5 2.6.16.46-0.14 (SLES 10),
6 2.6.18-8.1.10.EL5 (RHEL 5).
7 2.6.18.8 vanilla (kernel.org)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.9-42.0.10.EL (RHEL 4),
11 2.6.16 - 2.6.21 vanilla (kernel.org)
12 * Recommended e2fsprogs version: 1.40.2-cfs1
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
15 Severity : enhancement
17 Description: self-adjustable client's lru lists
18 Details : use adaptive algorithm for managing client cached locks lru
19 lists according to current server load, other client's work
20 pattern, memory activities, etc. Both, server and client
21 side namespaces provide number of proc tunables for controlling
24 Severity : enhancement
26 Description: light-weight GSS support
27 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
28 and reduce performance overhead.
30 Severity : enhancement
32 Description: Linux keyring support
33 Details : Support using service of Linux keyring for Lustre GSS internal
34 context refresh/cache mechanism.
38 Description: Fix errors in lfs documentation
39 Details : Fixes man pages
43 Description: don't use GFP_* in generic Lustre code.
44 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
48 Description: obdclass is limited by single OBD_ALLOC(idarray)
49 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
53 Description: warnings with build patchless client with vanila 2.6.19 and up
54 Details : change old ctl_table style and replace ctl_table/ctl_table_header
55 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
59 Frequency : Only for SLES
60 Description: sanity_quota fail test_1
61 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
62 file, which makes TSTID[2] inunique.
66 Frequency : Only for SLES9
67 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
68 Details : Update target/ChangeLog/which_patch .
72 Frequency : Only for test_10 in sanity-quota.sh
73 Description: a bug in quota test code
74 Details : checking a test flag in a wrong place
78 Frequency : Only for test_13 in sanity-quota.sh
79 Description: a bug in quota test script
80 Details : shouldn't check overfull allocated quota
84 Description: jbd statistics
85 Details : Port older jbd statistics patch for sles10
87 Severity : enhancement
89 Description: Add printing inode info into message about error in writepage.
94 Description: reply_lock_interpret crash due to race with it and lock cancel.
95 Details : Do not replay locks that are being cancelled. Do not reference
96 locks by their address during replay, just by their handle.
100 Frequency : When flocks are used.
101 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
102 Details : Flock locks might destroy just granted lock if it could be merged
103 with another existing flock, this is done in completion handler,
104 so teach ldlm_cli_enquque_fini that this is a valid case for
110 Description: Oops in read and write path when failing to allocate lock.
111 Details : Check if lock allocation failed and return error back.
115 Description: lstripe command fails for valid OST index
116 Details : The stripe offset is compared to lov->desc.ld_tgt_count
117 instead of lov->desc.ld_active_tgt_count.
121 Description: sanity.sh failed test 103
122 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
127 Description: df doesn't work properly if diskfs blocksize != 4K
128 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
132 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
133 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
137 Description: obd_health_check_timeout too short
138 Details : set obd_health_check_timeout as 1.5x of obd_timeout
142 Description: llapi_file_create() does not allow some changes
143 Details : add llapi_file_open() that allows specifying the mode and
144 open flags, and also returns an open file handle.
148 Description: merge and cleanup kernel patches.
149 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
153 Description: Add journal checksum support.(Kernel part)
154 Details : The journal checksum feature adds two new flags i.e
155 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
156 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
157 indicates that the commit block contains the checksum for
158 the blocks described by the descriptor blocks. Now commit
159 record can be sent to disk without waiting for descriptor
160 blocks to be written to disk. This behavior is controlled
161 using JBD2_FEATURE_ASYNC_COMMIT flag.
165 Description: OSS needs mutliple precreate threads
166 Details : Add ability to start more than one create thread per OSS.
170 Description: Sanity.sh test_65k failed: setstripe should have succeeded
171 Details : execute lfs setstripe on client
175 Description: mds_obd_create error creating tmp object
176 Details : When the user sets quota on root, llog will be affected and can't
177 create files and write files.
180 Frequency : Always on ia64 patchless client, and possibly others.
182 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
183 Details : This allows the patchless client to be loaded on architectures
188 Description: RedHat Update kernel for RHEL5
189 Details : Add the kernel config file for RHEL5.
193 Description: RedHat Update kernel for RHEL5
194 Details : Modify the kernel config file more closer RHEL5.
198 Description: Build failure against Centos5 (RHEL5)
199 Details : Define PAGE_SIZE when it isn't present.
203 Description: test_7 on sanity-quota.sh can't be used on separate machines
204 Details : add facet to handle it
208 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
209 Details : with NFS, the anon dentry's parent was set to itself in
210 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
211 corresponding dentry other than use rec->ur_name.
213 Severity : enhancement
215 Description: lfs setstripe enhancement
216 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
220 Description: testing performance impact of enabling checksumming
221 Details : enable checksum by default, allow --disable-checksum
222 configure option and "-o nochecksum" mount option
226 Description: lustre support for RHEL5
227 Details : Add support for RHEL5.
231 Description: Client eviction due to failover config
232 Details : after a connection loss, the lustre client should attempt to
233 reconnect to the last active server first before trying the
234 other potential connections.
238 Description: when mds and osts use different quota unit(32bit and 64bit),
239 quota will be released repeatly.
240 Details : void sending multiple quota reqs to mds, which will keep the
241 status between the reqs.
245 Description: osts not allocated evenly to files
246 Details : change the condition to increase offset_idx
249 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
251 Description: Data corruption for OSTs that are formatted larger than 2TB
253 Details : When generating the bio request for lustre file writes the
254 sector number would overflow a temporary variable before being
255 used for the IO. The data reads correctly from Lustre (which
256 will overflow in a similar manner) but other file data or
257 filesystem metadata may be corrupted in some cases.
261 Description: Tune the kernel for good SCSI performance.
262 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
263 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
268 Description: rewrite ext2-derived code in obdclass/uuid.c
269 Details : rewrite inherited code (uuid parsing code from ext2 utils)
270 from scratch preserving functionality.
274 Description: "lfs find -obd UUID" prints directories
275 Details : "lfs find -obd UUID" will return all directory names instead
276 of just file names. It is incorrect because the directories
277 do not reside on the OSTs.
281 Description: use i_size_read and i_size_write in 2.6 port
282 Details : replace inode->i_size access with i_size_read/write()
285 Frequency : when removing large files
287 Description: scheduling issue during removal of large Lustre files
288 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
289 It causes scheduling issues when removing large files (17TB in the
293 Frequency : only with liblustre clients on XT3
295 Description: evictions taking too long
296 Details : allow llrd to evict clients directly on OSTs
299 Frequency : only on ppc
301 Description: /proc/fs/lustre/devices broken on ppc
302 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
303 We should make sure the type of 'index' is loff_t and then cast
304 explicitly as needed below. Do not assign an explicitly cast
308 Frequency : only for rhel5
310 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
311 Details : Modify the target file & which_kernel.
313 Severity : enhancement
315 Description: omit set fsid for export NFS
316 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
317 as FS_REQUIRES_DEV for avoid problems with generate fsid.
321 Frequency : Always, for kernels after 2.6.16
322 Description: Fix warning idr_remove called for id=.. which is not
324 Details : Last kernels save old s_dev before kill super and not allow
325 to restore from callback - restore it before call kill_anon_super.
329 Description: buffer overruns could theoretically occur
330 Details : llapi_semantic_traverse() modifies the "path" argument by
331 appending values to the end of the origin string, and a
332 overrun may occur. Adding buffer overrun check in liblustreapi.
334 --------------------------------------------------------------------------------
336 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
338 * Support for kernels:
339 2.6.5-7.283 (SLES 9),
340 2.6.9-55.EL (RHEL 4),
341 2.6.16.46-0.14 (SLES 10),
342 2.6.18.8 vanilla (kernel.org)
343 * Client support for unpatched kernels:
344 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
345 2.6.16 - 2.6.22 vanilla (kernel.org)
346 * Due to recently discovered recovery problems, we do not recommend
347 using patchless RHEL 4 clients with this or any earlier release.
348 * Recommended e2fsprogs version: 1.39.cfs8
349 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
350 * Starting with this release, the ldiskfs backing filesystem required
351 by Lustre is now in its own package, lustre-ldiskfs. This package
352 should be installed. It is versioned separately from Lustre and
353 may be released separately in future.
357 Description: block reactivating mgc import until all deactivates complete
358 Details : Fix race when failing back MDT/MGS to itself (testing)
360 Severity : enhancement
362 Description: add optional extra BUILD_VERSION info
363 Details : add a new environment variable (namely LUSTRE_VERS) which allows
364 to override the lustre version.
367 Frequency : 2.6.18 servers only
369 Description: ll_kern_mount() doesn't release the module reference
370 Details : The ldiskfs module reference count never drops down to 0
371 because ll_kern_mount() doesn't release the module reference.
376 Description: server LBUG when using old ost_num_threads parameter
377 Details : Accept the old ost_num_threads parameter but warn that it
378 is deprecated, and fix an off-by-one error that caused an LBUG.
383 Description: Transient SCSI error results in persistent IO issue
384 Details : iobuf->dr_error is not reinitialized to 0 between two
388 Frequency : sometimes when underlying device returns I/O errors
390 Description: OSTs not going read-only during write failures
391 Details : OSTs are not remounted read-only when the journal commit threads
392 get I/O errors because fsfilt_ext3 calls journal_start/stop()
393 instead of the ext3 wrappers.
397 Description: poor connect scaling with increasing client count
398 Details : Don't run filter_grant_sanity_check for more than 100 exports
399 to improve scaling for large numbers of clients.
402 Frequency : SLES10 only
404 Description: sanity-quota.sh quotacheck failed: rc = -22
405 Details : Quotas cannot be enabled on SLES10.
408 Frequency : liblustre clients only
410 Description: getdirentries does not give error when run on compute nodes
411 Details : getdirentries does not fail when the size specified as an argument
412 is too small to contain at least one entry
414 Severity : enhancement
416 Description: Add LNET router traceability for debug purposes
417 Details : If a checksum failure occurs with a router as part of the
418 IO path, the NID of the last router that forwarded the bulk data
419 is printed so it can be identified.
424 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
425 Details : Due to a race condition, liblustre clients were occasionally
428 Severity : enhancement
430 Description: lfs setstripe use optional parameters instead of postional
433 Severity : enhancement
435 Description: Nanosecond timestamp support for ldiskfs
436 Details : The on-disk ldiskfs filesystem has added support for nanosecond
437 resolution timestamps. There is not yet support for this at
438 the Lustre filesystem level.
441 Frequency : during server recovery
443 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
444 Details : request with rq_no_resend flag not awake l_wait_event if they get a
448 Frequency : nfs export on patchless client
450 Description: connectathon hang when test nfs export over patchless client
451 Details : Disconnected dentry cannot be found with lookup, so we do not need
452 to unhash it or make it invalid
455 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
456 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
457 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
462 Description: open req refcounting wrong on reconnect
463 Details : If reconnect happened between getting open reply from server and
464 call to mdc_set_replay_data in ll_file_open, we will schedule
465 replay for unreferenced request that we are about to free.
466 Subsequent close will crash in variety of ways.
467 Check that request is still eligible for replay in
468 mdc_set_replay_data().
473 Description: disable writes to filesystem when reading health_check file
474 Details : the default for reading the health_check proc file has changed
475 to NOT do a journal transaction and write to disk, because this
476 can cause reads of the /proc file to hang and block HA state
477 checking on a healthy but otherwise heavily loaded system. It
478 is possible to return to the previous behaviour during configure
479 with --enable-health-write.
481 Severity : enhancement
483 Description: 64-bit inode version
484 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
485 to the inode. This will be required for version-based recovery.
490 Description: MDS fails to start if a duplicate client export is detected
491 Details : in some rare cases it was possible for a client to connect to
492 an MDS multiple times. Upon recovery the MDS would detect this
493 and fail during startup. Handle this more gracefully.
495 Severity : enhancement
497 Description: Add -o localflock option to simulate old noflock
499 Details : This will achieve local-only flock/fcntl locks
505 Description: log_commit_thread vs filter_destroy race leads to crash
506 Details : Take import reference before releasing llog record semaphore
511 Description: Wrong request locking in request set processing
512 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
513 imp_delayed_list, in this place should be used imp_lock.
516 Frequency : when reconnection
518 Description: Grant Leak when osc reconnect to OST
519 Details : When osc reconnect ost, OST(filter) should check whether it
520 should grant more space to client by comparing fed_grant and
521 cl_avail_grant, and return the granted space to client instead
522 of "new granted" space, because client will call osc_init_grant
523 to update the client grant space info.
526 Frequency : when client reconnect to OST
528 Description: Grant Leak when osc do resend and replay bulk write
529 Details : When osc reconnect to OST, OST(filter)should clear grant info of
530 bulk write request, because the grant info while be sync between
531 OSC and OST when reconnect, and we should ignore the grant info
532 these of resend/replay write req.
537 Description: Grant space more than avaiable left space sometimes.
538 Details : When then OST is about to be full, if two bulk writing from
539 different clients came to OST. Accord the avaliable space of the
540 OST, the first req should be permitted, and the second one
541 should be denied by ENOSPC. But if the seconde arrived before
542 the first one is commited. The OST might wrongly permit second
543 writing, which will cause grant space > avaiable space.
546 Frequency : when client is evicted
548 Description: Grant might be wrongly erased when osc is evicted by OST
549 Details : when the import is evicted by server, it will fork another
550 thread ptlrpc_invalidate_import_thread to invalidate the
551 import, where the grant will be set to 0. While the original
552 thread will update the grant it got when connecting. So if
553 the former happened latter, the grant will be wrongly errased
554 because of this race.
559 Description: Checking Stale with correct fid
560 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
561 to check whether it is stale, because sometimes, we want the
562 enqueue happened anyway, and op_data.fid2 will not be initialized.
564 Severity : enhancement
566 Description: update patchless client
567 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
570 Frequency : only with 2.4 kernel
572 Description: random memory corruption
573 Details : size of struct ll_inode_info is to big for union inode.u and this
574 can be cause of random memory corruption.
579 Description: Memory leak in recovery
580 Details : Lov_mds_md was not free in an error handler in mds_create_object.
581 It should also check obd_fail before fsfilt_start, otherwise if
582 fsfilt_start return -EROFS,(failover mds during mds recovery).
583 then the req will return with repmsg->transno = 0 and rc = EROFS.
584 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
585 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
586 be freed no matter whether fsfilt_commit success or not.
589 Frequency : only with huge count clients
591 Description: Prevents from taking the superblock lock in llap_from_page for
593 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
594 for llap_from_page call in ll_removepage prevents from taking the
595 superblock lock for a soon died page.
600 Description: Not check open intent error before release open handle
601 Details : in some rare cases, the open intent error is not checked before
602 release open handle, which may cause
603 ASSERTION(open_req->rq_transno != 0), because it tries to release
604 the failed open handle.
609 Description: Set cat log bitmap only after create log success.
610 Details : in some rare cases, the cat log bitmap is set too early. and it
611 should be set only after create log success.
615 Description: Accessing a block bevice can re-enable I/O when Lustre is
616 tearing down a device.
617 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
622 Description: service threads may hog cpus when there are a lot of requests
624 Details : Insert cond_resched to give other threads a chance to use some of
630 Description: the cat log was not initialized in recovery
631 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
632 unlink log on mds will not be initialized until mds post
633 recovery. And also in mds post recovery, the unlink log will
634 initialization will be done asynchronausly, so there will be race
635 between add unlink log and unlink log initialization.
639 Description: brw_stats were being printed incorrectly
640 Details : brw_stats were being printed as log2 but all of them were not
641 recorded as log2. Also remove some code duplication arising from
642 filter_tally_{read,write}.
646 Frequency : rare, only in recovery.
647 Description: ASSERTION(req->rq_type != LI_POISON) failed
648 Details : imp_lock should be held while iterating over imp_sending_list for
649 prevent destroy request after get timeout in ptlrpc_queue_wait.
653 Description: replay-single.sh test 52 fails
654 Details : A lock's skiplist need to be cleanup when it being unlinked
655 from its resource list.
659 Description: Short directio read returns full requested size rather than
661 Details : Direct I/O operations should return actual amount of bytes
662 transferred rather than requested size.
664 Severity : enhancement
666 Description: metadata RPC reduction (e.g. for rm performance)
667 Details : decrease the amount of synchronous RPC between clients and servers
668 by canceling conflicing lock before the operation on the client side
669 and packing thier handles into the main operation RPC to server.
671 Severity : enhancement
673 Description: Async OSC create to avoid the blocking unnecessarily.
674 Details : If a OST has no remain object, system will block on the creating
675 when need to create a new object on this OST. Now, ways use
676 pre-created objects when available, instead of blocking on an
677 empty osc while others are not empty. If we must block, we block
678 for the shortest possible period of time.
680 --------------------------------------------------------------------------------
682 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
687 Frequency : on some architectures
689 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
690 Details : Uninitialized flags sometimes cause configuration commands to
694 Frequency : patchless clients only
696 Description: missing __iget() symbol export
697 Details : The __iget() symbol export is missing. To avoid the need for
698 this on patchless clients the deathrow inode reaper is turned
699 off, and we depend on the VM to clean up old inodes. This
700 dependency was during via the fix for bug 12181.
702 --------------------------------------------------------------------------------
704 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
706 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
707 INTEROPERATE with older versions automatically. In many cases a
708 special upgrade step is needed. Please read the
709 user documentation before upgrading any part of a live system.
710 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
711 version will not interoperate with 1.6 betas before beta5 (1.5.95).
712 * WARNING: Lustre configuration and startup changes are required with
713 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
719 Severity : enhancement
721 Description: MountConf
722 Details : Lustre configuration is now managed via mkfs and mount
723 commands instead of lmc and lconf. New obd types (MGS, MGC)
724 are added for dynamic configuration management. See
725 https://mail.clusterfs.com/wikis/lustre/MountConf for
728 Severity : enhancement
730 Description: dynamic OST addition
731 Details : OSTs can now be added to a live filesystem
733 Severity : enhancement
735 Description: startup order invariance
736 Details : MDTs and OSTs can be started in any order. Clients only
737 require the MDT to complete startup.
739 Severity : enhancement
741 Description: parallel, asynchronous orphan cleanup
742 Details : orphan cleanup is now performed in separate threads for each
743 OST, allowing parallel non-blocking operation.
745 Severity : enhancement
747 Description: optimized stripe assignment
748 Details : stripe assignments are now made based on ost space available,
749 ost previous usage, and OSS previous usage, in order to try
750 to optimize storage space and networking resources.
752 Severity : enhancement
754 Description: Permanently set tunables
755 Details : All writable /proc/fs/lustre tunables can now be permanently
756 set on a per-server basis, at mkfs time or on a live
759 Severity : enhancement
761 Description: Lustre message v2
762 Details : Add lustre message format v2.
764 Severity : enhancement
766 Description: client OST exclusion list
767 Details : Clients can be started with a list of OSTs that should be
768 declared "inactive" for known non-responsive OSTs.
772 Description: SPEC SFS validation failure on NFS v2 over lustre.
773 Details : Changes the blocksize for regular files to be 2x RPC size,
774 and not depend on stripe size.
776 Severity : enhancement
778 Description: Multiple MD RPCs in flight.
779 Details : Further unserialise some read-only MDS RPCs - learn about intents.
780 To avoid overly-overloading MDS, introduce a limit on number of
781 MDS RPCs in flight for a single client and add /proc controls
782 to adjust this limit.
784 Severity : enhancement
786 Description: client read/write statistics
787 Details : Add client read/write call usage stats for performance
788 analysis of user processes.
789 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
790 file access. extents_stats shows chunk size distribution.
791 extents_stats_per_process show chunk size distribution per
794 Severity : enhancement
796 Description: mds statistics
797 Details : Add detailed mds operations statistics in
798 /proc/fs/lustre/mds/*/stats.
802 Description: Failure of copying files with lustre special EAs.
803 Details : Client side always return success for setxattr call for lustre
804 special xattr (currently only "trusted.lov").
809 Description: Refcount LNET uuids
810 Details : The global LNET uuid list grew linearly with every startup;
811 refcount repeated list entries instead of always adding to
814 Severity : enhancement
816 Description: Dynamic service threads
817 Details : Within a small range, start extra service threads
818 automatically when the request queue builds up.
821 Frequency : mixed-endian client/server environments
823 Description: mixed-endian crashes
824 Details : The new msg_v2 system had some failures in mixed-endian
827 Severity : enhancement
829 Description: Easy OST removal
830 Details : OSTs can be permanently deactivated with e.g. 'lctl
831 conf_param lustre-OST0001.osc.active=0'
833 Severity : enhancement
835 Description: MGS proc entries
836 Details : Added basic proc entries for the MGS showing what filesystems
839 Severity : enhancement
841 Description: provide MGS failover
842 Details : Added config lock reacquisition after MGS server failover.
844 Severity : enhancement
846 Description: add Linux 2.4 support
847 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
851 Description: a large application tries to do I/O to the same resource and dies
853 Details : Check the req->rq_arrival time after the call to
854 ost_brw_lock_get(), but before we do anything about
855 processing it & sending the BULK transfer request. This
856 should help move old stale pending locks off the queue as
857 quickly as obd_timeout.
860 Frequency : when an incorrect nid is specified during startup
862 Description: ptlrpc connect to non-existant node causes kernel crash
863 Details : LNET can't be re-entered from an event callback, which
864 happened when we expire a message after the export has been
865 cleaned up. Instead, hand the zombie cleanup off to another
868 Severity : enhancement
870 Description: plain/inodebits lock performance improvement
871 Details : Grouping plain/inodebits in granted list by their request modes
872 and bits policy, thus improving the performance of search through
876 Frequency : only if OST filesystem is corrupted
878 Description: client incorrectly hits assertion in ptlrpc_replay_req()
879 Details : for a short time RPCs with bulk IO are in the replay list,
880 but replay of bulk IOs is unimplemented. If the OST filesystem
881 is corrupted due to disk cache incoherency and then replay is
882 started it is possible to trip an assertion. Avoid putting
883 committed RPCs into the replay list at all to avoid this issue.
886 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
888 Description: System hang on startup
889 Details : This bug allowed the liblustre (e.g. catamount) client to
890 return to the app before handling all startup RPCs. This
891 could leave the node unresponsive to lustre network traffic
892 and manifested as a server ptllnd timeout.
894 Severity : enhancement
896 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
897 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
898 boolean to control whether to print peer debug info when a
899 client's RPC times out.
902 Frequency : only for kernels with patches from Lustre below 1.4.3
904 Description: Remove old rdonly API
905 Details : Remove old rdonly API which unused from at least lustre 1.4.3
908 Frequency : only for devices with external journals
910 Description: Set external device read-only also
911 Details : During a commanded failover stop, we set the disk device
912 read-only while the server shuts down. We now also set any
913 external journal device read-only at the same time.
916 Frequency : when upgrading from 1.4 while trying to change parameters
918 Description: The wrong (new) MDC name was used when setting parameters for
919 upgraded MDT's. Also allows changing of OSC (and MDC)
920 parameters if --writeconf is specified at tunefs upgrade time.
923 Frequency : when setting specific ost indicies
925 Description: QOS code breaks on skipped indicies
926 Details : Add checks for missing OST indicies in the QOS code, so OSTs
927 created with --index need not be sequential.
929 Severity : enhancement
931 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
932 Details : The uninit_groups feature works in conjunction with the kernel
933 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
934 up the pass1 processing of e2fsck. This is a read-only feature
935 in ldiskfs2 only, so older kernels and current ldiskfs cannot
936 mount filesystems that have had this feature enabled.
938 Severity : enhancement
940 Description: Improve multi-block allocation algorithm to avoid fragmentation
941 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
942 allocation locality and avoid filesystem fragmentation.
944 ------------------------------------------------------------------------------
946 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
948 * Support for kernels:
949 2.4.21-47.0.1.EL (RHEL 3)
951 2.6.9-42.0.10.EL (RHEL 4)
952 2.6.12.6 vanilla (kernel.org)
953 2.6.16.27-0.9 (SLES 10)
954 * Recommended e2fsprogs version: 1.39.cfs5
956 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
960 Frequency : occasional, depends on client load and configuration
961 Bugzilla : 12181, 12203
962 Description: data loss for recently-modified files
964 Details : In some cases it is possible that recently written or created
965 files may not be written to disk in a timely manner (this should
966 normally be within 30s unless client IO load is very high).
967 The problem appears as zero-length files or files that are a
968 multiple of 1MB in size after a client crash or client eviction
969 that are missing data at the end of the file.
971 This problem is more likely to be hit on clients where files are
972 repeatedly created and unlinked in the same directory, clients
973 have a large amount of RAM, have many CPUs, the filesystem has
974 many OSTs, the clients are rebooted frequently, and/or the files
975 are not accessed by other nodes after being written.
977 The presence of the problem can be detected by looking at
978 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
979 smaller than the second (nr_unused) then dirty files will not
980 be flushed automatically to disk. "sync; sleep 10" should be
981 run several times on the node before unmounting it to update
982 Lustre (this is also safe to run on nodes without this problem).
984 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
985 that can cause this same problem, so customers using that kernel
986 also need to update the kernel in addition to Lustre. In order
987 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
989 It is normal that files written just before a client crash (less
990 than 30s) may not yet have been flushed to disk, even for local
994 Frequency : frequent on thin XT3 nodes
996 Description: UUID collision on thin XT3 Linux nodes
997 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
998 randomly, since we relied on an insufficiently-seeded PRNG.
1003 Description: OSS hangs after "All ost request buffers busy"
1004 Details : A deadlock between quota and journal operations caused OSS
1005 hangs after printing "All ost request buffers busy."
1008 Frequency : always on liblustre builds
1010 Description: Cleanup compiler warnings on liblustre
1013 Frequency : always on liblustre builds on XT3
1015 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1019 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1020 > 32000 subdirectories
1021 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1022 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1023 subdirectory count crosses 32000. This will aid e2fsck to
1024 correctly handle more than 32000 subdirectories.
1027 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1029 Description: System hang on startup
1030 Details : This bug allowed the liblustre (e.g. catamount) client to
1031 return to the app before handling all startup RPCs. This
1032 could leave the node unresponsive to lustre network traffic
1033 and manifested as a server ptllnd timeout.
1035 Severity : enhancement
1037 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1038 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1039 boolean to control whether to print peer debug info when a
1040 client's RPC times out.
1045 Description: make O_SYNC working on 2.6 kernels
1046 Details : 2.6 kernels use different method for mark pages for write,
1047 so need add a code to lustre for O_SYNC work.
1052 Description: Failure to close file and release space on NFS
1053 Details : Put inode details into lock acquired in ll_intent_file_open.
1054 Use mdc_intent_lock in ll_intent_open to properly
1055 detect all kind of errors unhandled by mdc_enqueue.
1060 Description: proc file read during shutdown sometimes raced obd removal,
1062 Details : Add lock to prevent obd access after proc file removal.
1065 Frequency : Only for files larger than 4GB on 32-bit clients.
1067 Description: improperly doing page alignment of locks
1068 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1069 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1074 Description: RPCs being resent when they shouldn't be
1075 Details : Some RPCs that should not be resent are being resent. This
1076 can cause inconsistencies in the RPC state machine. Do not
1077 resend such requests.
1080 Frequency : rare, only with NFS export
1082 Description: Crash on NFS re-export node
1083 Details : under very unusual load conditions an assertion is hit in
1084 ll_intent_file_open()
1087 Frequency : only if OST filesystem is corrupted
1089 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1090 Details : for a short time RPCs with bulk IO are in the replay list,
1091 but replay of bulk IOs is unimplemented. If the OST filesystem
1092 is corrupted due to disk cache incoherency and then replay is
1093 started it is possible to trip an assertion. Avoid putting
1094 committed RPCs into the replay list at all to avoid this issue.
1099 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1100 Details : Large single O_DIRECT read and write calls can fail to allocate
1101 a sufficiently large buffer to process the request. In case of
1102 allocation failure the allocation is retried with a smaller
1103 buffer and broken into smaller requests.
1105 Severity : enhancement
1107 Description: Add -o localflock option to simulate old noflock behaviour.
1108 Details : This will achieve local-only flock/fcntl locks coherentness.
1113 Description: versioning check is incomplete
1114 Details : Checking the version difference of client vs. server, report
1115 error if the gap is too big.
1120 Description: add support PG_writeback bit
1121 Details : add support for PG_writeback bit for Lustre, for more carefull
1122 work with page cache in 2.6 kernel. This also fix some deadlocks
1123 and remove hack for work O_SYNC with 2.6 kernel.
1125 Severity : enhancement
1127 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1128 Details : The uninit_groups feature works in conjunction with the kernel
1129 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1130 up the pass1 processing of e2fsck. This is a read-only feature
1131 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1132 mount filesystems that have had this feature enabled.
1134 Severity : enhancement
1136 Description: Improve multi-block allocation algorithm to avoid fragmentation
1137 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1138 allocation locality and avoid filesystem fragmentation.
1140 ------------------------------------------------------------------------------
1142 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1144 * Support for kernels:
1145 2.6.9-42.0.3.EL (RHEL 4)
1146 2.6.5-7.276 (SLES 9)
1147 2.4.21-47.0.1.EL (RHEL 3)
1148 2.6.12.6 vanilla (kernel.org)
1149 2.6.16.21-0.8 (SLES10)
1150 * Recommended e2fsprogs version: 1.39.cfs2-0
1152 * The backwards-compatible /proc/sys/portals symlink has been removed
1153 in this release. Before upgrading, please ensure that you change
1154 any configuration scripts or /etc/sysctl.conf files that access
1155 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1156 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1157 in advance of the upgrade on any system running Lustre 1.4.6 or
1158 newer, since /proc/sys/lnet was added in that version.
1159 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1163 Frequency : only when quota is used
1165 Description: avoid scanning export list for quota master
1166 Details : Change the algorithms to avoid scanning export list in order
1167 to improve the efficiency.
1170 Frequency : MDS failover only, very rarely
1172 Description: "went back in time" messages on mds failover
1173 Details : The greatest transno may be lost when the current operation
1174 finishes with an error (transno==0) and the client's last_rcvd
1175 record is over-written. Save the greatest transno in the
1176 mds_last_transno for this case.
1179 Frequency : always for specific kernels and striping counts
1181 Description: client may get "Matching packet too big" without ACL support
1182 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1183 when trying to access files in certain configurations. The
1184 clients should in fact be denied when mounting because they do
1185 not understand ACLs.
1188 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1190 Description: many clients connecting with IO in progress causes connect timeouts
1191 Details : Avoid synchronous journal commits to avoid delays caused by many
1192 clients connecting/disconnecting when bulk IO is in progress.
1193 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1194 OST_IO_PORTAL to avoid delays behind potentially many pending
1198 Frequency : occasionally with multiple writers to a single file
1200 Description: shared writes to file may result in wrong size reported by stat()
1201 Details : Allow growing of kms when extent lock is cancelled
1204 Frequency : always with random mmap IO to multi-striped file
1206 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1207 Details : Only if the hole is at the end of OST object so that kms is too
1208 small. Fix is to increase kms accordingly in ll_nopage.
1211 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1213 Description: writes to a missing object would leak memory on the OST
1214 Details : If there is an inconsistency between the MDS and OST filesystems,
1215 such that the MDS references an object that doesn't exist, writes
1216 to that object will leak memory due to incorrect cleanup in the
1217 error handling path, eventually running out of memory on the OST.
1222 Description: Creating too long symlink causes lustre errors
1223 Details : Check symlink and name lengths before sending requests to MDS.
1226 Frequency : only if flock is enabled (not on by default)
1228 Description: posix locks not released on fd closure on 2.6.9+
1229 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1230 kernels, this caused such locks not to be released on fd close and
1231 then assertions on fs unmount about still used locks.
1234 Frequency : MDS failover only, very rarely
1236 Description: clients may get ASSERTION(granted_lock != NULL)
1237 Details : When request was taking a long time, and a client was resending
1238 a getattr by name lock request. The were multiple lock requests
1239 with the same client lock handle and
1240 mds_getattr_name->fixup_handle_for_resent_request found one of the
1241 lock handles but later failed with ASSERTION(granted_lock != NULL).
1246 Description: handle->h_buffer_credits > 0, assertion failure
1247 Details : h_buffer_credits is zero after truncate, causing assertion
1248 failure. This patch extends the transaction or creates a new
1252 Frequency : NFS re-export or patchless client
1253 Bugzilla : 11179, 10796
1254 Description: Crash on NFS re-export node (__d_move)
1255 Details : We do not want to hash the dentry if we don't have a lock.
1256 But if this dentry is later used in d_move, we'd hit uninitialised
1257 list head d_hash, so we just do this to init d_hash field but
1258 leave dentry unhashed.
1261 Frequency : NFS re-export or patchless client
1263 Description: NFS exports has problem with symbolic link
1264 Details : lustre client didn't properly install dentry when re-exported
1265 to NFS or running patchless client.
1268 Frequency : NFS re-export or patchless client
1270 Description: Various nfs/patchless fixes.
1271 Details : fixes reuse disconected alias for lookup process - this fixes
1272 warning "find_exported_dentry: npd != pd",
1273 fix permission error with open files at nfs.
1274 fix apply umask when do revalidate.
1277 Frequency : occasional
1279 Description: Crash on NFS re-export node
1280 Details : calling clear_page() on the wrong pointer triggered oops in
1281 generic_mapping_read().
1284 Frequency : rarely, using O_DIRECT IO
1286 Description: unaligned directio crashes client with LASSERT
1287 Details : check for unaligned buffers before trying any requests.
1290 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1292 Description: stale data returned from RAID cache
1293 Details : If only a small amount of IO is done to the RAID device before
1294 reading it again it is possible to get stale data from the RAID
1295 cache instead of reading it from disk.
1298 Frequency : always for sles10 kernel
1300 Description: sles10 support
1301 Details : ll_follow_link: compile fixes and using of nd_set_link
1302 under newer kernels.
1305 Frequency : depends on arch, kernel and compiler version, always on sles10
1308 Description: recursive or deep enough symlinks cause stack overflow
1309 Details : getting rid of large stack-allocated variable in
1313 Frequency : depends on hardware
1315 Description: lustre write performance loss in the SLES10 kernel
1316 Details : the performance loss is caused by using of write barriers in the
1317 ext3 code. The SLES10 kernel turns barrier support on by
1318 default. The fix is to undo that change for ldiskfs.
1320 ------------------------------------------------------------------------------
1322 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1324 * Support for kernels:
1325 2.6.9-42.0.3EL (RHEL 4)
1326 2.6.5-7.276 (SLES 9)
1327 2.4.21-47.0.1.EL (RHEL 3)
1328 2.6.12.6 vanilla (kernel.org)
1332 Frequency : quota enabled and large files being deleted
1334 Description: releasing more than 4GB of quota at once hangs OST
1335 Details : If a user deletes more than 4GB of files on a single OST it
1336 will cause the OST to spin in an infinite loop. Release
1337 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1342 Description: statfs data retrieved from /proc may be stale or zero
1343 Details : When reading per-device statfs data from /proc, in the
1344 {kbytes,files}_{total,free,avail} files, it may appear
1345 as zero or be out of date.
1348 Frequency : systems with MD RAID1 external journal devices
1350 Description: lconf's call to blkid is confused by RAID1 journal devices
1351 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1352 of returning all block devices that match the journal UUID.
1355 Frequency : always, for aggregate stripe size over 4GB
1357 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1358 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1359 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1362 Frequency : always if "lfs find" used on a local file/directory
1364 Description: "lfs find" segfaults if used on a local file/directory
1365 Details : The case where a directory component was not specified wasn't
1366 handled correctly. Handle this properly.
1369 Frequency : always on ppc64
1371 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1372 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1373 perform endianness conversion.
1376 Frequency : rarely (truncate to non-zero file size after write under load)
1377 Bugzilla : 10730, 10687
1378 Description: Files padded with zeros to next 4K multiple
1379 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1380 it is possible that files that are truncated to a non-zero size
1381 immediately after being written are filled with zero bytes beyond
1382 the truncated size. No file data is lost.
1384 Severity : enhancement
1386 Description: Allow recovery/failover for liblustre clients.
1387 Details : liblustre clients were unaware of failover configurations until
1390 Severity : enhancement
1392 Description: user file locks should fail when not mounting with flock option
1393 Details : Set up an error-returning stub in ll_file_operations.lock field
1394 to prevent incorrect behaviour when client is mounted without
1395 flock option. Also, set up properly f_op->flock field for
1399 Frequency : always on ia64
1401 Description: "lfs df" loops on printing out MDS statfs information
1402 Details : The obd_ioctl_data was not initialized and in some systems
1403 this caused a failure during the ioctl that did not return
1404 an error. Initialize the struct and return an error on failure.
1407 Frequency : SLES 9 only
1409 Description: Error of copying files with lustre special EAs as root
1410 Details : Client side always return success for setxattr call for lustre
1411 special xattr (currently only "trusted.lov").
1414 Frequency : rarely on clusters with both ia64+i386 clients
1416 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1417 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1418 clients are doing concurrent writes to the same file it is
1419 possible that the ia64 clients may LASSERT because the OST
1420 extent locks are not PAGE_SIZE aligned. Ensure that grown
1421 locks are always aligned on the request boundary.
1424 Frequency : specific use, occasional
1426 Description: Overwriting in use executable truncates on-disk binary image
1427 Details : If one node attempts to overwrite an executable in use by
1428 another node, we now correctly return ETXTBSY instead of
1429 truncating the file.
1431 Severity : enhancement
1433 Description: Async OSC create to avoid the blocking unnecessarily.
1434 Details : If a OST has no remain object, system will block on the creating
1435 when need to create a new object on this OST. Now, ways use
1436 pre-created objects when available, instead of blocking on an
1437 empty osc while others are not empty. If we must block, we block
1438 for the shortest possible period of time.
1443 Description: chmod on Lustre root is propagated to other clients
1444 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1445 invalid by the follow_mount time.
1448 Frequency : liblustre clients only
1450 Description: Race in 'instant cancel' lock handling could lead to such locks
1451 never to be granted in case of SMP MDS
1452 Details : Do not destroy not yet granted but cbpending locks in
1456 Frequency : replay/resend of open
1458 Description: non null lock assetion failure in mds_intent_policy
1459 Details : Trying to replay/resend lockless open requests resulted in
1460 mds_open() returning 0 with no lock. Now it sets a flag if
1461 a lock is going to be returned.
1463 Severity : enhancement
1465 Description: Checksum enhancements
1466 Details : New checksum enhancements allow for resending RPCs that failed
1469 Severity : enhancement
1471 Description: Tunables on number of dirty pages in cacche
1472 Details : Allow to set limit on number of dirty pages cached.
1477 Description: client crash on unmount - lock still has references
1478 Details : In some error handling cases it was possible to leak a lock
1479 reference on a client while accessing a file. This was not
1480 harmful to the client during operation, but would cause the
1481 client to crash when the filesystem is unmounted.
1484 Frequency : specific case, rare
1486 Description: ETXTBSY on mds though file not in use by client
1487 Details : ETXTBSY is no longer incorrectly returned when attempting to
1488 chmod or chown a directory that the user previously tried to
1489 execute or a currently-executing binary.
1492 Frequency : extremely rare except on liblustre-based clients
1494 Description: Lustre space not freed when files are deleted
1495 Details : Clean up open-unlinked files after client eviction. Previously
1496 the unlink was skipped and the files remained as orphans.
1501 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1502 Details : In some cases it was possible to send a blocking callback to
1503 a client doing a glimpse, even though that client didn't get
1504 a lock granted. When the glimpse lock is cancelled on the OST
1505 the freed lock is left on the waiting list and corrupted the list.
1508 Frequency : all core dumps
1510 Description: Broke core dumps to lustre
1511 Details : Negative dentry may be unhashed if parent does not have UPDATE
1512 lock, but some callers, e.g. do_coredump, expect dentry to be
1513 hashed after successful create, hash it in ll_create_it.
1515 ------------------------------------------------------------------------------
1517 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1519 * Support for kernels:
1520 2.6.9-42.0.2.EL (RHEL 4)
1521 2.6.5-7.276 (SLES 9)
1522 2.4.21-40.EL (RHEL 3)
1523 2.6.12.6 vanilla (kernel.org)
1527 Frequency : always on RHEL 3
1529 Description: Number of open files grows over time
1530 Details : The number of open files grows over time, whether or not
1531 Lustre is started. This was due to a filp leak introduced
1532 by one of our kernel patches.
1534 ------------------------------------------------------------------------------
1536 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1538 * Support for kernels:
1539 2.6.9-42.EL (RHEL 4)
1540 2.6.5-7.276 (SLES 9)
1541 2.4.21-40.EL (RHEL 3)
1542 2.6.12.6 vanilla (kernel.org)
1547 Bugzilla : 5719, 9635, 9792, 9684
1548 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1549 Details : If a server is under heavy load and cannot reply to new
1550 connection requests before the client resends the (re)connect,
1551 the connection handling code can behave badly if two service
1552 threads are concurrently handing separate (re)connections from
1553 the same client. Add better locking to the connection handling
1554 code, and ensure that only a single connection will be processed
1555 for a given client UUID, even if the lock is dropped.
1557 Severity : enhancement
1559 Description: add TCP zero-copy support to kernel
1560 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1561 sends if the hardware supports scatter-gather and checksumming.
1562 This allows socklnd to do client-write and server-read more
1563 efficiently and reduce CPU utilization from skbuf copying.
1566 Frequency : only if NFS exporting from client
1568 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1569 Details : If setting attributes on a file created under NFS that had
1570 never been opened it would be possible to oops the client
1571 if the file had no objects.
1574 Frequency : always for liblustre
1576 Description: liblustre client does MDS+OSTs setattr RPC for each write
1577 Details : When doing a write from a liblustre client, the client
1578 incorrectly issued an RPC to the MDS and each OST the file was
1579 striped over in order to update the timestamps. When writing
1580 with small chunks and many clients this could overwhelm the MDS
1581 with RPCs. In all cases it would slow down the write because
1582 these RPCs are unnecessary.
1584 Severity : enhancement
1586 Description: allow number of MDS service threads to be changed at module load
1587 Details : It is now possible to change the number of MDS service threads
1588 running. Adding "options mds mds_num_threads={N}" to the MDS's
1589 /etc/modprobe.conf will set the number of threads for the next
1590 time Lustre is restarted (assuming the "mds" module is also
1591 reloaded at that time). The default number of threads will
1592 stay the same, 32 for most systems.
1597 Description: OST crash if filesystem is unformatted or corrupt
1598 Details : If an OST is started on a device that has never been formatted
1599 or if the filesystem is corrupt and cannot even mount then the
1600 error handling cleanup routines would dereference a NULL pointer.
1605 Description: NULL pointer deref in llap_from_page.
1606 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1607 incorrect) mapping if the page was truncated/reclaimed while it was
1608 searched for. Check for this condition and skip such pages when
1609 doing readahead. Introduce extra check to llap_from_page() to
1610 verify page->mapping->host is non-NULL (so page is not anonymous).
1613 Frequency : Sometimes when using sys_sendfile
1615 Description: "page not covered by a lock" warnings from ll_readpage
1616 Details : sendfile called ll_readpage without right page locks present.
1617 Now we introduced ll_file_sendfile that does necessary locking
1618 around call to generic_file_sendfile() much like we do in
1622 Frequency : with certain MDS communication failures at client mount time
1624 Description: NULL pointer deref after failed client mount
1625 Details : a client connection request may delayed by the network layer
1626 and not be sent until after the PTLRPC layer has timed out the
1627 request. If the client fails the mount immediately it will try
1628 to clean up before the network times out the request. Add a
1629 reference from the request import to the obd device and delay
1630 the cleanup until the network drops the request.
1633 Frequency : occasionally during client (re)connect
1635 Description: assertion failure during client (re)connect
1636 Details : processing a client connection request may be delayed by the
1637 client or server longer than the client connect timeout. This
1638 causes the client to resend the connection request. If the
1639 original connection request is replied in this interval, the
1640 client may trip an assertion failure in ptlrpc_connect_interpret()
1641 which thought it would be the only running connect process.
1644 Frequency : only with obd_echo servers and clients that are rebooted
1646 Description: kernel BUG accessing uninitialized data structure
1647 Details : When running an obd_echo server it did not start the ping_evictor
1648 thread, and when a client was evicted an uninitialized data
1649 structure was accessed. Start the ping_evictor in the RPC
1650 service startup instead of the OBD startup.
1652 Severity : enhancement
1653 Bugzilla : 10193 (patchless)
1654 Description: Remove dependency on various unexported kernel interfaces.
1655 Details : No longer need reparent_to_init, exit_mm, exit_files,
1656 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1659 Frequency : rare (only users of deprecated and unsupported LDAP config)
1661 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1662 Details : LDAP apparently contains 'lustreName' attributes instead of
1663 'name'. A simple remapping of the name is sufficient.
1666 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1668 Description: waiting_locks_callback trips kernel BUG if client is evicted
1669 Details : Running with the dump_on_timeout debug flag turned on makes
1670 it possible that the waiting_locks_callback() can try to dump
1671 the Lustre kernel debug logs from an interrupt handler. Defer
1672 this log dumping to the expired_lock_main() thread.
1674 Severity : enhancement
1676 Description: Support NFS exporting on 2.6 kernels.
1677 Details : Implement non-rawops metadata methods for NFS server to use without
1678 changing NFS server code.
1681 Frequency : very rare (synthetic metadata workload only)
1683 Description: two racing renames might cause an MDS thread to deadlock
1684 Details : Running the "racer" program may cause one MDS thread to rename
1685 a file from being the source of a rename to being the target of
1686 a rename at exactly the same time that another thread is doing
1687 so, and the second thread has already enqueued these locks after
1688 doing a lookup of the target and is trying to relock them in
1689 order. Ensure that we don't try to re-lock the same resource.
1692 Frequency : only very large systems with liblustre clients
1694 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1695 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1696 all OSTs in parallel. This allows the eviction of stale liblustre
1697 clients to proceed much faster than if they were done in series,
1698 and also offers similar improvements for other set_info RPCs.
1703 Description: excessive CPU usage during initial read phase on client
1704 Details : During the initial read phase on a client, it would agressively
1705 retry readahead on the file, consuming too much CPU and impacting
1706 performance (since 1.4.5.8). Improve the readahead algorithm
1707 to avoid this, and also improve some other common cases (read
1708 of small files in particular, where "small" is files smaller than
1709 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1714 Description: MDS crash when receiving packet with unknown intent.
1715 Details : Do not LBUG in unknown intent case, just return -EFAULT
1717 Severity : enhancement
1718 Bugzilla : 9293, 9385
1719 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1720 Details : Do not serialize getattr (non-intent version) and statfs.
1723 Frequency : occasional, when OST network is overloaded/intermittent
1725 Description: client evicted by OST after bulk IO timeout
1726 Details : If a client sends a bulk IO request (read or write) the OST
1727 may evict the client if it is unresposive to its data GET/PUT
1728 request. This is incorrect if the network is overloaded (takes
1729 too long to transfer the RPC data) or dropped the OST GET/PUT
1730 request. There is no need to evict the client at all, since
1731 the pinger and/or lock callbacks will handle this, and the
1732 client can restart the bulk request.
1735 Frequency : Always when mmapping file with no objects
1737 Description: client crashes when mmapping file with no objects
1738 Details : Check that we actually have objects in a file before doing any
1739 operations on objects in ll_vm_open, ll_vm_close and
1745 Description: Request leak when working with deleted CWD
1746 Details : Introduce advanced request refcount tracking for requests
1747 referenced from lustre intent.
1749 Severity : Enhancement
1751 Description: Cache open file handles on client.
1752 Details : MDS now will return special lock along with openhandle, if
1753 requested and client is allowed to hold openhandle, even if unused,
1754 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1755 closing files for every read/write openration.
1757 Severity : Enhancement
1759 Description: Cache open negative dentries on client when possible.
1760 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1761 negative dentries on lock revocation.
1766 Description: Remounting a client read-only wasn't possible with a zconf mount
1767 Details : It wasn't possible to remount a client read-only with llmount.
1769 Severity : enhancement
1770 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1771 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1772 MPICH is included to add Lustre-specific ADIO interfaces.
1773 This is based closely on the UFS ADIO layer and only differs
1774 in file creation, in order to allow the OST striping to be set.
1775 This is user-contributed code and not supported by CFS.
1780 Description: extended inode attributes (immutable, append-only) work improperly
1781 when 2.4 and 2.6 kernels are used on client/server or vice versa
1782 Details : Introduce kernel-independent values for these flags.
1784 Severity : enhancement
1787 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1788 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1789 granularity. Now, Fractional MB granularity is be supported,
1790 this is very useful for low memory system.
1792 Severity : enhancement
1794 Description: Getattr by fid
1795 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1796 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1797 have locks and avoids getattr rpc requests for it on every stat.
1800 Frequency : Always, for filesystems larger than 2TB
1802 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1803 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1804 filesystems larger than 2TB to be mounted with mballoc enabled.
1807 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1809 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1810 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1811 filesystems larger than 2TB is mounted then the kernel will
1812 silently corrupt the start of the filesystem. CONFIG_LBD is
1813 enabled for all CFS-supported kernels, but the possibility of
1814 this happening with a modified kernel config exists.
1816 Severity : enhancement
1818 Description: add client O_DIRECT support for 2.6 kernels
1819 Details : It is now possible to do O_DIRECT reads and writes to files
1820 in the Lustre client mountpoint on 2.6 kernel clients.
1822 Severity : enhancement
1824 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1825 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1826 parallel, not waiting for response from every OST before sending
1827 a rpc to the next OST.
1832 Description: setattr vs write race when updating file timestamps
1833 Details : Client processes that update a file timestamp into the past
1834 right after writing to the file (e.g. tar) it is possible that
1835 the updated file modification time can be reset to the current
1836 time due to a race between processing the setattr and write RPC.
1838 Severity : enhancement
1840 Description: Bring 'lfs find' closer in line with regular Linux find.
1841 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1842 -print0 options and obtains all the needed info through the lustre
1845 Severity : enhancement
1847 Description: support up to 1024 configured devices on one node
1848 Details : change obd_dev array from statically allocated to dynamically
1849 allocated structs as they are first used to reduce memory usage
1854 Description: Flush dirty partially truncated pages during truncate
1855 Details : Immediatelly flush partially truncated pages in filter_setattr,
1856 this way we completely avoid having any pages in page cache on OST
1857 and can retire ugly workarounds during writes to flush such pages.
1862 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
1863 Details : protect inode from truncation within vfs_unlink() context
1864 just take a reference before calling vfs_unlink() and release it
1865 when parent's i_sem is free.
1870 Description: last_id value checked outside lock on OST caused LASSERT failure
1871 Details : If there were multiple MDS->OST object precreate requests in
1872 flight, it was possible that the OST's last object id was checked
1873 outside a lock and incorrectly tripped an assertion. Move checks
1874 inside locks, and discard old precreate requests.
1877 Frequency : always, if extents are used on OSTs
1879 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
1880 Details : OSTs using the extents format would not zero the high 16 bits of
1881 the index physical block number. This is not a problem for any
1882 OST filesystems smaller than 16TB, and no kernels support ext3
1883 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
1884 new/modified files) and can be fixed for existing filesystems
1885 with e2fsprogs-1.39-cfs1.
1890 Description: import connection selection may be incorrect if timer wraps
1891 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
1892 import connections to be ignored if the 32-bit jiffies counter
1893 wraps. Use a 64-bit jiffies counter.
1896 Frequency : very large clusters immediately after boot
1898 Description: LNET request buffers exhausted under heavy short-term load
1899 Details : If a large number of client requests are generated on a service
1900 that has previously never seen so many requests it is possible
1901 that the request buffer growth cannot keep up with the spike in
1902 demand. Instead of dropping incoming requests, they are held in
1903 the LND until the RPC service can accept more requests.
1906 Frequency : Sometimes during replay
1908 Description: Assertion failure in ll_local_open after replay.
1909 Details : If replay happened on an open request reply before we were able
1910 to set replay handler, reply will become not swabbed tripping the
1911 assertion in ll_local_open. Now we set the handler right after
1912 recognising of open request
1915 Frequency : very rare
1917 Description: kernel reports "badness in vsnprintf"
1918 Details : Reading from the "recovery_status" /proc file in small chunks
1919 may cause a negative length in lprocfs_obd_rd_recovery_status()
1920 call to vsnprintf() (which is otherwise harmless). Exit early
1921 if there is no more space in the output buffer.
1923 Severity : enhancement
1925 Description: clear OBD RPC statistics by writing to them
1926 Details : It is now possible to clear the OBD RPC statistics by writing
1927 to the "stats" file.
1932 Description: Client mtime is not the same on different clients after utimes
1933 Details : In some cases, the client was using the utimes() syscall on
1934 a file cached on another node. The clients now validate the
1935 ctime from the MDS + OSTs to determine which one is right.
1940 Description: Inability to activate failout mode
1941 Details : lconf script incorrectly assumed that in pythong string's numeric
1942 value is used in comparisons.
1945 Frequency : always with multiple stripes per file
1947 Description: Inefficient object allocation for mutli-stripe files
1948 Details : When selecting which OSTs to stripe files over, for files with
1949 a stripe count that divides evenly into the number of OSTs,
1950 the MDS is always picking the same starting OST for each file.
1951 Return the OST selection heuristic to the original design.
1956 Description: mount failures may take full timeout to return an error
1957 Details : Under some heavy load conditions it is possible that a
1958 failed mount can wait for the full obd_timeout interval,
1959 possibly several minutes, before reporting an error.
1960 Instead return an error as soon as the status is known.
1962 Frequency : quota enabled and large files being deleted
1964 Description: releasing more than 4GB of quota at once hangs OST
1965 Details : If a user deletes more than 4GB of files on a single OST it
1966 will cause the OST to spin in an infinite loop. Release
1967 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1972 Description: statfs data retrieved from /proc may be stale or zero
1973 Details : When reading per-device statfs data from /proc, in the
1974 {kbytes,files}_{total,free,avail} files, it may appear
1975 as zero or be out of date.
1978 Frequency : systems with MD RAID1 external journal devices
1980 Description: lconf's call to blkid is confused by RAID1 journal devices
1981 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1982 of returning all block devices that match the journal UUID.
1985 Frequency : always, for aggregate stripe size over 4GB
1987 Description: assertion fails when trying to use 4GB stripe size
1988 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
1989 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
1992 Frequency : always on ppc64
1994 Description: the first write on an ext3 filesystem with mballoc got stuck
1995 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1996 perform endianness conversion.
1998 ------------------------------------------------------------------------------
2000 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2002 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2003 INTEROPERATE with older versions automatically. Please read the
2004 user documentation before upgrading any part of a live system.
2005 * WARNING: Lustre networking configuration changes are required with
2006 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2009 * Support for kernels:
2010 2.6.9-22.0.2.EL (RHEL 4)
2011 2.6.5-7.244 (SLES 9)
2012 2.6.12.6 vanilla (kernel.org)
2015 Severity : enhancement
2016 Bugzilla : 7981/8208
2017 Description: Introduced Lustre Networking (LNET)
2018 Details : LNET is new networking infrastructure for Lustre, it includes
2019 a reorganized network configuration mode (see the user
2020 documentation for full details) as well as support for routing
2021 between different network fabrics. Lustre Networking Devices
2022 (LNDs) for the supported network fabrics have also been
2023 created for this new infrastructure.
2025 Severity : enhancement
2026 Description: Introduced Access control lists
2027 Details : clients can set ACLs on files and directories in order to have
2028 more fine-grained permissions than the standard Unix UGO+RWX.
2029 The MDS must be started with the "-o acl" mount option.
2031 Severity : enhancement
2032 Description: Introduced filesystem quotas
2033 Details : Administrators may now establish per-user quotas on the
2036 Severity : enhancement
2038 Description: Configuration change for the XT3
2039 The PTLLND is now used to run Lustre over Portals on the XT3
2040 The configure option(s) --with-cray-portals are no longer used.
2041 Rather --with-portals=<path-to-portals-includes> is used to
2042 enable building on the XT3. In addition to enable XT3 specific
2043 features the option --enable-cray-xt3 must be used.
2048 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2049 Details : A race between allocating a new llcd and re-getting the llcd_lock
2050 allowed another thread to grab newly-allocated llcd.
2052 Severity : enhancement
2054 Description: 2.6 OST async journal commit and locking fix to improve performance
2055 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2056 2.6 kernels are now async as they already were in 2.4 kernels so
2057 that they can commit concurrently with the network bulk transfer.
2058 For block-allocated files the filter allocation semaphore is held
2059 to avoid filesystem fragmentation during allocation. BKL lock
2060 removed for 2.6 xattr operations where it is no longer needed.
2065 Description: lconf incorrectly determined whether two IP networks could talk
2066 Details : In some more complicated routing and multiple-network
2067 configurations, lconf will avoid trying to make a network
2068 connection to a disjoint part of the IP space. It was doing the
2069 math incorrectly for one set of cases.
2074 Description: Fix for potential infinite loop processing records in an llog.
2075 Details : If an llog record is corrupted/zeroed, it is possible to loop
2076 forever in llog_process(). Validate the llog record length
2077 and skip the remainder of the block on error.
2080 Frequency : occasional (liblustre only)
2082 Description: liblustre could not open files whose last component is a symlink
2083 Details : sysio_path_walk() would incorrectly pass the open intent to
2084 intermediate path components.
2087 Frequency : rare (liblustre only with non-standard tuning)
2088 Bugzilla : 7201 (7350)
2089 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2090 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2091 after it has been released, only do this for referenced locks
2093 Severity : enhancement
2095 Description: specifying an (invalid) directory default stripe_size of -1
2096 would reset the directory default striping
2097 Details : stripe_size -1 was used internally to signal directory stripe
2098 removal, now use "all default" to signal dir stripe removal
2099 as a directory striping of "all default" is not useful
2102 Frequency : common for large clusters running liblustre clients
2104 Description: doing an ls when liblustre clients are running is slow
2105 Details : sending a glimpse AST to a liblustre client waits for every AST
2106 to time out, as liblustre clients will not respond. Since they
2107 cannot cache data we refresh the OST lock LVB from disk instead.
2109 Severity : enhancement
2111 Description: doing an ls at the same time as file IO can be slow
2112 Details : enqueue and other "small" requests can be blocked behind many
2113 large IO requests. Create a new OST IO portal for non-IO
2114 requests so they can be processed faster.
2117 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2119 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2120 created file with mode 000
2121 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2122 allow an ftruncate() as a non-root user to a file with mode 000.
2123 Lustre now respects this flag to disable mode checks when
2124 truncating a file owned by the user
2127 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2130 Description: Revoking locks from clients that went dead or catatonic might take
2132 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2133 cancellation of such locks instant on servers without waiting for
2134 any reply from clients. Clients drops these locks when cancel
2135 notification from server is received without replying.
2138 Frequency : liblustre-only, when liblustre client dies or becomes busy
2140 Description: Doing ls on Linux clients can take a long time with active
2142 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2143 granting such locks to it in the first place if possible. Locks
2144 are taken by proxy on the OST during the read or write and
2145 dropped immediately afterward. Add connect flags handling, do
2146 not grant locks to liblustre clients for glimpse ASTs.
2148 Severity : enhancement
2150 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2151 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2152 sequential reads and behaves badly with applications doing random
2153 reads. Improve it by reading ahead at least read region, and
2154 avoiding excessive large RPC for small reads.
2156 Severity : enhancement
2158 Description: Creating more than 1000 files for a single job may cause a load
2159 imbalance on the OSTs if there are also a large number of OSTs.
2160 Details : qos_prep_create() uses an OST index reseed value that is an
2161 even multiple of the number of available OSTs so that if the
2162 reseed happens in the middle of the object allocation it will
2163 still utilize the OSTs as uniformly as possible.
2168 Description: OST or MDS may oops in ping_evictor_main()
2169 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2170 but doesn't restart at beginning of obd_exports_timed list
2173 Severity : enhancement
2175 Description: improve by-nid export eviction on the MDS and OST
2176 Details : allow multiple exports with the same NID to be evicted at one
2177 time without re-searching the exports list.
2180 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2182 Description: MDS may oops in groups_free()
2183 Details : in rare race conditions a newly allocated group_info struct is
2184 freed again, and this can be NULL. The 2.4 compatibility code
2185 for groups_free() checked for a NULL pointer, but 2.6 did not.
2188 Frequency : common for liblustre clients doing little filesystem IO
2189 Bugzilla : 9352, 7313
2190 Description: server may evict liblustre clients accessing contended locks
2191 Details : if a client is granted a lock or receives a completion AST
2192 with a blocking AST already set it would not reply to the AST
2193 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2196 Frequency : lfs setstripe, only systems with more than 160 OSTs
2198 Description: unable to set striping with a starting offset beyond OST 160
2199 Details : llapi_create_file() incorrectly limited the starting stripe
2200 index to the maximum single-file stripe count.
2203 Frequency : LDAP users only
2205 Description: lconf did not handle in-kernel recovery with LDAP properly
2206 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2208 Severity : enhancement
2210 Description: bind OST threads to NUMA nodes to improve performance
2211 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2212 node and do their allocations there to localize memory access
2214 Severity : enhancement
2216 Description: llmount can determine client NID directly from Myrinet (GM)
2217 Details : the client NID code from gmnalnid was moved directly into
2218 llmount, removing the need to use this or specifying the
2219 client NID explicitly when mounting GM clients with zeroconf
2222 Frequency : if client is started with down MDS
2224 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2225 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2226 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2227 ptlrpc_import_delayed_req() because we didn't check if the
2228 request was interrupted, and we also didn't break out of the
2229 event loop if there was no timeout
2234 Description: data loss during non-page-aligned writes to a single file from
2235 both multiple nodes and multiple threads on one node at same time
2236 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2237 inconsistency led to false short-reads, that were cached and later
2238 used by ->prepare_write() to fill in partially written page,
2239 leading to data loss.
2242 Frequency : always, if lconf --abort_recovery used
2244 Description: lconf --abort_recovery fails with 'Operation not supported'
2245 Details : lconf was attempting to abort recovery on the MDT device and not
2248 Severity : enhancement
2250 Description: remove cleanup logs
2251 Details : replace lconf-generated cleanup logs with lustre internal
2252 cleanup routines. Eliminates the need for client-cleanup and
2255 Severity : enhancement
2257 Description: add support for EAs (user and system) on lustre filesystems
2258 Details : it is now possible to store extended attributes in the Lustre
2259 client filesystem, and with the user_xattr mount option it
2260 is possible to allow users to store EAs on their files also
2262 Severity : enhancement
2264 Description: Add possibility (config option) to show minimal available OST free
2266 Details : When compiled with --enable-mindf configure option, statfs(2)
2267 (and so, df) will return least minimal free space available from
2268 all OSTs as amount of free space on FS, instead of summary of
2269 free spaces of all OSTs.
2271 Severity : enhancement
2273 Description: do not expand extent locks acquired on OST-side
2274 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2275 by server: they are not cached anyway.
2278 Frequency : when mmap is used/binaries executed from Lustre
2280 Description: Unmmap pages before throwing them away from read cache.
2281 Details : llap_shrink cache now attempts to unmap pages before discarding
2282 them (if unmapping failed - do not discard). SLES9 kernel has
2283 extra checks that trigger if this unmapping is not done first.
2288 Description: lconf didn't resolve symlinks before checking to see whether a
2289 given mountpoint was already in use
2292 Frequency : when migrating failover services
2293 Bugzilla : 6395, 9514
2294 Description: When migrating a subset of services from a node (e.g. failback
2295 from a failover service node) the remaining services would
2296 time out and evict clients.
2297 Details : lconf --force (implied by --failover) sets the global obd_timeout
2298 to 5 seconds in order to quickly disconnect, but this caused
2299 other RPCs to time out too quickly. Do not change the global
2300 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2302 Severity : enhancement
2303 Frequency : if MDS is started with down OST
2304 Bugzilla : 9439,5706
2305 Description: Allow startup/shutdown of an MDS without depending on the
2306 availability of the OSTs.
2307 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2308 Add appropriate locking and lov-osc refcounts for safe
2309 cleaning. Add osc abort_inflight calls in case the
2310 synchronize never started.
2313 Frequency : occasional (Cray XT3 only)
2315 Description: root not authorized to access files in CRAY_PORTALS environment
2316 Details : The client process capabilities were not honoured on the MDS in
2317 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2318 been accessed by an authorized user then root was able to access
2319 the file on the local client also. The root user capabilities
2320 are now allowed on the MDS, as this environment has secure UID.
2323 Frequency : occasional
2325 Description: ldiskfs "too long searching" message happens too often
2326 Details : A debugging message (otherwise harmless) prints too often on
2327 the OST console. This has been reduced to only happen when
2328 there are fragmentation problems on the filesystem.
2333 Description: Division by zero in statfs when all OSCs are inactive
2334 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2335 lov_statfs divides by value returned by lov_get_stripecnt().
2339 Bugzilla : 9489, 3273
2340 Description: First write from each client to each OST was only 4kB in size,
2341 to initialize client writeback cache, which caused sub-optimal
2342 RPCs and poor layout on disk for the first writen file.
2343 Details : Clients now request an initial cache grant at (re)connect time
2344 and so that they can start streaming writes to the cache right
2345 away and always do full-sized RPCs if there is enough data.
2346 If the OST is rebooted the client also re-establishes its grant
2347 so that client cached writes will be honoured under the grant.
2352 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2353 Details : Now I/O RPCs go to different portal number and (presumably) fast
2354 lock requests (and glimses) and other RPCs get their own service
2355 threads pool that should be able to service those RPCs
2358 Severity : enhancement
2360 Description: Ability to exchange lustre version between client and servers and
2361 issue warnings at client side if client is too old. Also for
2362 liblustre clients there is ability to refuse connection of too old
2364 Details : New 'version' field is added to connect data structure that is
2365 filled with version info. That info is later checked by server and
2369 Frequency : rare, liblustre only.
2370 Bugzilla : 9296, 9581
2371 Description: Two simultaneous writes from liblustre at offset within same page
2372 might proceed at the same time overwriting eachother with stale
2374 Details : I/O lock withing llu_file_prwv was released too early, before data
2375 actually was hitting the wire. Extended lock-holding time until
2376 server acknowledges receiving data.
2379 Frequency : extremely rare. Never observed in practice.
2381 Description: avoid generating lustre_handle cookie of 0.
2382 Details : class_handle_hash() generates handle cookies by incrementing
2383 global counter, and can hit 0 occasionaly (this is unlikely, but
2384 not impossible, because initial value of cookie counter is
2385 selected randonly). Value of 0 is used as a sentinel meaning
2386 "unassigned handle" --- avoid it. Also coalesce two critical
2387 sections in this function into one.
2389 Severity : enhancement
2391 Description: allow liblustre clients to delegate truncate locking to OST
2392 Details : To avoid overhead of locking, liblustre client instructs OST to
2393 take extent lock in ost_punch() on client's behalf. New connection
2394 flag is added to handle backward compatibility.
2396 Severity : enhancement
2397 Bugzilla : 4928, 7341, 9758
2398 Description: allow number of OST service threads to be specified
2399 Details : a module parameter allows the number of OST service threads
2400 to be specified via "options ost ost_num_threads={N}" in the
2401 OSS's /etc/modules.conf or /etc/modprobe.conf.
2405 Bugzilla : 6146, 9635, 9895
2406 Description: servers crash with bad pointer in target_handle_connect()
2407 Details : In rare cases when a client is reconnecting it was possible that
2408 the connection request was the last reference for that export.
2409 We would temporarily drop the export reference and get a new
2410 one, but this may have been the last reference and the export
2411 was just destroyed. Get new reference before dropping old one.
2413 Severity : enhancement
2414 Frequency : if client is started with failover MDS
2416 Description: Allow multiple MDS hostnames in the mount command
2417 Details : Try to read the configuration from all specified MDS
2418 hostnames during a client mount in case the "primary"
2421 Severity : enhancement
2423 Description: Stop sending data to evicted clients as soon as possible.
2424 Details : Check if the client we are about to send or are sending data to
2425 was evicted already. (Check is done every second of waiting,
2426 for which l_wait_event interface was extended to allow checking
2427 of exit condition at specified intervals).
2430 Frequency : rare, normally only when NFS exporting is done from client
2432 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2434 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2437 Frequency : always, if srand() is called before liblustre initialization
2439 Description: Liblustre uses system PRNG disturbing its usage by user application
2440 Details : Introduce internal to lustre fast and high-quality PRNG for
2441 lustre usage and make liblustre and some other places in generic
2442 lustre code to use it.
2444 Severity : enhancement
2445 Bugzilla : 9477, 9557, 9870
2446 Description: Verify that the MDS configuration logs are updated when xml is
2447 Details : Check if the .xml configuration logs are newer than the config
2448 logs stored on the MDS and report an error if this is the case.
2449 Request --write-conf, or allow starting with --old_conf.
2451 Severity : enhancement
2453 Description: Handle symlinks in the path when checking if Lustre is mounted.
2454 Details : Resolve intermediate symlinks when checking if a client has
2455 mounted a filesystem to avoid duplicate client mounts.
2460 Description: lconf can hit an error exception but still return success.
2461 Details : The lconf command catches the Command error exception at the top
2462 level script context and will exit with the associated exit
2463 status, but doesn't ensure that this exit status is non-zero.
2468 Description: failure of ptlrpc thread startup can cause oops
2469 Details : Starting a ptlrpc service thread can fail if there are a large
2470 number of threads or the server memory is very fragmented.
2471 Handle this without oopsing.
2474 Frequency : always, only if liblustre and non-default acceptor port was used
2476 Description: liblustre cannot connect to servers with non-default acceptor port
2477 Details : tcpnal_set_default_params() was not called and was therefore
2478 ignoring the environment varaible TCPNAL_PORT, as well as other
2479 TCPNAL_ environment variables
2484 Description: two objects could be created on the same OST for a single file
2485 Details : If an OST is down, in some cases it was possible to create two
2486 objects on a single OST for a single file. No problems other
2487 than potential performance impact and spurious error messages.
2491 Bugzilla : 5681, 9562
2492 Description: Client may oops in ll_unhash_aliases
2493 Details : Client dcache may become inconsistent in race condition.
2494 In some cases "getcwd" can fail if the current directory is
2500 Description: Inode refcounting problems in NFS export code
2501 Details : link_raw functions used to call d_instantiate without obtaining
2502 extra inode reference first.
2506 Bugzilla : 9942, 9903
2507 Description: Referencing freed requests leading to crash, memleaks with NFS.
2508 Details : We used to require that call to ll_revalidate_it was always
2509 followed by ll_lookup_it. Also with revalidate_special() it is
2510 possible to call ll_revalidate_it() twice for the same dentry
2511 even if first occurence returned success. This fix changes semantic
2512 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2513 reference on a request referred from the intent.
2514 ll_intent_release() then releases such a request.
2517 Frequency : rare, normally benchmark loads only
2519 Description: unlinked inodes were kept in memory on the client
2520 Details : If a client is repeatedly creating and unlinking files it
2521 can accumulate a lot of stale inodes in the inode slab cache.
2522 If there is no other client load running this can cause the
2523 client node to run out of memory. Instead flush old inodes
2524 from client cache that have the same inode number as a new inode.
2527 Frequency : SLES9 2.6.5 kernel and long filenames only
2528 Bugzilla : 9969, 10379
2529 Description: utime reports stale NFS file handle
2530 Details : SLES9 uses out-of-dentry names in some cases, which confused
2531 the lustre dentry revalidation. Change it to always use the
2535 Frequency : rare, unless heavy write-truncate concurrency is continuous
2536 Bugzilla : 4180, 6984, 7171, 9963, 9331
2537 Description: OST becomes very slow and/or deadlocked during object unlink
2538 Details : filter_destroy() was holding onto the parent directory lock
2539 while truncating+unlinking objects. For very large objects this
2540 may block other threads for a long time and slow overall OST
2541 responsiveness. It may also be possible to get a lock ordering
2542 deadlock in this case, or run out of journal credits because of
2543 the combined truncate+unlink. Solution is to do object truncate
2544 first in one transaction without parent lock, and then do the
2545 final unlink in a new transaction with the parent lock. This
2546 reduces the lock hold time dramatically.
2549 Frequency : rare, 2.4 kernels only
2551 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2552 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2553 before unmounting their filesystem in order to avoid deadlock.
2554 The kernel_locked() function in 2.4 kernels only checks whether
2555 the kernel lock is held, not whether it is this process that is
2556 holding it as 2.6 kernels do.
2561 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2562 Details : The client ptlrpc code may be trying to reconnect to a down
2563 server before a previous connection attempt has timed out.
2564 Increase the reconnect interval to be longer than the connection
2565 timeout interval to avoid sending duplicate connections to
2569 Frequency : echo_client brw_test command
2571 Description: fix echo_client to work with OST preallocated code
2572 Details : OST preallocation code (5137) didn't take echo_client IO path
2573 into account: echo_client calls filter methods outside of any
2574 OST thread and, hence, there is no per-thread preallocated
2575 pages and buffers to use. Solution: hijack pga pages for IO. As
2576 a byproduct, this avoids unnecessary data copying.
2580 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2581 Description: Client can oops in mdc_commit_close() after open replay
2582 Details : It was possible for the MDS to return an open request with no
2583 transaction number in mds_finish_transno() if the client was
2584 evicted, but without actually returning an error. Clients
2585 would later try to replay that open and may trip an assertion
2586 Simplify the client close codepath, and always return an error
2587 from the MDS in case the open is not successful.
2590 Frequency : rare, 2.6 OSTs only
2592 Description: OST may deadlock under high load on fragmented files
2593 Details : If there was a heavy load and highly-fragmented OST filesystems
2594 it was possible to have all the OST threads deadlock waiting on
2595 allocation of biovecs, because the biovecs were not released
2596 until the entire RPC IO was completed. Instead, release biovecs
2597 as soon as they are complete to ensure forward IO progress.
2599 Severity : enhancement
2601 Description: Support for specifying external journal device at mount
2602 Details : If an OST or MDS device is formatted with an external journal
2603 device, this device major/minor is stored in the ext3 superblock
2604 and may not be valid for failover. Allow detecting and
2605 specifying the external journal at mount time.
2610 Description: Mounting an MDS with pending unlinked files may cause oops
2611 Details : target_finish_recovery() calls mds_postrecov() which returned
2612 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2613 considers this an error and immediately begins cleaning up the
2614 lov, just after starting the mds_lov process
2616 Severity : enhancement
2618 Description: Implement 'lfs df' to report actual free space on per-OST basis
2619 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2620 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2621 usage of objects; '-h' to report in human readable format.
2623 ------------------------------------------------------------------------------
2625 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2632 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2633 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2634 at startup during orphan unlinking. Instead, skip buddy update
2635 before setup, as it will be regenerated after recovery is complete.
2640 Description: If an OST is inactive, its locks might reference stale inodes.
2641 Details : lov_change_cbdata() must iterate over all namespaces, even if
2642 they are inactive to clear inode references from the lock.
2644 Severity : enhancement
2645 Frequency : occasional, if non-standard max_dirty_mb used
2647 Description: Client will block write RPCs if not enough grant
2648 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2649 then the client will block writes while waiting for another RPC
2650 to complete instead of consuming its dirty limit. With change
2651 we get improved performance when max_dirty_mb is small.
2653 Severity : enhancement
2654 Bugzilla : 3389, 6253
2655 Description: Add support for supplementary groups on the MDS.
2656 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2657 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2658 lookups for user supplementary groups into a cache.
2662 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2663 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2664 for liblustre O_EXCL is enforced by MDS.
2668 Description: readdir never returns NULL in liblustre.
2669 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2670 dentry in struct dirent.
2674 Frequency : liblustre only, depends on application IO pattern
2675 Description: liblustre clients evicted if not contacting servers
2676 Details : Don't put liblustre clients into the ping_evictor list, so
2677 they will not be evicted by the pinger ever.
2679 Severity : enhancement
2681 Description: Add ability to evict clients by NID from MDS.
2682 Details : By echoing "nid:$NID" string into
2683 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2684 $NID would be instantly evicted from this MDS and from all active
2685 OSTs connected to it.
2689 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2690 Details : lookup_it_finish() used to query file size from OSTs that was not
2695 Description: service threads change working directory to that of init
2696 Details : Starting lustre service threads may pin the working directory
2697 of the parent thread, making that filesystem busy. Threads
2698 now change to the working directory of init to avoid this.
2702 Frequency : during shutdown only
2703 Description: shutdown with a failed MDS or OST can cause unmount to hang
2704 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2709 Frequency : 2.6 only
2710 Description: chmod/chown may include an extra supplementary group
2711 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2712 supplementary group and if none is specified this is used.
2715 Bugzilla : 5479 (6816)
2717 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2718 Details : If lookup is in progress on a file that is unlinked we might try
2719 to revalidate the inode and fail in revalidate after lookup is
2720 complete and ll_file_open() enqueues the open again but
2721 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2724 Frequency : always, if lconf --abort_recovery used
2726 Description: lconf --abort_recovery fails with 'Operation not supported'
2727 Details : lconf was attempting to abort recovery on the MDT device and not
2730 ------------------------------------------------------------------------------
2732 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2737 Frequency : rare (only unsupported configurations with a node running as an
2739 Bugzilla : 6514, 5137
2740 Description: Mounting a Lustre file system on a node running as an OST could
2742 Details : OSTs now preallocates memory needed to write out data at
2743 startup, instead of when needed, to avoid having to
2744 allocate memory in possibly low memory situations.
2745 Specifically, if the file system is mounted on on OST,
2746 memory pressure could force it to try to write out data,
2747 which it needed to allocate memory to do. Due to the low
2748 memory, it would be unable to do so and the node would
2749 become unresponsive.
2751 Severity : enhancement
2753 Description: Addition of lconf --service command line option
2754 Details : lconf now accepts a '--service <arg>' option, which is
2755 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2757 Severity : enhancement
2759 Description: Failover mode is now the default for OSTs.
2760 Details : By default, OSTs will now run in failover mode. To return to
2761 the old behaviour, add '--failout' to the lmc line for OSTs.
2763 Severity : enhancement
2765 Description: Health checks are now provided for MDS and OSTs
2766 Details : Additional detailed health check information on MSD and OSTs
2767 is now provided through the procfs health_check value.
2770 Frequency : occasional, depends on IO load
2772 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2773 after numerous create/delete cycles
2774 Details : The ext3 inode allocation policy would not allocate new inodes
2775 very well on the OSTs because there are no new directories
2776 being created. Instead we look for groups with free space if
2777 the parent directories are nearly full.
2782 Description: Network or server problems during mount may cause partially
2783 mounted clients instead of returning an error.
2784 Details : The config llog parsing code may overwrite the error return
2785 code during mount error handling, returning success instead
2791 Description: MDS can fail to allocate large reply buffers
2792 Details : After long uptimes the MDS can fail to allocate large reply
2793 buffers (e.g. zconf client mount config records) due to memory
2794 fragmentation or consumption by the buffer cache. Preallocate
2795 some large reply buffers so that these replies can be sent even
2796 under memory pressure.
2800 Frequency : rare (liblustre)
2801 Description: fsx running with liblustre complained that using truncate() to
2802 extend the file doesn't work. This patch corrects that issue.
2803 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2804 ATTR_SIZE and lsm use in llu_setattr_raw.
2808 Frequency : rare, only 2.6 kernels
2809 Description: Unusual file access patterns on the MDS may result in inode
2810 data being lost in very rare circumstances.
2811 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2812 code in the 2.6 kernel caused the inode and/or EA data not to
2813 be read from disk, causing single-file corruption.
2817 Frequency : rare, only 2.6 filesystems using extents
2818 Description: Heavy concurrent write and delete load may cause data corruption.
2819 Details : It was possible under high-load situations to have an extent
2820 metadata block in the block device cache from a just-unlinked
2821 file overwrite a newly-allocated data block. We now unmap any
2822 metadata buffers that alias just-allocated data blocks.
2826 Frequency : filesystems with default stripe_count larger than 77
2827 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2828 Details : lconf specifies an inode size of 4096 bytes when the default
2829 stripe_count is larger than 77. This conflicts with the default
2830 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2831 this case to avoid pinning too much memory for large EAs.
2833 ------------------------------------------------------------------------------
2835 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2840 Frequency : rare (extremely heavy IO load with hundreds of clients)
2842 Description: Client is evicted, gets IO error writing to file
2843 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2844 caused clients to be evicted for AST timeouts. The fixes in
2845 bug 5192 mean we no longer need to have such short AST timeouts
2846 so ldlm_timeout has been increased.
2849 Frequency : occasional during --force or --failover shutdown under load
2850 Bugzilla : 5949, 4834
2851 Description: Server oops/LBUG if stopped with --force or --failover under load
2852 Details : a collection of import/export refcount and cleanup ordering
2853 issues fixed for safer force cleanup
2856 Frequency : only filesystems larger than 120 OSTs
2857 Bugzilla : 5990, 6223
2858 Description: lfs getstripe would oops on a very large filesystem
2859 Details : lov_getconfig used kfree on vmalloc'd memory
2862 Frequency : only filesystems exporting via NFS to Solaris 10 clients
2863 Bugzilla : 6242, 6243
2864 Description: reading from files that had been truncated to a non-zero size
2865 but never opened returned no data
2866 Details : ll_file_read() reads zeros from no-object files to EOF
2871 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
2872 Details : There's little harm in aborting MDS/OSS recovery and letting it
2873 try again, so I removed the LASSERT and return an error instead.
2875 Severity : enhancement
2877 Description: New debugging infrastructure for tracking down data corruption
2878 Details : The I/O checksum code was replaced to: (a) control it at runtime,
2879 (b) cover more of the client-side code path, and (c) try to narrow
2880 down where problems occurred
2884 Bugzilla : 3819, 4364, 4397, 6313
2885 Description: Racing close and eviction MDS could cause assertion in mds_close
2886 Details : It was possible to get multiple mfd references during close and
2887 client eviction, leading to one thread referencing a freed mfd.
2889 Severity: : enhancement
2890 Bugzilla : 3262, 6359
2891 Description: Attempts to reconnect to servers are now more aggressive.
2892 Details : This builds on the enhanced upcall-less recovery that was added
2893 in 1.4.2. When trying to reconnect to servers, clients will
2894 now try each server in the failover group every 10 seconds. By
2895 default, clients would previously try one server every 25 seconds.
2900 Description: After recovery, certain operations trigger a failed
2901 assertion on a client.
2902 Details : Failing over an mds, using lconf -d --failover, while a
2903 client was doing a readdir() call would cause the client to
2904 LBUG after recovery completed and the readdir() was resent.
2906 Severity : enhancement
2908 Description: Default groups are now added by lconf
2909 Details : You can now run lconf --group <servicename> without having to
2910 manually add groups with lmc.
2913 Frequency : occasional
2915 Description: Nodes with an elan id of 0 trigger a failed assertion
2918 Frequency : always when accessing e.g. tty/console device nodes
2920 Description: tty and some other devices nodes cannot be used on lustre
2921 Details : file's private_data field is used by device data and lustre
2922 values in there got lost. New field was added to struct file to
2923 store fs-specific private data.
2926 Frequency : when exporting Lustre via NFS
2928 Description: NFSD failed occasionally when looking up a path component
2929 Details : NFSD is looking up ".." which was broken in ext3 directories
2930 that had grown large enough to become hashed.
2933 Frequency : Clusters with multiple interfaces not on the same subnet
2935 Description: Nodes will repeatedly try to reconnect to an interface which it
2936 cannot reach and report an error to the log.
2937 Details : Extra peer list entries will be created by lconf with some peers
2938 unreachable. lconf now validates the peer before adding it.
2941 Frequency : Only if a default stripe is set on the filesystem root.
2943 Description: Setting a default stripe on the filesystem root prevented the
2944 filesystem from being remounted.
2945 Details : The client was sending extra request flags in the root getattr
2946 request and did not allocate a reply buffer for the dir EA.
2949 Frequency : occasional, higher if lots of files are accessed by one client
2950 Bugzilla : 6159, 6097
2951 Description: Client trips assertion regarding lsm mismatch/magic
2952 Details : While revalidating inodes the VFS looks up inodes with ifind()
2953 and in rare cases can find an inode that is being freed.
2954 The ll_test_inode() code will free the lsm during ifind()
2955 when it finds an existing inode and then the VFS later attaches
2956 this free lsm to a new inode.
2960 Bugzilla : 6422, 7030
2961 Description: MDS deadlock between mkdir and client eviction
2962 Details : Creating a new file via mkdir or mknod (starting a transaction
2963 and getting the ns lock) can deadlock with client eviction
2964 (gets ns lock and trying to finish a synchronous transaction).
2967 Frequency : occasional
2968 Description: While starting a server, the fsfilt_ext3 module could not be
2970 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
2971 kernels. Previously, lconf would still use the ext3 name
2972 when trying to load modules. Now, it will correctly use
2973 ext3 on 2.4 and ldiskfs on 2.6.
2975 Severity : enhancement
2976 Description: The default stripe count has been changed to 1
2977 Details : The interpretation of the default stripe count (0, to lfs
2978 or lmc) has been changed to mean striping across a single
2979 OST, rather than all available. For general usage we have
2980 found a stripe count of 1 or 2 works best.
2982 Severity : enhancement
2983 Description: Add support for compiling against Cray portals.
2984 Details : Conditional compiling for some areas that are different
2988 Frequency : occasional
2989 Bugzilla : 6409, 6834
2990 Description: Creating files with an explicit stripe count may lead to
2991 a failed assertion on the MDS
2992 Details : If some OSTs are full or unavailable, creating files may
2993 trigger a failed assertion on the MDS. Now, Lustre will
2994 try to use other servers or return an error to the
2998 Frequency : occasional
3000 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3001 caused lock timeout during readahead (since 1.4.2).
3002 Details : Processes doing readahead might match a lock that hasn't been
3003 granted yet if there are overlapping and conflicting lock
3004 requests. The readahead process waits on ungranted lock
3005 (original lock is CBPENDING), while OST waits for that process
3006 to cancel CBPENDING read lock and eventually evicts client.
3008 Severity : enhancement
3010 Description: Initial enabling of flock support for clients
3011 Details : Implements fcntl advisory locking and file status functions.
3012 This feature is provided as an optional mount flag (default
3013 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3014 locking are implemented yet, and those that are are not guaranteed
3015 to be completely correct in production environments.
3016 mount -t lustre -o [flock|noflock] ...
3019 Frequency : occasional
3021 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3022 assertion in the ext3 JBD (journaling) layer.
3023 Details : The b_committed_data struct is protected by the big kernel lock
3024 in 2.4 kernels, serializing journal_commit_transaction() and
3025 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3026 access to this struct. In 2.6 kernels there is finer grained
3027 locking to improve SMP performance of the JBD layer.
3031 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3036 Description: merge and cleanup kernel patches.
3039 -----------------------------------------------------------------------------
3041 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3043 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3044 versions of Lustre. Please update all servers and clients to
3045 version 1.4.2 or later at the same time. You must also run
3046 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3047 to update the configuration logs.
3049 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3050 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3051 - swab llog records of type '0' so we get proper header size/idx (5861)
3052 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3053 - fix rename of one directory over another leaking an inode (5953)
3054 - avoid SetPageDirty on 2.6 (5981)
3055 - don't re-add just-being-destroyed locks to the waiting list (5653)
3056 - when creating new directories, inherit the parent's custom
3057 striping settings if present parent (3048)
3058 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3059 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3060 - don't hold client locks on temporary worklist from l_lru (5666)
3061 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3062 - automatically evict dead clients (5921)
3063 - Update file size properly in create+truncate+fstat case (6196)
3064 - Do not unhash mountpoint dentries, do not allow removal of
3066 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3067 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3068 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3069 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3070 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3071 - Correctly handle reads of files with no objects (6243)
3072 - lctl recover will also mark a device active if deactivate used (5933)
3074 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3075 - allow --write-conf on an MDS with different nettype than client (5619)
3076 - don't write config llogs to MDS for mounts not from that MDS (5617)
3077 - lconf should create multiple TCP connections from a client (5201)
3078 - init scripts are now turned off by default; run chkconfig --on
3079 lustre and chkconfig --on lustrefs to use them
3080 - upcalls are no longer needed for clients to recover to failover
3082 - add --abort-recovery option to lconf to abort recovery on device
3084 - add support for an arbitrary number of OSTs (3026)
3085 - Quota support protocol changes.
3086 - forward compatibility changes to wire structs (6007)
3087 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3088 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3089 - improved "lustre status" script
3090 - initialize blocksize for non-regular files (6062)
3091 - added --disable-server and --disable-client configure options (5782)
3092 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3093 - Vanilla 2.4.29 support
3094 - increase maximum number of obd devices to 520 (6242)
3095 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3096 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3097 SLES 9 2.6.5 kernels
3098 - limit stripes per file to 160 (the maximum EA size) (6093)
3100 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3103 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3104 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3105 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3106 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3107 - small ext3 extents cleanups and fixes (5733)
3108 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3109 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3110 - avoid needless client->OST connect, fix handle mismatch (5317)
3111 - fix DLM error path that led to out-of-sync client, long delays (5779)
3112 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3113 - fix several locking issues related to i_size (5492,5624,5654,5672)
3114 - don't move pending lock onto export if it is already evicted (5683)
3115 - fix kernel oops when creating .foo in unlinked directory (5548)
3116 - fix deadlock in obdfilter statistics vs. object create (5811)
3117 - use time_{before,after} to avoid timer jiffies wrap (5882)
3118 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3119 - Do not leak request if server was not able to process it (5154)
3120 - If mds_open unable to find parent dir, make that negative lookup(5154)
3121 - don't create new directories with extent-mapping (5909, 5936)
3123 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3124 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3125 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3126 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3127 - support for RHEL4 kernel on i686 (5773)
3128 - provide error messages when incompatible logs are encountered (5898)
3130 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3131 * version 1.4.0.10 (1.4.1 release candidate 1)
3133 - don't keep a lock reference when lock is not granted (4238)
3134 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3135 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3136 - drop import inflight refcount on signal_completed_replay error (5255)
3137 - unlock page after async write error during send (3677)
3138 - handle missing objects in filter_preprw_read properly (5265)
3139 - no transno return for symlink open, don't save no-trasno open (3440)
3140 - don't try to complete elan receive that already failed (4012)
3141 - free RPC server reply state on error (5406)
3142 - clean up thread from ptlrpc_start_thread() on error (5160)
3143 - readahead could read extra page into cache that wasn't ejected (5388)
3144 - prevent races in class_attach/setup/cleanup/detach (5260)
3145 - don't dereference de->d_inode after l_dput of de (5458)
3146 - use "int" for stripe value returned from lock_to_stripe (5544)
3147 - mballoc allocation and error-checking fixes in 2.6 (5504)
3148 - block device patches to fix I/O request sizes in 2.6 (5482)
3149 - look up hostnames for IB nals (5602)
3150 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3151 - don't start multiple acceptors for the same port (5277)
3152 - fix incorrect LASSERT in mds_getattr_name (5635)
3153 - export a proc file for general "ping" checking (5628)
3154 - fix "lfs check" to not block when the MDS is down (5628)
3156 - service request history (4965)
3157 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3158 - create an "evict_client" /proc entry on OSTs, like the MDS has
3159 - fix mount usage message, return errors per mount(8) (5168)
3160 - change grep [] to grep "[]" in tests so they work in more UMLs
3161 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3162 - remove ancient LOV_MAGIC_V0 EA support (5047)
3163 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3164 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3165 - allow readahead allocations to fail when low on memory (5383)
3166 - mmap locking landed again, after considerable improvement (2828)
3167 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3169 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3172 - send OST transaction number in read/write reply to free req (4966)
3173 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3174 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3175 - the watchdog thread now runs as interruptible (5246)
3176 - flock/lockf fixes (but it's still disabled, pending 5135)
3177 - don't use EXT3 constants in llite code (5094)
3178 - memory shortage at startup could cause assertion (5176)
3180 - reorganization of lov code
3181 - single portals codebase
3183 - add extents/mballoc support (5025)
3184 - direct I/O reads in the obdfilter (4048)
3185 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3187 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3190 - send OST transaction number in read/write reply to free req (4966)
3191 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3192 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3193 - the watchdog thread now runs as interruptible (5246)
3194 - handle missing objects in filter_preprw_read properly (5265)
3195 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3196 - drop import inflight refcount on signal_completed_replay error (5255)
3197 - unlock page after async write error during send (3677)
3198 - return original error code on reconstructed replies (3761)
3199 - no transno return for symlink open, don't save no-trasno open (3440)
3201 - add pid to ldlm debugging output (4922)
3202 - bump the watchdog timeouts -- we can't handle 30sec yet
3203 - extra debugging for orphan dentry/inode bug (5259)
3205 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3208 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3209 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3210 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3211 - fix NULL dereference in /proc/sys/portals/routes (4827)
3212 - allow failed mdc_close() operations to be interrupted (4561)
3213 - stop precreate on OST before MDS would time out on it (4778)
3214 - don't send partial-page writes before EOF from client (4410)
3215 - discard client grant for sub-page writes on large-page clients (4520)
3216 - don't free dentries not owned by NFS code, check generation (4806)
3217 - fix lsm leak if mds_create_objects() fails (4801)
3218 - limit debug_daemon file size, always print CERROR messages (4789)
3219 - use transno after validating reply (3892)
3220 - process timed out requests if import state changes (3754)
3221 - update mtime on OST during writes, return in glimpse (4829)
3222 - add mkfsoptions to LDAP (4679)
3223 - use ->max_readahead method instead of zapping global ra (5039)
3224 - don't interrupt __l_wait_event() during strace
3226 - add software watchdogs to catch hung threads quickly (4941)
3227 - make lustrefs init script start after nfs is mounted
3228 - fix CWARN/ERROR duplication (4930)
3229 - return async write errors to application if possible (2248)
3230 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3231 - print NAL number in %x format (4645)
3232 - update barely-supported suse-2.4.21-171 series (4842)
3233 - support for sles 9 %post scripts
3234 - support for building 2.6 kernel-source packages
3235 - support for sles km_* packages
3237 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3240 - ignore -ENOENT errors in osc_destroy (3639)
3241 - notify osc create thread that OSC is being cleaned up (4600)
3242 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3243 - reconstruct ost_handle() like mds_handle() (4657)
3244 - create a new thread to do import eviction to avoid deadlock (3969)
3245 - let lconf resolve symlinked-to devices (4629)
3246 - don't unlink "objects" from directory with default EA (4554)
3247 - hold socknal file ref over connect in case target is down (4394)
3248 - allow more than 32000 subdirectories in a single directory (3244)
3249 - fix blocks count for O_DIRECT writes (3751)
3250 - OST returns ENOSPC from object create when no space left (4539)
3251 - don't send truncate RPC if file size isn't changing (4410)
3252 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3253 - bind to privileged port in socknal and tcpnal (3689)
3255 - rate limit CERROR/CWARN console message to avoid overload (4519)
3256 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3257 - basic mmap support (3918)
3258 - kernel patch series update from b1_4 (4711)
3260 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3263 - avoid crash during MDS cleanup with OST shut down (2775)
3264 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3265 - don't use bad inodes on the MDS (3744)
3266 - dynamic object preallocation to improve recovery speed (4236)
3267 - don't hold spinlock over lock dumping or change debug flags (4401)
3268 - don't zero obd_dev when it is force cleaned (3651)
3269 - print grants to console if they go negative (4431)
3270 - "lctl deactivate" will stop automatic recovery attempts (3406)
3271 - look for existing locks in ldlm_handle_enqueue() (3764)
3272 - don't resolve lock handle twice in recovery avoiding race (4401)
3273 - revalidate should check working dir is a directory (4134)
3275 - don't always mark "slow" obdfilter messages as errors (4418)
3277 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3280 - don't close LustreDB during write_conf until it is done (3860)
3281 - fix typo in lconf for_each_profile (3821)
3282 - allow dumping logs from multiple threads at one time (3820)
3283 - don't allow multiple threads in OSC recovery (3812)
3284 - fix debug_size parameters (3864)
3285 - fix mds_postrecov to initialize import for llog ctxt (3121)
3286 - replace config semaphore with spinlock (3306)
3287 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3288 - don't allow enqueue to complete on a destroyed export (3822)
3289 - down write_lock before checking llog header bitmap (3825)
3290 - recover from lock replay timeout (3764)
3291 - up llog sem before sending rpc (3652)
3292 - reduce ns lock hold times when setting kms (3267)
3293 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3294 - fix NULL deref and obd_dev leak on setup error (3312)
3295 - replace some LBUG about llog ops with error handling (3841)
3296 - don't match INVALID dentries from d_lookup and spin (3784)
3297 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3298 - fix invalid assertion in ptlrpc_set_wait (3880)
3300 - add libwrap support for the TCP acceptor (3996)
3301 - add /proc/sys/portals/routes for non-root route listing (3994)
3302 - allow setting MDS UUID in .xml (2580)
3303 - print the stack of a process that LBUGs (4228)
3305 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3308 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3309 - make sure to unset replay flag from failed open requests (3440)
3310 - if default stripe count is 0, use OST count for inode size (3636)
3311 - update parent mtime/ctime on client for create/unlink (2611)
3312 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3313 - free recovery state on server during a forced cleanup (3571)
3314 - unregister_reply for resent reqs (3063)
3315 - loop back devices mounting and status check on 2.6 (3563)
3316 - fix resource-creation race that can provoke i_size == 0 (3513)
3317 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3318 - more debugging for page-accounting assertion (3746)
3319 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3320 - avoid deadlock after precreation failure (3758)
3321 - fix race and lock order deadlock in orphan handling (3450, 3750)
3322 - add validity checks when grabbing inodes from l_ast_data (3599)
3324 - add /proc/.../recovery_status to obdfilter (3428)
3325 - lightweight CDEBUG infrastructure, debug daemon (3668)
3326 - change default OSC RPC parameters to be better on small clusters
3327 - turn off OST read cache for files smaller than 32MB
3328 - install man pages and include them in rpms (3100)
3329 - add new init script for (un)mounting lustre filesystems (2593)
3330 - run chkconfig in %post for init scripts (3701)
3331 - drop scimac NAL (unmaintained)
3333 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3336 - clean kiobufs before and after use (3485)
3337 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3338 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3339 - change init script to fail more clearly if not run as root (1528)
3340 - allow clients to reconnect during replay (1742)
3341 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3342 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3343 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3344 - fix cond_resched (3554)
3345 - only evict unfinished clients after recovery (3515)
3346 - allow bulk resend, prevent data loss (3570)
3347 - dynamic ptlrpc request buffer allocation (2102)
3348 - don't allow unlinking open directory if it isn't empty (2904)
3349 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3350 - remove extraneous obd dereference causing LASSERT failure (3334)
3351 - don't use get_cycles() when creating temp. files on the mds (3156)
3352 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3353 - handle EEXIST for set-stripe, set proper directory name (3336)
3355 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3356 - fix ksocknal_fmb_callback() error messages (2918)
3358 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3361 - don't copy lvb into (possibly NULL) reply on error (2983)
3362 - don't deref dentry after dput, don't free lvb on error (2922)
3363 - use the kms to determine writeback rpc length (2947)
3364 - increment oti_logcookies when osc is inactive (2948)
3365 - update client's i_blocks count via lvb messages (2543)
3366 - handle intent open/close of special files properly (1557)
3367 - mount MDS with errors=remount-ro, like obdfilter (2009)
3368 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3369 - don't use cancelling-locks' kms values (2947)
3370 - use highest lock extent for kms, not last one (2925)
3371 - don't dereference ERR_PTR() dentry in error handling path (3107)
3372 - fix thread race in portals_debug_dumplog() (3122)
3373 - create lprocfs device entries at setup instead of at attach (1519)
3374 - common AST error handler, don't evict client on completion race (3145)
3375 - zero nameidata in detach_mnt in 2.6 (3118)
3376 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3377 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3378 - initialize RPC timeout timer earlier for 2.6 (3219)
3379 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3380 - print nal/nid for unknown nid (3258)
3381 - additional checks for oscc recovery before doing precreate (3284)
3382 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3383 - don't crash in mdc_close for bad permissions on open (3285)
3384 - zero i_rdev for non-device files (3147)
3385 - clear page->private before handing to FS, better assertion (3119)
3386 - tune the read pipeline (3236)
3387 - fix incorrect decref of invalidated dentry (2350)
3388 - provide read-ahead stats and refine rpc in flight stats (3328)
3389 - don't hold journal transaction open across create RPC (3313)
3390 - update atime on MDS at close time (3265)
3391 - close LDAP connection when recovering to avoid server load (3315)
3392 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3393 - don't leak open file on MDS after open resend (3325)
3394 - serialize filter_precreate and filter_destroy_precreated (3329)
3395 - loop device shouldn't call sync_dev() for nul device (3092)
3396 - clear page cache after eviction (2766)
3397 - resynchronize MDS->OST in background (2824)
3398 - refuse to mount the same filesystem twice on same mountpoint (3394)
3399 - allow llmount to create routes for mounting behind routers (3320)
3400 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3401 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3402 - fix rare problem with rename on htree directories (3417)
3404 - allow default OST striping configuration per directory (1414)
3405 - fix compilation for qswnal for 2.6 kernels (3125)
3406 - increase maximum number of MDS request buffers for large systems
3407 - change liblustreapi to be useful for external progs like lfsck (3098)
3408 - increase local configuration timeout for slow disks (3353)
3409 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3411 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3414 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3415 - don't overwrite extent policy data in reply if lock was blocked (2901)
3416 - drop filter export grants atomically with removal from device (2663)
3417 - del obd_self_export from work_list in class_disconnect_exports (2908)
3418 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3419 - swab reply message in mdc_close, other PPC fixes (2464)
3420 - fix destroying of named logs (2325)
3421 - overwrite old logs when running lconf --write_conf (2264)
3422 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3423 - fix race in target_handle_connect (2898)
3424 - mds_reint_create() should take same inode create lock (2926)
3425 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3426 - don't close files for self_export to avoid uninitialized obd (2936)
3427 - allow MDS with the same name as client node (2939)
3428 - hold dentry reference for closed log files for unlink (2325)
3429 - reserve space for all logs during transactions (2059)
3430 - don't evict page beyond end of stripe extent (2925)
3431 - don't oops on a deleted current working directory (2399)
3432 - handle hard links to targets without a parent properly (2517)
3433 - don't dereference NULL lock when racing during eviction (2867)
3434 - don't grow lock extents when lots of conflicting locks (2919)
3436 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3439 - account for cache space usage on clients to avoid data loss (974)
3440 - lfsck support in lustre kernel code (2349)
3441 - reduce journal credits needed for BRW writes (2370)
3442 - orphan handling to avoid losing space on client/server crashes
3443 - ptlrpcd can be blocked, stopping ALL progress (2477)
3444 - use lock value blocks to assist in proper KMS, faster stat (1021)
3445 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3446 - recovery for initial connections (2355)
3447 - fixes for mds_cleanup_orphans (1934)
3448 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3449 - block all file creations until orphan recovery completes (1901)
3450 - client remove rq_connection from request struct (2423)
3451 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3452 - recovery timer race (2670)
3453 - mdc_close recovey bug (2532)
3454 - ptlrpc cleanup bug (2710)
3455 - mds timeout on local locks (2588)
3456 - namespace lock held during RPCs (2431)
3457 - handle interrupted sync write properly (2503)
3458 - don't try to handle a message that hasn't been replied to (2699)
3459 - client assert failure during cleanup after abort recovery (2701)
3460 - leak mdc device after failed mount (2712)
3461 - ptlrpc_check_set allows timedout requests to complete (2714)
3462 - wait for inflight reqs when ptlrpcd finishes (2710)
3463 - make sure unregistered services are removed from the srv_list
3464 - reset bulk XID's when resending them (caught by 1138 test)
3465 - unregister_bulk after timeout
3466 - fix lconf error (2694)
3467 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3468 - readahead locks pages, leaves pending causing memory pressure (2673)
3469 - increase OST request buffers to 4096 on large machines (2729)
3470 - fix up permission of existing directories in simple_mkdir (2661)
3471 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3472 - don't assign transno to errored transactions (2742)
3473 - don't delete objects on OST if given a bogus objid from MDS (2751)
3474 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3475 - if rq_no_resend, then timeout request after recovery (2432)
3476 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3477 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3478 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3479 - MDS recovery shouldn't delete objects that it has given out (2730)
3480 - if enqueue arrives after completion, don't clobber LVB (2819)
3481 - don't unlock pages twice when trigger_group_io returns error (2814)
3482 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3483 - don't write pages to disk if there was an error (1450)
3484 - don't ping imports that have recovery disabled (2676)
3485 - take buffered bytes into account when balancing socknal conn (2817)
3486 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3487 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3488 - remove little-endian swabbing of llog records (1987)
3489 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3490 - retry reposting request buffers if they fail (1191)
3491 - grow extent at grant time to avoid granting a revoked lock (2809)
3492 - lock revoke doesn't evict page if covered by a second lock (2765)
3493 - disable VM readahead to avoid reading outside lock extents (2805)
3495 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3496 - updated kernel patches for hp-2.4.20 kernel (2681)
3498 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3501 - fix truncated write corruption (2366)
3502 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3504 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3505 - ensure len(uuid) < 37 in lmc (1171)
3506 - fix ia64 OOPS in llog_test (2255)
3507 - zero end of page at obdfilter for partial page writes (2648)
3508 - don't leave stale dentries around after renames (bug 2428)
3509 - fix timeouts when evicting a client with a single lock held (2642)
3510 - set deadline for the initial HELLO message to drain (2634)
3511 - print out dotted-quad IP addresses in the socknal (2302)
3513 - additional debugging for MDS client eviction problem (2443)
3514 - fix mkfsoptions support for osts (2603, 2604)
3516 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3519 - add series for the vanilla 2.6.0 kernel
3520 - add series for the vanilla 2.4.24 kernel
3521 - add series for a cray x86/64 UL kernel drop
3522 - fix xattr patches for the vanilla 2.4.19 series
3524 - generate true UUIDs in lmc (1171)
3525 - have portals stack dumping break in UML (2466)
3526 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3527 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3528 - wake callback waiting threads on client eviction (2460)
3529 - Add --ptldebug and --subsystem to lmc (1719)
3530 - update assertion to allow safe interrupt allocation
3531 - set rq_no_resend for cancel requests (2432)
3532 - recalculate ptlrpcd timeout after resend (2494)
3533 - call vfs_rmdir when removing pending directories (2368)
3534 - fix renaming a file to itself (2429)
3535 - lmc creates a default one-stripe lov (2454)
3536 - expand procfs space to handle large clusters (2326)
3537 - increase UML stack to avoid overflow
3538 - update lconf's list of debug and subsystem masks
3539 - fix lfs find --obd (2510)
3540 - /proc tunable for disabling filter read caching (2591)
3541 - stop rpm packages from altering slapd.conf (2301)
3542 - disable nagle in the socknal under 0conf (2578)
3543 - choose mds inode size based on stripe count (2572)
3544 - fix kernel-source rpm problems (2516)
3546 - add --disable-doc to avoid pdf generation (2421)
3547 - update documentation, tests, type-os, comments
3548 - avoid format warnings on ia64
3549 - remove the TOE NAL
3550 - tiny code cleanups by removing unused fields
3552 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3555 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3556 - remove the most blatant lies from BUILDING file (2371)
3557 - change default debug level to reasonable production setting
3558 - reduce client side cache size to reduce cache flush time
3559 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3560 - make TCP zerocopy and pinger support enabled by default (2476)
3561 - sync writes completed after process exits caused crashes (2319)
3562 - maintain correct mount count on the MDS (2356)
3563 - backout 1557, because 2316 wasn't really fixed
3564 - better file I/O statistics gathering in /proc
3565 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3566 - another kernel patch to fix zero-copy TCP function export
3567 - don't take duplicate lock when processing re-sent getattr (2420)
3568 - lctl uses obd_self_export instead of creating new conn (2353)
3569 - MDS/OST recovery case which requires object creation asserted (2425)
3570 - move lfs from /usr/sbin to /usr/bin in packages
3571 - fix race between mds_client_add and mds_client_free (2417)
3572 - use kmalloc instead of slabs in portals (2430)
3573 - don't create duplicate records when a failover MDS is present (2442)
3574 - remove unnecessary mount age check (2332)
3575 - don't remove directory inodes from locks prematurely (2451)
3576 - don't break if MDS service name is the same as hostname (2103)
3577 - fix races in client write RPC generation when cache full (2482)
3579 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3582 - remove now-unused request->rq_obd (278)
3583 - if an allocation fails, print out how much memory we've used (1933)
3584 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3585 - add the "configurable stack size" patch to most series files (1256)
3586 - ability to write large log records, for 100+ OST configs (2306)
3587 - fix NULL deref when filter_prep fails (2314)
3588 - fix operator precedence error in filter_sync
3589 - dynamic allocation of socknal TX descriptors (2315)
3590 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3591 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3592 - ensure that CWARN messages in Portals always get to the syslog
3593 - __init/__exit are not for prototype decls (ldlm_init/exit)
3594 - x86-64 compile warning fixes
3595 - fix gateway LMC keyword conflict (2318)
3596 - fix MDS lock inversions in getattr/reint paths (1844)
3597 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3598 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3599 - DLM race condition prevented some lock evictions (2328)
3600 - ENOMEM detection and retry on socknal sends (2230)
3601 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3602 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3603 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3604 - better error messages when a client is rejected during recovery (1505)
3605 - avoid cancelling locks which were never granted, after failure (2330)
3606 - fix i_sem/journal inversion in mds_client_add (2333)
3607 - fix truncate/getattr lock cycle deadlock (2334)
3608 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3609 - fix two rare exit paths which could leak an l_lock() ref (2321)
3610 - fencepost error in MDS/OST orphan recovery (2226)
3611 - make log record alignment 8 bytes (1988)
3612 - lstripe now fails when requested offset > ost_count (2237)
3613 - ensure that all kernel series have a complete list.h (1607)
3614 - fix crashes in special-file operations (2316)
3615 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3617 - allow configurable automake binary, for testing new versions
3618 - small update to the lfs documentation
3620 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3622 * fix negative export reference count in fsfilt_sync (2312)
3624 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3625 * release candidate 0.9.1
3627 - orphans are moved into the PENDING directory for possible recovery
3628 - replayed opens now open by fid for orphan/rename safety (1042)
3629 - last close of an orphan inode generates a transno (683)
3630 - chdir() and mount() now pin the directory entry (1020)
3631 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3632 - discard very old requests without processing them (1502)
3633 - remove some common, well-understood CERRORs (1505)
3634 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3635 - clear "grant" flags in OST replies until OST grant code lands (1644)
3636 - fix read performance by not clobbering i_blksize on client (1598)
3637 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3638 - make LRU size a /proc tunable, clears locks when reduced (707)
3639 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3640 - server locks take references on exports now (1558)
3641 - build fixes for 2.4.20-rh trees (1663)
3642 - return an error from lov_create if all OSCs are inactive (1751)
3643 - fix import levels when a reconnect happens without a timeout (1597)
3644 - exit early from mds_open if we get a lookup error (1749)
3645 - partial page read at EOF wouldn't wait for disk before sending (1642)
3646 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3647 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3648 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3649 - write/truncate lock inversion (1639)
3650 - Don't auto-load obdclass, portals modules during cleanup (1495)
3651 - fix timestamps from jumping to "now" (1763)
3652 - extra journal assertions (1648)
3653 - add an extra multiunlink test (1771)
3654 - fix read_record/write_record API (1776)
3655 - fix leak of offset_extent, possible incorrect i_size later (1772)
3656 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3657 - Debugging for the kqswnal_get_idle_tx problems (1820)
3658 - Allow recovery to be attempted multiple times (1536)
3659 - Write out MDS last_rcvd file after it is first created (1600)
3660 - Fix tx_descriptor leak in failed transmit situations (1827)
3661 - ext3 journaling fixes for assertion failure after IO error (1871)
3662 - class_export_put() on freed export after completion AST error (1896)
3663 - Fix revalidate looping in VFS (1322)
3664 - Don't access a freed export during MDS_REINT timeout (1521)
3665 - Add open-unlink recovery support on the MDS (1673,1764)
3666 - Return an error if no MDS data was read from last_rcvd (1946)
3667 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3668 - Don't setup a disk device that doesn't match exported UUID (317)
3669 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3670 - avoid committing NULL handle in force close
3671 - local.sh is now a one-stripe LOV configuration
3672 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3673 - fix ext3 htree duplicate directory entry corruption (1516)
3674 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3675 - update to new LOV EA format (2097)
3676 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3678 2003-06-15 Phil Schwan <phil@clusterfs.com>
3681 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3682 - per-import/export recovery handling (958, 931, 959)
3683 - multiple last-rcvd slots, for serving multiple FSes (949)
3684 - connections are again shared between multiple imp/exports (963, 964)
3685 - "umount -f" would hang if any requests needed to be sent (393, 978)
3686 - avoid pinning large req buffer by copying for queued messages (989)
3687 - add "uuid" to "lctl device" command to help upcalls (991)
3688 - "open" RPCs with transnos would confuse recovery counters (1037)
3689 - do proper endian conversion of all wire messages (288, 340, 891)
3690 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3691 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3692 - fix ptlrpc_connection leak in target_handle_connect (1174)
3693 - fix import refcounting bug in OST and MDS cleanup (1134)
3694 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3695 - fix very unlikely obd_types race condition (501)
3696 - remove osc_open hack for echo_client (1187)
3697 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3698 - a failure in read_inode2 leads to deadlock (1139)
3699 - cancel ack-locks as soon as transaction is committed (1072)
3700 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3701 - make sure to commitrw after any preprw to avoid deadlock (1162)
3702 - failing to execute a file in a lustre FS would lock inode (1203)
3703 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3704 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3705 - obd_uuid2tgt would walk off the end of the list (1255)
3706 - on IA64 the osc would give portals incorrect bulk size (1258)
3707 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3708 - fix lock inversion caused by new llite matching code (1282)
3709 - limit the number of dirty pages on a client to 10MB (1286)
3710 - timed out locks were not being corrected cancelled (1289)
3711 - fix O_DIRECT above 4GB on IA-32 (1292)
3712 * major user-visible changes
3713 - fail out/fail over policy now controlled by the upcall (993)
3715 - add OBD_PING to check server availability and failure (954)
3716 - lustre messages are now sent in sending host order (288, 340, 891)
3717 - add eadatalen to MDS getattr reply (340)
3718 - OST read replies may contain second buffer, with per-page status (593)
3720 2003-03-11 Phil Schwan <phil@clusterfs.com>
3723 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3724 - failed open()s could cause deadlock; fixed (867, 869)
3725 - stop cancelling OST locks when files are closed (481)
3726 - overlapping XID spaces caused network corruption (851, 853)
3727 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3728 - setattr_raw added, to do single-RPC, server-side setattrs
3729 - lmc/lconf syntax change for OST UUIDs
3730 - fix crashy race condition between ptlrpc_free_req and osc_close
3731 - don't use request in mdc_enqueue if we hit a timeout (889)
3732 - don't set the inode i_size for regular files from the MDS (896)
3733 - handle out of order completion AST (842)
3734 - don't LBUG if a lock request times out after receiving AST (913)
3735 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3736 - if a bad lock AST arrives, send an error instead of dropping entirely
3737 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3738 - fix leak in bulk IO when only partially completed (899, 900, 926)
3739 - fix O_DIRECT for ia64 (55)
3740 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3741 - C-z after timeout could hang a process forever; fixed (977)
3743 - client-side I/O cache (678, 924, 929, 941, 970)
3745 - READPAGE and SETATTRs which don't take server-side locks get
3748 2003-02-11 Phil Schwan <phil@clusterfs.com>
3751 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3752 - Fix fsfilt_extN_readpage() to read a full page of directory
3753 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3754 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3755 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3756 - Most intent operations were removed, in favour of a new RPC mode
3757 that does a single RPC to the server and bypasses most of the VFS
3758 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3759 - Aggressively cancel local locks on DLM servers
3760 - mds_reint_unlink sends EA to the client if it's the last nlink.
3761 client uses that EA to unlink OST objects.
3762 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3763 - recursive symlinks were fixed (439)
3764 - fixed NULL deref in DEBUG_REQ
3765 - filter_update_lastobjid no longer calls sync, which annoyed extN
3766 - fixed multi-client small-writes to a single file problem (445)
3767 - fixed mtime updates during file writes (607)
3768 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3769 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3770 - fixed Linux OST size reporting problem (444, 656)
3771 - OST now updates object mtime with writes or setattr (607, 619)
3772 - client verifies file size before zeroing page past EOF (445)
3773 - OST now writes last allocated objid to disk with allocation (108)
3774 - LOV on echo now works (409)
3776 - mds_reint_unlink sends a new buffer, with the EA included. this
3777 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3778 set if a regular file was being unlinked, and it was the last link
3779 - use PtlGet from the target for bulk writes (315)
3780 - OST now updates object mtime with writes or setattr (607, 619)
3781 - LDLM now has a grant-time callback to revalidate locked items, if
3783 - Many MDS operations were reorganized to combat race conditions
3785 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3786 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3788 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3791 - Fully reactivate OST imports after reconnection (512, others)
3792 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3793 - More graceful error handling for truncating on dead OST (515)
3794 - Don't error out unless we're actually accessing dead stripes (474)
3795 - Fix garbage sizes when stripes are missing (410)
3796 - LRU counters were broken, causing constant lock purge (433, 432)
3797 - garbage on read from stripes with failed OSTs (441)
3798 - mark OSCs as active before reconnecting during recovery (438)
3799 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3800 - lfind did not preserve OST order in output (443)
3801 - symlinks cause hung clients, incorrect data (439)
3802 - stop dereferencing request after dropping refcount (457)
3803 - don't LASSERT(spin_is_locked) on non-SMP (455)
3804 - fixes for many rename() bugs
3805 - fstat didn't correctly synchronize attributes (399)
3806 - server must handle lock cancellation during blocking AST prep (487)
3807 - bulk descriptors were free()d too soon (511)
3808 - fix paths in lconf, which would load incorrect modules (451, 507)
3809 - fix confusing lconf 'host not found' error message (386)
3810 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3811 - fix race condition in mdc_blocking_ast() for inode access (526)
3812 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3813 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3814 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3815 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3816 - fix lstripe in several ways (kernel side) (527)
3817 - fix request leak in ldlm_cli_enqueue (262)
3818 - incorrect OSC was marked inactive after OST failure
3819 - call mds_fs_cleanup before unmounting filesystem (524)
3820 - fix races between taking ns_lock and ldlm_lock_change_resource
3821 - fix races updating LOV export open file list
3822 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3823 - fix recovery NULL deref in ldlm_cli_cancel_unused
3824 - fix some DLM races by using new hash table for lock handles (419)
3825 - permit the client to specify desired inodes, at replay
3826 - duplicate requests when we queue them for replay reintegration
3827 - fix last_rcvd offset calculation
3828 - sync after each recovered transaction, so we always make progress
3829 - never, not always, ERESTART requests without transnos
3830 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3832 - skip replay if the MDS says that the client is already connected
3833 - don't check for a recovery-enabled export to match lctl's UUID
3834 - don't INC_USE_COUNT for phantom exports
3835 - don't crash when cleaning up phantom exports (567)
3836 - don't double-finish or set replay data for errored mdc_open requests
3837 - abort requests when they time out, so we don't get old replies
3838 - send/receive replies for AST messages again
3839 - if the client says that it doesn't have the lock, cancel it on the
3841 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3842 mark it as destroyed, it will all work out when decref is called
3843 - fix module use counts (22, 581)
3845 - ASTs now expect a reply (server cancels lock on error reply)
3847 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3850 - fix many simultaneous client startup (392)
3851 - fix dentry->d_it clobbering
3852 - credentials weren't being shipped for readdir/getattr operations
3853 - remove invalid assertions triggered during some concurrent MD
3855 - proper Lustre versions added (336, 389)
3856 - fix memory leak for create error case (398)
3857 - fix LOV locking bug that would get cli/srv out of sync
3858 - fix echo client over LOV (409)
3859 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
3860 - fix double-O_EXCL intent crash (424)
3861 - avoid sending multiple lock CANCELs (352)
3863 - MDS can do multi-client recovery (modulo bugs in new code)
3865 - many updates, edits, cleanups
3867 2002-11-18 Phil Schwan <phil@clusterfs.com>
3870 - fix null d_it dereference (346)
3871 - fix full OST/dbench hang (333)
3872 - fix permission problem with file removal (286)
3873 - fix removal of OSCs from LOV when they fail
3874 - fix NULL deref during bulk timeout (214)
3875 - fix problems related to multiple filesystems on one MDS (241)
3876 - fixed serious subtle metadata locking bugs
3877 - free locks on clients when inodes are removed due to memory
3879 - fix inode pointer in lock data (285)
3880 - partial support for multiple MDS on a single host (241)
3881 - data locks weren't cancelled at clear_inode time (290, 311)
3882 - intent locks could lead to unbounded lock growth (205)
3883 - added a maximum lock count, an LRU list, and a flusher
3884 - fix multiple rename (365)
3885 - properly abstracted the echo client
3886 - OSC locked 1 byte too many; fixed
3887 - rewrote brw callback code:
3888 - fixed recovery bugs related to LOVs (306)
3889 - fixed too-many-pages-in-one-write crash (191)
3890 - fixed (again) crash in sync_io_timeout (214)
3891 - probably fixed callback-related race (385)
3893 - Add capability to MDS protocol
3894 - LDLM cancellations and callbacks on different portals
3896 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
3899 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
3900 - defer open object destruction to close time (601981, 138)
3901 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
3902 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
3903 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
3904 - don't call obd_setattr in ll_file_release for destroyed objects
3905 * protocol change to lustre_msg: move |version| and add |flags|
3906 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
3907 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
3908 * added replay of create, unlink, link and rename operations during
3909 MDS failover; recovery should be much more robust now
3910 * remove failed OSCs from LOVs (only lov_create uses this so far)
3911 * the lustre-HOWTO was brought (more) up to date (582544)
3913 2002-10-23 Phil Schwan <phil@clusterfs.com>
3916 - in-use dentries weren't being reused properly (617851)
3917 - prevent multiple LDLM setup (599178)
3918 - fix LOV size calculations for truncate (617853)
3919 - fix client handling of MDS intent errors (POSIX)
3920 - fix permission bug in lovstripe.c test (624321)
3921 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
3922 - truncate past end of file could corrupt data
3923 - proper cleanup after timeouts, crashes, etc (592524, 550815)
3924 - a race in recovery could return ETIMEDOUT to apps (623947)
3925 - building outside the source directory was fixed
3926 * the lustre-HOWTO was brought (more) up to date (582544)
3927 * major progress was made on recovery functionality
3929 2002-10-10 Phil Schwan <phil@clusterfs.com>
3932 - recovery deadlock fix
3933 - rm -rf causes LBUG fix (617817)
3934 - file open by multiple tasks fix (618962)
3935 - directory permissions bugs (602707 and 620007)
3936 - journal_stop fixed with locking (611313)
3937 - O_APPEND failures resolved (618273, perhaps 614459)
3938 - lconf PATH fix (619770)
3939 - IA64 build fix (621450)
3940 - RPC buffer sizes scale with amount of memory
3942 2002-10-01 Phil Schwan <phil@clusterfs.com>
3945 - locks would be cancelled without throwing away data pages,
3946 resulting in inconsistent data (605627)
3947 - inode attributes were not always being refreshed (605627, 612449)
3948 - lconf now continues to cleanup after lctl reports an error
3949 - MDS now enforces user permissions (602707)
3950 - lprocfs cleanup fixed, but not yet enabled (614157)
3951 - fixed infinite server hang, should a client not respond to an AST
3952 - avoid going into recovery if user calls readlink() with a buffer
3953 that's too small (613941)
3954 - AST RPCs no longer require replies (614867) -- this may be changed
3955 - don't crash server if client sends an IOV that's too big (611336)
3956 - fixed lock conversion deadlock (611892)
3957 - fixed the following of symlinks (614622)
3958 * recovery: the server can remove locks from a client that dies, other
3959 clients can make progress
3960 * more extN patch fixes
3961 * compile-time configurable ptlrpc buffer allocations
3963 - collaborative read cache document
3964 - Lustre Lite Performance CDR document-in-progress
3966 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3969 - fix typo in patch-2.4.18
3971 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
3974 - clear ptlrpc request each time in handle_incoming_request()
3975 - unlink of files now destroys the object on the OST
3977 2002-09-19 Peter Braam <braam@clusterfs.com>
3979 * add hard link support
3980 * change obdfile creation method
3981 * kernel patch changed
3983 2002-09-19 Peter Braam <braam@clusterfs.com>
3986 - stack overflow bug in extN fixed
3988 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
3990 * documentation updates
3991 - add man pages for config tools
3992 - update tests/README to describe testing with new config tools
3993 - finish metadata API descriptions
3994 * bug fixes and cleanups
3995 - statfs workaround for 16TB limit
3996 - LOV stripe allocation improved, can stripe on subset of OSTs
3997 - LOV file size/IO offset was wrong for files > 4GB in size
3998 - object EA data was being dropped, caused files to be unreadable
3999 - memory overflow with non-LOV OST caused memory corruption
4000 - fixed regression tests to work with new config tools, obdfilter
4001 - fixed bug when directory size became larger than 1 block
4002 - fixed bug (for single client case) when PWD was deleted
4003 - invalidate local directory pages when doing intent-based ops
4004 - avoid LDLM oops when lock callback contained bad data
4006 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4008 * documentation updates
4009 * bug fixes and cleanups
4010 - configuration tools
4013 - 64-bit compile warnings
4014 - 64-bit internal statfs data
4016 * test_brw on persistent OST devices
4018 * lprocfs (disabled)
4020 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4022 * documentation updates
4023 * bug fixes and cleanups
4024 * configuration tools
4026 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4029 * many small fixes to 0_5_4
4030 * io/network handling
4031 * thinkos in MDS operations
4033 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4036 * crucial basic fixes to 0.5.3
4037 * IOR, Iozone work over Elan
4040 2002-08-07 Phil Schwan <phil@clusterfs.com>
4041 * version 0_5_3, our first alpha
4042 * we use the new Portals iovs
4043 * documentation updates
4044 * bug fixes and cleanups
4045 * small changes in the DLM wire protocol
4047 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4048 * version 0_5_1 with some initial stability,
4049 * locking on MD and file I/O.
4050 * documentation updates
4051 * several bug fixes since 0.5.0
4052 * small changes in wire protocol
4054 2002-07-18 Phil Schwan <phil@clusterfs.com>
4056 * delivered as Lustre Light Alpha
4057 * fixed a crash after handling invalid MDS requests
4058 * fixed directory pages for architectures with non-4k pages sizes
4060 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4061 * release version v0_4_4
4062 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4064 * Fixes a number of interruption problems with OST operations.
4065 * Update documentation for portals header changes
4066 * Move all wire protocol structs/defines to lustre_idl.h
4067 * Fixes symlink length bug.
4068 * Add tcpdump to repository.
4070 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4071 * release version v0_4_3
4072 * Fixes statfs for inodes on extN.
4073 * Fixes bug in runtests which would delete /etc/hosts.
4074 * Use 64-bit object IDs wherever possible (not into VFS though)
4075 Remove ost_get_info, which is unused by lustre, and out of date.
4077 2002-07-03 Peter Braam <braam@clusterfs.com>
4078 * release version v0_4_2 Fixes a lookup error (type not passed)
4079 * move forward to head of Portals
4080 * move forward to latest Lustre kernel
4082 2002-06-25 Peter Braam <braam@clusterfs.com>
4083 * release version v0_4_1. Hopefully stable on single node use.