1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4 2.4.21-47.0.1.EL (RHEL 3)
6 2.6.9-42.0.10.EL (RHEL 4)
7 2.6.12.6 vanilla (kernel.org)
8 2.6.16.27-0.9 (SLES 10)
9 * Client support for unpatched kernels:
10 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
11 2.6.16 - 2.6.19 vanilla (kernel.org)
12 2.6.9-42.0.8.EL (RHEL 4)
13 * Recommended e2fsprogs version: 1.39.cfs6
14 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
18 Frequency : liblustre clients only
20 Description: getdirentries does not give error when run on compute nodes
21 Details : getdirentries does not fail when the size specified as an argument
22 is too small to contain at least one entry
24 Severity : enhancement
26 Description: Add LNET router traceability for debug purposes
27 Details : If a checksum failure occurs with a router as part of the
28 IO path, the NID of the last router that forwarded the bulk data
29 is printed so it can be identified.
34 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
35 Details : Due to a race condition, liblustre clients were occasionally
38 Severity : enhancement
40 Description: lfs setstripe use optional parameters instead of postional
43 Severity : enhancement
45 Description: Nanosecond timestamp support for ldiskfs
46 Details : The on-disk ldiskfs filesystem has added support for nanosecond
47 resolution timestamps. There is not yet support for this at
48 the Lustre filesystem level.
51 Frequency : during server recovery
53 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
54 Details : request with rq_no_resend flag not awake l_wait_event if they get a
58 Frequency : nfs export on patchless client
60 Description: connectathon hang when test nfs export over patchless client
61 Details : Disconnected dentry cannot be found with lookup, so we do not need
62 to unhash it or make it invalid
65 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
66 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
67 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
72 Description: open req refcounting wrong on reconnect
73 Details : If reconnect happened between getting open reply from server and
74 call to mdc_set_replay_data in ll_file_open, we will schedule
75 replay for unreferenced request that we are about to free.
76 Subsequent close will crash in variety of ways.
77 Check that request is still eligible for replay in
78 mdc_set_replay_data().
83 Description: disable writes to filesystem when reading health_check file
84 Details : the default for reading the health_check proc file has changed
85 to NOT do a journal transaction and write to disk, because this
86 can cause reads of the /proc file to hang and block HA state
87 checking on a healthy but otherwise heavily loaded system. It
88 is possible to return to the previous behaviour during configure
89 with --enable-health-write.
91 --------------------------------------------------------------------------------
93 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
98 Frequency : on some architectures
100 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
101 Details : Uninitialized flags sometimes cause configuration commands to
104 --------------------------------------------------------------------------------
106 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
108 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
109 INTEROPERATE with older versions automatically. In many cases a
110 special upgrade step is needed. Please read the
111 user documentation before upgrading any part of a 1.4.x system.
112 * WARNING: Lustre configuration and startup changes are required with
113 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
115 * Support for kernels:
116 2.4.21-47.0.1.EL (RHEL 3)
118 2.6.9-42.0.10.EL (RHEL 4)
119 2.6.12.6 vanilla (kernel.org)
120 2.6.16.27-0.9 (SLES10)
121 * Client support for unpatched kernels:
122 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
123 2.6.16 - 2.6.19 vanilla (kernel.org)
124 2.6.9-42.0.8EL (RHEL 4)
125 * Recommended e2fsprogs version: 1.39.cfs6
126 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
129 Severity : enhancement
131 Description: MountConf
132 Details : Lustre configuration is now managed via mkfs and mount
133 commands instead of lmc and lconf. New obd types (MGS, MGC)
134 are added for dynamic configuration management. See
135 https://mail.clusterfs.com/wikis/lustre/MountConf for
138 Severity : enhancement
140 Description: dynamic OST addition
141 Details : OSTs can now be added to a live filesystem
143 Severity : enhancement
145 Description: startup order invariance
146 Details : MDTs and OSTs can be started in any order. Clients only
147 require the MDT to complete startup.
149 Severity : enhancement
151 Description: parallel, asynchronous orphan cleanup
152 Details : orphan cleanup is now performed in separate threads for each
153 OST, allowing parallel non-blocking operation.
155 Severity : enhancement
157 Description: optimized stripe assignment
158 Details : stripe assignments are now made based on ost space available,
159 ost previous usage, and OSS previous usage, in order to try
160 to optimize storage space and networking resources.
162 Severity : enhancement
164 Description: Permanently set tunables
165 Details : All writable /proc/fs/lustre tunables can now be permanently
166 set on a per-server basis, at mkfs time or on a live
169 Severity : enhancement
171 Description: Lustre message v2
172 Details : Add lustre message format v2.
174 Severity : enhancement
176 Description: client OST exclusion list
177 Details : Clients can be started with a list of OSTs that should be
178 declared "inactive" for known non-responsive OSTs.
182 Description: ENOENT returned for valid filehandle during dbench.
183 Details : Check if a directory has children when invalidating dentries
184 associated with an inode during lock cancellation. This fixes
185 an incorrect ENOENT sometimes seen for valid filehandles during
189 Frequency : SFS test only (otherwise harmless)
191 Description: SPEC SFS validation failure on NFS v2 over lustre.
192 Details : Changes the blocksize for regular files to be 2x RPC size,
193 and not depend on stripe size.
195 Severity : enhancement
197 Description: fine-grained SMP locking inside DLM
198 Details : Improve DLM performance on SMP systems by removing the single
199 per-namespace lock and replace it with per-resource locks.
201 Severity : enhancement
203 Description: don't hold multiple extent locks at one time
204 Details : To avoid client eviction during large writes, locks are not
205 held on multiple stripes at one time or for very large writes.
206 Otherwise, clients can block waiting for a lock on a failed OST
207 while holding locks on other OSTs and be evicted.
209 Severity : enhancement
211 Description: Multiple MD RPCs in flight.
212 Details : Further unserialise some read-only MDT RPCs - learn about intents.
213 To avoid overly-overloading MDT, introduce a limit on number of
214 MDT RPCs in flight for a single client and add /proc controls
215 to adjust this limit.
217 Severity : enhancement
219 Description: client read/write statistics
220 Details : Add client read/write call usage stats for performance
221 analysis of user processes.
222 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
223 file access. extents_stats shows chunk size distribution.
224 extents_stats_per_process show chunk size distribution per
227 Severity : enhancement
229 Description: per-client statistics on server
230 Details : Add ldlm and operations statistics for each client in
231 /proc/fs/lustre/mds|obdfilter/*/exports/
233 Severity : enhancement
235 Description: improved MDT statistics
236 Details : Add detailed MDT operations statistics in
237 /proc/fs/lustre/mds/*/stats
239 Severity : enhancement
241 Description: VFS operations stats
242 Details : Add client VFS call stats, trackable by pid, ppid, or gid
243 /proc/fs/lustre/llite/*/vfs_ops_stats
244 /proc/fs/lustre/llite/*/vfs_track_[pid|ppid|gid]
249 Description: Fix client-side osc byte counters
250 Details : The osc read/write byte counters in
251 /proc/fs/lustre/osc/*/stats are now working
254 Frequency : always as root on SLES
256 Description: Failure of copying files with lustre special EAs.
257 Details : Client side always return success for setxattr call for lustre
258 special xattr (currently only "trusted.lov").
263 Description: Refcount LNET uuids
264 Details : The global LNET uuid list grew linearly with every startup;
265 refcount repeated list entries instead of always adding to
268 Severity : enhancement
270 Description: Dynamic service threads
271 Details : Within a small range, start extra service threads
272 automatically when the request queue builds up.
275 Frequency : mixed-endian client/server environments
277 Description: mixed-endian crashes
278 Details : The new msg_v2 system had some failures in mixed-endian
281 Severity : enhancement
283 Description: Easy OST removal
284 Details : OSTs can be permanently deactivated with e.g. 'lctl
285 conf_param lustre-OST0001.osc.active=0'
287 Severity : enhancement
289 Description: MGS proc entries
290 Details : Added basic proc entries for the MGS showing what filesystems
293 Severity : enhancement
295 Description: provide MGS failover
296 Details : Added config lock reacquisition after MGS server failover.
298 Severity : enhancement
300 Description: add Linux 2.4 support
301 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
305 Description: a large application tries to do I/O to the same resource and dies
307 Details : Check the req->rq_arrival time after the call to
308 ost_brw_lock_get(), but before we do anything about
309 processing it & sending the BULK transfer request. This
310 should help move old stale pending locks off the queue as
311 quickly as obd_timeout.
314 Frequency : when an incorrect nid is specified during startup
316 Description: ptlrpc connect to non-existant node causes kernel crash
317 Details : LNET can't be re-entered from an event callback, which
318 happened when we expire a message after the export has been
319 cleaned up. Instead, hand the zombie cleanup off to another
322 Severity : enhancement
324 Description: plain/inodebits lock performance improvement
325 Details : Grouping plain/inodebits in granted list by their request modes
326 and bits policy, thus improving the performance of search through
330 Frequency : only if OST filesystem is corrupted
332 Description: client incorrectly hits assertion in ptlrpc_replay_req()
333 Details : for a short time RPCs with bulk IO are in the replay list,
334 but replay of bulk IOs is unimplemented. If the OST filesystem
335 is corrupted due to disk cache incoherency and then replay is
336 started it is possible to trip an assertion. Avoid putting
337 committed RPCs into the replay list at all to avoid this issue.
340 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
342 Description: System hang on startup
343 Details : This bug allowed the liblustre (e.g. catamount) client to
344 return to the app before handling all startup RPCs. This
345 could leave the node unresponsive to lustre network traffic
346 and manifested as a server ptllnd timeout.
348 Severity : enhancement
350 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
351 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
352 boolean to control whether to print peer debug info when a
353 client's RPC times out.
356 Frequency : only for kernels with patches from Lustre below 1.4.3
358 Description: Remove old rdonly API
359 Details : Remove old rdonly API which unused from at least lustre 1.4.3
362 Frequency : only for devices with external journals
364 Description: Set external device read-only also
365 Details : During a commanded failover stop, we set the disk device
366 read-only while the server shuts down. We now also set any
367 external journal device read-only at the same time.
370 Frequency : when upgrading from 1.4 while trying to change parameters
372 Description: The wrong (new) MDC name was used when setting parameters for
373 upgraded MDT's. Also allows changing of OSC (and MDC)
374 parameters if --writeconf is specified at tunefs upgrade time.
377 Frequency : when setting specific ost indicies
379 Description: QOS code breaks on skipped indicies
380 Details : Add checks for missing OST indicies in the QOS code, so OSTs
381 created with --index need not be sequential.
383 Severity : enhancement
385 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
386 Details : The uninit_groups feature works in conjunction with the kernel
387 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
388 up the pass1 processing of e2fsck. This is a read-only feature
389 in ldiskfs2 only, so older kernels and current ldiskfs cannot
390 mount filesystems that have had this feature enabled.
392 Severity : enhancement
394 Description: Improve multi-block allocation algorithm to avoid fragmentation
395 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
396 allocation locality and avoid filesystem fragmentation.
398 ------------------------------------------------------------------------------
400 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
402 * Support for kernels:
403 2.4.21-47.0.1.EL (RHEL 3)
405 2.6.9-42.0.10.EL (RHEL 4)
406 2.6.12.6 vanilla (kernel.org)
407 2.6.16.27-0.9 (SLES 10)
408 * Recommended e2fsprogs version: 1.39.cfs5
410 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
414 Frequency : occasional, depends on client load and configuration
415 Bugzilla : 12181, 12203
416 Description: data loss for recently-modified files
418 Details : In some cases it is possible that recently written or created
419 files may not be written to disk in a timely manner (this should
420 normally be within 30s unless client IO load is very high).
421 The problem appears as zero-length files or files that are a
422 multiple of 1MB in size after a client crash or client eviction
423 that are missing data at the end of the file.
425 This problem is more likely to be hit on clients where files are
426 repeatedly created and unlinked in the same directory, clients
427 have a large amount of RAM, have many CPUs, the filesystem has
428 many OSTs, the clients are rebooted frequently, and/or the files
429 are not accessed by other nodes after being written.
431 The presence of the problem can be detected by looking at
432 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
433 smaller than the second (nr_unused) then dirty files will not
434 be flushed automatically to disk. "sync; sleep 10" should be
435 run several times on the node before unmounting it to update
436 Lustre (this is also safe to run on nodes without this problem).
438 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
439 that can cause this same problem, so customers using that kernel
440 also need to update the kernel in addition to Lustre. In order
441 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
443 It is normal that files written just before a client crash (less
444 than 30s) may not yet have been flushed to disk, even for local
448 Frequency : frequent on thin XT3 nodes
450 Description: UUID collision on thin XT3 Linux nodes
451 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
452 randomly, since we relied on an insufficiently-seeded PRNG.
457 Description: OSS hangs after "All ost request buffers busy"
458 Details : A deadlock between quota and journal operations caused OSS
459 hangs after printing "All ost request buffers busy."
462 Frequency : always on liblustre builds
464 Description: Cleanup compiler warnings on liblustre
467 Frequency : always on liblustre builds on XT3
469 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
473 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
474 > 32000 subdirectories
475 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
476 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
477 subdirectory count crosses 32000. This will aid e2fsck to
478 correctly handle more than 32000 subdirectories.
481 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
483 Description: System hang on startup
484 Details : This bug allowed the liblustre (e.g. catamount) client to
485 return to the app before handling all startup RPCs. This
486 could leave the node unresponsive to lustre network traffic
487 and manifested as a server ptllnd timeout.
489 Severity : enhancement
491 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
492 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
493 boolean to control whether to print peer debug info when a
494 client's RPC times out.
499 Description: make O_SYNC working on 2.6 kernels
500 Details : 2.6 kernels use different method for mark pages for write,
501 so need add a code to lustre for O_SYNC work.
506 Description: Failure to close file and release space on NFS
507 Details : Put inode details into lock acquired in ll_intent_file_open.
508 Use mdc_intent_lock in ll_intent_open to properly
509 detect all kind of errors unhandled by mdc_enqueue.
514 Description: proc file read during shutdown sometimes raced obd removal,
516 Details : Add lock to prevent obd access after proc file removal.
519 Frequency : Only for files larger than 4GB on 32-bit clients.
521 Description: improperly doing page alignment of locks
522 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
523 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
528 Description: RPCs being resent when they shouldn't be
529 Details : Some RPCs that should not be resent are being resent. This
530 can cause inconsistencies in the RPC state machine. Do not
531 resend such requests.
534 Frequency : rare, only with NFS export
536 Description: Crash on NFS re-export node
537 Details : under very unusual load conditions an assertion is hit in
538 ll_intent_file_open()
541 Frequency : only if OST filesystem is corrupted
543 Description: client incorrectly hits assertion in ptlrpc_replay_req()
544 Details : for a short time RPCs with bulk IO are in the replay list,
545 but replay of bulk IOs is unimplemented. If the OST filesystem
546 is corrupted due to disk cache incoherency and then replay is
547 started it is possible to trip an assertion. Avoid putting
548 committed RPCs into the replay list at all to avoid this issue.
553 Description: large O_DIRECT requests fail under memory pressure/fragmentation
554 Details : Large single O_DIRECT read and write calls can fail to allocate
555 a sufficiently large buffer to process the request. In case of
556 allocation failure the allocation is retried with a smaller
557 buffer and broken into smaller requests.
559 Severity : enhancement
561 Description: Add -o localflock option to simulate old noflock behaviour.
562 Details : This will achieve local-only flock/fcntl locks coherentness.
567 Description: versioning check is incomplete
568 Details : Checking the version difference of client vs. server, report
569 error if the gap is too big.
574 Description: add support PG_writeback bit
575 Details : add support for PG_writeback bit for Lustre, for more carefull
576 work with page cache in 2.6 kernel. This also fix some deadlocks
577 and remove hack for work O_SYNC with 2.6 kernel.
579 Severity : enhancement
581 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
582 Details : The uninit_groups feature works in conjunction with the kernel
583 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
584 up the pass1 processing of e2fsck. This is a read-only feature
585 in ldiskfs2 only, so older kernels and current ldiskfs cannot
586 mount filesystems that have had this feature enabled.
588 Severity : enhancement
590 Description: Improve multi-block allocation algorithm to avoid fragmentation
591 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
592 allocation locality and avoid filesystem fragmentation.
594 ------------------------------------------------------------------------------
596 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
598 * Support for kernels:
599 2.6.9-42.0.3.EL (RHEL 4)
601 2.4.21-47.0.1.EL (RHEL 3)
602 2.6.12.6 vanilla (kernel.org)
603 2.6.16.21-0.8 (SLES10)
604 * Recommended e2fsprogs version: 1.39.cfs2-0
606 * The backwards-compatible /proc/sys/portals symlink has been removed
607 in this release. Before upgrading, please ensure that you change
608 any configuration scripts or /etc/sysctl.conf files that access
609 /proc/sys/portals/* or sysctl portals.* to use the corresponding
610 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
611 in advance of the upgrade on any system running Lustre 1.4.6 or
612 newer, since /proc/sys/lnet was added in that version.
613 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
617 Frequency : only when quota is used
619 Description: avoid scanning export list for quota master
620 Details : Change the algorithms to avoid scanning export list in order
621 to improve the efficiency.
624 Frequency : MDS failover only, very rarely
626 Description: "went back in time" messages on mds failover
627 Details : The greatest transno may be lost when the current operation
628 finishes with an error (transno==0) and the client's last_rcvd
629 record is over-written. Save the greatest transno in the
630 mds_last_transno for this case.
633 Frequency : always for specific kernels and striping counts
635 Description: client may get "Matching packet too big" without ACL support
636 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
637 when trying to access files in certain configurations. The
638 clients should in fact be denied when mounting because they do
642 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
644 Description: many clients connecting with IO in progress causes connect timeouts
645 Details : Avoid synchronous journal commits to avoid delays caused by many
646 clients connecting/disconnecting when bulk IO is in progress.
647 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
648 OST_IO_PORTAL to avoid delays behind potentially many pending
652 Frequency : occasionally with multiple writers to a single file
654 Description: shared writes to file may result in wrong size reported by stat()
655 Details : Allow growing of kms when extent lock is cancelled
658 Frequency : always with random mmap IO to multi-striped file
660 Description: mmap write might be lost if we are writing to a 'hole' in stripe
661 Details : Only if the hole is at the end of OST object so that kms is too
662 small. Fix is to increase kms accordingly in ll_nopage.
665 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
667 Description: writes to a missing object would leak memory on the OST
668 Details : If there is an inconsistency between the MDS and OST filesystems,
669 such that the MDS references an object that doesn't exist, writes
670 to that object will leak memory due to incorrect cleanup in the
671 error handling path, eventually running out of memory on the OST.
676 Description: Creating too long symlink causes lustre errors
677 Details : Check symlink and name lengths before sending requests to MDS.
680 Frequency : only if flock is enabled (not on by default)
682 Description: posix locks not released on fd closure on 2.6.9+
683 Details : We failed to add posix locks to list of inode locks on 2.6.9+
684 kernels, this caused such locks not to be released on fd close and
685 then assertions on fs unmount about still used locks.
688 Frequency : MDS failover only, very rarely
690 Description: clients may get ASSERTION(granted_lock != NULL)
691 Details : When request was taking a long time, and a client was resending
692 a getattr by name lock request. The were multiple lock requests
693 with the same client lock handle and
694 mds_getattr_name->fixup_handle_for_resent_request found one of the
695 lock handles but later failed with ASSERTION(granted_lock != NULL).
700 Description: handle->h_buffer_credits > 0, assertion failure
701 Details : h_buffer_credits is zero after truncate, causing assertion
702 failure. This patch extends the transaction or creates a new
706 Frequency : NFS re-export or patchless client
707 Bugzilla : 11179, 10796
708 Description: Crash on NFS re-export node (__d_move)
709 Details : We do not want to hash the dentry if we don't have a lock.
710 But if this dentry is later used in d_move, we'd hit uninitialised
711 list head d_hash, so we just do this to init d_hash field but
712 leave dentry unhashed.
715 Frequency : NFS re-export or patchless client
717 Description: NFS exports has problem with symbolic link
718 Details : lustre client didn't properly install dentry when re-exported
719 to NFS or running patchless client.
722 Frequency : NFS re-export or patchless client
724 Description: Various nfs/patchless fixes.
725 Details : fixes reuse disconected alias for lookup process - this fixes
726 warning "find_exported_dentry: npd != pd",
727 fix permission error with open files at nfs.
728 fix apply umask when do revalidate.
731 Frequency : occasional
733 Description: Crash on NFS re-export node
734 Details : calling clear_page() on the wrong pointer triggered oops in
735 generic_mapping_read().
738 Frequency : rarely, using O_DIRECT IO
740 Description: unaligned directio crashes client with LASSERT
741 Details : check for unaligned buffers before trying any requests.
744 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
746 Description: stale data returned from RAID cache
747 Details : If only a small amount of IO is done to the RAID device before
748 reading it again it is possible to get stale data from the RAID
749 cache instead of reading it from disk.
752 Frequency : always for sles10 kernel
754 Description: sles10 support
755 Details : ll_follow_link: compile fixes and using of nd_set_link
759 Frequency : depends on arch, kernel and compiler version, always on sles10
762 Description: recursive or deep enough symlinks cause stack overflow
763 Details : getting rid of large stack-allocated variable in
767 Frequency : depends on hardware
769 Description: lustre write performance loss in the SLES10 kernel
770 Details : the performance loss is caused by using of write barriers in the
771 ext3 code. The SLES10 kernel turns barrier support on by
772 default. The fix is to undo that change for ldiskfs.
775 ------------------------------------------------------------------------------
777 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
779 * Support for kernels:
780 2.6.9-42.0.3EL (RHEL 4)
782 2.4.21-47.0.1.EL (RHEL 3)
783 2.6.12.6 vanilla (kernel.org)
787 Frequency : quota enabled and large files being deleted
789 Description: releasing more than 4GB of quota at once hangs OST
790 Details : If a user deletes more than 4GB of files on a single OST it
791 will cause the OST to spin in an infinite loop. Release
792 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
797 Description: statfs data retrieved from /proc may be stale or zero
798 Details : When reading per-device statfs data from /proc, in the
799 {kbytes,files}_{total,free,avail} files, it may appear
800 as zero or be out of date.
803 Frequency : systems with MD RAID1 external journal devices
805 Description: lconf's call to blkid is confused by RAID1 journal devices
806 Details : Use the "blkid -l" flag to locate the MD RAID device instead
807 of returning all block devices that match the journal UUID.
810 Frequency : always, for aggregate stripe size over 4GB
812 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
813 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
814 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
817 Frequency : always if "lfs find" used on a local file/directory
819 Description: "lfs find" segfaults if used on a local file/directory
820 Details : The case where a directory component was not specified wasn't
821 handled correctly. Handle this properly.
824 Frequency : always on ppc64
826 Description: the write to an ext3 filesystem mounted with mballoc got stuck
827 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
828 perform endianness conversion.
831 Frequency : rarely (truncate to non-zero file size after write under load)
832 Bugzilla : 10730, 10687
833 Description: Files padded with zeros to next 4K multiple
834 Details : With filesystems mounted using the "extents" option (2.6 kernels)
835 it is possible that files that are truncated to a non-zero size
836 immediately after being written are filled with zero bytes beyond
837 the truncated size. No file data is lost.
839 Severity : enhancement
841 Description: Allow recovery/failover for liblustre clients.
842 Details : liblustre clients were unaware of failover configurations until
845 Severity : enhancement
847 Description: user file locks should fail when not mounting with flock option
848 Details : Set up an error-returning stub in ll_file_operations.lock field
849 to prevent incorrect behaviour when client is mounted without
850 flock option. Also, set up properly f_op->flock field for
854 Frequency : always on ia64
856 Description: "lfs df" loops on printing out MDS statfs information
857 Details : The obd_ioctl_data was not initialized and in some systems
858 this caused a failure during the ioctl that did not return
859 an error. Initialize the struct and return an error on failure.
862 Frequency : SLES 9 only
864 Description: Error of copying files with lustre special EAs as root
865 Details : Client side always return success for setxattr call for lustre
866 special xattr (currently only "trusted.lov").
869 Frequency : rarely on clusters with both ia64+i386 clients
871 Description: ia64+i686 clients doing shared IO on the same file may LBUG
872 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
873 clients are doing concurrent writes to the same file it is
874 possible that the ia64 clients may LASSERT because the OST
875 extent locks are not PAGE_SIZE aligned. Ensure that grown
876 locks are always aligned on the request boundary.
879 Frequency : specific use, occasional
881 Description: Overwriting in use executable truncates on-disk binary image
882 Details : If one node attempts to overwrite an executable in use by
883 another node, we now correctly return ETXTBSY instead of
886 Severity : enhancement
888 Description: Async OSC create to avoid the blocking unnecessarily.
889 Details : If a OST has no remain object, system will block on the creating
890 when need to create a new object on this OST. Now, ways use
891 pre-created objects when available, instead of blocking on an
892 empty osc while others are not empty. If we must block, we block
893 for the shortest possible period of time.
898 Description: chmod on Lustre root is propagated to other clients
899 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
900 invalid by the follow_mount time.
903 Frequency : liblustre clients only
905 Description: Race in 'instant cancel' lock handling could lead to such locks
906 never to be granted in case of SMP MDS
907 Details : Do not destroy not yet granted but cbpending locks in
911 Frequency : replay/resend of open
913 Description: non null lock assetion failure in mds_intent_policy
914 Details : Trying to replay/resend lockless open requests resulted in
915 mds_open() returning 0 with no lock. Now it sets a flag if
916 a lock is going to be returned.
918 Severity : enhancement
920 Description: Checksum enhancements
921 Details : New checksum enhancements allow for resending RPCs that failed
924 Severity : enhancement
926 Description: Tunables on number of dirty pages in cacche
927 Details : Allow to set limit on number of dirty pages cached.
932 Description: client crash on unmount - lock still has references
933 Details : In some error handling cases it was possible to leak a lock
934 reference on a client while accessing a file. This was not
935 harmful to the client during operation, but would cause the
936 client to crash when the filesystem is unmounted.
939 Frequency : specific case, rare
941 Description: ETXTBSY on mds though file not in use by client
942 Details : ETXTBSY is no longer incorrectly returned when attempting to
943 chmod or chown a directory that the user previously tried to
944 execute or a currently-executing binary.
947 Frequency : extremely rare except on liblustre-based clients
949 Description: Lustre space not freed when files are deleted
950 Details : Clean up open-unlinked files after client eviction. Previously
951 the unlink was skipped and the files remained as orphans.
956 Description: OST failure "would be an LBUG" in waiting_locks_callback()
957 Details : In some cases it was possible to send a blocking callback to
958 a client doing a glimpse, even though that client didn't get
959 a lock granted. When the glimpse lock is cancelled on the OST
960 the freed lock is left on the waiting list and corrupted the list.
963 Frequency : all core dumps
965 Description: Broke core dumps to lustre
966 Details : Negative dentry may be unhashed if parent does not have UPDATE
967 lock, but some callers, e.g. do_coredump, expect dentry to be
968 hashed after successful create, hash it in ll_create_it.
970 ------------------------------------------------------------------------------
972 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
974 * Support for kernels:
975 2.6.9-42.0.2.EL (RHEL 4)
977 2.4.21-40.EL (RHEL 3)
978 2.6.12.6 vanilla (kernel.org)
982 Frequency : always on RHEL 3
984 Description: Number of open files grows over time
985 Details : The number of open files grows over time, whether or not
986 Lustre is started. This was due to a filp leak introduced
987 by one of our kernel patches.
989 ------------------------------------------------------------------------------
991 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
993 * Support for kernels:
996 2.4.21-40.EL (RHEL 3)
997 2.6.12.6 vanilla (kernel.org)
1002 Bugzilla : 5719, 9635, 9792, 9684
1003 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1004 Details : If a server is under heavy load and cannot reply to new
1005 connection requests before the client resends the (re)connect,
1006 the connection handling code can behave badly if two service
1007 threads are concurrently handing separate (re)connections from
1008 the same client. Add better locking to the connection handling
1009 code, and ensure that only a single connection will be processed
1010 for a given client UUID, even if the lock is dropped.
1012 Severity : enhancement
1014 Description: add TCP zero-copy support to kernel
1015 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1016 sends if the hardware supports scatter-gather and checksumming.
1017 This allows socklnd to do client-write and server-read more
1018 efficiently and reduce CPU utilization from skbuf copying.
1021 Frequency : only if NFS exporting from client
1023 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1024 Details : If setting attributes on a file created under NFS that had
1025 never been opened it would be possible to oops the client
1026 if the file had no objects.
1030 Bugzilla : 9326, 10402, 10897
1031 Description: client crash in ptlrpcd_wake() thread when sending async RPC
1032 Details : It is possible that ptlrpcd_wake() dereferences a freed async
1033 RPC. In rare cases the ptlrpcd thread alread processed the RPC
1034 before ptlrpcd_wake() was called and the request was freed.
1037 Frequency : always for liblustre
1039 Description: liblustre client does MDS+OSTs setattr RPC for each write
1040 Details : When doing a write from a liblustre client, the client
1041 incorrectly issued an RPC to the MDS and each OST the file was
1042 striped over in order to update the timestamps. When writing
1043 with small chunks and many clients this could overwhelm the MDS
1044 with RPCs. In all cases it would slow down the write because
1045 these RPCs are unnecessary.
1047 Severity : enhancement
1049 Description: allow number of MDS service threads to be changed at module load
1050 Details : It is now possible to change the number of MDS service threads
1051 running. Adding "options mds mds_num_threads={N}" to the MDS's
1052 /etc/modprobe.conf will set the number of threads for the next
1053 time Lustre is restarted (assuming the "mds" module is also
1054 reloaded at that time). The default number of threads will
1055 stay the same, 32 for most systems.
1060 Description: OST crash if filesystem is unformatted or corrupt
1061 Details : If an OST is started on a device that has never been formatted
1062 or if the filesystem is corrupt and cannot even mount then the
1063 error handling cleanup routines would dereference a NULL pointer.
1068 Description: NULL pointer deref in llap_from_page.
1069 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1070 incorrect) mapping if the page was truncated/reclaimed while it was
1071 searched for. Check for this condition and skip such pages when
1072 doing readahead. Introduce extra check to llap_from_page() to
1073 verify page->mapping->host is non-NULL (so page is not anonymous).
1076 Frequency : Sometimes when using sys_sendfile
1078 Description: "page not covered by a lock" warnings from ll_readpage
1079 Details : sendfile called ll_readpage without right page locks present.
1080 Now we introduced ll_file_sendfile that does necessary locking
1081 around call to generic_file_sendfile() much like we do in
1085 Frequency : with certain MDS communication failures at client mount time
1087 Description: NULL pointer deref after failed client mount
1088 Details : a client connection request may delayed by the network layer
1089 and not be sent until after the PTLRPC layer has timed out the
1090 request. If the client fails the mount immediately it will try
1091 to clean up before the network times out the request. Add a
1092 reference from the request import to the obd device and delay
1093 the cleanup until the network drops the request.
1096 Frequency : occasionally during client (re)connect
1098 Description: assertion failure during client (re)connect
1099 Details : processing a client connection request may be delayed by the
1100 client or server longer than the client connect timeout. This
1101 causes the client to resend the connection request. If the
1102 original connection request is replied in this interval, the
1103 client may trip an assertion failure in ptlrpc_connect_interpret()
1104 which thought it would be the only running connect process.
1107 Frequency : only with obd_echo servers and clients that are rebooted
1109 Description: kernel BUG accessing uninitialized data structure
1110 Details : When running an obd_echo server it did not start the ping_evictor
1111 thread, and when a client was evicted an uninitialized data
1112 structure was accessed. Start the ping_evictor in the RPC
1113 service startup instead of the OBD startup.
1115 Severity : enhancement
1116 Bugzilla : 10193 (patchless)
1117 Description: Remove dependency on various unexported kernel interfaces.
1118 Details : No longer need reparent_to_init, exit_mm, exit_files,
1119 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1122 Frequency : rare (only users of deprecated and unsupported LDAP config)
1124 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1125 Details : LDAP apparently contains 'lustreName' attributes instead of
1126 'name'. A simple remapping of the name is sufficient.
1129 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1131 Description: waiting_locks_callback trips kernel BUG if client is evicted
1132 Details : Running with the dump_on_timeout debug flag turned on makes
1133 it possible that the waiting_locks_callback() can try to dump
1134 the Lustre kernel debug logs from an interrupt handler. Defer
1135 this log dumping to the expired_lock_main() thread.
1137 Severity : enhancement
1139 Description: Support NFS exporting on 2.6 kernels.
1140 Details : Implement non-rawops metadata methods for NFS server to use without
1141 changing NFS server code.
1144 Frequency : very rare (synthetic metadata workload only)
1146 Description: two racing renames might cause an MDS thread to deadlock
1147 Details : Running the "racer" program may cause one MDS thread to rename
1148 a file from being the source of a rename to being the target of
1149 a rename at exactly the same time that another thread is doing
1150 so, and the second thread has already enqueued these locks after
1151 doing a lookup of the target and is trying to relock them in
1152 order. Ensure that we don't try to re-lock the same resource.
1155 Frequency : only very large systems with liblustre clients
1157 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1158 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1159 all OSTs in parallel. This allows the eviction of stale liblustre
1160 clients to proceed much faster than if they were done in series,
1161 and also offers similar improvements for other set_info RPCs.
1166 Description: excessive CPU usage during initial read phase on client
1167 Details : During the initial read phase on a client, it would agressively
1168 retry readahead on the file, consuming too much CPU and impacting
1169 performance (since 1.4.5.8). Improve the readahead algorithm
1170 to avoid this, and also improve some other common cases (read
1171 of small files in particular, where "small" is files smaller than
1172 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1177 Description: MDS crash when receiving packet with unknown intent.
1178 Details : Do not LBUG in unknown intent case, just return -EFAULT
1180 Severity : enhancement
1181 Bugzilla : 9293, 9385
1182 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1183 Details : Do not serialize getattr (non-intent version) and statfs.
1186 Frequency : occasional, when OST network is overloaded/intermittent
1188 Description: client evicted by OST after bulk IO timeout
1189 Details : If a client sends a bulk IO request (read or write) the OST
1190 may evict the client if it is unresposive to its data GET/PUT
1191 request. This is incorrect if the network is overloaded (takes
1192 too long to transfer the RPC data) or dropped the OST GET/PUT
1193 request. There is no need to evict the client at all, since
1194 the pinger and/or lock callbacks will handle this, and the
1195 client can restart the bulk request.
1198 Frequency : Always when mmapping file with no objects
1200 Description: client crashes when mmapping file with no objects
1201 Details : Check that we actually have objects in a file before doing any
1202 operations on objects in ll_vm_open, ll_vm_close and
1208 Description: Request leak when working with deleted CWD
1209 Details : Introduce advanced request refcount tracking for requests
1210 referenced from lustre intent.
1212 Severity : Enhancement
1214 Description: Cache open file handles on client.
1215 Details : MDS now will return special lock along with openhandle, if
1216 requested and client is allowed to hold openhandle, even if unused,
1217 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1218 closing files for every read/write openration.
1220 Severity : Enhancement
1222 Description: Cache open negative dentries on client when possible.
1223 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1224 negative dentries on lock revocation.
1229 Description: Remounting a client read-only wasn't possible with a zconf mount
1230 Details : It wasn't possible to remount a client read-only with llmount.
1232 Severity : enhancement
1233 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1234 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1235 MPICH is included to add Lustre-specific ADIO interfaces.
1236 This is based closely on the UFS ADIO layer and only differs
1237 in file creation, in order to allow the OST striping to be set.
1238 This is user-contributed code and not supported by CFS.
1243 Description: extended inode attributes (immutable, append-only) work improperly
1244 when 2.4 and 2.6 kernels are used on client/server or vice versa
1245 Details : Introduce kernel-independent values for these flags.
1247 Severity : enhancement
1250 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1251 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1252 granularity. Now, Fractional MB granularity is be supported,
1253 this is very useful for low memory system.
1255 Severity : enhancement
1257 Description: Getattr by fid
1258 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1259 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1260 have locks and avoids getattr rpc requests for it on every stat.
1263 Frequency : Always, for filesystems larger than 2TB
1265 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1266 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1267 filesystems larger than 2TB to be mounted with mballoc enabled.
1270 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1272 Description: filesystem corruption for non-standard kernels and very large OSTs
1273 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1274 filesystems larger than 2TB is mounted then the kernel will
1275 silently corrupt the start of the filesystem. CONFIG_LBD is
1276 enabled for all CFS-supported kernels, but the possibility of
1277 this happening with a modified kernel config exists.
1279 Severity : enhancement
1281 Description: add client O_DIRECT support for 2.6 kernels
1282 Details : It is now possible to do O_DIRECT reads and writes to files
1283 in the Lustre client mountpoint on 2.6 kernel clients.
1285 Severity : enhancement
1287 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1288 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1289 parallel, not waiting for response from every OST before sending
1290 a rpc to the next OST.
1295 Description: setattr vs write race when updating file timestamps
1296 Details : Client processes that update a file timestamp into the past
1297 right after writing to the file (e.g. tar) it is possible that
1298 the updated file modification time can be reset to the current
1299 time due to a race between processing the setattr and write RPC.
1301 Severity : enhancement
1303 Description: Bring 'lfs find' closer in line with regular Linux find.
1304 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1305 -print0 options and obtains all the needed info through the lustre
1308 Severity : enhancement
1310 Description: support up to 1024 configured devices on one node
1311 Details : change obd_dev array from statically allocated to dynamically
1312 allocated structs as they are first used to reduce memory usage
1317 Description: Flush dirty partially truncated pages during truncate
1318 Details : Immediatelly flush partially truncated pages in filter_setattr,
1319 this way we completely avoid having any pages in page cache on OST
1320 and can retire ugly workarounds during writes to flush such pages.
1325 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1326 Details : protect inode from truncation within vfs_unlink() context
1327 just take a reference before calling vfs_unlink() and release it
1328 when parent's i_sem is free.
1331 Frequency : always, if extents are used on OSTs
1333 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1334 Details : OSTs using the extents format would not zero the high 16 bits of
1335 the index physical block number. This is not a problem for any
1336 OST filesystems smaller than 16TB, and no kernels support ext3
1337 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1338 new/modified files) and can be fixed for existing filesystems
1339 with e2fsprogs-1.39-cfs1.
1344 Description: import connection selection may be incorrect if timer wraps
1345 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1346 import connections to be ignored if the 32-bit jiffies counter
1347 wraps. Use a 64-bit jiffies counter.
1350 Frequency : during server recovery
1352 Description: crash after server is denying duplicate export
1353 Details : If clients are resending connect requests to the server, the
1354 server refuses to allow a client to connect multiple times.
1355 Fixed a bug in the handling of this case.
1358 Frequency : very large clusters immediately after boot
1360 Description: LNET request buffers exhausted under heavy short-term load
1361 Details : If a large number of client requests are generated on a service
1362 that has previously never seen so many requests it is possible
1363 that the request buffer growth cannot keep up with the spike in
1364 demand. Instead of dropping incoming requests, they are held in
1365 the LND until the RPC service can accept more requests.
1368 Frequency : Sometimes during replay
1370 Description: Assertion failure in ll_local_open after replay.
1371 Details : If replay happened on an open request reply before we were able
1372 to set replay handler, reply will become not swabbed tripping the
1373 assertion in ll_local_open. Now we set the handler right after
1374 recognising of open request
1377 Frequency : very rare
1379 Description: kernel reports "badness in vsnprintf"
1380 Details : Reading from the "recovery_status" /proc file in small chunks
1381 may cause a negative length in lprocfs_obd_rd_recovery_status()
1382 call to vsnprintf() (which is otherwise harmless). Exit early
1383 if there is no more space in the output buffer.
1385 Severity : enhancement
1387 Description: clear OBD RPC statistics by writing to them
1388 Details : It is now possible to clear the OBD RPC statistics by writing
1389 to the "stats" file.
1394 Description: Client mtime is not the same on different clients after utimes
1395 Details : In some cases, the client was using the utimes() syscall on
1396 a file cached on another node. The clients now validate the
1397 ctime from the MDS + OSTs to determine which one is right.
1402 Description: Inability to activate failout mode
1403 Details : lconf script incorrectly assumed that in python string's numeric
1404 value is used in comparisons.
1407 Frequency : always with multiple stripes per file
1409 Description: Inefficient object allocation for mutli-stripe files
1410 Details : When selecting which OSTs to stripe files over, for files with
1411 a stripe count that divides evenly into the number of OSTs,
1412 the MDS is always picking the same starting OST for each file.
1413 Return the OST selection heuristic to the original design.
1418 Description: mount failures may take full timeout to return an error
1419 Details : Under some heavy load conditions it is possible that a
1420 failed mount can wait for the full obd_timeout interval,
1421 possibly several minutes, before reporting an error.
1422 Instead return an error as soon as the status is known.
1424 ------------------------------------------------------------------------------
1426 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
1428 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
1429 INTEROPERATE with older versions automatically. Please read the
1430 user documentation before upgrading any part of a live system.
1431 * WARNING: Lustre networking configuration changes are required with
1432 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
1435 * Support for kernels:
1436 2.6.9-22.0.2.EL (RHEL 4)
1437 2.6.5-7.244 (SLES 9)
1438 2.6.12.6 vanilla (kernel.org)
1441 Severity : enhancement
1442 Bugzilla : 7981/8208
1443 Description: Introduced Lustre Networking (LNET)
1444 Details : LNET is new networking infrastructure for Lustre, it includes
1445 a reorganized network configuration mode (see the user
1446 documentation for full details) as well as support for routing
1447 between different network fabrics. Lustre Networking Devices
1448 (LNDs) for the supported network fabrics have also been
1449 created for this new infrastructure.
1451 Severity : enhancement
1452 Description: Introduced Access control lists
1453 Details : clients can set ACLs on files and directories in order to have
1454 more fine-grained permissions than the standard Unix UGO+RWX.
1455 The MDS must be started with the "-o acl" mount option.
1457 Severity : enhancement
1458 Description: Introduced filesystem quotas
1459 Details : Administrators may now establish per-user quotas on the
1462 Severity : enhancement
1464 Description: Configuration change for the XT3
1465 The PTLLND is now used to run Lustre over Portals on the XT3
1466 The configure option(s) --with-cray-portals are no longer used.
1467 Rather --with-portals=<path-to-portals-includes> is used to
1468 enable building on the XT3. In addition to enable XT3 specific
1469 features the option --enable-cray-xt3 must be used.
1474 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
1475 Details : A race between allocating a new llcd and re-getting the llcd_lock
1476 allowed another thread to grab newly-allocated llcd.
1478 Severity : enhancement
1480 Description: 2.6 OST async journal commit and locking fix to improve performance
1481 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
1482 2.6 kernels are now async as they already were in 2.4 kernels so
1483 that they can commit concurrently with the network bulk transfer.
1484 For block-allocated files the filter allocation semaphore is held
1485 to avoid filesystem fragmentation during allocation. BKL lock
1486 removed for 2.6 xattr operations where it is no longer needed.
1491 Description: lconf incorrectly determined whether two IP networks could talk
1492 Details : In some more complicated routing and multiple-network
1493 configurations, lconf will avoid trying to make a network
1494 connection to a disjoint part of the IP space. It was doing the
1495 math incorrectly for one set of cases.
1500 Description: Fix for potential infinite loop processing records in an llog.
1501 Details : If an llog record is corrupted/zeroed, it is possible to loop
1502 forever in llog_process(). Validate the llog record length
1503 and skip the remainder of the block on error.
1506 Frequency : occasional (liblustre only)
1508 Description: liblustre could not open files whose last component is a symlink
1509 Details : sysio_path_walk() would incorrectly pass the open intent to
1510 intermediate path components.
1513 Frequency : rare (liblustre only with non-standard tuning)
1514 Bugzilla : 7201 (7350)
1515 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
1516 Details : llu_lookup_finish_locks() tries to set lock data on a lock
1517 after it has been released, only do this for referenced locks
1519 Severity : enhancement
1521 Description: specifying an (invalid) directory default stripe_size of -1
1522 would reset the directory default striping
1523 Details : stripe_size -1 was used internally to signal directory stripe
1524 removal, now use "all default" to signal dir stripe removal
1525 as a directory striping of "all default" is not useful
1528 Frequency : common for large clusters running liblustre clients
1530 Description: doing an ls when liblustre clients are running is slow
1531 Details : sending a glimpse AST to a liblustre client waits for every AST
1532 to time out, as liblustre clients will not respond. Since they
1533 cannot cache data we refresh the OST lock LVB from disk instead.
1535 Severity : enhancement
1537 Description: doing an ls at the same time as file IO can be slow
1538 Details : enqueue and other "small" requests can be blocked behind many
1539 large IO requests. Create a new OST IO portal for non-IO
1540 requests so they can be processed faster.
1543 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
1545 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
1546 created file with mode 000
1547 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
1548 allow an ftruncate() as a non-root user to a file with mode 000.
1549 Lustre now respects this flag to disable mode checks when
1550 truncating a file owned by the user
1553 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
1556 Description: Revoking locks from clients that went dead or catatonic might take
1558 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
1559 cancellation of such locks instant on servers without waiting for
1560 any reply from clients. Clients drops these locks when cancel
1561 notification from server is received without replying.
1564 Frequency : liblustre-only, when liblustre client dies or becomes busy
1566 Description: Doing ls on Linux clients can take a long time with active
1568 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
1569 granting such locks to it in the first place if possible. Locks
1570 are taken by proxy on the OST during the read or write and
1571 dropped immediately afterward. Add connect flags handling, do
1572 not grant locks to liblustre clients for glimpse ASTs.
1574 Severity : enhancement
1576 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
1577 Details : Existing read-ahead algorithm is tuned for the case of streamlined
1578 sequential reads and behaves badly with applications doing random
1579 reads. Improve it by reading ahead at least read region, and
1580 avoiding excessive large RPC for small reads.
1582 Severity : enhancement
1584 Description: Creating more than 1000 files for a single job may cause a load
1585 imbalance on the OSTs if there are also a large number of OSTs.
1586 Details : qos_prep_create() uses an OST index reseed value that is an
1587 even multiple of the number of available OSTs so that if the
1588 reseed happens in the middle of the object allocation it will
1589 still utilize the OSTs as uniformly as possible.
1594 Description: OST or MDS may oops in ping_evictor_main()
1595 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
1596 but doesn't restart at beginning of obd_exports_timed list
1599 Severity : enhancement
1601 Description: improve by-nid export eviction on the MDS and OST
1602 Details : allow multiple exports with the same NID to be evicted at one
1603 time without re-searching the exports list.
1606 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
1608 Description: MDS may oops in groups_free()
1609 Details : in rare race conditions a newly allocated group_info struct is
1610 freed again, and this can be NULL. The 2.4 compatibility code
1611 for groups_free() checked for a NULL pointer, but 2.6 did not.
1614 Frequency : common for liblustre clients doing little filesystem IO
1615 Bugzilla : 9352, 7313
1616 Description: server may evict liblustre clients accessing contended locks
1617 Details : if a client is granted a lock or receives a completion AST
1618 with a blocking AST already set it would not reply to the AST
1619 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
1622 Frequency : lfs setstripe, only systems with more than 160 OSTs
1624 Description: unable to set striping with a starting offset beyond OST 160
1625 Details : llapi_create_file() incorrectly limited the starting stripe
1626 index to the maximum single-file stripe count.
1629 Frequency : LDAP users only
1631 Description: lconf did not handle in-kernel recovery with LDAP properly
1632 Details : lconf/LustreDB get_refs() is searching the wrong namespace
1634 Severity : enhancement
1636 Description: bind OST threads to NUMA nodes to improve performance
1637 Details : all OST threads are uniformly bound to CPUs on a single NUMA
1638 node and do their allocations there to localize memory access
1640 Severity : enhancement
1642 Description: llmount can determine client NID directly from Myrinet (GM)
1643 Details : the client NID code from gmnalnid was moved directly into
1644 llmount, removing the need to use this or specifying the
1645 client NID explicitly when mounting GM clients with zeroconf
1648 Frequency : if client is started with down MDS
1650 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
1651 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
1652 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
1653 ptlrpc_import_delayed_req() because we didn't check if the
1654 request was interrupted, and we also didn't break out of the
1655 event loop if there was no timeout
1660 Description: data loss during non-page-aligned writes to a single file from
1661 both multiple nodes and multiple threads on one node at same time
1662 Details : updates to KMS and lsm weren't protected by common lock. Resulting
1663 inconsistency led to false short-reads, that were cached and later
1664 used by ->prepare_write() to fill in partially written page,
1665 leading to data loss.
1668 Frequency : always, if lconf --abort_recovery used
1670 Description: lconf --abort_recovery fails with 'Operation not supported'
1671 Details : lconf was attempting to abort recovery on the MDT device and not
1674 Severity : enhancement
1676 Description: remove cleanup logs
1677 Details : replace lconf-generated cleanup logs with lustre internal
1678 cleanup routines. Eliminates the need for client-cleanup and
1681 Severity : enhancement
1683 Description: add support for EAs (user and system) on lustre filesystems
1684 Details : it is now possible to store extended attributes in the Lustre
1685 client filesystem, and with the user_xattr mount option it
1686 is possible to allow users to store EAs on their files also
1688 Severity : enhancement
1690 Description: Add possibility (config option) to show minimal available OST free
1692 Details : When compiled with --enable-mindf configure option, statfs(2)
1693 (and so, df) will return least minimal free space available from
1694 all OSTs as amount of free space on FS, instead of summary of
1695 free spaces of all OSTs.
1697 Severity : enhancement
1699 Description: do not expand extent locks acquired on OST-side
1700 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
1701 by server: they are not cached anyway.
1704 Frequency : when mmap is used/binaries executed from Lustre
1706 Description: Unmmap pages before throwing them away from read cache.
1707 Details : llap_shrink cache now attempts to unmap pages before discarding
1708 them (if unmapping failed - do not discard). SLES9 kernel has
1709 extra checks that trigger if this unmapping is not done first.
1714 Description: lconf didn't resolve symlinks before checking to see whether a
1715 given mountpoint was already in use
1718 Frequency : when migrating failover services
1719 Bugzilla : 6395, 9514
1720 Description: When migrating a subset of services from a node (e.g. failback
1721 from a failover service node) the remaining services would
1722 time out and evict clients.
1723 Details : lconf --force (implied by --failover) sets the global obd_timeout
1724 to 5 seconds in order to quickly disconnect, but this caused
1725 other RPCs to time out too quickly. Do not change the global
1726 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
1728 Severity : enhancement
1729 Frequency : if MDS is started with down OST
1730 Bugzilla : 9439,5706
1731 Description: Allow startup/shutdown of an MDS without depending on the
1732 availability of the OSTs.
1733 Details : Asynchronously call mds_lov_synchronize during MDS startup.
1734 Add appropriate locking and lov-osc refcounts for safe
1735 cleaning. Add osc abort_inflight calls in case the
1736 synchronize never started.
1739 Frequency : occasional (Cray XT3 only)
1741 Description: root not authorized to access files in CRAY_PORTALS environment
1742 Details : The client process capabilities were not honoured on the MDS in
1743 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
1744 been accessed by an authorized user then root was able to access
1745 the file on the local client also. The root user capabilities
1746 are now allowed on the MDS, as this environment has secure UID.
1749 Frequency : occasional
1751 Description: ldiskfs "too long searching" message happens too often
1752 Details : A debugging message (otherwise harmless) prints too often on
1753 the OST console. This has been reduced to only happen when
1754 there are fragmentation problems on the filesystem.
1759 Description: Division by zero in statfs when all OSCs are inactive
1760 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
1761 lov_statfs divides by value returned by lov_get_stripecnt().
1765 Bugzilla : 9489, 3273
1766 Description: First write from each client to each OST was only 4kB in size,
1767 to initialize client writeback cache, which caused sub-optimal
1768 RPCs and poor layout on disk for the first writen file.
1769 Details : Clients now request an initial cache grant at (re)connect time
1770 and so that they can start streaming writes to the cache right
1771 away and always do full-sized RPCs if there is enough data.
1772 If the OST is rebooted the client also re-establishes its grant
1773 so that client cached writes will be honoured under the grant.
1778 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
1779 Details : Now I/O RPCs go to different portal number and (presumably) fast
1780 lock requests (and glimses) and other RPCs get their own service
1781 threads pool that should be able to service those RPCs
1784 Severity : enhancement
1786 Description: Ability to exchange lustre version between client and servers and
1787 issue warnings at client side if client is too old. Also for
1788 liblustre clients there is ability to refuse connection of too old
1790 Details : New 'version' field is added to connect data structure that is
1791 filled with version info. That info is later checked by server and
1795 Frequency : rare, liblustre only.
1796 Bugzilla : 9296, 9581
1797 Description: Two simultaneous writes from liblustre at offset within same page
1798 might proceed at the same time overwriting eachother with stale
1800 Details : I/O lock withing llu_file_prwv was released too early, before data
1801 actually was hitting the wire. Extended lock-holding time until
1802 server acknowledges receiving data.
1805 Frequency : extremely rare. Never observed in practice.
1807 Description: avoid generating lustre_handle cookie of 0.
1808 Details : class_handle_hash() generates handle cookies by incrementing
1809 global counter, and can hit 0 occasionaly (this is unlikely, but
1810 not impossible, because initial value of cookie counter is
1811 selected randonly). Value of 0 is used as a sentinel meaning
1812 "unassigned handle" --- avoid it. Also coalesce two critical
1813 sections in this function into one.
1815 Severity : enhancement
1817 Description: allow liblustre clients to delegate truncate locking to OST
1818 Details : To avoid overhead of locking, liblustre client instructs OST to
1819 take extent lock in ost_punch() on client's behalf. New connection
1820 flag is added to handle backward compatibility.
1822 Severity : enhancement
1823 Bugzilla : 4928, 7341, 9758
1824 Description: allow number of OST service threads to be specified
1825 Details : a module parameter allows the number of OST service threads
1826 to be specified via "options ost ost_num_threads={N}" in the
1827 OSS's /etc/modules.conf or /etc/modprobe.conf.
1831 Bugzilla : 6146, 9635, 9895
1832 Description: servers crash with bad pointer in target_handle_connect()
1833 Details : In rare cases when a client is reconnecting it was possible that
1834 the connection request was the last reference for that export.
1835 We would temporarily drop the export reference and get a new
1836 one, but this may have been the last reference and the export
1837 was just destroyed. Get new reference before dropping old one.
1839 Severity : enhancement
1840 Frequency : if client is started with failover MDS
1842 Description: Allow multiple MDS hostnames in the mount command
1843 Details : Try to read the configuration from all specified MDS
1844 hostnames during a client mount in case the "primary"
1847 Severity : enhancement
1849 Description: Stop sending data to evicted clients as soon as possible.
1850 Details : Check if the client we are about to send or are sending data to
1851 was evicted already. (Check is done every second of waiting,
1852 for which l_wait_event interface was extended to allow checking
1853 of exit condition at specified intervals).
1856 Frequency : rare, normally only when NFS exporting is done from client
1858 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
1860 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
1863 Frequency : always, if srand() is called before liblustre initialization
1865 Description: Liblustre uses system PRNG disturbing its usage by user application
1866 Details : Introduce internal to lustre fast and high-quality PRNG for
1867 lustre usage and make liblustre and some other places in generic
1868 lustre code to use it.
1870 Severity : enhancement
1871 Bugzilla : 9477, 9557, 9870
1872 Description: Verify that the MDS configuration logs are updated when xml is
1873 Details : Check if the .xml configuration logs are newer than the config
1874 logs stored on the MDS and report an error if this is the case.
1875 Request --write-conf, or allow starting with --old_conf.
1877 Severity : enhancement
1879 Description: Handle symlinks in the path when checking if Lustre is mounted.
1880 Details : Resolve intermediate symlinks when checking if a client has
1881 mounted a filesystem to avoid duplicate client mounts.
1886 Description: lconf can hit an error exception but still return success.
1887 Details : The lconf command catches the Command error exception at the top
1888 level script context and will exit with the associated exit
1889 status, but doesn't ensure that this exit status is non-zero.
1894 Description: failure of ptlrpc thread startup can cause oops
1895 Details : Starting a ptlrpc service thread can fail if there are a large
1896 number of threads or the server memory is very fragmented.
1897 Handle this without oopsing.
1900 Frequency : always, only if liblustre and non-default acceptor port was used
1902 Description: liblustre cannot connect to servers with non-default acceptor port
1903 Details : tcpnal_set_default_params() was not called and was therefore
1904 ignoring the environment varaible TCPNAL_PORT, as well as other
1905 TCPNAL_ environment variables
1910 Description: two objects could be created on the same OST for a single file
1911 Details : If an OST is down, in some cases it was possible to create two
1912 objects on a single OST for a single file. No problems other
1913 than potential performance impact and spurious error messages.
1917 Bugzilla : 5681, 9562
1918 Description: Client may oops in ll_unhash_aliases
1919 Details : Client dcache may become inconsistent in race condition.
1920 In some cases "getcwd" can fail if the current directory is
1926 Description: Inode refcounting problems in NFS export code
1927 Details : link_raw functions used to call d_instantiate without obtaining
1928 extra inode reference first.
1932 Bugzilla : 9942, 9903
1933 Description: Referencing freed requests leading to crash, memleaks with NFS.
1934 Details : We used to require that call to ll_revalidate_it was always
1935 followed by ll_lookup_it. Also with revalidate_special() it is
1936 possible to call ll_revalidate_it() twice for the same dentry
1937 even if first occurence returned success. This fix changes semantic
1938 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
1939 reference on a request referred from the intent.
1940 ll_intent_release() then releases such a request.
1943 Frequency : rare, normally benchmark loads only
1945 Description: unlinked inodes were kept in memory on the client
1946 Details : If a client is repeatedly creating and unlinking files it
1947 can accumulate a lot of stale inodes in the inode slab cache.
1948 If there is no other client load running this can cause the
1949 client node to run out of memory. Instead flush old inodes
1950 from client cache that have the same inode number as a new inode.
1953 Frequency : SLES9 2.6.5 kernel and long filenames only
1954 Bugzilla : 9969, 10379
1955 Description: utime reports stale NFS file handle
1956 Details : SLES9 uses out-of-dentry names in some cases, which confused
1957 the lustre dentry revalidation. Change it to always use the
1961 Frequency : rare, unless heavy write-truncate concurrency is continuous
1962 Bugzilla : 4180, 6984, 7171, 9963, 9331
1963 Description: OST becomes very slow and/or deadlocked during object unlink
1964 Details : filter_destroy() was holding onto the parent directory lock
1965 while truncating+unlinking objects. For very large objects this
1966 may block other threads for a long time and slow overall OST
1967 responsiveness. It may also be possible to get a lock ordering
1968 deadlock in this case, or run out of journal credits because of
1969 the combined truncate+unlink. Solution is to do object truncate
1970 first in one transaction without parent lock, and then do the
1971 final unlink in a new transaction with the parent lock. This
1972 reduces the lock hold time dramatically.
1975 Frequency : rare, 2.4 kernels only
1977 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
1978 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
1979 before unmounting their filesystem in order to avoid deadlock.
1980 The kernel_locked() function in 2.4 kernels only checks whether
1981 the kernel lock is held, not whether it is this process that is
1982 holding it as 2.6 kernels do.
1987 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
1988 Details : The client ptlrpc code may be trying to reconnect to a down
1989 server before a previous connection attempt has timed out.
1990 Increase the reconnect interval to be longer than the connection
1991 timeout interval to avoid sending duplicate connections to
1995 Frequency : echo_client brw_test command
1997 Description: fix echo_client to work with OST preallocated code
1998 Details : OST preallocation code (5137) didn't take echo_client IO path
1999 into account: echo_client calls filter methods outside of any
2000 OST thread and, hence, there is no per-thread preallocated
2001 pages and buffers to use. Solution: hijack pga pages for IO. As
2002 a byproduct, this avoids unnecessary data copying.
2006 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2007 Description: Client can oops in mdc_commit_close() after open replay
2008 Details : It was possible for the MDS to return an open request with no
2009 transaction number in mds_finish_transno() if the client was
2010 evicted, but without actually returning an error. Clients
2011 would later try to replay that open and may trip an assertion
2012 Simplify the client close codepath, and always return an error
2013 from the MDS in case the open is not successful.
2016 Frequency : rare, 2.6 OSTs only
2018 Description: OST may deadlock under high load on fragmented files
2019 Details : If there was a heavy load and highly-fragmented OST filesystems
2020 it was possible to have all the OST threads deadlock waiting on
2021 allocation of biovecs, because the biovecs were not released
2022 until the entire RPC IO was completed. Instead, release biovecs
2023 as soon as they are complete to ensure forward IO progress.
2025 Severity : enhancement
2027 Description: Support for specifying external journal device at mount
2028 Details : If an OST or MDS device is formatted with an external journal
2029 device, this device major/minor is stored in the ext3 superblock
2030 and may not be valid for failover. Allow detecting and
2031 specifying the external journal at mount time.
2036 Description: Mounting an MDS with pending unlinked files may cause oops
2037 Details : target_finish_recovery() calls mds_postrecov() which returned
2038 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2039 considers this an error and immediately begins cleaning up the
2040 lov, just after starting the mds_lov process
2042 Severity : enhancement
2044 Description: Implement 'lfs df' to report actual free space on per-OST basis
2045 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2046 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2047 usage of objects; '-h' to report in human readable format.
2049 ------------------------------------------------------------------------------
2051 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
2058 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2059 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2060 at startup during orphan unlinking. Instead, skip buddy update
2061 before setup, as it will be regenerated after recovery is complete.
2066 Description: If an OST is inactive, its locks might reference stale inodes.
2067 Details : lov_change_cbdata() must iterate over all namespaces, even if
2068 they are inactive to clear inode references from the lock.
2070 Severity : enhancement
2071 Frequency : occasional, if non-standard max_dirty_mb used
2073 Description: Client will block write RPCs if not enough grant
2074 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2075 then the client will block writes while waiting for another RPC
2076 to complete instead of consuming its dirty limit. With change
2077 we get improved performance when max_dirty_mb is small.
2079 Severity : enhancement
2080 Bugzilla : 3389, 6253
2081 Description: Add support for supplementary groups on the MDS.
2082 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2083 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2084 lookups for user supplementary groups into a cache.
2088 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2089 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2090 for liblustre O_EXCL is enforced by MDS.
2094 Description: readdir never returns NULL in liblustre.
2095 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2096 dentry in struct dirent.
2100 Frequency : liblustre only, depends on application IO pattern
2101 Description: liblustre clients evicted if not contacting servers
2102 Details : Don't put liblustre clients into the ping_evictor list, so
2103 they will not be evicted by the pinger ever.
2105 Severity : enhancement
2107 Description: Add ability to evict clients by NID from MDS.
2108 Details : By echoing "nid:$NID" string into
2109 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2110 $NID would be instantly evicted from this MDS and from all active
2111 OSTs connected to it.
2115 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2116 Details : lookup_it_finish() used to query file size from OSTs that was not
2121 Description: service threads change working directory to that of init
2122 Details : Starting lustre service threads may pin the working directory
2123 of the parent thread, making that filesystem busy. Threads
2124 now change to the working directory of init to avoid this.
2128 Frequency : during shutdown only
2129 Description: shutdown with a failed MDS or OST can cause unmount to hang
2130 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2135 Frequency : 2.6 only
2136 Description: chmod/chown may include an extra supplementary group
2137 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2138 supplementary group and if none is specified this is used.
2141 Bugzilla : 5479 (6816)
2143 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2144 Details : If lookup is in progress on a file that is unlinked we might try
2145 to revalidate the inode and fail in revalidate after lookup is
2146 complete and ll_file_open() enqueues the open again but
2147 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2150 Frequency : always, if lconf --abort_recovery used
2152 Description: lconf --abort_recovery fails with 'Operation not supported'
2153 Details : lconf was attempting to abort recovery on the MDT device and not
2156 ------------------------------------------------------------------------------
2158 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2163 Frequency : rare (only unsupported configurations with a node running as an
2165 Bugzilla : 6514, 5137
2166 Description: Mounting a Lustre file system on a node running as an OST could
2168 Details : OSTs now preallocates memory needed to write out data at
2169 startup, instead of when needed, to avoid having to
2170 allocate memory in possibly low memory situations.
2171 Specifically, if the file system is mounted on on OST,
2172 memory pressure could force it to try to write out data,
2173 which it needed to allocate memory to do. Due to the low
2174 memory, it would be unable to do so and the node would
2175 become unresponsive.
2177 Severity : enhancement
2179 Description: Addition of lconf --service command line option
2180 Details : lconf now accepts a '--service <arg>' option, which is
2181 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2183 Severity : enhancement
2185 Description: Failover mode is now the default for OSTs.
2186 Details : By default, OSTs will now run in failover mode. To return to
2187 the old behaviour, add '--failout' to the lmc line for OSTs.
2189 Severity : enhancement
2191 Description: Health checks are now provided for MDS and OSTs
2192 Details : Additional detailed health check information on MSD and OSTs
2193 is now provided through the procfs health_check value.
2196 Frequency : occasional, depends on IO load
2198 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2199 after numerous create/delete cycles
2200 Details : The ext3 inode allocation policy would not allocate new inodes
2201 very well on the OSTs because there are no new directories
2202 being created. Instead we look for groups with free space if
2203 the parent directories are nearly full.
2208 Description: Network or server problems during mount may cause partially
2209 mounted clients instead of returning an error.
2210 Details : The config llog parsing code may overwrite the error return
2211 code during mount error handling, returning success instead
2217 Description: MDS can fail to allocate large reply buffers
2218 Details : After long uptimes the MDS can fail to allocate large reply
2219 buffers (e.g. zconf client mount config records) due to memory
2220 fragmentation or consumption by the buffer cache. Preallocate
2221 some large reply buffers so that these replies can be sent even
2222 under memory pressure.
2226 Frequency : rare (liblustre)
2227 Description: fsx running with liblustre complained that using truncate() to
2228 extend the file doesn't work. This patch corrects that issue.
2229 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2230 ATTR_SIZE and lsm use in llu_setattr_raw.
2234 Frequency : rare, only 2.6 kernels
2235 Description: Unusual file access patterns on the MDS may result in inode
2236 data being lost in very rare circumstances.
2237 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2238 code in the 2.6 kernel caused the inode and/or EA data not to
2239 be read from disk, causing single-file corruption.
2243 Frequency : rare, only 2.6 filesystems using extents
2244 Description: Heavy concurrent write and delete load may cause data corruption.
2245 Details : It was possible under high-load situations to have an extent
2246 metadata block in the block device cache from a just-unlinked
2247 file overwrite a newly-allocated data block. We now unmap any
2248 metadata buffers that alias just-allocated data blocks.
2252 Frequency : filesystems with default stripe_count larger than 77
2253 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2254 Details : lconf specifies an inode size of 4096 bytes when the default
2255 stripe_count is larger than 77. This conflicts with the default
2256 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2257 this case to avoid pinning too much memory for large EAs.
2259 ------------------------------------------------------------------------------
2261 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2266 Frequency : rare (extremely heavy IO load with hundreds of clients)
2268 Description: Client is evicted, gets IO error writing to file
2269 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2270 caused clients to be evicted for AST timeouts. The fixes in
2271 bug 5192 mean we no longer need to have such short AST timeouts
2272 so ldlm_timeout has been increased.
2275 Frequency : occasional during --force or --failover shutdown under load
2276 Bugzilla : 5949, 4834
2277 Description: Server oops/LBUG if stopped with --force or --failover under load
2278 Details : a collection of import/export refcount and cleanup ordering
2279 issues fixed for safer force cleanup
2282 Frequency : only filesystems larger than 120 OSTs
2283 Bugzilla : 5990, 6223
2284 Description: lfs getstripe would oops on a very large filesystem
2285 Details : lov_getconfig used kfree on vmalloc'd memory
2288 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2289 Bugzilla : 6242, 6243
2290 Description: reading from files that had been truncated to a non-zero size
2291 but never opened returned no data
2292 Details : ll_file_read() reads zeros from no-object files to EOF
2297 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2298 Details : There's little harm in aborting MDS/OSS recovery and letting it
2299 try again, so I removed the LASSERT and return an error instead.
2301 Severity : enhancement
2303 Description: New debugging infrastructure for tracking down data corruption
2304 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2305 (b) cover more of the client-side code path, and (c) try to narrow
2306 down where problems occurred
2310 Bugzilla : 3819, 4364, 4397, 6313
2311 Description: Racing close and eviction MDS could cause assertion in mds_close
2312 Details : It was possible to get multiple mfd references during close and
2313 client eviction, leading to one thread referencing a freed mfd.
2315 Severity: : enhancement
2316 Bugzilla : 3262, 6359
2317 Description: Attempts to reconnect to servers are now more aggressive.
2318 Details : This builds on the enhanced upcall-less recovery that was added
2319 in 1.4.2. When trying to reconnect to servers, clients will
2320 now try each server in the failover group every 10 seconds. By
2321 default, clients would previously try one server every 25 seconds.
2326 Description: After recovery, certain operations trigger a failed
2327 assertion on a client.
2328 Details : Failing over an mds, using lconf -d --failover, while a
2329 client was doing a readdir() call would cause the client to
2330 LBUG after recovery completed and the readdir() was resent.
2332 Severity : enhancement
2334 Description: Default groups are now added by lconf
2335 Details : You can now run lconf --group <servicename> without having to
2336 manually add groups with lmc.
2339 Frequency : occasional
2341 Description: Nodes with an elan id of 0 trigger a failed assertion
2344 Frequency : always when accessing e.g. tty/console device nodes
2346 Description: tty and some other devices nodes cannot be used on lustre
2347 Details : file's private_data field is used by device data and lustre
2348 values in there got lost. New field was added to struct file to
2349 store fs-specific private data.
2352 Frequency : when exporting Lustre via NFS
2354 Description: NFSD failed occasionally when looking up a path component
2355 Details : NFSD is looking up ".." which was broken in ext3 directories
2356 that had grown large enough to become hashed.
2359 Frequency : Clusters with multiple interfaces not on the same subnet
2361 Description: Nodes will repeatedly try to reconnect to an interface which it
2362 cannot reach and report an error to the log.
2363 Details : Extra peer list entries will be created by lconf with some peers
2364 unreachable. lconf now validates the peer before adding it.
2367 Frequency : Only if a default stripe is set on the filesystem root.
2369 Description: Setting a default stripe on the filesystem root prevented the
2370 filesystem from being remounted.
2371 Details : The client was sending extra request flags in the root getattr
2372 request and did not allocate a reply buffer for the dir EA.
2375 Frequency : occasional, higher if lots of files are accessed by one client
2376 Bugzilla : 6159, 6097
2377 Description: Client trips assertion regarding lsm mismatch/magic
2378 Details : While revalidating inodes the VFS looks up inodes with ifind()
2379 and in rare cases can find an inode that is being freed.
2380 The ll_test_inode() code will free the lsm during ifind()
2381 when it finds an existing inode and then the VFS later attaches
2382 this free lsm to a new inode.
2386 Bugzilla : 6422, 7030
2387 Description: MDS deadlock between mkdir and client eviction
2388 Details : Creating a new file via mkdir or mknod (starting a transaction
2389 and getting the ns lock) can deadlock with client eviction
2390 (gets ns lock and trying to finish a synchronous transaction).
2393 Frequency : occasional
2394 Description: While starting a server, the fsfilt_ext3 module could not be
2396 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2397 kernels. Previously, lconf would still use the ext3 name
2398 when trying to load modules. Now, it will correctly use
2399 ext3 on 2.4 and ldiskfs on 2.6.
2401 Severity : enhancement
2402 Description: The default stripe count has been changed to 1
2403 Details : The interpretation of the default stripe count (0, to lfs
2404 or lmc) has been changed to mean striping across a single
2405 OST, rather than all available. For general usage we have
2406 found a stripe count of 1 or 2 works best.
2408 Severity : enhancement
2409 Description: Add support for compiling against Cray portals.
2410 Details : Conditional compiling for some areas that are different
2414 Frequency : occasional
2415 Bugzilla : 6409, 6834
2416 Description: Creating files with an explicit stripe count may lead to
2417 a failed assertion on the MDS
2418 Details : If some OSTs are full or unavailable, creating files may
2419 trigger a failed assertion on the MDS. Now, Lustre will
2420 try to use other servers or return an error to the
2424 Frequency : occasional
2426 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
2427 caused lock timeout during readahead (since 1.4.2).
2428 Details : Processes doing readahead might match a lock that hasn't been
2429 granted yet if there are overlapping and conflicting lock
2430 requests. The readahead process waits on ungranted lock
2431 (original lock is CBPENDING), while OST waits for that process
2432 to cancel CBPENDING read lock and eventually evicts client.
2434 Severity : enhancement
2436 Description: Initial enabling of flock support for clients
2437 Details : Implements fcntl advisory locking and file status functions.
2438 This feature is provided as an optional mount flag (default
2439 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
2440 locking are implemented yet, and those that are are not guaranteed
2441 to be completely correct in production environments.
2442 mount -t lustre -o [flock|noflock] ...
2445 Frequency : occasional
2447 Description: OSTs running 2.4 kernels but with extents enabled might trip an
2448 assertion in the ext3 JBD (journaling) layer.
2449 Details : The b_committed_data struct is protected by the big kernel lock
2450 in 2.4 kernels, serializing journal_commit_transaction() and
2451 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
2452 access to this struct. In 2.6 kernels there is finer grained
2453 locking to improve SMP performance of the JBD layer.
2457 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
2459 -----------------------------------------------------------------------------
2461 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
2463 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
2464 versions of Lustre. Please update all servers and clients to
2465 version 1.4.2 or later at the same time. You must also run
2466 "lconf --write-conf {config}.xml" on the MDS while it is stopped
2467 to update the configuration logs.
2469 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
2470 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
2471 - swab llog records of type '0' so we get proper header size/idx (5861)
2472 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
2473 - fix rename of one directory over another leaking an inode (5953)
2474 - avoid SetPageDirty on 2.6 (5981)
2475 - don't re-add just-being-destroyed locks to the waiting list (5653)
2476 - when creating new directories, inherit the parent's custom
2477 striping settings if present parent (3048)
2478 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
2479 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
2480 - don't hold client locks on temporary worklist from l_lru (5666)
2481 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
2482 - automatically evict dead clients (5921)
2483 - Update file size properly in create+truncate+fstat case (6196)
2484 - Do not unhash mountpoint dentries, do not allow removal of
2486 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
2487 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
2488 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
2489 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
2490 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
2491 - Correctly handle reads of files with no objects (6243)
2492 - lctl recover will also mark a device active if deactivate used (5933)
2494 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
2495 - allow --write-conf on an MDS with different nettype than client (5619)
2496 - don't write config llogs to MDS for mounts not from that MDS (5617)
2497 - lconf should create multiple TCP connections from a client (5201)
2498 - init scripts are now turned off by default; run chkconfig --on
2499 lustre and chkconfig --on lustrefs to use them
2500 - upcalls are no longer needed for clients to recover to failover
2502 - add --abort-recovery option to lconf to abort recovery on device
2504 - add support for an arbitrary number of OSTs (3026)
2505 - Quota support protocol changes.
2506 - forward compatibility changes to wire structs (6007)
2507 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
2508 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
2509 - improved "lustre status" script
2510 - initialize blocksize for non-regular files (6062)
2511 - added --disable-server and --disable-client configure options (5782)
2512 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
2513 - Vanilla 2.4.29 support
2514 - increase maximum number of obd devices to 520 (6242)
2515 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
2516 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
2517 SLES 9 2.6.5 kernels
2518 - limit stripes per file to 160 (the maximum EA size) (6093)
2520 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2523 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
2524 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
2525 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
2526 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
2527 - small ext3 extents cleanups and fixes (5733)
2528 - improved mballoc code, several small races and bugs fixed (5733, 5638)
2529 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
2530 - avoid needless client->OST connect, fix handle mismatch (5317)
2531 - fix DLM error path that led to out-of-sync client, long delays (5779)
2532 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
2533 - fix several locking issues related to i_size (5492,5624,5654,5672)
2534 - don't move pending lock onto export if it is already evicted (5683)
2535 - fix kernel oops when creating .foo in unlinked directory (5548)
2536 - fix deadlock in obdfilter statistics vs. object create (5811)
2537 - use time_{before,after} to avoid timer jiffies wrap (5882)
2538 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
2539 - Do not leak request if server was not able to process it (5154)
2540 - If mds_open unable to find parent dir, make that negative lookup(5154)
2541 - don't create new directories with extent-mapping (5909, 5936)
2543 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
2544 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
2545 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
2546 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
2547 - support for RHEL4 kernel on i686 (5773)
2548 - provide error messages when incompatible logs are encountered (5898)
2550 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
2551 * version 1.4.0.10 (1.4.1 release candidate 1)
2553 - don't keep a lock reference when lock is not granted (4238)
2554 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2555 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
2556 - drop import inflight refcount on signal_completed_replay error (5255)
2557 - unlock page after async write error during send (3677)
2558 - handle missing objects in filter_preprw_read properly (5265)
2559 - no transno return for symlink open, don't save no-trasno open (3440)
2560 - don't try to complete elan receive that already failed (4012)
2561 - free RPC server reply state on error (5406)
2562 - clean up thread from ptlrpc_start_thread() on error (5160)
2563 - readahead could read extra page into cache that wasn't ejected (5388)
2564 - prevent races in class_attach/setup/cleanup/detach (5260)
2565 - don't dereference de->d_inode after l_dput of de (5458)
2566 - use "int" for stripe value returned from lock_to_stripe (5544)
2567 - mballoc allocation and error-checking fixes in 2.6 (5504)
2568 - block device patches to fix I/O request sizes in 2.6 (5482)
2569 - look up hostnames for IB nals (5602)
2570 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
2571 - don't start multiple acceptors for the same port (5277)
2572 - fix incorrect LASSERT in mds_getattr_name (5635)
2573 - export a proc file for general "ping" checking (5628)
2574 - fix "lfs check" to not block when the MDS is down (5628)
2576 - service request history (4965)
2577 - put {ll,lov,osc}_async_page structs in a single slab (4699)
2578 - create an "evict_client" /proc entry on OSTs, like the MDS has
2579 - fix mount usage message, return errors per mount(8) (5168)
2580 - change grep [] to grep "[]" in tests so they work in more UMLs
2581 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
2582 - remove ancient LOV_MAGIC_V0 EA support (5047)
2583 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
2584 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
2585 - allow readahead allocations to fail when low on memory (5383)
2586 - mmap locking landed again, after considerable improvement (2828)
2587 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
2589 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
2592 - send OST transaction number in read/write reply to free req (4966)
2593 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2594 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2595 - the watchdog thread now runs as interruptible (5246)
2596 - flock/lockf fixes (but it's still disabled, pending 5135)
2597 - don't use EXT3 constants in llite code (5094)
2598 - memory shortage at startup could cause assertion (5176)
2600 - reorganization of lov code
2601 - single portals codebase
2603 - add extents/mballoc support (5025)
2604 - direct I/O reads in the obdfilter (4048)
2605 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
2607 tbd Cluster File Systems, Inc. <info@clusterfs.com>
2610 - send OST transaction number in read/write reply to free req (4966)
2611 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
2612 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
2613 - the watchdog thread now runs as interruptible (5246)
2614 - handle missing objects in filter_preprw_read properly (5265)
2615 - unsafe list practices (rarely) led to infinite eviction loop (4908)
2616 - drop import inflight refcount on signal_completed_replay error (5255)
2617 - unlock page after async write error during send (3677)
2618 - return original error code on reconstructed replies (3761)
2619 - no transno return for symlink open, don't save no-trasno open (3440)
2621 - add pid to ldlm debugging output (4922)
2622 - bump the watchdog timeouts -- we can't handle 30sec yet
2623 - extra debugging for orphan dentry/inode bug (5259)
2625 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
2628 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
2629 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
2630 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
2631 - fix NULL dereference in /proc/sys/portals/routes (4827)
2632 - allow failed mdc_close() operations to be interrupted (4561)
2633 - stop precreate on OST before MDS would time out on it (4778)
2634 - don't send partial-page writes before EOF from client (4410)
2635 - discard client grant for sub-page writes on large-page clients (4520)
2636 - don't free dentries not owned by NFS code, check generation (4806)
2637 - fix lsm leak if mds_create_objects() fails (4801)
2638 - limit debug_daemon file size, always print CERROR messages (4789)
2639 - use transno after validating reply (3892)
2640 - process timed out requests if import state changes (3754)
2641 - update mtime on OST during writes, return in glimpse (4829)
2642 - add mkfsoptions to LDAP (4679)
2643 - use ->max_readahead method instead of zapping global ra (5039)
2644 - don't interrupt __l_wait_event() during strace
2646 - add software watchdogs to catch hung threads quickly (4941)
2647 - make lustrefs init script start after nfs is mounted
2648 - fix CWARN/ERROR duplication (4930)
2649 - return async write errors to application if possible (2248)
2650 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
2651 - print NAL number in %x format (4645)
2652 - update barely-supported suse-2.4.21-171 series (4842)
2653 - support for sles 9 %post scripts
2654 - support for building 2.6 kernel-source packages
2655 - support for sles km_* packages
2657 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
2660 - ignore -ENOENT errors in osc_destroy (3639)
2661 - notify osc create thread that OSC is being cleaned up (4600)
2662 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
2663 - reconstruct ost_handle() like mds_handle() (4657)
2664 - create a new thread to do import eviction to avoid deadlock (3969)
2665 - let lconf resolve symlinked-to devices (4629)
2666 - don't unlink "objects" from directory with default EA (4554)
2667 - hold socknal file ref over connect in case target is down (4394)
2668 - allow more than 32000 subdirectories in a single directory (3244)
2669 - fix blocks count for O_DIRECT writes (3751)
2670 - OST returns ENOSPC from object create when no space left (4539)
2671 - don't send truncate RPC if file size isn't changing (4410)
2672 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
2673 - bind to privileged port in socknal and tcpnal (3689)
2675 - rate limit CERROR/CWARN console message to avoid overload (4519)
2676 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
2677 - basic mmap support (3918)
2678 - kernel patch series update from b1_4 (4711)
2680 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
2683 - avoid crash during MDS cleanup with OST shut down (2775)
2684 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
2685 - don't use bad inodes on the MDS (3744)
2686 - dynamic object preallocation to improve recovery speed (4236)
2687 - don't hold spinlock over lock dumping or change debug flags (4401)
2688 - don't zero obd_dev when it is force cleaned (3651)
2689 - print grants to console if they go negative (4431)
2690 - "lctl deactivate" will stop automatic recovery attempts (3406)
2691 - look for existing locks in ldlm_handle_enqueue() (3764)
2692 - don't resolve lock handle twice in recovery avoiding race (4401)
2693 - revalidate should check working dir is a directory (4134)
2695 - don't always mark "slow" obdfilter messages as errors (4418)
2697 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
2700 - don't close LustreDB during write_conf until it is done (3860)
2701 - fix typo in lconf for_each_profile (3821)
2702 - allow dumping logs from multiple threads at one time (3820)
2703 - don't allow multiple threads in OSC recovery (3812)
2704 - fix debug_size parameters (3864)
2705 - fix mds_postrecov to initialize import for llog ctxt (3121)
2706 - replace config semaphore with spinlock (3306)
2707 - be sure to send a reply for a CANCEL rpc with bad export (3863)
2708 - don't allow enqueue to complete on a destroyed export (3822)
2709 - down write_lock before checking llog header bitmap (3825)
2710 - recover from lock replay timeout (3764)
2711 - up llog sem before sending rpc (3652)
2712 - reduce ns lock hold times when setting kms (3267)
2713 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
2714 - fix NULL deref and obd_dev leak on setup error (3312)
2715 - replace some LBUG about llog ops with error handling (3841)
2716 - don't match INVALID dentries from d_lookup and spin (3784)
2717 - hold dcache_lock while marking dentries INVALID and hashing (4255)
2718 - fix invalid assertion in ptlrpc_set_wait (3880)
2720 - add libwrap support for the TCP acceptor (3996)
2721 - add /proc/sys/portals/routes for non-root route listing (3994)
2722 - allow setting MDS UUID in .xml (2580)
2723 - print the stack of a process that LBUGs (4228)
2725 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
2728 - don't cleanup request in ll_file_open() on failed MDS open (3430)
2729 - make sure to unset replay flag from failed open requests (3440)
2730 - if default stripe count is 0, use OST count for inode size (3636)
2731 - update parent mtime/ctime on client for create/unlink (2611)
2732 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
2733 - free recovery state on server during a forced cleanup (3571)
2734 - unregister_reply for resent reqs (3063)
2735 - loop back devices mounting and status check on 2.6 (3563)
2736 - fix resource-creation race that can provoke i_size == 0 (3513)
2737 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
2738 - more debugging for page-accounting assertion (3746)
2739 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
2740 - avoid deadlock after precreation failure (3758)
2741 - fix race and lock order deadlock in orphan handling (3450, 3750)
2742 - add validity checks when grabbing inodes from l_ast_data (3599)
2744 - add /proc/.../recovery_status to obdfilter (3428)
2745 - lightweight CDEBUG infrastructure, debug daemon (3668)
2746 - change default OSC RPC parameters to be better on small clusters
2747 - turn off OST read cache for files smaller than 32MB
2748 - install man pages and include them in rpms (3100)
2749 - add new init script for (un)mounting lustre filesystems (2593)
2750 - run chkconfig in %post for init scripts (3701)
2751 - drop scimac NAL (unmaintained)
2753 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
2756 - clean kiobufs before and after use (3485)
2757 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
2758 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
2759 - change init script to fail more clearly if not run as root (1528)
2760 - allow clients to reconnect during replay (1742)
2761 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
2762 - don't do DNS lookups on NIDs too small for IP addresses (3442)
2763 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
2764 - fix cond_resched (3554)
2765 - only evict unfinished clients after recovery (3515)
2766 - allow bulk resend, prevent data loss (3570)
2767 - dynamic ptlrpc request buffer allocation (2102)
2768 - don't allow unlinking open directory if it isn't empty (2904)
2769 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
2770 - remove extraneous obd dereference causing LASSERT failure (3334)
2771 - don't use get_cycles() when creating temp. files on the mds (3156)
2772 - hold i_sem when setting i_size in ll_extent_lock() (3564)
2773 - handle EEXIST for set-stripe, set proper directory name (3336)
2775 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
2776 - fix ksocknal_fmb_callback() error messages (2918)
2778 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
2781 - don't copy lvb into (possibly NULL) reply on error (2983)
2782 - don't deref dentry after dput, don't free lvb on error (2922)
2783 - use the kms to determine writeback rpc length (2947)
2784 - increment oti_logcookies when osc is inactive (2948)
2785 - update client's i_blocks count via lvb messages (2543)
2786 - handle intent open/close of special files properly (1557)
2787 - mount MDS with errors=remount-ro, like obdfilter (2009)
2788 - initialize lock handle to avoid ASSERT on error cleanup (3057)
2789 - don't use cancelling-locks' kms values (2947)
2790 - use highest lock extent for kms, not last one (2925)
2791 - don't dereference ERR_PTR() dentry in error handling path (3107)
2792 - fix thread race in portals_debug_dumplog() (3122)
2793 - create lprocfs device entries at setup instead of at attach (1519)
2794 - common AST error handler, don't evict client on completion race (3145)
2795 - zero nameidata in detach_mnt in 2.6 (3118)
2796 - verify d_inode after revalidate_special is valid in 2.6 (3116)
2797 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
2798 - initialize RPC timeout timer earlier for 2.6 (3219)
2799 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
2800 - print nal/nid for unknown nid (3258)
2801 - additional checks for oscc recovery before doing precreate (3284)
2802 - fix ll_extent_lock() error return code for 64-bit systems (3043)
2803 - don't crash in mdc_close for bad permissions on open (3285)
2804 - zero i_rdev for non-device files (3147)
2805 - clear page->private before handing to FS, better assertion (3119)
2806 - tune the read pipeline (3236)
2807 - fix incorrect decref of invalidated dentry (2350)
2808 - provide read-ahead stats and refine rpc in flight stats (3328)
2809 - don't hold journal transaction open across create RPC (3313)
2810 - update atime on MDS at close time (3265)
2811 - close LDAP connection when recovering to avoid server load (3315)
2812 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
2813 - don't leak open file on MDS after open resend (3325)
2814 - serialize filter_precreate and filter_destroy_precreated (3329)
2815 - loop device shouldn't call sync_dev() for nul device (3092)
2816 - clear page cache after eviction (2766)
2817 - resynchronize MDS->OST in background (2824)
2818 - refuse to mount the same filesystem twice on same mountpoint (3394)
2819 - allow llmount to create routes for mounting behind routers (3320)
2820 - push lock cancellation to blocking thread for glimpse ASTs (3409)
2821 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
2822 - fix rare problem with rename on htree directories (3417)
2824 - allow default OST striping configuration per directory (1414)
2825 - fix compilation for qswnal for 2.6 kernels (3125)
2826 - increase maximum number of MDS request buffers for large systems
2827 - change liblustreapi to be useful for external progs like lfsck (3098)
2828 - increase local configuration timeout for slow disks (3353)
2829 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
2831 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
2834 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
2835 - don't overwrite extent policy data in reply if lock was blocked (2901)
2836 - drop filter export grants atomically with removal from device (2663)
2837 - del obd_self_export from work_list in class_disconnect_exports (2908)
2838 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
2839 - swab reply message in mdc_close, other PPC fixes (2464)
2840 - fix destroying of named logs (2325)
2841 - overwrite old logs when running lconf --write_conf (2264)
2842 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
2843 - fix race in target_handle_connect (2898)
2844 - mds_reint_create() should take same inode create lock (2926)
2845 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
2846 - don't close files for self_export to avoid uninitialized obd (2936)
2847 - allow MDS with the same name as client node (2939)
2848 - hold dentry reference for closed log files for unlink (2325)
2849 - reserve space for all logs during transactions (2059)
2850 - don't evict page beyond end of stripe extent (2925)
2851 - don't oops on a deleted current working directory (2399)
2852 - handle hard links to targets without a parent properly (2517)
2853 - don't dereference NULL lock when racing during eviction (2867)
2854 - don't grow lock extents when lots of conflicting locks (2919)
2856 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
2859 - account for cache space usage on clients to avoid data loss (974)
2860 - lfsck support in lustre kernel code (2349)
2861 - reduce journal credits needed for BRW writes (2370)
2862 - orphan handling to avoid losing space on client/server crashes
2863 - ptlrpcd can be blocked, stopping ALL progress (2477)
2864 - use lock value blocks to assist in proper KMS, faster stat (1021)
2865 - takes i_sem instead of DLM locks internally on obdfilter (2720)
2866 - recovery for initial connections (2355)
2867 - fixes for mds_cleanup_orphans (1934)
2868 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
2869 - block all file creations until orphan recovery completes (1901)
2870 - client remove rq_connection from request struct (2423)
2871 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
2872 - recovery timer race (2670)
2873 - mdc_close recovey bug (2532)
2874 - ptlrpc cleanup bug (2710)
2875 - mds timeout on local locks (2588)
2876 - namespace lock held during RPCs (2431)
2877 - handle interrupted sync write properly (2503)
2878 - don't try to handle a message that hasn't been replied to (2699)
2879 - client assert failure during cleanup after abort recovery (2701)
2880 - leak mdc device after failed mount (2712)
2881 - ptlrpc_check_set allows timedout requests to complete (2714)
2882 - wait for inflight reqs when ptlrpcd finishes (2710)
2883 - make sure unregistered services are removed from the srv_list
2884 - reset bulk XID's when resending them (caught by 1138 test)
2885 - unregister_bulk after timeout
2886 - fix lconf error (2694)
2887 - handle write after unfinished setstripe, stripe-only getstripe (2388)
2888 - readahead locks pages, leaves pending causing memory pressure (2673)
2889 - increase OST request buffers to 4096 on large machines (2729)
2890 - fix up permission of existing directories in simple_mkdir (2661)
2891 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
2892 - don't assign transno to errored transactions (2742)
2893 - don't delete objects on OST if given a bogus objid from MDS (2751)
2894 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
2895 - if rq_no_resend, then timeout request after recovery (2432)
2896 - fix MDS llog_logid record size, 64-bit array alignment (2733)
2897 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
2898 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
2899 - MDS recovery shouldn't delete objects that it has given out (2730)
2900 - if enqueue arrives after completion, don't clobber LVB (2819)
2901 - don't unlock pages twice when trigger_group_io returns error (2814)
2902 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
2903 - don't write pages to disk if there was an error (1450)
2904 - don't ping imports that have recovery disabled (2676)
2905 - take buffered bytes into account when balancing socknal conn (2817)
2906 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
2907 - reconnect unlink llog connection after MDS reconnects to OST (2816)
2908 - remove little-endian swabbing of llog records (1987)
2909 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
2910 - retry reposting request buffers if they fail (1191)
2911 - grow extent at grant time to avoid granting a revoked lock (2809)
2912 - lock revoke doesn't evict page if covered by a second lock (2765)
2913 - disable VM readahead to avoid reading outside lock extents (2805)
2915 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
2916 - updated kernel patches for hp-2.4.20 kernel (2681)
2918 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
2921 - fix truncated write corruption (2366)
2922 - fix for failed assertion in iopen_connect_dentry (1792,2517)
2924 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
2925 - ensure len(uuid) < 37 in lmc (1171)
2926 - fix ia64 OOPS in llog_test (2255)
2927 - zero end of page at obdfilter for partial page writes (2648)
2928 - don't leave stale dentries around after renames (bug 2428)
2929 - fix timeouts when evicting a client with a single lock held (2642)
2930 - set deadline for the initial HELLO message to drain (2634)
2931 - print out dotted-quad IP addresses in the socknal (2302)
2933 - additional debugging for MDS client eviction problem (2443)
2934 - fix mkfsoptions support for osts (2603, 2604)
2936 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
2939 - add series for the vanilla 2.6.0 kernel
2940 - add series for the vanilla 2.4.24 kernel
2941 - add series for a cray x86/64 UL kernel drop
2942 - fix xattr patches for the vanilla 2.4.19 series
2944 - generate true UUIDs in lmc (1171)
2945 - have portals stack dumping break in UML (2466)
2946 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
2947 - allocate with _NOFS in ldlm to avoid deadlock (1933)
2948 - wake callback waiting threads on client eviction (2460)
2949 - Add --ptldebug and --subsystem to lmc (1719)
2950 - update assertion to allow safe interrupt allocation
2951 - set rq_no_resend for cancel requests (2432)
2952 - recalculate ptlrpcd timeout after resend (2494)
2953 - call vfs_rmdir when removing pending directories (2368)
2954 - fix renaming a file to itself (2429)
2955 - lmc creates a default one-stripe lov (2454)
2956 - expand procfs space to handle large clusters (2326)
2957 - increase UML stack to avoid overflow
2958 - update lconf's list of debug and subsystem masks
2959 - fix lfs find --obd (2510)
2960 - /proc tunable for disabling filter read caching (2591)
2961 - stop rpm packages from altering slapd.conf (2301)
2962 - disable nagle in the socknal under 0conf (2578)
2963 - choose mds inode size based on stripe count (2572)
2964 - fix kernel-source rpm problems (2516)
2966 - add --disable-doc to avoid pdf generation (2421)
2967 - update documentation, tests, type-os, comments
2968 - avoid format warnings on ia64
2969 - remove the TOE NAL
2970 - tiny code cleanups by removing unused fields
2972 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
2975 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
2976 - remove the most blatant lies from BUILDING file (2371)
2977 - change default debug level to reasonable production setting
2978 - reduce client side cache size to reduce cache flush time
2979 - reduce max RPCs in flight to avoid unnecessary file fragmentation
2980 - make TCP zerocopy and pinger support enabled by default (2476)
2981 - sync writes completed after process exits caused crashes (2319)
2982 - maintain correct mount count on the MDS (2356)
2983 - backout 1557, because 2316 wasn't really fixed
2984 - better file I/O statistics gathering in /proc
2985 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
2986 - another kernel patch to fix zero-copy TCP function export
2987 - don't take duplicate lock when processing re-sent getattr (2420)
2988 - lctl uses obd_self_export instead of creating new conn (2353)
2989 - MDS/OST recovery case which requires object creation asserted (2425)
2990 - move lfs from /usr/sbin to /usr/bin in packages
2991 - fix race between mds_client_add and mds_client_free (2417)
2992 - use kmalloc instead of slabs in portals (2430)
2993 - don't create duplicate records when a failover MDS is present (2442)
2994 - remove unnecessary mount age check (2332)
2995 - don't remove directory inodes from locks prematurely (2451)
2996 - don't break if MDS service name is the same as hostname (2103)
2997 - fix races in client write RPC generation when cache full (2482)
2999 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3002 - remove now-unused request->rq_obd (278)
3003 - if an allocation fails, print out how much memory we've used (1933)
3004 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3005 - add the "configurable stack size" patch to most series files (1256)
3006 - ability to write large log records, for 100+ OST configs (2306)
3007 - fix NULL deref when filter_prep fails (2314)
3008 - fix operator precedence error in filter_sync
3009 - dynamic allocation of socknal TX descriptors (2315)
3010 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3011 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3012 - ensure that CWARN messages in Portals always get to the syslog
3013 - __init/__exit are not for prototype decls (ldlm_init/exit)
3014 - x86-64 compile warning fixes
3015 - fix gateway LMC keyword conflict (2318)
3016 - fix MDS lock inversions in getattr/reint paths (1844)
3017 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3018 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3019 - DLM race condition prevented some lock evictions (2328)
3020 - ENOMEM detection and retry on socknal sends (2230)
3021 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3022 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3023 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3024 - better error messages when a client is rejected during recovery (1505)
3025 - avoid cancelling locks which were never granted, after failure (2330)
3026 - fix i_sem/journal inversion in mds_client_add (2333)
3027 - fix truncate/getattr lock cycle deadlock (2334)
3028 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3029 - fix two rare exit paths which could leak an l_lock() ref (2321)
3030 - fencepost error in MDS/OST orphan recovery (2226)
3031 - make log record alignment 8 bytes (1988)
3032 - lstripe now fails when requested offset > ost_count (2237)
3033 - ensure that all kernel series have a complete list.h (1607)
3034 - fix crashes in special-file operations (2316)
3035 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3037 - allow configurable automake binary, for testing new versions
3038 - small update to the lfs documentation
3040 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3042 * fix negative export reference count in fsfilt_sync (2312)
3044 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3045 * release candidate 0.9.1
3047 - orphans are moved into the PENDING directory for possible recovery
3048 - replayed opens now open by fid for orphan/rename safety (1042)
3049 - last close of an orphan inode generates a transno (683)
3050 - chdir() and mount() now pin the directory entry (1020)
3051 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3052 - discard very old requests without processing them (1502)
3053 - remove some common, well-understood CERRORs (1505)
3054 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3055 - clear "grant" flags in OST replies until OST grant code lands (1644)
3056 - fix read performance by not clobbering i_blksize on client (1598)
3057 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3058 - make LRU size a /proc tunable, clears locks when reduced (707)
3059 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3060 - server locks take references on exports now (1558)
3061 - build fixes for 2.4.20-rh trees (1663)
3062 - return an error from lov_create if all OSCs are inactive (1751)
3063 - fix import levels when a reconnect happens without a timeout (1597)
3064 - exit early from mds_open if we get a lookup error (1749)
3065 - partial page read at EOF wouldn't wait for disk before sending (1642)
3066 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3067 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3068 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3069 - write/truncate lock inversion (1639)
3070 - Don't auto-load obdclass, portals modules during cleanup (1495)
3071 - fix timestamps from jumping to "now" (1763)
3072 - extra journal assertions (1648)
3073 - add an extra multiunlink test (1771)
3074 - fix read_record/write_record API (1776)
3075 - fix leak of offset_extent, possible incorrect i_size later (1772)
3076 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3077 - Debugging for the kqswnal_get_idle_tx problems (1820)
3078 - Allow recovery to be attempted multiple times (1536)
3079 - Write out MDS last_rcvd file after it is first created (1600)
3080 - Fix tx_descriptor leak in failed transmit situations (1827)
3081 - ext3 journaling fixes for assertion failure after IO error (1871)
3082 - class_export_put() on freed export after completion AST error (1896)
3083 - Fix revalidate looping in VFS (1322)
3084 - Don't access a freed export during MDS_REINT timeout (1521)
3085 - Add open-unlink recovery support on the MDS (1673,1764)
3086 - Return an error if no MDS data was read from last_rcvd (1946)
3087 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3088 - Don't setup a disk device that doesn't match exported UUID (317)
3089 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3090 - avoid committing NULL handle in force close
3091 - local.sh is now a one-stripe LOV configuration
3092 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3093 - fix ext3 htree duplicate directory entry corruption (1516)
3094 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3095 - update to new LOV EA format (2097)
3096 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3098 2003-06-15 Phil Schwan <phil@clusterfs.com>
3101 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3102 - per-import/export recovery handling (958, 931, 959)
3103 - multiple last-rcvd slots, for serving multiple FSes (949)
3104 - connections are again shared between multiple imp/exports (963, 964)
3105 - "umount -f" would hang if any requests needed to be sent (393, 978)
3106 - avoid pinning large req buffer by copying for queued messages (989)
3107 - add "uuid" to "lctl device" command to help upcalls (991)
3108 - "open" RPCs with transnos would confuse recovery counters (1037)
3109 - do proper endian conversion of all wire messages (288, 340, 891)
3110 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3111 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3112 - fix ptlrpc_connection leak in target_handle_connect (1174)
3113 - fix import refcounting bug in OST and MDS cleanup (1134)
3114 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3115 - fix very unlikely obd_types race condition (501)
3116 - remove osc_open hack for echo_client (1187)
3117 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3118 - a failure in read_inode2 leads to deadlock (1139)
3119 - cancel ack-locks as soon as transaction is committed (1072)
3120 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3121 - make sure to commitrw after any preprw to avoid deadlock (1162)
3122 - failing to execute a file in a lustre FS would lock inode (1203)
3123 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3124 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3125 - obd_uuid2tgt would walk off the end of the list (1255)
3126 - on IA64 the osc would give portals incorrect bulk size (1258)
3127 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3128 - fix lock inversion caused by new llite matching code (1282)
3129 - limit the number of dirty pages on a client to 10MB (1286)
3130 - timed out locks were not being corrected cancelled (1289)
3131 - fix O_DIRECT above 4GB on IA-32 (1292)
3132 * major user-visible changes
3133 - fail out/fail over policy now controlled by the upcall (993)
3135 - add OBD_PING to check server availability and failure (954)
3136 - lustre messages are now sent in sending host order (288, 340, 891)
3137 - add eadatalen to MDS getattr reply (340)
3138 - OST read replies may contain second buffer, with per-page status (593)
3140 2003-03-11 Phil Schwan <phil@clusterfs.com>
3143 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3144 - failed open()s could cause deadlock; fixed (867, 869)
3145 - stop cancelling OST locks when files are closed (481)
3146 - overlapping XID spaces caused network corruption (851, 853)
3147 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3148 - setattr_raw added, to do single-RPC, server-side setattrs
3149 - lmc/lconf syntax change for OST UUIDs
3150 - fix crashy race condition between ptlrpc_free_req and osc_close
3151 - don't use request in mdc_enqueue if we hit a timeout (889)
3152 - don't set the inode i_size for regular files from the MDS (896)
3153 - handle out of order completion AST (842)
3154 - don't LBUG if a lock request times out after receiving AST (913)
3155 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3156 - if a bad lock AST arrives, send an error instead of dropping entirely
3157 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3158 - fix leak in bulk IO when only partially completed (899, 900, 926)
3159 - fix O_DIRECT for ia64 (55)
3160 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3161 - C-z after timeout could hang a process forever; fixed (977)
3163 - client-side I/O cache (678, 924, 929, 941, 970)
3165 - READPAGE and SETATTRs which don't take server-side locks get
3168 2003-02-11 Phil Schwan <phil@clusterfs.com>
3171 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3172 - Fix fsfilt_extN_readpage() to read a full page of directory
3173 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3174 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3175 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3176 - Most intent operations were removed, in favour of a new RPC mode
3177 that does a single RPC to the server and bypasses most of the VFS
3178 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3179 - Aggressively cancel local locks on DLM servers
3180 - mds_reint_unlink sends EA to the client if it's the last nlink.
3181 client uses that EA to unlink OST objects.
3182 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3183 - recursive symlinks were fixed (439)
3184 - fixed NULL deref in DEBUG_REQ
3185 - filter_update_lastobjid no longer calls sync, which annoyed extN
3186 - fixed multi-client small-writes to a single file problem (445)
3187 - fixed mtime updates during file writes (607)
3188 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3189 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3190 - fixed Linux OST size reporting problem (444, 656)
3191 - OST now updates object mtime with writes or setattr (607, 619)
3192 - client verifies file size before zeroing page past EOF (445)
3193 - OST now writes last allocated objid to disk with allocation (108)
3194 - LOV on echo now works (409)
3196 - mds_reint_unlink sends a new buffer, with the EA included. this
3197 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3198 set if a regular file was being unlinked, and it was the last link
3199 - use PtlGet from the target for bulk writes (315)
3200 - OST now updates object mtime with writes or setattr (607, 619)
3201 - LDLM now has a grant-time callback to revalidate locked items, if
3203 - Many MDS operations were reorganized to combat race conditions
3205 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3206 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3208 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3211 - Fully reactivate OST imports after reconnection (512, others)
3212 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3213 - More graceful error handling for truncating on dead OST (515)
3214 - Don't error out unless we're actually accessing dead stripes (474)
3215 - Fix garbage sizes when stripes are missing (410)
3216 - LRU counters were broken, causing constant lock purge (433, 432)
3217 - garbage on read from stripes with failed OSTs (441)
3218 - mark OSCs as active before reconnecting during recovery (438)
3219 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3220 - lfind did not preserve OST order in output (443)
3221 - symlinks cause hung clients, incorrect data (439)
3222 - stop dereferencing request after dropping refcount (457)
3223 - don't LASSERT(spin_is_locked) on non-SMP (455)
3224 - fixes for many rename() bugs
3225 - fstat didn't correctly synchronize attributes (399)
3226 - server must handle lock cancellation during blocking AST prep (487)
3227 - bulk descriptors were free()d too soon (511)
3228 - fix paths in lconf, which would load incorrect modules (451, 507)
3229 - fix confusing lconf 'host not found' error message (386)
3230 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3231 - fix race condition in mdc_blocking_ast() for inode access (526)
3232 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3233 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3234 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3235 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3236 - fix lstripe in several ways (kernel side) (527)
3237 - fix request leak in ldlm_cli_enqueue (262)
3238 - incorrect OSC was marked inactive after OST failure
3239 - call mds_fs_cleanup before unmounting filesystem (524)
3240 - fix races between taking ns_lock and ldlm_lock_change_resource
3241 - fix races updating LOV export open file list
3242 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3243 - fix recovery NULL deref in ldlm_cli_cancel_unused
3244 - fix some DLM races by using new hash table for lock handles (419)
3245 - permit the client to specify desired inodes, at replay
3246 - duplicate requests when we queue them for replay reintegration
3247 - fix last_rcvd offset calculation
3248 - sync after each recovered transaction, so we always make progress
3249 - never, not always, ERESTART requests without transnos
3250 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3252 - skip replay if the MDS says that the client is already connected
3253 - don't check for a recovery-enabled export to match lctl's UUID
3254 - don't INC_USE_COUNT for phantom exports
3255 - don't crash when cleaning up phantom exports (567)
3256 - don't double-finish or set replay data for errored mdc_open requests
3257 - abort requests when they time out, so we don't get old replies
3258 - send/receive replies for AST messages again
3259 - if the client says that it doesn't have the lock, cancel it on the
3261 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3262 mark it as destroyed, it will all work out when decref is called
3263 - fix module use counts (22, 581)
3265 - ASTs now expect a reply (server cancels lock on error reply)
3267 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3270 - fix many simultaneous client startup (392)
3271 - fix dentry->d_it clobbering
3272 - credentials weren't being shipped for readdir/getattr operations
3273 - remove invalid assertions triggered during some concurrent MD
3275 - proper Lustre versions added (336, 389)
3276 - fix memory leak for create error case (398)
3277 - fix LOV locking bug that would get cli/srv out of sync
3278 - fix echo client over LOV (409)
3279 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3280 - fix double-O_EXCL intent crash (424)
3281 - avoid sending multiple lock CANCELs (352)
3283 - MDS can do multi-client recovery (modulo bugs in new code)
3285 - many updates, edits, cleanups
3287 2002-11-18 Phil Schwan <phil@clusterfs.com>
3290 - fix null d_it dereference (346)
3291 - fix full OST/dbench hang (333)
3292 - fix permission problem with file removal (286)
3293 - fix removal of OSCs from LOV when they fail
3294 - fix NULL deref during bulk timeout (214)
3295 - fix problems related to multiple filesystems on one MDS (241)
3296 - fixed serious subtle metadata locking bugs
3297 - free locks on clients when inodes are removed due to memory
3299 - fix inode pointer in lock data (285)
3300 - partial support for multiple MDS on a single host (241)
3301 - data locks weren't cancelled at clear_inode time (290, 311)
3302 - intent locks could lead to unbounded lock growth (205)
3303 - added a maximum lock count, an LRU list, and a flusher
3304 - fix multiple rename (365)
3305 - properly abstracted the echo client
3306 - OSC locked 1 byte too many; fixed
3307 - rewrote brw callback code:
3308 - fixed recovery bugs related to LOVs (306)
3309 - fixed too-many-pages-in-one-write crash (191)
3310 - fixed (again) crash in sync_io_timeout (214)
3311 - probably fixed callback-related race (385)
3313 - Add capability to MDS protocol
3314 - LDLM cancellations and callbacks on different portals
3316 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3319 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3320 - defer open object destruction to close time (601981, 138)
3321 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3322 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3323 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3324 - don't call obd_setattr in ll_file_release for destroyed objects
3325 * protocol change to lustre_msg: move |version| and add |flags|
3326 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3327 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3328 * added replay of create, unlink, link and rename operations during
3329 MDS failover; recovery should be much more robust now
3330 * remove failed OSCs from LOVs (only lov_create uses this so far)
3331 * the lustre-HOWTO was brought (more) up to date (582544)
3333 2002-10-23 Phil Schwan <phil@clusterfs.com>
3336 - in-use dentries weren't being reused properly (617851)
3337 - prevent multiple LDLM setup (599178)
3338 - fix LOV size calculations for truncate (617853)
3339 - fix client handling of MDS intent errors (POSIX)
3340 - fix permission bug in lovstripe.c test (624321)
3341 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3342 - truncate past end of file could corrupt data
3343 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3344 - a race in recovery could return ETIMEDOUT to apps (623947)
3345 - building outside the source directory was fixed
3346 * the lustre-HOWTO was brought (more) up to date (582544)
3347 * major progress was made on recovery functionality
3349 2002-10-10 Phil Schwan <phil@clusterfs.com>
3352 - recovery deadlock fix
3353 - rm -rf causes LBUG fix (617817)
3354 - file open by multiple tasks fix (618962)
3355 - directory permissions bugs (602707 and 620007)
3356 - journal_stop fixed with locking (611313)
3357 - O_APPEND failures resolved (618273, perhaps 614459)
3358 - lconf PATH fix (619770)
3359 - IA64 build fix (621450)
3360 - RPC buffer sizes scale with amount of memory
3362 2002-10-01 Phil Schwan <phil@clusterfs.com>
3365 - locks would be cancelled without throwing away data pages,
3366 resulting in inconsistent data (605627)
3367 - inode attributes were not always being refreshed (605627, 612449)
3368 - lconf now continues to cleanup after lctl reports an error
3369 - MDS now enforces user permissions (602707)
3370 - lprocfs cleanup fixed, but not yet enabled (614157)
3371 - fixed infinite server hang, should a client not respond to an AST
3372 - avoid going into recovery if user calls readlink() with a buffer
3373 that's too small (613941)
3374 - AST RPCs no longer require replies (614867) -- this may be changed
3375 - don't crash server if client sends an IOV that's too big (611336)
3376 - fixed lock conversion deadlock (611892)
3377 - fixed the following of symlinks (614622)
3378 * recovery: the server can remove locks from a client that dies, other
3379 clients can make progress
3380 * more extN patch fixes
3381 * compile-time configurable ptlrpc buffer allocations
3383 - collaborative read cache document
3384 - Lustre Lite Performance CDR document-in-progress
3386 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3389 - fix typo in patch-2.4.18
3391 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3394 - clear ptlrpc request each time in handle_incoming_request()
3395 - unlink of files now destroys the object on the OST
3397 2002-09-19 Peter Braam <braam@clusterfs.com>
3399 * add hard link support
3400 * change obdfile creation method
3401 * kernel patch changed
3403 2002-09-19 Peter Braam <braam@clusterfs.com>
3406 - stack overflow bug in extN fixed
3408 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3410 * documentation updates
3411 - add man pages for config tools
3412 - update tests/README to describe testing with new config tools
3413 - finish metadata API descriptions
3414 * bug fixes and cleanups
3415 - statfs workaround for 16TB limit
3416 - LOV stripe allocation improved, can stripe on subset of OSTs
3417 - LOV file size/IO offset was wrong for files > 4GB in size
3418 - object EA data was being dropped, caused files to be unreadable
3419 - memory overflow with non-LOV OST caused memory corruption
3420 - fixed regression tests to work with new config tools, obdfilter
3421 - fixed bug when directory size became larger than 1 block
3422 - fixed bug (for single client case) when PWD was deleted
3423 - invalidate local directory pages when doing intent-based ops
3424 - avoid LDLM oops when lock callback contained bad data
3426 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
3428 * documentation updates
3429 * bug fixes and cleanups
3430 - configuration tools
3433 - 64-bit compile warnings
3434 - 64-bit internal statfs data
3436 * test_brw on persistent OST devices
3438 * lprocfs (disabled)
3440 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
3442 * documentation updates
3443 * bug fixes and cleanups
3444 * configuration tools
3446 2002-08-30 Peter J. Braam <braam@clusterfs.com>
3449 * many small fixes to 0_5_4
3450 * io/network handling
3451 * thinkos in MDS operations
3453 2002-08-24 Peter J. Braam <braam@clusterfs.com>
3456 * crucial basic fixes to 0.5.3
3457 * IOR, Iozone work over Elan
3460 2002-08-07 Phil Schwan <phil@clusterfs.com>
3461 * version 0_5_3, our first alpha
3462 * we use the new Portals iovs
3463 * documentation updates
3464 * bug fixes and cleanups
3465 * small changes in the DLM wire protocol
3467 2002-07-25 Peter J. Braam <braam@clusterfs.com>
3468 * version 0_5_1 with some initial stability,
3469 * locking on MD and file I/O.
3470 * documentation updates
3471 * several bug fixes since 0.5.0
3472 * small changes in wire protocol
3474 2002-07-18 Phil Schwan <phil@clusterfs.com>
3476 * delivered as Lustre Light Alpha
3477 * fixed a crash after handling invalid MDS requests
3478 * fixed directory pages for architectures with non-4k pages sizes
3480 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
3481 * release version v0_4_4
3482 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
3484 * Fixes a number of interruption problems with OST operations.
3485 * Update documentation for portals header changes
3486 * Move all wire protocol structs/defines to lustre_idl.h
3487 * Fixes symlink length bug.
3488 * Add tcpdump to repository.
3490 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
3491 * release version v0_4_3
3492 * Fixes statfs for inodes on extN.
3493 * Fixes bug in runtests which would delete /etc/hosts.
3494 * Use 64-bit object IDs wherever possible (not into VFS though)
3495 Remove ost_get_info, which is unused by lustre, and out of date.
3497 2002-07-03 Peter Braam <braam@clusterfs.com>
3498 * release version v0_4_2 Fixes a lookup error (type not passed)
3499 * move forward to head of Portals
3500 * move forward to latest Lustre kernel
3502 2002-06-25 Peter Braam <braam@clusterfs.com>
3503 * release version v0_4_1. Hopefully stable on single node use.