 (SLES 10)
376 * Recommended e2fsprogs version: 1.39.cfs5
378 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
382 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
384 Description: System hang on startup
385 Details : This bug allowed the liblustre (e.g. catamount) client to
386 return to the app before handling all startup RPCs. This
387 could leave the node unresponsive to lustre network traffic
388 and manifested as a server ptllnd timeout.
390 Severity : enhancement
392 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
393 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
394 boolean to control whether to print peer debug info when a
395 client's RPC times out.
400 Description: make O_SYNC working on 2.6 kernels
401 Details : 2.6 kernels use different method for mark pages for write,
402 so need add a code to lustre for O_SYNC work.
407 Description: Failure to close file and release space on NFS
408 Details : Put inode details into lock acquired in ll_intent_file_open.
409 Use mdc_intent_lock in ll_intent_open to properly
410 detect all kind of errors unhandled by mdc_enqueue.
415 Description: proc file read during shutdown sometimes raced obd removal,
417 Details : Add lock to prevent obd access after proc file removal.
420 Frequency : Only for files larger than 4GB on 32-bit clients.
422 Description: improperly doing page alignment of locks
423 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
424 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
429 Description: RPCs being resent when they shouldn't be
430 Details : Some RPCs that should not be resent are being resent. This
431 can cause inconsistencies in the RPC state machine. Do not
432 resend such requests.
435 Frequency : rare, only with NFS export
437 Description: Crash on NFS re-export node
438 Details : under very unusual load conditions an assertion is hit in
439 ll_intent_file_open()
442 Frequency : only if OST filesystem is corrupted
444 Description: client incorrectly hits assertion in ptlrpc_replay_req()
445 Details : for a short time RPCs with bulk IO are in the replay list,
446 but replay of bulk IOs is unimplemented. If the OST filesystem
447 is corrupted due to disk cache incoherency and then replay is
448 started it is possible to trip an assertion. Avoid putting
449 committed RPCs into the replay list at all to avoid this issue.
454 Description: large O_DIRECT requests fail under memory pressure/fragmentation
455 Details : Large single O_DIRECT read and write calls can fail to allocate
456 a sufficiently large buffer to process the request. In case of
457 allocation failure the allocation is retried with a smaller
458 buffer and broken into smaller requests.
460 Severity : enhancement
462 Description: Add -o localflock option to simulate old noflock behaviour.
463 Details : This will achieve local-only flock/fcntl locks coherentness.
468 Description: versioning check is incomplete
469 Details : Checking the version difference of client vs. server, report
470 error if the gap is too big.
475 Description: add support PG_writeback bit
476 Details : add support for PG_writeback bit for Lustre, for more carefull
477 work with page cache in 2.6 kernel. This also fix some deadlocks
478 and remove hack for work O_SYNC with 2.6 kernel.
480 Severity : enhancement
482 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
483 Details : The uninit_groups feature works in conjunction with the kernel
484 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
485 up the pass1 processing of e2fsck. This is a read-only feature
486 in ldiskfs2 only, so older kernels and current ldiskfs cannot
487 mount filesystems that have had this feature enabled.
489 Severity : enhancement
491 Description: Improve multi-block allocation algorithm to avoid fragmentation
492 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
493 allocation locality and avoid filesystem fragmentation.
495 ------------------------------------------------------------------------------
497 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
499 * Support for kernels:
500 2.6.9-42.0.3EL (RHEL 4)
502 2.4.21-47.0.1.EL (RHEL 3)
503 2.6.12.6 vanilla (kernel.org)
504 * Recommended e2fsprogs version: 1.39.cfs2-0
506 * The backwards-compatible /proc/sys/portals symlink has been removed
507 in this release. Before upgrading, please ensure that you change
508 any configuration scripts or /etc/sysctl.conf files that access
509 /proc/sys/portals/* or sysctl portals.* to use the corresponding
510 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
511 in advance of the upgrade on any system running Lustre 1.4.6 or
512 newer, since /proc/sys/lnet was added in that version.
513 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
517 Frequency : only when quota is used
519 Description: avoid scanning export list for quota master
520 Details : Change the algorithms to avoid scanning export list in order
521 to improve the efficiency.
524 Frequency : MDS failover only, very rarely
526 Description: "went back in time" messages on mds failover
527 Details : The greatest transno may be lost when the current operation
528 finishes with an error (transno==0) and the client's last_rcvd
529 record is over-written. Save the greatest transno in the
530 mds_last_transno for this case.
533 Frequency : always for specific kernels and striping counts
535 Description: client may get "Matching packet too big" without ACL support
536 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
537 when trying to access files in certain configurations. The
538 clients should in fact be denied when mounting because they do
542 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
544 Description: many clients connecting with IO in progress causes connect timeouts
545 Details : Avoid synchronous journal commits to avoid delays caused by many
546 clients connecting/disconnecting when bulk IO is in progress.
547 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
548 OST_IO_PORTAL to avoid delays behind potentially many pending
552 Frequency : occasionally with multiple writers to a single file
554 Description: shared writes to file may result in wrong size reported by stat()
555 Details : Allow growing of kms when extent lock is cancelled
558 Frequency : always with random mmap IO to multi-striped file
560 Description: mmap write might be lost if we are writing to a 'hole' in stripe
561 Details : Only if the hole is at the end of OST object so that kms is too
562 small. Fix is to increase kms accordingly in ll_nopage.
565 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
567 Description: writes to a missing object would leak memory on the OST
568 Details : If there is an inconsistency between the MDS and OST filesystems,
569 such that the MDS references an object that doesn't exist, writes
570 to that object will leak memory due to incorrect cleanup in the
571 error handling path, eventually running out of memory on the OST.
576 Description: Creating too long symlink causes lustre errors
577 Details : Check symlink and name lengths before sending requests to MDS.
580 Frequency : only if flock is enabled (not on by default)
582 Description: posix locks not released on fd closure on 2.6.9+
583 Details : We failed to add posix locks to list of inode locks on 2.6.9+
584 kernels, this caused such locks not to be released on fd close and
585 then assertions on fs unmount about still used locks.
588 Frequency : MDS failover only, very rarely
590 Description: clients may get ASSERTION(granted_lock != NULL)
591 Details : When request was taking a long time, and a client was resending
592 a getattr by name lock request. The were multiple lock requests
593 with the same client lock handle and
594 mds_getattr_name->fixup_handle_for_resent_request found one of the
595 lock handles but later failed with ASSERTION(granted_lock != NULL).
600 Description: handle->h_buffer_credits > 0, assertion failure
601 Details : h_buffer_credits is zero after truncate, causing assertion
602 failure. This patch extends the transaction or creates a new
606 Frequency : NFS re-export or patchless client
607 Bugzilla : 11179, 10796
608 Description: Crash on NFS re-export node (__d_move)
609 Details : We do not want to hash the dentry if we don't have a lock.
610 But if this dentry is later used in d_move, we'd hit uninitialised
611 list head d_hash, so we just do this to init d_hash field but
612 leave dentry unhashed.
615 Frequency : NFS re-export or patchless client
617 Description: NFS exports has problem with symbolic link
618 Details : lustre client didn't properly install dentry when re-exported
619 to NFS or running patchless client.
622 Frequency : NFS re-export or patchless client
624 Description: Various nfs/patchless fixes.
625 Details : fixes reuse disconected alias for lookup process - this fixes
626 warning "find_exported_dentry: npd != pd",
627 fix permission error with open files at nfs.
628 fix apply umask when do revalidate.
631 Frequency : occasional
633 Description: Crash on NFS re-export node
634 Details : calling clear_page() on the wrong pointer triggered oops in
635 generic_mapping_read().
638 Frequency : rarely, using O_DIRECT IO
640 Description: unaligned directio crashes client with LASSERT
641 Details : check for unaligned buffers before trying any requests.
644 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
646 Description: stale data returned from RAID cache
647 Details : If only a small amount of IO is done to the RAID device before
648 reading it again it is possible to get stale data from the RAID
649 cache instead of reading it from disk.
652 Frequency : always for sles10 kernel
654 Description: sles10 support
655 Details : ll_follow_link: compile fixes and using of nd_set_link
659 Frequency : depends on arch, kernel and compiler version, always on sles10
662 Description: recursive or deep enough symlinks cause stack overflow
663 Details : getting rid of large stack-allocated variable in
667 Frequency : depends on hardware
669 Description: lustre write performance loss in the SLES10 kernel
670 Details : the performance loss is caused by using of write barriers in the
671 ext3 code. The SLES10 kernel turns barrier support on by
672 default. The fix is to undo that change for ldiskfs.
675 ------------------------------------------------------------------------------
677 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
679 * Support for kernels:
680 2.6.9-42.0.3EL (RHEL 4)
682 2.4.21-47.0.1.EL (RHEL 3)
683 2.6.12.6 vanilla (kernel.org)
687 Frequency : quota enabled and large files being deleted
689 Description: releasing more than 4GB of quota at once hangs OST
690 Details : If a user deletes more than 4GB of files on a single OST it
691 will cause the OST to spin in an infinite loop. Release
692 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
697 Description: statfs data retrieved from /proc may be stale or zero
698 Details : When reading per-device statfs data from /proc, in the
699 {kbytes,files}_{total,free,avail} files, it may appear
700 as zero or be out of date.
703 Frequency : systems with MD RAID1 external journal devices
705 Description: lconf's call to blkid is confused by RAID1 journal devices
706 Details : Use the "blkid -l" flag to locate the MD RAID device instead
707 of returning all block devices that match the journal UUID.
710 Frequency : always, for aggregate stripe size over 4GB
712 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
713 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
714 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
717 Frequency : always if "lfs find" used on a local file/directory
719 Description: "lfs find" segfaults if used on a local file/directory
720 Details : The case where a directory component was not specified wasn't
721 handled correctly. Handle this properly.
724 Frequency : always on ppc64
726 Description: the write to an ext3 filesystem mounted with mballoc got stuck
727 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
728 perform endianness conversion.
731 Frequency : rarely (truncate to non-zero file size after write under load)
732 Bugzilla : 10730, 10687
733 Description: Files padded with zeros to next 4K multiple
734 Details : With filesystems mounted using the "extents" option (2.6 kernels)
735 it is possible that files that are truncated to a non-zero size
736 immediately after being written are filled with zero bytes beyond
737 the truncated size. No file data is lost.
739 Severity : enhancement
741 Description: Allow recovery/failover for liblustre clients.
742 Details : liblustre clients were unaware of failover configurations until
745 Severity : enhancement
747 Description: user file locks should fail when not mounting with flock option
748 Details : Set up an error-returning stub in ll_file_operations.lock field
749 to prevent incorrect behaviour when client is mounted without
750 flock option. Also, set up properly f_op->flock field for
754 Frequency : always on ia64
756 Description: "lfs df" loops on printing out MDS statfs information
757 Details : The obd_ioctl_data was not initialized and in some systems
758 this caused a failure during the ioctl that did not return
759 an error. Initialize the struct and return an error on failure.
762 Frequency : SLES 9 only
764 Description: Error of copying files with lustre special EAs as root
765 Details : Client side always return success for setxattr call for lustre
766 special xattr (currently only "trusted.lov").
769 Frequency : rarely on clusters with both ia64+i386 clients
771 Description: ia64+i686 clients doing shared IO on the same file may LBUG
772 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
773 clients are doing concurrent writes to the same file it is
774 possible that the ia64 clients may LASSERT because the OST
775 extent locks are not PAGE_SIZE aligned. Ensure that grown
776 locks are always aligned on the request boundary.
779 Frequency : specific use, occasional
781 Description: Overwriting in use executable truncates on-disk binary image
782 Details : If one node attempts to overwrite an executable in use by
783 another node, we now correctly return ETXTBSY instead of
786 Severity : enhancement
788 Description: Async OSC create to avoid the blocking unnecessarily.
789 Details : If a OST has no remain object, system will block on the creating
790 when need to create a new object on this OST. Now, ways use
791 pre-created objects when available, instead of blocking on an
792 empty osc while others are not empty. If we must block, we block
793 for the shortest possible period of time.
798 Description: chmod on Lustre root is propagated to other clients
799 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
800 invalid by the follow_mount time.
803 Frequency : liblustre clients only
805 Description: Race in 'instant cancel' lock handling could lead to such locks
806 never to be granted in case of SMP MDS
807 Details : Do not destroy not yet granted but cbpending locks in
811 Frequency : replay/resend of open
813 Description: non null lock assetion failure in mds_intent_policy
814 Details : Trying to replay/resend lockless open requests resulted in
815 mds_open() returning 0 with no lock. Now it sets a flag if
816 a lock is going to be returned.
818 Severity : enhancement
820 Description: Checksum enhancements
821 Details : New checksum enhancements allow for resending RPCs that failed
824 Severity : enhancement
826 Description: Tunables on number of dirty pages in cacche
827 Details : Allow to set limit on number of dirty pages cached.
832 Description: client crash on unmount - lock still has references
833 Details : In some error handling cases it was possible to leak a lock
834 reference on a client while accessing a file. This was not
835 harmful to the client during operation, but would cause the
836 client to crash when the filesystem is unmounted.
839 Frequency : specific case, rare
841 Description: ETXTBSY on mds though file not in use by client
842 Details : ETXTBSY is no longer incorrectly returned when attempting to
843 chmod or chown a directory that the user previously tried to
844 execute or a currently-executing binary.
847 Frequency : extremely rare except on liblustre-based clients
849 Description: Lustre space not freed when files are deleted
850 Details : Clean up open-unlinked files after client eviction. Previously
851 the unlink was skipped and the files remained as orphans.
856 Description: OST failure "would be an LBUG" in waiting_locks_callback()
857 Details : In some cases it was possible to send a blocking callback to
858 a client doing a glimpse, even though that client didn't get
859 a lock granted. When the glimpse lock is cancelled on the OST
860 the freed lock is left on the waiting list and corrupted the list.
863 Frequency : all core dumps
865 Description: Broke core dumps to lustre
866 Details : Negative dentry may be unhashed if parent does not have UPDATE
867 lock, but some callers, e.g. do_coredump, expect dentry to be
868 hashed after successful create, hash it in ll_create_it.
870 ------------------------------------------------------------------------------
872 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
874 * Support for kernels:
875 2.6.9-42.0.2.EL (RHEL 4)
877 2.4.21-40.EL (RHEL 3)
878 2.6.12.6 vanilla (kernel.org)
882 Frequency : always on RHEL 3
884 Description: Number of open files grows over time
885 Details : The number of open files grows over time, whether or not
886 Lustre is started. This was due to a filp leak introduced
887 by one of our kernel patches.
889 ------------------------------------------------------------------------------
891 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
893 * Support for kernels:
896 2.4.21-40.EL (RHEL 3)
897 2.6.12.6 vanilla (kernel.org)
902 Bugzilla : 5719, 9635, 9792, 9684
903 Description: OST (or MDS) trips assertions in (re)connection under heavy load
904 Details : If a server is under heavy load and cannot reply to new
905 connection requests before the client resends the (re)connect,
906 the connection handling code can behave badly if two service
907 threads are concurrently handing separate (re)connections from
908 the same client. Add better locking to the connection handling
909 code, and ensure that only a single connection will be processed
910 for a given client UUID, even if the lock is dropped.
912 Severity : enhancement
914 Description: add TCP zero-copy support to kernel
915 Details : Add support to the kernel TCP stack to allow zero-copy bulk
916 sends if the hardware supports scatter-gather and checksumming.
917 This allows socklnd to do client-write and server-read more
918 efficiently and reduce CPU utilization from skbuf copying.
921 Frequency : only if NFS exporting from client
923 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
924 Details : If setting attributes on a file created under NFS that had
925 never been opened it would be possible to oops the client
926 if the file had no objects.
930 Bugzilla : 9326, 10402, 10897
931 Description: client crash in ptlrpcd_wake() thread when sending async RPC
932 Details : It is possible that ptlrpcd_wake() dereferences a freed async
933 RPC. In rare cases the ptlrpcd thread alread processed the RPC
934 before ptlrpcd_wake() was called and the request was freed.
937 Frequency : always for liblustre
939 Description: liblustre client does MDS+OSTs setattr RPC for each write
940 Details : When doing a write from a liblustre client, the client
941 incorrectly issued an RPC to the MDS and each OST the file was
942 striped over in order to update the timestamps. When writing
943 with small chunks and many clients this could overwhelm the MDS
944 with RPCs. In all cases it would slow down the write because
945 these RPCs are unnecessary.
947 Severity : enhancement
949 Description: allow number of MDS service threads to be changed at module load
950 Details : It is now possible to change the number of MDS service threads
951 running. Adding "options mds mds_num_threads={N}" to the MDS's
952 /etc/modprobe.conf will set the number of threads for the next
953 time Lustre is restarted (assuming the "mds" module is also
954 reloaded at that time). The default number of threads will
955 stay the same, 32 for most systems.
960 Description: OST crash if filesystem is unformatted or corrupt
961 Details : If an OST is started on a device that has never been formatted
962 or if the filesystem is corrupt and cannot even mount then the
963 error handling cleanup routines would dereference a NULL pointer.
968 Description: NULL pointer deref in llap_from_page.
969 Details : get_cache_page_nowait can return a page with NULL (or otherwise
970 incorrect) mapping if the page was truncated/reclaimed while it was
971 searched for. Check for this condition and skip such pages when
972 doing readahead. Introduce extra check to llap_from_page() to
973 verify page->mapping->host is non-NULL (so page is not anonymous).
976 Frequency : Sometimes when using sys_sendfile
978 Description: "page not covered by a lock" warnings from ll_readpage
979 Details : sendfile called ll_readpage without right page locks present.
980 Now we introduced ll_file_sendfile that does necessary locking
981 around call to generic_file_sendfile() much like we do in
985 Frequency : with certain MDS communication failures at client mount time
987 Description: NULL pointer deref after failed client mount
988 Details : a client connection request may delayed by the network layer
989 and not be sent until after the PTLRPC layer has timed out the
990 request. If the client fails the mount immediately it will try
991 to clean up before the network times out the request. Add a
992 reference from the request import to the obd device and delay
993 the cleanup until the network drops the request.
996 Frequency : occasionally during client (re)connect
998 Description: assertion failure during client (re)connect
999 Details : processing a client connection request may be delayed by the
1000 client or server longer than the client connect timeout. This
1001 causes the client to resend the connection request. If the
1002 original connection request is replied in this interval, the
1003 client may trip an assertion failure in ptlrpc_connect_interpret()
1004 which thought it would be the only running connect process.
1007 Frequency : only with obd_echo servers and clients that are rebooted
1009 Description: kernel BUG accessing uninitialized data structure
1010 Details : When running an obd_echo server it did not start the ping_evictor
1011 thread, and when a client was evicted an uninitialized data
1012 structure was accessed. Start the ping_evictor in the RPC
1013 service startup instead of the OBD startup.
1015 Severity : enhancement
1016 Bugzilla : 10193 (patchless)
1017 Description: Remove dependency on various unexported kernel interfaces.
1018 Details : No longer need reparent_to_init, exit_mm, exit_files,
1019 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1022 Frequency : rare (only users of deprecated and unsupported LDAP config)
1024 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1025 Details : LDAP apparently contains 'lustreName' attributes instead of
1026 'name'. A simple remapping of the name is sufficient.
1029 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1031 Description: waiting_locks_callback trips kernel BUG if client is evicted
1032 Details : Running with the dump_on_timeout debug flag turned on makes
1033 it possible that the waiting_locks_callback() can try to dump
1034 the Lustre kernel debug logs from an interrupt handler. Defer
1035 this log dumping to the expired_lock_main() thread.
1037 Severity : enhancement
1039 Description: Support NFS exporting on 2.6 kernels.
1040 Details : Implement non-rawops metadata methods for NFS server to use without
1041 changing NFS server code.
1044 Frequency : very rare (synthetic metadata workload only)
1046 Description: two racing renames might cause an MDS thread to deadlock
1047 Details : Running the "racer" program may cause one MDS thread to rename
1048 a file from being the source of a rename to being the target of
1049 a rename at exactly the same time that another thread is doing
1050 so, and the second thread has already enqueued these locks after
1051 doing a lookup of the target and is trying to relock them in
1052 order. Ensure that we don't try to re-lock the same resource.
1055 Frequency : only very large systems with liblustre clients
1057 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1058 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1059 all OSTs in parallel. This allows the eviction of stale liblustre
1060 clients to proceed much faster than if they were done in series,
1061 and also offers similar improvements for other set_info RPCs.
1066 Description: excessive CPU usage during initial read phase on client
1067 Details : During the initial read phase on a client, it would agressively
1068 retry readahead on the file, consuming too much CPU and impacting
1069 performance (since 1.4.5.8). Improve the readahead algorithm
1070 to avoid this, and also improve some other common cases (read
1071 of small files in particular, where "small" is files smaller than
1072 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1077 Description: MDS crash when receiving packet with unknown intent.
1078 Details : Do not LBUG in unknown intent case, just return -EFAULT
1080 Severity : enhancement
1081 Bugzilla : 9293, 9385
1082 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1083 Details : Do not serialize getattr (non-intent version) and statfs.
1086 Frequency : occasional, when OST network is overloaded/intermittent
1088 Description: client evicted by OST after bulk IO timeout
1089 Details : If a client sends a bulk IO request (read or write) the OST
1090 may evict the client if it is unresposive to its data GET/PUT
1091 request. This is incorrect if the network is overloaded (takes
1092 too long to transfer the RPC data) or dropped the OST GET/PUT
1093 request. There is no need to evict the client at all, since
1094 the pinger and/or lock callbacks will handle this, and the
1095 client can restart the bulk request.
1098 Frequency : Always when mmapping file with no objects
1100 Description: client crashes when mmapping file with no objects
1101 Details : Check that we actually have objects in a file before doing any
1102 operations on objects in ll_vm_open, ll_vm_close and
1108 Description: Request leak when working with deleted CWD
1109 Details : Introduce advanced request refcount tracking for requests
1110 referenced from lustre intent.
1112 Severity : Enhancement
1114 Description: Cache open file handles on client.
1115 Details : MDS now will return special lock along with openhandle, if
1116 requested and client is allowed to hold openhandle, even if unused,
1117 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1118 closing files for every read/write openration.
1120 Severity : Enhancement
1122 Description: Cache open negative dentries on client when possible.
1123 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1124 negative dentries on lock revocation.
1129 Description: Remounting a client read-only wasn't possible with a zconf mount
1130 Details : It wasn't possible to remount a client read-only with llmount.
1132 Severity : enhancement
1133 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1134 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1135 MPICH is included to add Lustre-specific ADIO interfaces.
1136 This is based closely on the UFS ADIO layer and only differs
1137 in file creation, in order to allow the OST striping to be set.
1138 This is user-contributed code and not supported by CFS.
1143 Description: extended inode attributes (immutable, append-only) work improperly
1144 when 2.4 and 2.6 kernels are used on client/server or vice versa
1145 Details : Introduce kernel-independent values for these flags.
1147 Severity : enhancement
1150 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1151 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1152 granularity. Now, Fractional MB granularity is be supported,
1153 this is very useful for low memory system.
1155 Severity : enhancement
1157 Description: Getattr by fid
1158 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1159 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1160 have locks and avoids getattr rpc requests for it on every stat.
1163 Frequency : Always, for filesystems larger than 2TB
1165 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1166 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1167 filesystems larger than 2TB to be mounted with mballoc enabled.
1170 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1172 Description: filesystem corruption for non-standard kernels and very large OSTs
1173 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1174 filesystems larger than 2TB is mounted then the kernel will
1175 silently corrupt the start of the filesystem. CONFIG_LBD is
1176 enabled for all CFS-supported kernels, but the possibility of
1177 this happening with a modified kernel config exists.
1179 Severity : enhancement
1181 Description: add client O_DIRECT support for 2.6 kernels
1182 Details : It is now possible to do O_DIRECT reads and writes to files
1183 in the Lustre client mountpoint on 2.6 kernel clients.
1185 Severity : enhancement
1187 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1188 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1189 parallel, not waiting for response from every OST before sending
1190 a rpc to the next OST.
1195 Description: setattr vs write race when updating file timestamps
1196 Details : Client processes that update a file timestamp into the past
1197 right after writing to the file (e.g. tar) it is possible that
1198 the updated file modification time can be reset to the current
1199 time due to a race between processing the setattr and write RPC.
1201 Severity : enhancement
1203 Description: Bring 'lfs find' closer in line with regular Linux find.
1204 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1205 -print0 options and obtains all the needed info through the lustre
1208 Severity : enhancement
1210 Description: support up to 1024 configured devices on one node
1211 Details : change obd_dev array from statically allocated to dynamically
1212 allocated structs as they are first used to reduce memory usage
1217 Description: Flush dirty partially truncated pages during truncate
1218 Details : Immediatelly flush partially truncated pages in filter_setattr,
1219 this way we completely avoid having any pages in page cache on OST
1220 and can retire ugly workarounds during writes to flush such pages.
1225 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1226 Details : protect inode from truncation within vfs_unlink() context
1227 just take a reference before calling vfs_unlink() and release it
1228 when parent's i_sem is free.
1231 Frequency : always, if extents are used on OSTs
1233 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1234 Details : OSTs using the extents format would not zero the high 16 bits of
1235 the index physical block number. This is not a problem for any
1236 OST filesystems smaller than 16TB, and no kernels support ext3
1237 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1238 new/modified files) and can be fixed for existing filesystems
1239 with e2fsprogs-1.39-cfs1.
1244 Description: import connection selection may be incorrect if timer wraps
1245 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1246 import connections to be ignored if the 32-bit jiffies counter
1247 wraps. Use a 64-bit jiffies counter.
1250 Frequency : during server recovery
1252 Description: crash after server is denying duplicate export
1253 Details : If clients are resending connect requests to the server, the
1254 server refuses to allow a client to connect multiple times.
1255 Fixed a bug in the handling of this case.
1258 Frequency : very large clusters immediately after boot
1260 Description: LNET request buffers exhausted under heavy short-term load
1261 Details : If a large number of client requests are generated on a service
1262 that has previously never seen so many requests it is possible
1263 that the request buffer growth cannot keep up with the spike in
1264 demand. Instead of dropping incoming requests, they are held in
1265 the LND until the RPC service can accept more requests.
1268 Frequency : Sometimes during replay
1270 Description: Assertion failure in ll_local_open after replay.
1271 Details : If replay happened on an open request reply before we were able
1272 to set replay handler, reply will become not swabbed tripping the
1273 assertion in ll_local_open. Now we set the handler right after
1274 recognising of open request
1277 Frequency : very rare
1279 Description: kernel reports "badness in vsnprintf"
1280 Details : Reading from the "recovery_status" /proc file in small chunks
1281 may cause a negative length in lprocfs_obd_rd_recovery_status()
1282 call to vsnprintf() (which is otherwise harmless). Exit early
1283 if there is no more space in the output buffer.
1285 Severity : enhancement
1287 Description: clear OBD RPC statistics by writing to them
1288 Details : It is now possible to clear the OBD RPC statistics by writing
1289 to the "stats" file.
1294 Description: Client mtime is not the same on different clients after utimes
1295 Details : In some cases, the client was using the utimes() syscall on
1296 a file cached on another node. The clients now validate the
1297 ctime from the MDS + OSTs to determine which one is right.
1302 Description: Inability to activate failout mode
1303 Details : lconf script incorrectly assumed that in python string's numeric
1304 value is used in comparisons.
1307 Frequency : always with multiple stripes per file
1309 Description: Inefficient object allocation for mutli-stripe files
1310 Details : When selecting which OSTs to stripe files over, for files with
1311 a stripe count that divides evenly into the number of OSTs,
1312 the MDS is always picking the same starting OST for each file.
1313 Return the OST selection heuristic to the original design.
1318 Description: mount failures may take full timeout to return an error
1319 Details : Under some heavy load conditions it is possible that a
1320 failed mount can wait for the full obd_timeout interval,
1321 possibly several minutes, before reporting an error.
1322 Instead return an error as soon as the status is known.
1324 ------------------------------------------------------------------------------
1326 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
1328 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1329 INTEROPERATE with older versions automatically. Please read the
1330 user documentation before upgrading any part of a live system.
1331 * WARNING: Lustre networking configuration changes are required with
1332 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1335 * Support for kernels:
1336 2.6.9-22.0.2.EL (RHEL 4)
1337 2.6.5-7.244 (SLES 9)
1338 2.6.12.6 vanilla (kernel.org)
1341 Severity : enhancement
1342 Bugzilla : 7981/8208
1343 Description: Introduced Lustre Networking (LNET)
1344 Details : LNET is new networking infrastructure for Lustre, it includes
1345 a reorganized network configuration mode (see the user
1346 documentation for full details) as well as support for routing
1347 between different network fabrics. Lustre Networking Devices
1348 (LNDs) for the supported network fabrics have also been
1349 created for this new infrastructure.
1351 Severity : enhancement
1352 Description: Introduced Access control lists
1353 Details : clients can set ACLs on files and directories in order to have
1354 more fine-grained permissions than the standard Unix UGO+RWX.
1355 The MDS must be started with the "-o acl" mount option.
1357 Severity : enhancement
1358 Description: Introduced filesystem quotas
1359 Details : Administrators may now establish per-user quotas on the
1362 Severity : enhancement
1364 Description: Configuration change for the XT3
1365 The PTLLND is now used to run Lustre over Portals on the XT3
1366 The configure option(s) --with-cray-portals are no longer used.
1367 Rather --with-portals=<path-to-portals-includes> is used to
1368 enable building on the XT3. In addition to enable XT3 specific
1369 features the option --enable-cray-xt3 must be used.
1374 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1375 Details : A race between allocating a new llcd and re-getting the llcd_lock
1376 allowed another thread to grab newly-allocated llcd.
1378 Severity : enhancement
1380 Description: 2.6 OST async journal commit and locking fix to improve performance
1381 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1382 2.6 kernels are now async as they already were in 2.4 kernels so
1383 that they can commit concurrently with the network bulk transfer.
1384 For block-allocated files the filter allocation semaphore is held
1385 to avoid filesystem fragmentation during allocation. BKL lock
1386 removed for 2.6 xattr operations where it is no longer needed.
1391 Description: lconf incorrectly determined whether two IP networks could talk
1392 Details : In some more complicated routing and multiple-network
1393 configurations, lconf will avoid trying to make a network
1394 connection to a disjoint part of the IP space. It was doing the
1395 math incorrectly for one set of cases.
1400 Description: Fix for potential infinite loop processing records in an llog.
1401 Details : If an llog record is corrupted/zeroed, it is possible to loop
1402 forever in llog_process(). Validate the llog record length
1403 and skip the remainder of the block on error.
1406 Frequency : occasional (liblustre only)
1408 Description: liblustre could not open files whose last component is a symlink
1409 Details : sysio_path_walk() would incorrectly pass the open intent to
1410 intermediate path components.
1413 Frequency : rare (liblustre only with non-standard tuning)
1414 Bugzilla : 7201 (7350)
1415 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1416 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1417 after it has been released, only do this for referenced locks
1419 Severity : enhancement
1421 Description: specifying an (invalid) directory default stripe_size of -1
1422 would reset the directory default striping
1423 Details : stripe_size -1 was used internally to signal directory stripe
1424 removal, now use "all default" to signal dir stripe removal
1425 as a directory striping of "all default" is not useful
1428 Frequency : common for large clusters running liblustre clients
1430 Description: doing an ls when liblustre clients are running is slow
1431 Details : sending a glimpse AST to a liblustre client waits for every AST
1432 to time out, as liblustre clients will not respond. Since they
1433 cannot cache data we refresh the OST lock LVB from disk instead.
1435 Severity : enhancement
1437 Description: doing an ls at the same time as file IO can be slow
1438 Details : enqueue and other "small" requests can be blocked behind many
1439 large IO requests. Create a new OST IO portal for non-IO
1440 requests so they can be processed faster.
1443 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1445 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1446 created file with mode 000
1447 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1448 allow an ftruncate() as a non-root user to a file with mode 000.
1449 Lustre now respects this flag to disable mode checks when
1450 truncating a file owned by the user
1453 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1456 Description: Revoking locks from clients that went dead or catatonic might take
1458 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1459 cancellation of such locks instant on servers without waiting for
1460 any reply from clients. Clients drops these locks when cancel
1461 notification from server is received without replying.
1464 Frequency : liblustre-only, when liblustre client dies or becomes busy
1466 Description: Doing ls on Linux clients can take a long time with active
1468 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1469 granting such locks to it in the first place if possible. Locks
1470 are taken by proxy on the OST during the read or write and
1471 dropped immediately afterward. Add connect flags handling, do
1472 not grant locks to liblustre clients for glimpse ASTs.
1474 Severity : enhancement
1476 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1477 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1478 sequential reads and behaves badly with applications doing random
1479 reads. Improve it by reading ahead at least read region, and
1480 avoiding excessive large RPC for small reads.
1482 Severity : enhancement
1484 Description: Creating more than 1000 files for a single job may cause a load
1485 imbalance on the OSTs if there are also a large number of OSTs.
1486 Details : qos_prep_create() uses an OST index reseed value that is an
1487 even multiple of the number of available OSTs so that if the
1488 reseed happens in the middle of the object allocation it will
1489 still utilize the OSTs as uniformly as possible.
1494 Description: OST or MDS may oops in ping_evictor_main()
1495 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1496 but doesn't restart at beginning of obd_exports_timed list
1499 Severity : enhancement
1501 Description: improve by-nid export eviction on the MDS and OST
1502 Details : allow multiple exports with the same NID to be evicted at one
1503 time without re-searching the exports list.
1506 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1508 Description: MDS may oops in groups_free()
1509 Details : in rare race conditions a newly allocated group_info struct is
1510 freed again, and this can be NULL. The 2.4 compatibility code
1511 for groups_free() checked for a NULL pointer, but 2.6 did not.
1514 Frequency : common for liblustre clients doing little filesystem IO
1515 Bugzilla : 9352, 7313
1516 Description: server may evict liblustre clients accessing contended locks
1517 Details : if a client is granted a lock or receives a completion AST
1518 with a blocking AST already set it would not reply to the AST
1519 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1522 Frequency : lfs setstripe, only systems with more than 160 OSTs
1524 Description: unable to set striping with a starting offset beyond OST 160
1525 Details : llapi_create_file() incorrectly limited the starting stripe
1526 index to the maximum single-file stripe count.
1529 Frequency : LDAP users only
1531 Description: lconf did not handle in-kernel recovery with LDAP properly
1532 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1534 Severity : enhancement
1536 Description: bind OST threads to NUMA nodes to improve performance
1537 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1538 node and do their allocations there to localize memory access
1540 Severity : enhancement
1542 Description: llmount can determine client NID directly from Myrinet (GM)
1543 Details : the client NID code from gmnalnid was moved directly into
1544 llmount, removing the need to use this or specifying the
1545 client NID explicitly when mounting GM clients with zeroconf
1548 Frequency : if client is started with down MDS
1550 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1551 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1552 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1553 ptlrpc_import_delayed_req() because we didn't check if the
1554 request was interrupted, and we also didn't break out of the
1555 event loop if there was no timeout
1560 Description: data loss during non-page-aligned writes to a single file from
1561 both multiple nodes and multiple threads on one node at same time
1562 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1563 inconsistency led to false short-reads, that were cached and later
1564 used by ->prepare_write() to fill in partially written page,
1565 leading to data loss.
1568 Frequency : always, if lconf --abort_recovery used
1570 Description: lconf --abort_recovery fails with 'Operation not supported'
1571 Details : lconf was attempting to abort recovery on the MDT device and not
1574 Severity : enhancement
1576 Description: remove cleanup logs
1577 Details : replace lconf-generated cleanup logs with lustre internal
1578 cleanup routines. Eliminates the need for client-cleanup and
1581 Severity : enhancement
1583 Description: add support for EAs (user and system) on lustre filesystems
1584 Details : it is now possible to store extended attributes in the Lustre
1585 client filesystem, and with the user_xattr mount option it
1586 is possible to allow users to store EAs on their files also
1588 Severity : enhancement
1590 Description: Add possibility (config option) to show minimal available OST free
1592 Details : When compiled with --enable-mindf configure option, statfs(2)
1593 (and so, df) will return least minimal free space available from
1594 all OSTs as amount of free space on FS, instead of summary of
1595 free spaces of all OSTs.
1597 Severity : enhancement
1599 Description: do not expand extent locks acquired on OST-side
1600 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1601 by server: they are not cached anyway.
1604 Frequency : when mmap is used/binaries executed from Lustre
1606 Description: Unmmap pages before throwing them away from read cache.
1607 Details : llap_shrink cache now attempts to unmap pages before discarding
1608 them (if unmapping failed - do not discard). SLES9 kernel has
1609 extra checks that trigger if this unmapping is not done first.
1614 Description: lconf didn't resolve symlinks before checking to see whether a
1615 given mountpoint was already in use
1618 Frequency : when migrating failover services
1619 Bugzilla : 6395, 9514
1620 Description: When migrating a subset of services from a node (e.g. failback
1621 from a failover service node) the remaining services would
1622 time out and evict clients.
1623 Details : lconf --force (implied by --failover) sets the global obd_timeout
1624 to 5 seconds in order to quickly disconnect, but this caused
1625 other RPCs to time out too quickly. Do not change the global
1626 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1628 Severity : enhancement
1629 Frequency : if MDS is started with down OST
1630 Bugzilla : 9439,5706
1631 Description: Allow startup/shutdown of an MDS without depending on the
1632 availability of the OSTs.
1633 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1634 Add appropriate locking and lov-osc refcounts for safe
1635 cleaning. Add osc abort_inflight calls in case the
1636 synchronize never started.
1639 Frequency : occasional (Cray XT3 only)
1641 Description: root not authorized to access files in CRAY_PORTALS environment
1642 Details : The client process capabilities were not honoured on the MDS in
1643 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1644 been accessed by an authorized user then root was able to access
1645 the file on the local client also. The root user capabilities
1646 are now allowed on the MDS, as this environment has secure UID.
1649 Frequency : occasional
1651 Description: ldiskfs "too long searching" message happens too often
1652 Details : A debugging message (otherwise harmless) prints too often on
1653 the OST console. This has been reduced to only happen when
1654 there are fragmentation problems on the filesystem.
1659 Description: Division by zero in statfs when all OSCs are inactive
1660 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
1661 lov_statfs divides by value returned by lov_get_stripecnt().
1665 Bugzilla : 9489, 3273
1666 Description: First write from each client to each OST was only 4kB in size,
1667 to initialize client writeback cache, which caused sub-optimal
1668 RPCs and poor layout on disk for the first writen file.
1669 Details : Clients now request an initial cache grant at (re)connect time
1670 and so that they can start streaming writes to the cache right
1671 away and always do full-sized RPCs if there is enough data.
1672 If the OST is rebooted the client also re-establishes its grant
1673 so that client cached writes will be honoured under the grant.
1678 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
1679 Details : Now I/O RPCs go to different portal number and (presumably) fast
1680 lock requests (and glimses) and other RPCs get their own service
1681 threads pool that should be able to service those RPCs
1684 Severity : enhancement
1686 Description: Ability to exchange lustre version between client and servers and
1687 issue warnings at client side if client is too old. Also for
1688 liblustre clients there is ability to refuse connection of too old
1690 Details : New 'version' field is added to connect data structure that is
1691 filled with version info. That info is later checked by server and
1695 Frequency : rare, liblustre only.
1696 Bugzilla : 9296, 9581
1697 Description: Two simultaneous writes from liblustre at offset within same page
1698 might proceed at the same time overwriting eachother with stale
1700 Details : I/O lock withing llu_file_prwv was released too early, before data
1701 actually was hitting the wire. Extended lock-holding time until
1702 server acknowledges receiving data.
1705 Frequency : extremely rare. Never observed in practice.
1707 Description: avoid generating lustre_handle cookie of 0.
1708 Details : class_handle_hash() generates handle cookies by incrementing
1709 global counter, and can hit 0 occasionaly (this is unlikely, but
1710 not impossible, because initial value of cookie counter is
1711 selected randonly). Value of 0 is used as a sentinel meaning
1712 "unassigned handle" --- avoid it. Also coalesce two critical
1713 sections in this function into one.
1715 Severity : enhancement
1717 Description: allow liblustre clients to delegate truncate locking to OST
1718 Details : To avoid overhead of locking, liblustre client instructs OST to
1719 take extent lock in ost_punch() on client's behalf. New connection
1720 flag is added to handle backward compatibility.
1722 Severity : enhancement
1723 Bugzilla : 4928, 7341, 9758
1724 Description: allow number of OST service threads to be specified
1725 Details : a module parameter allows the number of OST service threads
1726 to be specified via "options ost ost_num_threads={N}" in the
1727 OSS's /etc/modules.conf or /etc/modprobe.conf.
1731 Bugzilla : 6146, 9635, 9895
1732 Description: servers crash with bad pointer in target_handle_connect()
1733 Details : In rare cases when a client is reconnecting it was possible that
1734 the connection request was the last reference for that export.
1735 We would temporarily drop the export reference and get a new
1736 one, but this may have been the last reference and the export
1737 was just destroyed. Get new reference before dropping old one.
1739 Severity : enhancement
1740 Frequency : if client is started with failover MDS
1742 Description: Allow multiple MDS hostnames in the mount command
1743 Details : Try to read the configuration from all specified MDS
1744 hostnames during a client mount in case the "primary"
1747 Severity : enhancement
1749 Description: Stop sending data to evicted clients as soon as possible.
1750 Details : Check if the client we are about to send or are sending data to
1751 was evicted already. (Check is done every second of waiting,
1752 for which l_wait_event interface was extended to allow checking
1753 of exit condition at specified intervals).
1756 Frequency : rare, normally only when NFS exporting is done from client
1758 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
1760 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
1763 Frequency : always, if srand() is called before liblustre initialization
1765 Description: Liblustre uses system PRNG disturbing its usage by user application
1766 Details : Introduce internal to lustre fast and high-quality PRNG for
1767 lustre usage and make liblustre and some other places in generic
1768 lustre code to use it.
1770 Severity : enhancement
1771 Bugzilla : 9477, 9557, 9870
1772 Description: Verify that the MDS configuration logs are updated when xml is
1773 Details : Check if the .xml configuration logs are newer than the config
1774 logs stored on the MDS and report an error if this is the case.
1775 Request --write-conf, or allow starting with --old_conf.
1777 Severity : enhancement
1779 Description: Handle symlinks in the path when checking if Lustre is mounted.
1780 Details : Resolve intermediate symlinks when checking if a client has
1781 mounted a filesystem to avoid duplicate client mounts.
1786 Description: lconf can hit an error exception but still return success.
1787 Details : The lconf command catches the Command error exception at the top
1788 level script context and will exit with the associated exit
1789 status, but doesn't ensure that this exit status is non-zero.
1794 Description: failure of ptlrpc thread startup can cause oops
1795 Details : Starting a ptlrpc service thread can fail if there are a large
1796 number of threads or the server memory is very fragmented.
1797 Handle this without oopsing.
1800 Frequency : always, only if liblustre and non-default acceptor port was used
1802 Description: liblustre cannot connect to servers with non-default acceptor port
1803 Details : tcpnal_set_default_params() was not called and was therefore
1804 ignoring the environment varaible TCPNAL_PORT, as well as other
1805 TCPNAL_ environment variables
1810 Description: two objects could be created on the same OST for a single file
1811 Details : If an OST is down, in some cases it was possible to create two
1812 objects on a single OST for a single file. No problems other
1813 than potential performance impact and spurious error messages.
1817 Bugzilla : 5681, 9562
1818 Description: Client may oops in ll_unhash_aliases
1819 Details : Client dcache may become inconsistent in race condition.
1820 In some cases "getcwd" can fail if the current directory is
1826 Description: Inode refcounting problems in NFS export code
1827 Details : link_raw functions used to call d_instantiate without obtaining
1828 extra inode reference first.
1832 Bugzilla : 9942, 9903
1833 Description: Referencing freed requests leading to crash, memleaks with NFS.
1834 Details : We used to require that call to ll_revalidate_it was always
1835 followed by ll_lookup_it. Also with revalidate_special() it is
1836 possible to call ll_revalidate_it() twice for the same dentry
1837 even if first occurence returned success. This fix changes semantic
1838 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
1839 reference on a request referred from the intent.
1840 ll_intent_release() then releases such a request.
1843 Frequency : rare, normally benchmark loads only
1845 Description: unlinked inodes were kept in memory on the client
1846 Details : If a client is repeatedly creating and unlinking files it
1847 can accumulate a lot of stale inodes in the inode slab cache.
1848 If there is no other client load running this can cause the
1849 client node to run out of memory. Instead flush old inodes
1850 from client cache that have the same inode number as a new inode.
1853 Frequency : SLES9 2.6.5 kernel and long filenames only
1854 Bugzilla : 9969, 10379
1855 Description: utime reports stale NFS file handle
1856 Details : SLES9 uses out-of-dentry names in some cases, which confused
1857 the lustre dentry revalidation. Change it to always use the
1861 Frequency : rare, unless heavy write-truncate concurrency is continuous
1862 Bugzilla : 4180, 6984, 7171, 9963, 9331
1863 Description: OST becomes very slow and/or deadlocked during object unlink
1864 Details : filter_destroy() was holding onto the parent directory lock
1865 while truncating+unlinking objects. For very large objects this
1866 may block other threads for a long time and slow overall OST
1867 responsiveness. It may also be possible to get a lock ordering
1868 deadlock in this case, or run out of journal credits because of
1869 the combined truncate+unlink. Solution is to do object truncate
1870 first in one transaction without parent lock, and then do the
1871 final unlink in a new transaction with the parent lock. This
1872 reduces the lock hold time dramatically.
1875 Frequency : rare, 2.4 kernels only
1877 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
1878 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
1879 before unmounting their filesystem in order to avoid deadlock.
1880 The kernel_locked() function in 2.4 kernels only checks whether
1881 the kernel lock is held, not whether it is this process that is
1882 holding it as 2.6 kernels do.
1887 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
1888 Details : The client ptlrpc code may be trying to reconnect to a down
1889 server before a previous connection attempt has timed out.
1890 Increase the reconnect interval to be longer than the connection
1891 timeout interval to avoid sending duplicate connections to
1895 Frequency : echo_client brw_test command
1897 Description: fix echo_client to work with OST preallocated code
1898 Details : OST preallocation code (5137) didn't take echo_client IO path
1899 into account: echo_client calls filter methods outside of any
1900 OST thread and, hence, there is no per-thread preallocated
1901 pages and buffers to use. Solution: hijack pga pages for IO. As
1902 a byproduct, this avoids unnecessary data copying.
1906 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
1907 Description: Client can oops in mdc_commit_close() after open replay
1908 Details : It was possible for the MDS to return an open request with no
1909 transaction number in mds_finish_transno() if the client was
1910 evicted, but without actually returning an error. Clients
1911 would later try to replay that open and may trip an assertion
1912 Simplify the client close codepath, and always return an error
1913 from the MDS in case the open is not successful.
1916 Frequency : rare, 2.6 OSTs only
1918 Description: OST may deadlock under high load on fragmented files
1919 Details : If there was a heavy load and highly-fragmented OST filesystems
1920 it was possible to have all the OST threads deadlock waiting on
1921 allocation of biovecs, because the biovecs were not released
1922 until the entire RPC IO was completed. Instead, release biovecs
1923 as soon as they are complete to ensure forward IO progress.
1925 Severity : enhancement
1927 Description: Support for specifying external journal device at mount
1928 Details : If an OST or MDS device is formatted with an external journal
1929 device, this device major/minor is stored in the ext3 superblock
1930 and may not be valid for failover. Allow detecting and
1931 specifying the external journal at mount time.
1936 Description: Mounting an MDS with pending unlinked files may cause oops
1937 Details : target_finish_recovery() calls mds_postrecov() which returned
1938 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
1939 considers this an error and immediately begins cleaning up the
1940 lov, just after starting the mds_lov process
1942 Severity : enhancement
1944 Description: Implement 'lfs df' to report actual free space on per-OST basis
1945 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
1946 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
1947 usage of objects; '-h' to report in human readable format.
1949 ------------------------------------------------------------------------------
1951 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
1958 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
1959 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
1960 at startup during orphan unlinking. Instead, skip buddy update
1961 before setup, as it will be regenerated after recovery is complete.
1966 Description: If an OST is inactive, its locks might reference stale inodes.
1967 Details : lov_change_cbdata() must iterate over all namespaces, even if
1968 they are inactive to clear inode references from the lock.
1970 Severity : enhancement
1971 Frequency : occasional, if non-standard max_dirty_mb used
1973 Description: Client will block write RPCs if not enough grant
1974 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
1975 then the client will block writes while waiting for another RPC
1976 to complete instead of consuming its dirty limit. With change
1977 we get improved performance when max_dirty_mb is small.
1979 Severity : enhancement
1980 Bugzilla : 3389, 6253
1981 Description: Add support for supplementary groups on the MDS.
1982 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
1983 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
1984 lookups for user supplementary groups into a cache.
1988 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
1989 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
1990 for liblustre O_EXCL is enforced by MDS.
1994 Description: readdir never returns NULL in liblustre.
1995 Details : Corrected llu_iop_getdirentries logic, to return offset of next
1996 dentry in struct dirent.
2000 Frequency : liblustre only, depends on application IO pattern
2001 Description: liblustre clients evicted if not contacting servers
2002 Details : Don't put liblustre clients into the ping_evictor list, so
2003 they will not be evicted by the pinger ever.
2005 Severity : enhancement
2007 Description: Add ability to evict clients by NID from MDS.
2008 Details : By echoing "nid:$NID" string into
2009 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2010 $NID would be instantly evicted from this MDS and from all active
2011 OSTs connected to it.
2015 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2016 Details : lookup_it_finish() used to query file size from OSTs that was not
2021 Description: service threads change working directory to that of init
2022 Details : Starting lustre service threads may pin the working directory
2023 of the parent thread, making that filesystem busy. Threads
2024 now change to the working directory of init to avoid this.
2028 Frequency : during shutdown only
2029 Description: shutdown with a failed MDS or OST can cause unmount to hang
2030 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2035 Frequency : 2.6 only
2036 Description: chmod/chown may include an extra supplementary group
2037 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2038 supplementary group and if none is specified this is used.
2041 Bugzilla : 5479 (6816)
2043 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2044 Details : If lookup is in progress on a file that is unlinked we might try
2045 to revalidate the inode and fail in revalidate after lookup is
2046 complete and ll_file_open() enqueues the open again but
2047 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2050 Frequency : always, if lconf --abort_recovery used
2052 Description: lconf --abort_recovery fails with 'Operation not supported'
2053 Details : lconf was attempting to abort recovery on the MDT device and not
2056 ------------------------------------------------------------------------------
2058 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2063 Frequency : rare (only unsupported configurations with a node running as an
2065 Bugzilla : 6514, 5137
2066 Description: Mounting a Lustre file system on a node running as an OST could
2068 Details : OSTs now preallocates memory needed to write out data at
2069 startup, instead of when needed, to avoid having to
2070 allocate memory in possibly low memory situations.
2071 Specifically, if the file system is mounted on on OST,
2072 memory pressure could force it to try to write out data,
2073 which it needed to allocate memory to do. Due to the low
2074 memory, it would be unable to do so and the node would
2075 become unresponsive.
2077 Severity : enhancement
2079 Description: Addition of lconf --service command line option
2080 Details : lconf now accepts a '--service <arg>' option, which is
2081 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2083 Severity : enhancement
2085 Description: Failover mode is now the default for OSTs.
2086 Details : By default, OSTs will now run in failover mode. To return to
2087 the old behaviour, add '--failout' to the lmc line for OSTs.
2089 Severity : enhancement
2091 Description: Health checks are now provided for MDS and OSTs
2092 Details : Additional detailed health check information on MSD and OSTs
2093 is now provided through the procfs health_check value.
2096 Frequency : occasional, depends on IO load
2098 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2099 after numerous create/delete cycles
2100 Details : The ext3 inode allocation policy would not allocate new inodes
2101 very well on the OSTs because there are no new directories
2102 being created. Instead we look for groups with free space if
2103 the parent directories are nearly full.
2108 Description: Network or server problems during mount may cause partially
2109 mounted clients instead of returning an error.
2110 Details : The config llog parsing code may overwrite the error return
2111 code during mount error handling, returning success instead
2117 Description: MDS can fail to allocate large reply buffers
2118 Details : After long uptimes the MDS can fail to allocate large reply
2119 buffers (e.g. zconf client mount config records) due to memory
2120 fragmentation or consumption by the buffer cache. Preallocate
2121 some large reply buffers so that these replies can be sent even
2122 under memory pressure.
2126 Frequency : rare (liblustre)
2127 Description: fsx running with liblustre complained that using truncate() to
2128 extend the file doesn't work. This patch corrects that issue.
2129 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2130 ATTR_SIZE and lsm use in llu_setattr_raw.
2134 Frequency : rare, only 2.6 kernels
2135 Description: Unusual file access patterns on the MDS may result in inode
2136 data being lost in very rare circumstances.
2137 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2138 code in the 2.6 kernel caused the inode and/or EA data not to
2139 be read from disk, causing single-file corruption.
2143 Frequency : rare, only 2.6 filesystems using extents
2144 Description: Heavy concurrent write and delete load may cause data corruption.
2145 Details : It was possible under high-load situations to have an extent
2146 metadata block in the block device cache from a just-unlinked
2147 file overwrite a newly-allocated data block. We now unmap any
2148 metadata buffers that alias just-allocated data blocks.
2152 Frequency : filesystems with default stripe_count larger than 77
2153 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2154 Details : lconf specifies an inode size of 4096 bytes when the default
2155 stripe_count is larger than 77. This conflicts with the default
2156 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2157 this case to avoid pinning too much memory for large EAs.
2159 ------------------------------------------------------------------------------
2161 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2166 Frequency : rare (extremely heavy IO load with hundreds of clients)
2168 Description: Client is evicted, gets IO error writing to file
2169 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2170 caused clients to be evicted for AST timeouts. The fixes in
2171 bug 5192 mean we no longer need to have such short AST timeouts
2172 so ldlm_timeout has been increased.
2175 Frequency : occasional during --force or --failover shutdown under load
2176 Bugzilla : 5949, 4834
2177 Description: Server oops/LBUG if stopped with --force or --failover under load
2178 Details : a collection of import/export refcount and cleanup ordering
2179 issues fixed for safer force cleanup
2182 Frequency : only filesystems larger than 120 OSTs
2183 Bugzilla : 5990, 6223
2184 Description: lfs getstripe would oops on a very large filesystem
2185 Details : lov_getconfig used kfree on vmalloc'd memory
2188 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2189 Bugzilla : 6242, 6243
2190 Description: reading from files that had been truncated to a non-zero size
2191 but never opened returned no data
2192 Details : ll_file_read() reads zeros from no-object files to EOF
2197 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2198 Details : There's little harm in aborting MDS/OSS recovery and letting it
2199 try again, so I removed the LASSERT and return an error instead.
2201 Severity : enhancement
2203 Description: New debugging infrastructure for tracking down data corruption
2204 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2205 (b) cover more of the client-side code path, and (c) try to narrow
2206 down where problems occurred
2210 Bugzilla : 3819, 4364, 4397, 6313
2211 Description: Racing close and eviction MDS could cause assertion in mds_close
2212 Details : It was possible to get multiple mfd references during close and
2213 client eviction, leading to one thread referencing a freed mfd.
2215 Severity: : enhancement
2216 Bugzilla : 3262, 6359
2217 Description: Attempts to reconnect to servers are now more aggressive.
2218 Details : This builds on the enhanced upcall-less recovery that was added
2219 in 1.4.2. When trying to reconnect to servers, clients will
2220 now try each server in the failover group every 10 seconds. By
2221 default, clients would previously try one server every 25 seconds.
2226 Description: After recovery, certain operations trigger a failed
2227 assertion on a client.
2228 Details : Failing over an mds, using lconf -d --failover, while a
2229 client was doing a readdir() call would cause the client to
2230 LBUG after recovery completed and the readdir() was resent.
2232 Severity : enhancement
2234 Description: Default groups are now added by lconf
2235 Details : You can now run lconf --group <servicename> without having to
2236 manually add groups with lmc.
2239 Frequency : occasional
2241 Description: Nodes with an elan id of 0 trigger a failed assertion
2244 Frequency : always when accessing e.g. tty/console device nodes
2246 Description: tty and some other devices nodes cannot be used on lustre
2247 Details : file's private_data field is used by device data and lustre
2248 values in there got lost. New field was added to struct file to
2249 store fs-specific private data.
2252 Frequency : when exporting Lustre via NFS
2254 Description: NFSD failed occasionally when looking up a path component
2255 Details : NFSD is looking up ".." which was broken in ext3 directories
2256 that had grown large enough to become hashed.
2259 Frequency : Clusters with multiple interfaces not on the same subnet
2261 Description: Nodes will repeatedly try to reconnect to an interface which it
2262 cannot reach and report an error to the log.
2263 Details : Extra peer list entries will be created by lconf with some peers
2264 unreachable. lconf now validates the peer before adding it.
2267 Frequency : Only if a default stripe is set on the filesystem root.
2269 Description: Setting a default stripe on the filesystem root prevented the
2270 filesystem from being remounted.
2271 Details : The client was sending extra request flags in the root getattr
2272 request and did not allocate a reply buffer for the dir EA.
2275 Frequency : occasional, higher if lots of files are accessed by one client
2276 Bugzilla : 6159, 6097
2277 Description: Client trips assertion regarding lsm mismatch/magic
2278 Details : While revalidating inodes the VFS looks up inodes with ifind()
2279 and in rare cases can find an inode that is being freed.
2280 The ll_test_inode() code will free the lsm during ifind()
2281 when it finds an existing inode and then the VFS later attaches
2282 this free lsm to a new inode.
2286 Bugzilla : 6422, 7030
2287 Description: MDS deadlock between mkdir and client eviction
2288 Details : Creating a new file via mkdir or mknod (starting a transaction
2289 and getting the ns lock) can deadlock with client eviction
2290 (gets ns lock and trying to finish a synchronous transaction).
2293 Frequency : occasional
2294 Description: While starting a server, the fsfilt_ext3 module could not be
2296 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2297 kernels. Previously, lconf would still use the ext3 name
2298 when trying to load modules. Now, it will correctly use
2299 ext3 on 2.4 and ldiskfs on 2.6.
2301 Severity : enhancement
2302 Description: The default stripe count has been changed to 1
2303 Details : The interpretation of the default stripe count (0, to lfs
2304 or lmc) has been changed to mean striping across a single
2305 OST, rather than all available. For general usage we have
2306 found a stripe count of 1 or 2 works best.
2308 Severity : enhancement
2309 Description: Add support for compiling against Cray portals.
2310 Details : Conditional compiling for some areas that are different
2314 Frequency : occasional
2315 Bugzilla : 6409, 6834
2316 Description: Creating files with an explicit stripe count may lead to
2317 a failed assertion on the MDS
2318 Details : If some OSTs are full or unavailable, creating files may
2319 trigger a failed assertion on the MDS. Now, Lustre will
2320 try to use other servers or return an error to the
2324 Frequency : occasional
2326 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2327 caused lock timeout during readahead (since 1.4.2).
2328 Details : Processes doing readahead might match a lock that hasn't been
2329 granted yet if there are overlapping and conflicting lock
2330 requests. The readahead process waits on ungranted lock
2331 (original lock is CBPENDING), while OST waits for that process
2332 to cancel CBPENDING read lock and eventually evicts client.
2334 Severity : enhancement
2336 Description: Initial enabling of flock support for clients
2337 Details : Implements fcntl advisory locking and file status functions.
2338 This feature is provided as an optional mount flag (default
2339 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2340 locking are implemented yet, and those that are are not guaranteed
2341 to be completely correct in production environments.
2342 mount -t lustre -o [flock|noflock] ...
2345 Frequency : occasional
2347 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2348 assertion in the ext3 JBD (journaling) layer.
2349 Details : The b_committed_data struct is protected by the big kernel lock
2350 in 2.4 kernels, serializing journal_commit_transaction() and
2351 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2352 access to this struct. In 2.6 kernels there is finer grained
2353 locking to improve SMP performance of the JBD layer.
2357 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2359 -----------------------------------------------------------------------------
2361 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2363 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2364 versions of Lustre. Please update all servers and clients to
2365 version 1.4.2 or later at the same time. You must also run
2366 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2367 to update the configuration logs.
2369 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2370 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2371 - swab llog records of type '0' so we get proper header size/idx (5861)
2372 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2373 - fix rename of one directory over another leaking an inode (5953)
2374 - avoid SetPageDirty on 2.6 (5981)
2375 - don't re-add just-being-destroyed locks to the waiting list (5653)
2376 - when creating new directories, inherit the parent's custom
2377 striping settings if present parent (3048)
2378 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2379 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2380 - don't hold client locks on temporary worklist from l_lru (5666)
2381 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2382 - automatically evict dead clients (5921)
2383 - Update file size properly in create+truncate+fstat case (6196)
2384 - Do not unhash mountpoint dentries, do not allow removal of
2386 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2387 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2388 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2389 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2390 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2391 - Correctly handle reads of files with no objects (6243)
2392 - lctl recover will also mark a device active if deactivate used (5933)
2394 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2395 - allow --write-conf on an MDS with different nettype than client (5619)
2396 - don't write config llogs to MDS for mounts not from that MDS (5617)
2397 - lconf should create multiple TCP connections from a client (5201)
2398 - init scripts are now turned off by default; run chkconfig --on
2399 lustre and chkconfig --on lustrefs to use them
2400 - upcalls are no longer needed for clients to recover to failover
2402 - add --abort-recovery option to lconf to abort recovery on device
2404 - add support for an arbitrary number of OSTs (3026)
2405 - Quota support protocol changes.
2406 - forward compatibility changes to wire structs (6007)
2407 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2408 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2409 - improved "lustre status" script
2410 - initialize blocksize for non-regular files (6062)
2411 - added --disable-server and --disable-client configure options (5782)
2412 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2413 - Vanilla 2.4.29 support
2414 - increase maximum number of obd devices to 520 (6242)
2415 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2416 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2417 SLES 9 2.6.5 kernels
2418 - limit stripes per file to 160 (the maximum EA size) (6093)
2420 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2423 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2424 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2425 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2426 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2427 - small ext3 extents cleanups and fixes (5733)
2428 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2429 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2430 - avoid needless client->OST connect, fix handle mismatch (5317)
2431 - fix DLM error path that led to out-of-sync client, long delays (5779)
2432 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2433 - fix several locking issues related to i_size (5492,5624,5654,5672)
2434 - don't move pending lock onto export if it is already evicted (5683)
2435 - fix kernel oops when creating .foo in unlinked directory (5548)
2436 - fix deadlock in obdfilter statistics vs. object create (5811)
2437 - use time_{before,after} to avoid timer jiffies wrap (5882)
2438 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2439 - Do not leak request if server was not able to process it (5154)
2440 - If mds_open unable to find parent dir, make that negative lookup(5154)
2441 - don't create new directories with extent-mapping (5909, 5936)
2443 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2444 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2445 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2446 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2447 - support for RHEL4 kernel on i686 (5773)
2448 - provide error messages when incompatible logs are encountered (5898)
2450 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2451 * version 1.4.0.10 (1.4.1 release candidate 1)
2453 - don't keep a lock reference when lock is not granted (4238)
2454 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2455 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2456 - drop import inflight refcount on signal_completed_replay error (5255)
2457 - unlock page after async write error during send (3677)
2458 - handle missing objects in filter_preprw_read properly (5265)
2459 - no transno return for symlink open, don't save no-trasno open (3440)
2460 - don't try to complete elan receive that already failed (4012)
2461 - free RPC server reply state on error (5406)
2462 - clean up thread from ptlrpc_start_thread() on error (5160)
2463 - readahead could read extra page into cache that wasn't ejected (5388)
2464 - prevent races in class_attach/setup/cleanup/detach (5260)
2465 - don't dereference de->d_inode after l_dput of de (5458)
2466 - use "int" for stripe value returned from lock_to_stripe (5544)
2467 - mballoc allocation and error-checking fixes in 2.6 (5504)
2468 - block device patches to fix I/O request sizes in 2.6 (5482)
2469 - look up hostnames for IB nals (5602)
2470 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2471 - don't start multiple acceptors for the same port (5277)
2472 - fix incorrect LASSERT in mds_getattr_name (5635)
2473 - export a proc file for general "ping" checking (5628)
2474 - fix "lfs check" to not block when the MDS is down (5628)
2476 - service request history (4965)
2477 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2478 - create an "evict_client" /proc entry on OSTs, like the MDS has
2479 - fix mount usage message, return errors per mount(8) (5168)
2480 - change grep [] to grep "[]" in tests so they work in more UMLs
2481 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2482 - remove ancient LOV_MAGIC_V0 EA support (5047)
2483 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2484 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2485 - allow readahead allocations to fail when low on memory (5383)
2486 - mmap locking landed again, after considerable improvement (2828)
2487 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2489 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2492 - send OST transaction number in read/write reply to free req (4966)
2493 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2494 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2495 - the watchdog thread now runs as interruptible (5246)
2496 - flock/lockf fixes (but it's still disabled, pending 5135)
2497 - don't use EXT3 constants in llite code (5094)
2498 - memory shortage at startup could cause assertion (5176)
2500 - reorganization of lov code
2501 - single portals codebase
2503 - add extents/mballoc support (5025)
2504 - direct I/O reads in the obdfilter (4048)
2505 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2507 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2510 - send OST transaction number in read/write reply to free req (4966)
2511 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2512 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2513 - the watchdog thread now runs as interruptible (5246)
2514 - handle missing objects in filter_preprw_read properly (5265)
2515 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2516 - drop import inflight refcount on signal_completed_replay error (5255)
2517 - unlock page after async write error during send (3677)
2518 - return original error code on reconstructed replies (3761)
2519 - no transno return for symlink open, don't save no-trasno open (3440)
2521 - add pid to ldlm debugging output (4922)
2522 - bump the watchdog timeouts -- we can't handle 30sec yet
2523 - extra debugging for orphan dentry/inode bug (5259)
2525 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2528 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2529 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2530 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2531 - fix NULL dereference in /proc/sys/portals/routes (4827)
2532 - allow failed mdc_close() operations to be interrupted (4561)
2533 - stop precreate on OST before MDS would time out on it (4778)
2534 - don't send partial-page writes before EOF from client (4410)
2535 - discard client grant for sub-page writes on large-page clients (4520)
2536 - don't free dentries not owned by NFS code, check generation (4806)
2537 - fix lsm leak if mds_create_objects() fails (4801)
2538 - limit debug_daemon file size, always print CERROR messages (4789)
2539 - use transno after validating reply (3892)
2540 - process timed out requests if import state changes (3754)
2541 - update mtime on OST during writes, return in glimpse (4829)
2542 - add mkfsoptions to LDAP (4679)
2543 - use ->max_readahead method instead of zapping global ra (5039)
2544 - don't interrupt __l_wait_event() during strace
2546 - add software watchdogs to catch hung threads quickly (4941)
2547 - make lustrefs init script start after nfs is mounted
2548 - fix CWARN/ERROR duplication (4930)
2549 - return async write errors to application if possible (2248)
2550 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2551 - print NAL number in %x format (4645)
2552 - update barely-supported suse-2.4.21-171 series (4842)
2553 - support for sles 9 %post scripts
2554 - support for building 2.6 kernel-source packages
2555 - support for sles km_* packages
2557 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2560 - ignore -ENOENT errors in osc_destroy (3639)
2561 - notify osc create thread that OSC is being cleaned up (4600)
2562 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2563 - reconstruct ost_handle() like mds_handle() (4657)
2564 - create a new thread to do import eviction to avoid deadlock (3969)
2565 - let lconf resolve symlinked-to devices (4629)
2566 - don't unlink "objects" from directory with default EA (4554)
2567 - hold socknal file ref over connect in case target is down (4394)
2568 - allow more than 32000 subdirectories in a single directory (3244)
2569 - fix blocks count for O_DIRECT writes (3751)
2570 - OST returns ENOSPC from object create when no space left (4539)
2571 - don't send truncate RPC if file size isn't changing (4410)
2572 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2573 - bind to privileged port in socknal and tcpnal (3689)
2575 - rate limit CERROR/CWARN console message to avoid overload (4519)
2576 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2577 - basic mmap support (3918)
2578 - kernel patch series update from b1_4 (4711)
2580 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2583 - avoid crash during MDS cleanup with OST shut down (2775)
2584 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2585 - don't use bad inodes on the MDS (3744)
2586 - dynamic object preallocation to improve recovery speed (4236)
2587 - don't hold spinlock over lock dumping or change debug flags (4401)
2588 - don't zero obd_dev when it is force cleaned (3651)
2589 - print grants to console if they go negative (4431)
2590 - "lctl deactivate" will stop automatic recovery attempts (3406)
2591 - look for existing locks in ldlm_handle_enqueue() (3764)
2592 - don't resolve lock handle twice in recovery avoiding race (4401)
2593 - revalidate should check working dir is a directory (4134)
2595 - don't always mark "slow" obdfilter messages as errors (4418)
2597 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2600 - don't close LustreDB during write_conf until it is done (3860)
2601 - fix typo in lconf for_each_profile (3821)
2602 - allow dumping logs from multiple threads at one time (3820)
2603 - don't allow multiple threads in OSC recovery (3812)
2604 - fix debug_size parameters (3864)
2605 - fix mds_postrecov to initialize import for llog ctxt (3121)
2606 - replace config semaphore with spinlock (3306)
2607 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2608 - don't allow enqueue to complete on a destroyed export (3822)
2609 - down write_lock before checking llog header bitmap (3825)
2610 - recover from lock replay timeout (3764)
2611 - up llog sem before sending rpc (3652)
2612 - reduce ns lock hold times when setting kms (3267)
2613 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2614 - fix NULL deref and obd_dev leak on setup error (3312)
2615 - replace some LBUG about llog ops with error handling (3841)
2616 - don't match INVALID dentries from d_lookup and spin (3784)
2617 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2618 - fix invalid assertion in ptlrpc_set_wait (3880)
2620 - add libwrap support for the TCP acceptor (3996)
2621 - add /proc/sys/portals/routes for non-root route listing (3994)
2622 - allow setting MDS UUID in .xml (2580)
2623 - print the stack of a process that LBUGs (4228)
2625 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2628 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2629 - make sure to unset replay flag from failed open requests (3440)
2630 - if default stripe count is 0, use OST count for inode size (3636)
2631 - update parent mtime/ctime on client for create/unlink (2611)
2632 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2633 - free recovery state on server during a forced cleanup (3571)
2634 - unregister_reply for resent reqs (3063)
2635 - loop back devices mounting and status check on 2.6 (3563)
2636 - fix resource-creation race that can provoke i_size == 0 (3513)
2637 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2638 - more debugging for page-accounting assertion (3746)
2639 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2640 - avoid deadlock after precreation failure (3758)
2641 - fix race and lock order deadlock in orphan handling (3450, 3750)
2642 - add validity checks when grabbing inodes from l_ast_data (3599)
2644 - add /proc/.../recovery_status to obdfilter (3428)
2645 - lightweight CDEBUG infrastructure, debug daemon (3668)
2646 - change default OSC RPC parameters to be better on small clusters
2647 - turn off OST read cache for files smaller than 32MB
2648 - install man pages and include them in rpms (3100)
2649 - add new init script for (un)mounting lustre filesystems (2593)
2650 - run chkconfig in %post for init scripts (3701)
2651 - drop scimac NAL (unmaintained)
2653 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
2656 - clean kiobufs before and after use (3485)
2657 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
2658 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
2659 - change init script to fail more clearly if not run as root (1528)
2660 - allow clients to reconnect during replay (1742)
2661 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
2662 - don't do DNS lookups on NIDs too small for IP addresses (3442)
2663 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
2664 - fix cond_resched (3554)
2665 - only evict unfinished clients after recovery (3515)
2666 - allow bulk resend, prevent data loss (3570)
2667 - dynamic ptlrpc request buffer allocation (2102)
2668 - don't allow unlinking open directory if it isn't empty (2904)
2669 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
2670 - remove extraneous obd dereference causing LASSERT failure (3334)
2671 - don't use get_cycles() when creating temp. files on the mds (3156)
2672 - hold i_sem when setting i_size in ll_extent_lock() (3564)
2673 - handle EEXIST for set-stripe, set proper directory name (3336)
2675 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
2676 - fix ksocknal_fmb_callback() error messages (2918)
2678 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
2681 - don't copy lvb into (possibly NULL) reply on error (2983)
2682 - don't deref dentry after dput, don't free lvb on error (2922)
2683 - use the kms to determine writeback rpc length (2947)
2684 - increment oti_logcookies when osc is inactive (2948)
2685 - update client's i_blocks count via lvb messages (2543)
2686 - handle intent open/close of special files properly (1557)
2687 - mount MDS with errors=remount-ro, like obdfilter (2009)
2688 - initialize lock handle to avoid ASSERT on error cleanup (3057)
2689 - don't use cancelling-locks' kms values (2947)
2690 - use highest lock extent for kms, not last one (2925)
2691 - don't dereference ERR_PTR() dentry in error handling path (3107)
2692 - fix thread race in portals_debug_dumplog() (3122)
2693 - create lprocfs device entries at setup instead of at attach (1519)
2694 - common AST error handler, don't evict client on completion race (3145)
2695 - zero nameidata in detach_mnt in 2.6 (3118)
2696 - verify d_inode after revalidate_special is valid in 2.6 (3116)
2697 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
2698 - initialize RPC timeout timer earlier for 2.6 (3219)
2699 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
2700 - print nal/nid for unknown nid (3258)
2701 - additional checks for oscc recovery before doing precreate (3284)
2702 - fix ll_extent_lock() error return code for 64-bit systems (3043)
2703 - don't crash in mdc_close for bad permissions on open (3285)
2704 - zero i_rdev for non-device files (3147)
2705 - clear page->private before handing to FS, better assertion (3119)
2706 - tune the read pipeline (3236)
2707 - fix incorrect decref of invalidated dentry (2350)
2708 - provide read-ahead stats and refine rpc in flight stats (3328)
2709 - don't hold journal transaction open across create RPC (3313)
2710 - update atime on MDS at close time (3265)
2711 - close LDAP connection when recovering to avoid server load (3315)
2712 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
2713 - don't leak open file on MDS after open resend (3325)
2714 - serialize filter_precreate and filter_destroy_precreated (3329)
2715 - loop device shouldn't call sync_dev() for nul device (3092)
2716 - clear page cache after eviction (2766)
2717 - resynchronize MDS->OST in background (2824)
2718 - refuse to mount the same filesystem twice on same mountpoint (3394)
2719 - allow llmount to create routes for mounting behind routers (3320)
2720 - push lock cancellation to blocking thread for glimpse ASTs (3409)
2721 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
2722 - fix rare problem with rename on htree directories (3417)
2724 - allow default OST striping configuration per directory (1414)
2725 - fix compilation for qswnal for 2.6 kernels (3125)
2726 - increase maximum number of MDS request buffers for large systems
2727 - change liblustreapi to be useful for external progs like lfsck (3098)
2728 - increase local configuration timeout for slow disks (3353)
2729 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
2731 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2734 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
2735 - don't overwrite extent policy data in reply if lock was blocked (2901)
2736 - drop filter export grants atomically with removal from device (2663)
2737 - del obd_self_export from work_list in class_disconnect_exports (2908)
2738 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
2739 - swab reply message in mdc_close, other PPC fixes (2464)
2740 - fix destroying of named logs (2325)
2741 - overwrite old logs when running lconf --write_conf (2264)
2742 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
2743 - fix race in target_handle_connect (2898)
2744 - mds_reint_create() should take same inode create lock (2926)
2745 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
2746 - don't close files for self_export to avoid uninitialized obd (2936)
2747 - allow MDS with the same name as client node (2939)
2748 - hold dentry reference for closed log files for unlink (2325)
2749 - reserve space for all logs during transactions (2059)
2750 - don't evict page beyond end of stripe extent (2925)
2751 - don't oops on a deleted current working directory (2399)
2752 - handle hard links to targets without a parent properly (2517)
2753 - don't dereference NULL lock when racing during eviction (2867)
2754 - don't grow lock extents when lots of conflicting locks (2919)
2756 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
2759 - account for cache space usage on clients to avoid data loss (974)
2760 - lfsck support in lustre kernel code (2349)
2761 - reduce journal credits needed for BRW writes (2370)
2762 - orphan handling to avoid losing space on client/server crashes
2763 - ptlrpcd can be blocked, stopping ALL progress (2477)
2764 - use lock value blocks to assist in proper KMS, faster stat (1021)
2765 - takes i_sem instead of DLM locks internally on obdfilter (2720)
2766 - recovery for initial connections (2355)
2767 - fixes for mds_cleanup_orphans (1934)
2768 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
2769 - block all file creations until orphan recovery completes (1901)
2770 - client remove rq_connection from request struct (2423)
2771 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
2772 - recovery timer race (2670)
2773 - mdc_close recovey bug (2532)
2774 - ptlrpc cleanup bug (2710)
2775 - mds timeout on local locks (2588)
2776 - namespace lock held during RPCs (2431)
2777 - handle interrupted sync write properly (2503)
2778 - don't try to handle a message that hasn't been replied to (2699)
2779 - client assert failure during cleanup after abort recovery (2701)
2780 - leak mdc device after failed mount (2712)
2781 - ptlrpc_check_set allows timedout requests to complete (2714)
2782 - wait for inflight reqs when ptlrpcd finishes (2710)
2783 - make sure unregistered services are removed from the srv_list
2784 - reset bulk XID's when resending them (caught by 1138 test)
2785 - unregister_bulk after timeout
2786 - fix lconf error (2694)
2787 - handle write after unfinished setstripe, stripe-only getstripe (2388)
2788 - readahead locks pages, leaves pending causing memory pressure (2673)
2789 - increase OST request buffers to 4096 on large machines (2729)
2790 - fix up permission of existing directories in simple_mkdir (2661)
2791 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
2792 - don't assign transno to errored transactions (2742)
2793 - don't delete objects on OST if given a bogus objid from MDS (2751)
2794 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
2795 - if rq_no_resend, then timeout request after recovery (2432)
2796 - fix MDS llog_logid record size, 64-bit array alignment (2733)
2797 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
2798 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
2799 - MDS recovery shouldn't delete objects that it has given out (2730)
2800 - if enqueue arrives after completion, don't clobber LVB (2819)
2801 - don't unlock pages twice when trigger_group_io returns error (2814)
2802 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
2803 - don't write pages to disk if there was an error (1450)
2804 - don't ping imports that have recovery disabled (2676)
2805 - take buffered bytes into account when balancing socknal conn (2817)
2806 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
2807 - reconnect unlink llog connection after MDS reconnects to OST (2816)
2808 - remove little-endian swabbing of llog records (1987)
2809 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
2810 - retry reposting request buffers if they fail (1191)
2811 - grow extent at grant time to avoid granting a revoked lock (2809)
2812 - lock revoke doesn't evict page if covered by a second lock (2765)
2813 - disable VM readahead to avoid reading outside lock extents (2805)
2815 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
2816 - updated kernel patches for hp-2.4.20 kernel (2681)
2818 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
2821 - fix truncated write corruption (2366)
2822 - fix for failed assertion in iopen_connect_dentry (1792,2517)
2824 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
2825 - ensure len(uuid) < 37 in lmc (1171)
2826 - fix ia64 OOPS in llog_test (2255)
2827 - zero end of page at obdfilter for partial page writes (2648)
2828 - don't leave stale dentries around after renames (bug 2428)
2829 - fix timeouts when evicting a client with a single lock held (2642)
2830 - set deadline for the initial HELLO message to drain (2634)
2831 - print out dotted-quad IP addresses in the socknal (2302)
2833 - additional debugging for MDS client eviction problem (2443)
2834 - fix mkfsoptions support for osts (2603, 2604)
2836 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
2839 - add series for the vanilla 2.6.0 kernel
2840 - add series for the vanilla 2.4.24 kernel
2841 - add series for a cray x86/64 UL kernel drop
2842 - fix xattr patches for the vanilla 2.4.19 series
2844 - generate true UUIDs in lmc (1171)
2845 - have portals stack dumping break in UML (2466)
2846 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
2847 - allocate with _NOFS in ldlm to avoid deadlock (1933)
2848 - wake callback waiting threads on client eviction (2460)
2849 - Add --ptldebug and --subsystem to lmc (1719)
2850 - update assertion to allow safe interrupt allocation
2851 - set rq_no_resend for cancel requests (2432)
2852 - recalculate ptlrpcd timeout after resend (2494)
2853 - call vfs_rmdir when removing pending directories (2368)
2854 - fix renaming a file to itself (2429)
2855 - lmc creates a default one-stripe lov (2454)
2856 - expand procfs space to handle large clusters (2326)
2857 - increase UML stack to avoid overflow
2858 - update lconf's list of debug and subsystem masks
2859 - fix lfs find --obd (2510)
2860 - /proc tunable for disabling filter read caching (2591)
2861 - stop rpm packages from altering slapd.conf (2301)
2862 - disable nagle in the socknal under 0conf (2578)
2863 - choose mds inode size based on stripe count (2572)
2864 - fix kernel-source rpm problems (2516)
2866 - add --disable-doc to avoid pdf generation (2421)
2867 - update documentation, tests, type-os, comments
2868 - avoid format warnings on ia64
2869 - remove the TOE NAL
2870 - tiny code cleanups by removing unused fields
2872 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
2875 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
2876 - remove the most blatant lies from BUILDING file (2371)
2877 - change default debug level to reasonable production setting
2878 - reduce client side cache size to reduce cache flush time
2879 - reduce max RPCs in flight to avoid unnecessary file fragmentation
2880 - make TCP zerocopy and pinger support enabled by default (2476)
2881 - sync writes completed after process exits caused crashes (2319)
2882 - maintain correct mount count on the MDS (2356)
2883 - backout 1557, because 2316 wasn't really fixed
2884 - better file I/O statistics gathering in /proc
2885 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
2886 - another kernel patch to fix zero-copy TCP function export
2887 - don't take duplicate lock when processing re-sent getattr (2420)
2888 - lctl uses obd_self_export instead of creating new conn (2353)
2889 - MDS/OST recovery case which requires object creation asserted (2425)
2890 - move lfs from /usr/sbin to /usr/bin in packages
2891 - fix race between mds_client_add and mds_client_free (2417)
2892 - use kmalloc instead of slabs in portals (2430)
2893 - don't create duplicate records when a failover MDS is present (2442)
2894 - remove unnecessary mount age check (2332)
2895 - don't remove directory inodes from locks prematurely (2451)
2896 - don't break if MDS service name is the same as hostname (2103)
2897 - fix races in client write RPC generation when cache full (2482)
2899 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
2902 - remove now-unused request->rq_obd (278)
2903 - if an allocation fails, print out how much memory we've used (1933)
2904 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
2905 - add the "configurable stack size" patch to most series files (1256)
2906 - ability to write large log records, for 100+ OST configs (2306)
2907 - fix NULL deref when filter_prep fails (2314)
2908 - fix operator precedence error in filter_sync
2909 - dynamic allocation of socknal TX descriptors (2315)
2910 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
2911 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
2912 - ensure that CWARN messages in Portals always get to the syslog
2913 - __init/__exit are not for prototype decls (ldlm_init/exit)
2914 - x86-64 compile warning fixes
2915 - fix gateway LMC keyword conflict (2318)
2916 - fix MDS lock inversions in getattr/reint paths (1844)
2917 - fix a rare lock re-ordering bug, which caused deadlock (2322)
2918 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
2919 - DLM race condition prevented some lock evictions (2328)
2920 - ENOMEM detection and retry on socknal sends (2230)
2921 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
2922 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
2923 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
2924 - better error messages when a client is rejected during recovery (1505)
2925 - avoid cancelling locks which were never granted, after failure (2330)
2926 - fix i_sem/journal inversion in mds_client_add (2333)
2927 - fix truncate/getattr lock cycle deadlock (2334)
2928 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
2929 - fix two rare exit paths which could leak an l_lock() ref (2321)
2930 - fencepost error in MDS/OST orphan recovery (2226)
2931 - make log record alignment 8 bytes (1988)
2932 - lstripe now fails when requested offset > ost_count (2237)
2933 - ensure that all kernel series have a complete list.h (1607)
2934 - fix crashes in special-file operations (2316)
2935 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
2937 - allow configurable automake binary, for testing new versions
2938 - small update to the lfs documentation
2940 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
2942 * fix negative export reference count in fsfilt_sync (2312)
2944 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
2945 * release candidate 0.9.1
2947 - orphans are moved into the PENDING directory for possible recovery
2948 - replayed opens now open by fid for orphan/rename safety (1042)
2949 - last close of an orphan inode generates a transno (683)
2950 - chdir() and mount() now pin the directory entry (1020)
2951 - avoid CERROR in normal ll_setattr_raw() error case (1500)
2952 - discard very old requests without processing them (1502)
2953 - remove some common, well-understood CERRORs (1505)
2954 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
2955 - clear "grant" flags in OST replies until OST grant code lands (1644)
2956 - fix read performance by not clobbering i_blksize on client (1598)
2957 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
2958 - make LRU size a /proc tunable, clears locks when reduced (707)
2959 - fix some lprocfs rot that prevented ptlbd from loading (1732)
2960 - server locks take references on exports now (1558)
2961 - build fixes for 2.4.20-rh trees (1663)
2962 - return an error from lov_create if all OSCs are inactive (1751)
2963 - fix import levels when a reconnect happens without a timeout (1597)
2964 - exit early from mds_open if we get a lookup error (1749)
2965 - partial page read at EOF wouldn't wait for disk before sending (1642)
2966 - avoid NULL deref in obdfilter when reading page past EOF (1592)
2967 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
2968 - fix LBUG in ll_it_open_error with rc = -2 (1861)
2969 - write/truncate lock inversion (1639)
2970 - Don't auto-load obdclass, portals modules during cleanup (1495)
2971 - fix timestamps from jumping to "now" (1763)
2972 - extra journal assertions (1648)
2973 - add an extra multiunlink test (1771)
2974 - fix read_record/write_record API (1776)
2975 - fix leak of offset_extent, possible incorrect i_size later (1772)
2976 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
2977 - Debugging for the kqswnal_get_idle_tx problems (1820)
2978 - Allow recovery to be attempted multiple times (1536)
2979 - Write out MDS last_rcvd file after it is first created (1600)
2980 - Fix tx_descriptor leak in failed transmit situations (1827)
2981 - ext3 journaling fixes for assertion failure after IO error (1871)
2982 - class_export_put() on freed export after completion AST error (1896)
2983 - Fix revalidate looping in VFS (1322)
2984 - Don't access a freed export during MDS_REINT timeout (1521)
2985 - Add open-unlink recovery support on the MDS (1673,1764)
2986 - Return an error if no MDS data was read from last_rcvd (1946)
2987 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
2988 - Don't setup a disk device that doesn't match exported UUID (317)
2989 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
2990 - avoid committing NULL handle in force close
2991 - local.sh is now a one-stripe LOV configuration
2992 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
2993 - fix ext3 htree duplicate directory entry corruption (1516)
2994 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
2995 - update to new LOV EA format (2097)
2996 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
2998 2003-06-15 Phil Schwan <phil@clusterfs.com>
3001 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3002 - per-import/export recovery handling (958, 931, 959)
3003 - multiple last-rcvd slots, for serving multiple FSes (949)
3004 - connections are again shared between multiple imp/exports (963, 964)
3005 - "umount -f" would hang if any requests needed to be sent (393, 978)
3006 - avoid pinning large req buffer by copying for queued messages (989)
3007 - add "uuid" to "lctl device" command to help upcalls (991)
3008 - "open" RPCs with transnos would confuse recovery counters (1037)
3009 - do proper endian conversion of all wire messages (288, 340, 891)
3010 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3011 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3012 - fix ptlrpc_connection leak in target_handle_connect (1174)
3013 - fix import refcounting bug in OST and MDS cleanup (1134)
3014 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3015 - fix very unlikely obd_types race condition (501)
3016 - remove osc_open hack for echo_client (1187)
3017 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3018 - a failure in read_inode2 leads to deadlock (1139)
3019 - cancel ack-locks as soon as transaction is committed (1072)
3020 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3021 - make sure to commitrw after any preprw to avoid deadlock (1162)
3022 - failing to execute a file in a lustre FS would lock inode (1203)
3023 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3024 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3025 - obd_uuid2tgt would walk off the end of the list (1255)
3026 - on IA64 the osc would give portals incorrect bulk size (1258)
3027 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3028 - fix lock inversion caused by new llite matching code (1282)
3029 - limit the number of dirty pages on a client to 10MB (1286)
3030 - timed out locks were not being corrected cancelled (1289)
3031 - fix O_DIRECT above 4GB on IA-32 (1292)
3032 * major user-visible changes
3033 - fail out/fail over policy now controlled by the upcall (993)
3035 - add OBD_PING to check server availability and failure (954)
3036 - lustre messages are now sent in sending host order (288, 340, 891)
3037 - add eadatalen to MDS getattr reply (340)
3038 - OST read replies may contain second buffer, with per-page status (593)
3040 2003-03-11 Phil Schwan <phil@clusterfs.com>
3043 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3044 - failed open()s could cause deadlock; fixed (867, 869)
3045 - stop cancelling OST locks when files are closed (481)
3046 - overlapping XID spaces caused network corruption (851, 853)
3047 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3048 - setattr_raw added, to do single-RPC, server-side setattrs
3049 - lmc/lconf syntax change for OST UUIDs
3050 - fix crashy race condition between ptlrpc_free_req and osc_close
3051 - don't use request in mdc_enqueue if we hit a timeout (889)
3052 - don't set the inode i_size for regular files from the MDS (896)
3053 - handle out of order completion AST (842)
3054 - don't LBUG if a lock request times out after receiving AST (913)
3055 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3056 - if a bad lock AST arrives, send an error instead of dropping entirely
3057 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3058 - fix leak in bulk IO when only partially completed (899, 900, 926)
3059 - fix O_DIRECT for ia64 (55)
3060 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3061 - C-z after timeout could hang a process forever; fixed (977)
3063 - client-side I/O cache (678, 924, 929, 941, 970)
3065 - READPAGE and SETATTRs which don't take server-side locks get
3068 2003-02-11 Phil Schwan <phil@clusterfs.com>
3071 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3072 - Fix fsfilt_extN_readpage() to read a full page of directory
3073 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3074 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3075 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3076 - Most intent operations were removed, in favour of a new RPC mode
3077 that does a single RPC to the server and bypasses most of the VFS
3078 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3079 - Aggressively cancel local locks on DLM servers
3080 - mds_reint_unlink sends EA to the client if it's the last nlink.
3081 client uses that EA to unlink OST objects.
3082 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3083 - recursive symlinks were fixed (439)
3084 - fixed NULL deref in DEBUG_REQ
3085 - filter_update_lastobjid no longer calls sync, which annoyed extN
3086 - fixed multi-client small-writes to a single file problem (445)
3087 - fixed mtime updates during file writes (607)
3088 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3089 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3090 - fixed Linux OST size reporting problem (444, 656)
3091 - OST now updates object mtime with writes or setattr (607, 619)
3092 - client verifies file size before zeroing page past EOF (445)
3093 - OST now writes last allocated objid to disk with allocation (108)
3094 - LOV on echo now works (409)
3096 - mds_reint_unlink sends a new buffer, with the EA included. this
3097 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3098 set if a regular file was being unlinked, and it was the last link
3099 - use PtlGet from the target for bulk writes (315)
3100 - OST now updates object mtime with writes or setattr (607, 619)
3101 - LDLM now has a grant-time callback to revalidate locked items, if
3103 - Many MDS operations were reorganized to combat race conditions
3105 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3106 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3108 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3111 - Fully reactivate OST imports after reconnection (512, others)
3112 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3113 - More graceful error handling for truncating on dead OST (515)
3114 - Don't error out unless we're actually accessing dead stripes (474)
3115 - Fix garbage sizes when stripes are missing (410)
3116 - LRU counters were broken, causing constant lock purge (433, 432)
3117 - garbage on read from stripes with failed OSTs (441)
3118 - mark OSCs as active before reconnecting during recovery (438)
3119 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3120 - lfind did not preserve OST order in output (443)
3121 - symlinks cause hung clients, incorrect data (439)
3122 - stop dereferencing request after dropping refcount (457)
3123 - don't LASSERT(spin_is_locked) on non-SMP (455)
3124 - fixes for many rename() bugs
3125 - fstat didn't correctly synchronize attributes (399)
3126 - server must handle lock cancellation during blocking AST prep (487)
3127 - bulk descriptors were free()d too soon (511)
3128 - fix paths in lconf, which would load incorrect modules (451, 507)
3129 - fix confusing lconf 'host not found' error message (386)
3130 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3131 - fix race condition in mdc_blocking_ast() for inode access (526)
3132 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3133 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3134 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3135 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3136 - fix lstripe in several ways (kernel side) (527)
3137 - fix request leak in ldlm_cli_enqueue (262)
3138 - incorrect OSC was marked inactive after OST failure
3139 - call mds_fs_cleanup before unmounting filesystem (524)
3140 - fix races between taking ns_lock and ldlm_lock_change_resource
3141 - fix races updating LOV export open file list
3142 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3143 - fix recovery NULL deref in ldlm_cli_cancel_unused
3144 - fix some DLM races by using new hash table for lock handles (419)
3145 - permit the client to specify desired inodes, at replay
3146 - duplicate requests when we queue them for replay reintegration
3147 - fix last_rcvd offset calculation
3148 - sync after each recovered transaction, so we always make progress
3149 - never, not always, ERESTART requests without transnos
3150 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3152 - skip replay if the MDS says that the client is already connected
3153 - don't check for a recovery-enabled export to match lctl's UUID
3154 - don't INC_USE_COUNT for phantom exports
3155 - don't crash when cleaning up phantom exports (567)
3156 - don't double-finish or set replay data for errored mdc_open requests
3157 - abort requests when they time out, so we don't get old replies
3158 - send/receive replies for AST messages again
3159 - if the client says that it doesn't have the lock, cancel it on the
3161 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3162 mark it as destroyed, it will all work out when decref is called
3163 - fix module use counts (22, 581)
3165 - ASTs now expect a reply (server cancels lock on error reply)
3167 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3170 - fix many simultaneous client startup (392)
3171 - fix dentry->d_it clobbering
3172 - credentials weren't being shipped for readdir/getattr operations
3173 - remove invalid assertions triggered during some concurrent MD
3175 - proper Lustre versions added (336, 389)
3176 - fix memory leak for create error case (398)
3177 - fix LOV locking bug that would get cli/srv out of sync
3178 - fix echo client over LOV (409)
3179 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3180 - fix double-O_EXCL intent crash (424)
3181 - avoid sending multiple lock CANCELs (352)
3183 - MDS can do multi-client recovery (modulo bugs in new code)
3185 - many updates, edits, cleanups
3187 2002-11-18 Phil Schwan <phil@clusterfs.com>
3190 - fix null d_it dereference (346)
3191 - fix full OST/dbench hang (333)
3192 - fix permission problem with file removal (286)
3193 - fix removal of OSCs from LOV when they fail
3194 - fix NULL deref during bulk timeout (214)
3195 - fix problems related to multiple filesystems on one MDS (241)
3196 - fixed serious subtle metadata locking bugs
3197 - free locks on clients when inodes are removed due to memory
3199 - fix inode pointer in lock data (285)
3200 - partial support for multiple MDS on a single host (241)
3201 - data locks weren't cancelled at clear_inode time (290, 311)
3202 - intent locks could lead to unbounded lock growth (205)
3203 - added a maximum lock count, an LRU list, and a flusher
3204 - fix multiple rename (365)
3205 - properly abstracted the echo client
3206 - OSC locked 1 byte too many; fixed
3207 - rewrote brw callback code:
3208 - fixed recovery bugs related to LOVs (306)
3209 - fixed too-many-pages-in-one-write crash (191)
3210 - fixed (again) crash in sync_io_timeout (214)
3211 - probably fixed callback-related race (385)
3213 - Add capability to MDS protocol
3214 - LDLM cancellations and callbacks on different portals
3216 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3219 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3220 - defer open object destruction to close time (601981, 138)
3221 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3222 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3223 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3224 - don't call obd_setattr in ll_file_release for destroyed objects
3225 * protocol change to lustre_msg: move |version| and add |flags|
3226 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3227 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3228 * added replay of create, unlink, link and rename operations during
3229 MDS failover; recovery should be much more robust now
3230 * remove failed OSCs from LOVs (only lov_create uses this so far)
3231 * the lustre-HOWTO was brought (more) up to date (582544)
3233 2002-10-23 Phil Schwan <phil@clusterfs.com>
3236 - in-use dentries weren't being reused properly (617851)
3237 - prevent multiple LDLM setup (599178)
3238 - fix LOV size calculations for truncate (617853)
3239 - fix client handling of MDS intent errors (POSIX)
3240 - fix permission bug in lovstripe.c test (624321)
3241 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3242 - truncate past end of file could corrupt data
3243 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3244 - a race in recovery could return ETIMEDOUT to apps (623947)
3245 - building outside the source directory was fixed
3246 * the lustre-HOWTO was brought (more) up to date (582544)
3247 * major progress was made on recovery functionality
3249 2002-10-10 Phil Schwan <phil@clusterfs.com>
3252 - recovery deadlock fix
3253 - rm -rf causes LBUG fix (617817)
3254 - file open by multiple tasks fix (618962)
3255 - directory permissions bugs (602707 and 620007)
3256 - journal_stop fixed with locking (611313)
3257 - O_APPEND failures resolved (618273, perhaps 614459)
3258 - lconf PATH fix (619770)
3259 - IA64 build fix (621450)
3260 - RPC buffer sizes scale with amount of memory
3262 2002-10-01 Phil Schwan <phil@clusterfs.com>
3265 - locks would be cancelled without throwing away data pages,
3266 resulting in inconsistent data (605627)
3267 - inode attributes were not always being refreshed (605627, 612449)
3268 - lconf now continues to cleanup after lctl reports an error
3269 - MDS now enforces user permissions (602707)
3270 - lprocfs cleanup fixed, but not yet enabled (614157)
3271 - fixed infinite server hang, should a client not respond to an AST
3272 - avoid going into recovery if user calls readlink() with a buffer
3273 that's too small (613941)
3274 - AST RPCs no longer require replies (614867) -- this may be changed
3275 - don't crash server if client sends an IOV that's too big (611336)
3276 - fixed lock conversion deadlock (611892)
3277 - fixed the following of symlinks (614622)
3278 * recovery: the server can remove locks from a client that dies, other
3279 clients can make progress
3280 * more extN patch fixes
3281 * compile-time configurable ptlrpc buffer allocations
3283 - collaborative read cache document
3284 - Lustre Lite Performance CDR document-in-progress
3286 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3289 - fix typo in patch-2.4.18
3291 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3294 - clear ptlrpc request each time in handle_incoming_request()
3295 - unlink of files now destroys the object on the OST
3297 2002-09-19 Peter Braam <braam@clusterfs.com>
3299 * add hard link support
3300 * change obdfile creation method
3301 * kernel patch changed
3303 2002-09-19 Peter Braam <braam@clusterfs.com>
3306 - stack overflow bug in extN fixed
3308 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3310 * documentation updates
3311 - add man pages for config tools
3312 - update tests/README to describe testing with new config tools
3313 - finish metadata API descriptions
3314 * bug fixes and cleanups
3315 - statfs workaround for 16TB limit
3316 - LOV stripe allocation improved, can stripe on subset of OSTs
3317 - LOV file size/IO offset was wrong for files > 4GB in size
3318 - object EA data was being dropped, caused files to be unreadable
3319 - memory overflow with non-LOV OST caused memory corruption
3320 - fixed regression tests to work with new config tools, obdfilter
3321 - fixed bug when directory size became larger than 1 block
3322 - fixed bug (for single client case) when PWD was deleted
3323 - invalidate local directory pages when doing intent-based ops
3324 - avoid LDLM oops when lock callback contained bad data
3326 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3328 * documentation updates
3329 * bug fixes and cleanups
3330 - configuration tools
3333 - 64-bit compile warnings
3334 - 64-bit internal statfs data
3336 * test_brw on persistent OST devices
3338 * lprocfs (disabled)
3340 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3342 * documentation updates
3343 * bug fixes and cleanups
3344 * configuration tools
3346 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3349 * many small fixes to 0_5_4
3350 * io/network handling
3351 * thinkos in MDS operations
3353 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3356 * crucial basic fixes to 0.5.3
3357 * IOR, Iozone work over Elan
3360 2002-08-07 Phil Schwan <phil@clusterfs.com>
3361 * version 0_5_3, our first alpha
3362 * we use the new Portals iovs
3363 * documentation updates
3364 * bug fixes and cleanups
3365 * small changes in the DLM wire protocol
3367 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3368 * version 0_5_1 with some initial stability,
3369 * locking on MD and file I/O.
3370 * documentation updates
3371 * several bug fixes since 0.5.0
3372 * small changes in wire protocol
3374 2002-07-18 Phil Schwan <phil@clusterfs.com>
3376 * delivered as Lustre Light Alpha
3377 * fixed a crash after handling invalid MDS requests
3378 * fixed directory pages for architectures with non-4k pages sizes
3380 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3381 * release version v0_4_4
3382 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3384 * Fixes a number of interruption problems with OST operations.
3385 * Update documentation for portals header changes
3386 * Move all wire protocol structs/defines to lustre_idl.h
3387 * Fixes symlink length bug.
3388 * Add tcpdump to repository.
3390 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3391 * release version v0_4_3
3392 * Fixes statfs for inodes on extN.
3393 * Fixes bug in runtests which would delete /etc/hosts.
3394 * Use 64-bit object IDs wherever possible (not into VFS though)
3395 Remove ost_get_info, which is unused by lustre, and out of date.
3397 2002-07-03 Peter Braam <braam@clusterfs.com>
3398 * release version v0_4_2 Fixes a lookup error (type not passed)
3399 * move forward to head of Portals
3400 * move forward to latest Lustre kernel
3402 2002-06-25 Peter Braam <braam@clusterfs.com>
3403 * release version v0_4_1. Hopefully stable on single node use.