1 xxxx-xx-xx Oracle, Inc.
4 2.6.16.60-0.42.8 (SLES 10),
5 2.6.27.39-0.3.1 (SLES11),
6 2.6.18-194.3.1.el5 (RHEL 5)
7 2.6.18-194.3.1.0.1.el5 (OEL 5)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.16 - 2.6.30 vanilla (kernel.org)
11 * Recommended e2fsprogs version: 1.41.10-sun2
12 * The async journal commit feature (bug 19128) and the cancel
13 lock before replay feature (bug 16774) are disabled by default.
17 Description: llap pages are moved to head of LRU list instead of tail
21 Description: disable statahead by default due to important races found in the
22 statahead implementation.
26 Description: release request ref through ptlrpc_server_drop_request()
27 Details : make sure the request is protected by rq_refcount while
28 being processed in ptlrpc_server_handle_req_in().
32 Description: correct check for transno value in filter_finish_transno()
36 Description: Set body->eadatasize in mdc_getattr_pack()
40 Description: Handle unsent requests with rq_net_err in ptlrpc_check_set()
44 Description: apps stuck in ptlrpc_check_set() during direct I/O
45 Details : start async bulk unregistering at the same time at reply unlink
49 Description: lfs setstripe --pool broken in 1.8.4
50 Details : fix llapi_search_fsname() to handle relative pathnames
54 Description: Only force the mode change if we're changing the size as well
55 Details : Fix regression introduced in 1.8.0. The offending code was added
56 by commit 77ba4b2141d04180211efa8a75c11ab0abf7fafb to remove
57 setgid/setuid bits when do_truncate() is called on the file.
58 We should only force the change when that occurs, similarly to
59 ll_setattr() in lustre/llite/llite_lib.c
61 Severity : enhancement
63 Description: add procfs tunable to enable/disable lockless direct I/O
64 Details : llite.lustre-*.lockless_direct_io=0 will disable default semantics
65 of direct I/O that forces it to be lockless. lockless_direct_io
66 value, however, will be ignored if per-file LL_FILE_LOCKED_DIRECTIO
71 Description: Lustre servers crashing with NULL pointer errors
72 Details : Make sure the request is protected by rq_refcount while being
73 processed in ptlrpc_server_handle_req_in().
77 Description: kernel BUG at drivers/scsi/sd.c
78 Details : Remove sd iostats patch from sles11 patch series.
82 Description: Modified value of at_min is not taken into account
86 Description: lustre grants flock exclusive locks to two fd's in the same process
88 -------------------------------------------------------------------------------
90 2010-10-29 Oracle, Inc.
92 * Support for kernels:
93 2.6.16.60-0.69.1 (SLES 10),
94 2.6.32.19-0.2.1 (SLES11),
95 2.6.18-194.17.1.el5 (RHEL 5)
96 2.6.18-194.17.1.0.1.el5 (OEL 5)
97 * Client support for unpatched kernels:
98 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
99 2.6.16 - 2.6.32 vanilla (kernel.org)
100 * Recommended e2fsprogs version: 1.41.10-sun2
101 * The async journal commit feature (bug 19128) and the cancel
102 lock before replay feature (bug 16774) are disabled by default.
106 Description: atime is not properly updated on an MDS
108 Severity : enhancement
110 Description: Update to RHEL5.5 kernel 2.6.19-194.17.1.el5.
111 Update to OEL5.5 kernel 2.6.19-194.17.1.0.1.el5.
113 Severity : enhancement
115 Description: Update to SLES10 SP3 kernel 2.6.16.60-0.69.1.
118 Frequency : only with SLES10
120 Description: Use OFED "KMP" provided by Novell
121 Details : SLES10 SP3 ships with OFED in a separate "KMP" package.
122 Lustre is now built against this package. That means you need to
123 install the ofed-kmp package from Novell for the patchless client
124 and from our download site for the server. Note that the ofed-kmp
125 that Novell ships may not exactly match the kernel version but
126 should still be compatible.
128 Severity : enhancement
130 Description: Update SLES11 SP1 kernel to 2.6.32.19-0.2.1.
134 Description: Enabling quotas fails with non-consecutive OST numbering.
138 Description: Fix kernel warning due to lookup_on_len() called without i_mutex
143 Description: Account direct i/o inflight rpcs separately from non-direct i/o so
144 that direct i/o, which is limited by max_rpcs_in_flight, should not
145 block non-direct i/o, which is not limited by max_rpcs_in_flight.
149 Description: Fix per-NID reporting on outstanding writes
153 Description: Reduce stack pressure by uninlining some mds and ptlrpc functions.
157 Description: Remove LASSERT in lprocfs_rd_conn_uuid() since conn == NULL is a
162 Description: fix obdo leak issue in ll_setattr_raw()
166 Description: limit MMP interval
168 Severity : enhancement
170 Description: add several lfs ost enhancements
174 Description: Too many default ACLs break directory access on new directories
178 Description: Lustre inode size is not coherent across nodes.
179 Details : Update lvbo from disk when AST fails with EINVAL. Lvbo is updated
180 on EINVAL error in ldlm_handle_ast_error(). The updates in
181 filter_intent_policy() and ldlm_cb_interpret() have been removed as
186 Description: Oops at __percpu_counter_add+0x1b
187 Details : Use bdi_init()/bdi_destroy() to proper initialize backing_dev_info
192 Description: add mount option to generate 32bit ino, this can be used for 32bit
193 application compatibility.
197 Description: keep reference count for "lli_sai" to prevent it to be released
198 when "statahead_enter()"
202 Description: allow quotacheck over OSTs with sparse indices
206 Description: Objects not getting deleted for files which have been removed
207 Details : ll_have_md_lock() should differentiate between CR and CW OPEN
212 Description: pin object's inode in memory to avoid certain timeouts
216 Description: fix LBUG when obdfilter-survey is interrupted.
218 -------------------------------------------------------------------------------
220 2010-07-31 Oracle, Inc.
222 * Support for kernels:
223 2.6.16.60-0.42.8 (SLES 10),
224 2.6.27.39-0.3.1 (SLES11),
225 2.6.18-194.3.1.el5 (RHEL 5)
226 2.6.18-194.3.1.0.1.el5 (OEL 5)
227 * Client support for unpatched kernels:
228 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
229 2.6.16 - 2.6.30 vanilla (kernel.org)
230 * Recommended e2fsprogs version: 1.41.10-sun2
231 * The async journal commit feature (bug 19128) and the cancel
232 lock before replay feature (bug 16774) are disabled by default.
234 Severity : enhancement
236 Description: Update RHEL5.5 kernel to 2.6.18-194.8.1.el5 and OEL5.5 kernel to
237 2.6.18-194.8.1.0.1.el5.
239 Severity : enhancement
241 Description: using inkernel OFED stack for rhel5 & oel5.
243 Severity : enhancement
245 Description: Add "lfs_migrate" script from manual into lustre/scripts and RPMs
246 Details : lfs_migrate does a "poor man's" migration of files from their
247 current OST layout to a new OST layout as chosen by the MDS.
251 Description: mds_orphan_add_link()) error linking orphan to PENDING
252 Details : quota limits might disallow linking orphans to PENDING
253 when unlinking a file - temporary raise threads' privileges
254 when processing unlinks.
256 Severity : enhancement
258 Description: add conf-param -d option to remove permanent settings.
259 Details : Add the ability to remove permanent lctl conf_param settings.
260 (Previously conf_param settings could only be changed, not
261 removed.) This also provides a method to change failover
262 nid locations. Improve lctl man page.
264 Severity : enhancement
266 Description: add list_param to b1_8 and add "-R" option to list params
269 Severity : enhancement
271 Description: lfs quota output is not very convenient for awk/sed-parsing
272 Details : Some positions in lfs quota output table could be empty or
273 non-empty which made it hard to parse it with scripts, now a dash
274 is put instead of space where there is not supposed to be any data.
276 Severity : enhancement
278 Description: fix obdfilter-survey script to work properly with remote oss-s
280 Severity : enhancement
282 Description: add new OBDFILTER_SURVEY test suite
284 Severity : enhancement
286 Description: add new multiple mount protection (MMP) test suite
288 Severity : enhancement
290 Description: add support for async journal commit in echo client
292 Severity : enhancement
294 Description: allow userland programs to include <lustre/lustre_idl.h>
295 from stardard include directories
297 Severity : enhancement
299 Description: The prune-icache-use-trylock is no longer needed now that
300 the patch from bug 20008 is landed.
304 Description: The shrink grant feature is still active on the client although the
305 connect flag is not set.
309 Description: Don't leak grant space if the write failed with quota exceeded.
313 Description: Don't consume grant space twice on recoverable resent.
317 Description: a race condition could lead to SIGBUS being sent to an
318 application using mmap-ped files from Lustre
319 Details : truncate_complete_page implementation for the patchless
320 client could arbitrarily unset PG_Uptodate flag for a
321 page being kicked from the page cache, an uptodate check
322 right after a readpage call in filemap_fault could fail
323 because of that as though the page read had been unsuccessful.
327 Description: dlm lock slab shrinking is not efficient
328 Details : The dlm_locks slab can grow significantly and consumes a lot of
329 memory on the server. Set a hardlimit to grant_plan.
333 Description: Lustre does not do 1MB IOs to HW RAID
334 Details : Bump MAX_PHYS/HW_SEGMENTS and SG_ALL to 256 in the RHEL5 kernel.
335 This is what we do already for SLES kernels.
339 Description: bump maximum number of phys/hw segments in the SLES11 kernel
340 until s/g chaining works properly.
345 Description: LSI Fusion MPT driver hacks to improve performance
346 Details : Set CONFIG_FUSION_MAX_SGE to 256 for RHEL5
348 Severity : enhancement
350 Description: increase default md stripe_cache_size to 16k
353 Bugzilla : 15587/21439
354 Description: don't handle security.capability xattr
355 Details : CONFIG_SECURITY_FILE_CAPABILITIES is enabled by default on SLES11.
356 This results in additional getxattr calls, causing VBR test
357 failures as well as a preformance drop when writing.
361 Description: obdfilter-survey is no longer working
362 Details : revert patch from bug 20355 to resolve an issue with lctl --threads
363 not working correctly with $(PTHREAD_LIBS) being linked to lctl.
367 Description: ll_shrink_cache does not handle __GFP_FS properly
371 Description: lfs getstripe shows wrong info for directories
372 Details : Set correct LOVEA default values for filesystem-wide.
376 Description: FSX checksum false positves due to mmap IO
377 Details : Use OBD_FL_MMAP flag for IOs on a memory mapped file. Do not print
378 checksum errors, if the flag is set on a request.
382 Description: file operations after eviction have successful return values
383 Details : use vfs ->flush callback to return any pending async errors
388 Description: mdsrate fails to write after 1.3+M files opened
389 Details : decrease memory usage on clients by recycling dentries and
394 Description: obdfilter-survey gives unreasonably high numbers
395 Details : Wait for all threads to complete when running test_brw.
399 Description: do not set lustre read_only device when server umount and keep
400 client records for recoverable ones
404 Description: move sync_on_lock_cancel tunable to the obdfilter layer
405 Details : move the tunable to trigger a journal flush on lock cancel from
406 the ost layer to the obdfilter layer. This tunable is useful
407 when using the async journal commit feature.
411 Description: exp->exp_nid_stats == NULL in filter_tally()
412 Details : fix race with per-nid stats by delaying procfs cleanup until
417 Description: extent lock cancellation on client can keep the cpu busy for too
422 Description: Do not fail OST activation when a llog is not found, just
423 issue an error message.
427 Description: Don't enable extents by default for MDT.
431 Description: Protect bitfield access to ptlrpc_request's rq_flags, since
432 the AT code can access it concurrently while sending early
437 Description: Disable lockless truncate by default since it is sometimes flawed
438 and causes the write_disjoint test to fail.
442 Description: OSSs which don't have the patch from bug 20278 can trigger an
445 Severity : enhancement
447 Description: don't print message to the console when we have not managed to
452 Description: The MDS fails to synchronize OSTs which registered with the MGS
453 after the MDT. The problem is that OBD_NOTIFY_CREATE events are
454 raised too early and thus discarded by the MDT stack.
455 The fix consists of issuing OBD_NOTIFY_CREATE event in the lov
460 Description: Fix race when the ping evictor and a service thread execute
461 target_recovery_check_and_stop() concurrently.
465 Description: quota broadcast can trigger a LBUG on the MDT if there are
470 Description: Resetting the lov_objid values to last_id reported by the OST
471 during orphan recovery is incorrect and can cause the same
472 objects to be allocated twice.
474 Severity : enhancement
476 Description: "weak-modules" support
477 Details : Implement "weak-modules" support which enables kernel modules
478 to be used with any kernel that implements the same kABI. In
479 order to achieve this modules are now installed in
480 /lib/modules/$(uname -r)/updates/kernel on all distributions.
482 Severity : enhancement
484 Description: add writeconf as mount option
486 Severity : enhancement
488 Description: produce debuginfo packages for SLES.
490 Severity : enhancement
492 Description: add failover nidlist to the import proc file.
494 Severity : enhancement
496 Description: fix LUSTRE_SEQ_MAX_WIDTH for interoperability between 1.8
497 clients and 2.0 servers.
499 Severity : enhancement
501 Description: lfs find -s does not work correctly because of a bug in
506 Description: ll_read_ahead_page() must validate the dlm lock before using
511 Description: Prevent failover nids from registering with MGS first.
515 Description: fix lock inversion in ll_setattr_raw().
519 Description: object allocation is not balanced across OSTs.
520 Details : osc_precreate() should return 0, if there are enough objects left.
522 -------------------------------------------------------------------------------
524 2010-04-30 Oracle, Inc.
526 * Support for kernels:
527 2.6.16.60-0.42.8 (SLES 10),
528 2.6.27.39-0.3.1 (SLES11),
529 2.6.18-164.11.1.el5 (RHEL 5)
530 2.6.18-164.11.1.0.1.el5 (OEL 5)
531 * Client support for unpatched kernels:
532 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
533 2.6.16 - 2.6.30 vanilla (kernel.org)
534 * Recommended e2fsprogs version: 1.41.10-sun2
535 * The async journal commit feature (bug 19128) and the cancel
536 lock before replay feature (bug 16774) are disabled by default.
540 Description: open-unlinked directories trigger MDS LBUG
541 Details : Fix regression introduced by the patch from bug 19640.
542 ext3_inc_count() can reset nlink to 1 when the directory
543 is indexed and inode->i_nlink == 2. Work around the problem
544 by incrementing nlink by 2 instead of 1.
548 Description: Reconnects are not throttled
549 Details : Don't wake up pinger on reconnect failures and rely on regular
550 pings to trigger the next reconnection. Please note that the
551 pinger already uses a smaller interval if the import is
555 Frequency : only with NFS export
557 Description: Console flooded with error message from ll_inode_from_lock()
559 Details : in mds_open, initialize the child_res_id before enqueuing
560 the OPEN lock for the child inode, then to avoid senting
561 wrong ldlm_res_id to client.
565 Description: allow multiple instances of the same nid in NID hash
566 Details : Case of multiple separate clients from the same NID (as
567 with liblustre) is legitimate and so we should allow
568 multiple instances of the same NID in nid hash.
572 Description: the readahead code can sleep on a semaphore while holding a
574 Details : in ras_update, "lov_get_info" could be called during increasing
575 readahead windows, which tries to get the mutex lock "lov_lock"
576 while holding the spin_lock "ras_lock", then causes system
581 Description: ASSERTION(cli->cl_avail_grant >= 0) failed
582 Details : fix assertion failure in the grant code.
586 Description: Use CNETERR (which is rate limited) in specific places in
587 the portal's LNET driver to avoid flooding the console.
591 Description: include last created object in precreate slow case
595 Description: don't do rep-ack if not created anything
596 Details : mds_open currently always put a lock into a rep-ack regardless
597 if something was created or not. This is pointless and only
598 creates needless contention. In fact the entire idea was to do
599 this for real creates as a recovery protection.
603 Description: Spurious error messages from smp_processor_id() on preemptible
605 Details : Disable a preemption by grabbing the lock in fs_trace_get_tcd()
606 first. The function fs_trace_get_tcd() was moved up.
610 Description: interval_erase() fix
611 Details : interval_erase() calls update_maxhigh() properly when child
616 Description: "lfs df" does not print stats for all mountpoints
617 Details : Print all mounted lustre filesystems with "lfs df"
621 Description: lfs setstripe -p not longer work with a relative pathname
622 Details : Use realpath() to provide absolute pathname.
626 Description: fix for truncated reply buffer
627 Details : reply buffer could be referred by reply_in_callback after released
632 Description: lustre.lov error when backing up symlinks with extended attributes
633 Details : Improved logic in ll_listxattr()
637 Description: properly handle null value for setattr -n lustre.lov
638 Details : Running "setfattr -n trusted.lov ." causes a NULL dereference
639 in ll_setxattr() due to no checking if "value" is NULL.
640 This command now resets to the default striping when executed
645 Description: stack overflow on lock cancellation due to fsync call
646 Details : sync_on_lock_cancel is needed for recovery when async journal
647 is enabled, but we actually just need to make sure that
648 metadata blocks have hit the journal, so doing a fs sync
649 should be enough and should consume less stack (just create an
650 empty handle and commmit it).
654 Description: using current->journal_info to store per-thread data leads
655 to problem under memory pressure
656 Details : disable the per-thread data (current->journal_info) containing
657 the lock info during I/O to work around the issue for short tem
661 Description: control DCACHE_LUSTRE_INVALID flag with MDS_INODELOCK_LOOKUP lock
662 Details : DCACHE_LUSTRE_INVALID is controlled by MDS_INODELOCK_LOOKUP
663 lock which is corresponding to "IT_LOOKUP", do not skip invalidate
668 Description: Cannot send after transport shutdown
669 Details : Clear imp_vbr_failed flag upon eviction
673 Description: soft lock in request set code during recovery
674 Details : during recovery, uses req->rq_set itself to replay the request
675 instead of ptlrpcd_recovery_pc
679 Description: Use CFS_ALLOC_IO instead of _STD in llap_from_page_with_lockh
680 Details : During an ll_readahead under ll_readpage, we have seen the the
681 OBD_SLAB_ALLOC hang under ldlm_pools_shrink when trying to lock
682 a page that is already locked by the readahead code.
687 Description: stop waitting for next replay transno if shutdown
688 Details : if the system is shutting down, wake up service thread blocked
689 to wait for next replay transno during recovery, then all the
690 references held by queued requests can be dropped and device
695 Description: lov_merge_lvb()) ASSERTION(spin_is_locked(&lsm->lsm_lock)) failed
696 Details : Protect lli->lli_smd pointer updates with lli->lli_lock.
700 Description: per-nid stats should not access lustre-hash internal structures
705 Description: mount.lustre fails to pass some options to mount()
710 Description: ext4 extent allocation is slower than in ext3
711 Details : Increase the default value of MB_DEFAULT_ORDER2_REQS to 8,
712 enlarge ext4 preallocation table for 2048 4K blocks extents
717 Description: incorrect triggering of synchronous IO
718 Details : The OSC can mistakenly fall back to synchronous IO when the
719 max_dirty_mb limit is reached and no write requests have yet
720 been issued. This can occur when the dirty pages are spread
721 over many files all of which are below the optimal request size.
725 Description: Optimize quota_ctl operations by sending requests in parallel
726 Details : Send MDS->OST quota_ctl requests in parallel, do not resend.
727 Compiled from two attachments in the ticket.
731 Description: ordering issue between transaction start & i_mutex
732 Details : start the transaction earlier in llog_lvfs_destroy to get
733 transaction start and inode mutex lock nested properly.
737 Description: lru resize SLV can get stuck
738 Details : calculate SLV with a greater precision to not lose small
739 changes due to interger math truncation; round up SLV only
740 if the amount of granted locks less than the limit to not
741 get stuck with this SLV
745 Description: avoid divide-by-zero in lprocfs_rd_import()
749 Description: lfs quota failed when OSTs are down
750 Details : really return approximate block/inode usage when OSTs are down
754 Description: llobdstat fix and enhancement
755 Details : add a counter to set a limit to how many samples will be returned
756 fix a wildcard in the path to limit to obdfilter stats only
760 Description: recovery-small 51 hang - MDT can not stop, Mount still busy with
762 Details : abort lock enqueue processing sooner when we umount is in progress.
764 Severity : enhancement
766 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and OEL5.4 kernel to
767 2.6.18-164.11.1.0.1.el5.
769 Severity : enhancement
771 Description: error message improvements
772 Details : Use INFO/WARN instead of WARN/ERROR for the slow messages.
773 Simplify MDT/OST service start message.
774 Suppress "changing the import ..." warning.
777 Bugzilla : 21961/17914
778 Description: ignore trailing -mdc when determining index number
781 Frequency : only with SLES11
783 Description: fix for a race condition in linux quotas implementation
784 Details : dq_flags(struct dquot) access is not properly locked which could
785 lead to certain inconsistencies when accessing it using non-atomic
786 bit operations like __set_bit in do_set_dqblk.
790 Description: return any pending async errors in close(2) using flush callback
792 -------------------------------------------------------------------------------
794 2010-01-29 Sun Microsystems, Inc.
796 * Support for kernels:
797 2.6.16.60-0.42.8 (SLES 10),
798 2.6.27.39-0.3.1 (SLES11),
799 2.6.18-164.11.1.el5 (RHEL 5)
800 2.6.18-164.6.1.0.1.el5 (OEL 5)
801 * Client support for unpatched kernels:
802 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
803 2.6.16 - 2.6.30 vanilla (kernel.org)
804 * Recommended e2fsprogs version: 1.41.6.sun1
805 * The async journal commit feature (bug 19128) and the cancel
806 lock before replay feature (bug 16774) are disabled by default.
808 Severity : enhancement
810 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and
811 OEL5.4 kernel to 2.6.18-164.11.1.0.1.el5.
813 Severity : enhancement
814 Bugzilla : 21511/19848
815 Description: Update SLES11 kernel to 2.6.27.39-0.3.1.
817 Severity : enhancement
819 Description: Update supported SLES10 kernel to 2.6.16.60-0.42.8.
821 Severity : enhancement
823 Description: Update kernel to RHEL5.4 2.6.18-164.6.1.el5 and
824 OEL5 2.6.18-164.6.1.0.1.el5(Both in-kernel OFED enabled).
826 Severity : enhancement
828 Description: Build kernels (RHEL5, OEL5 and SLES10/11) using the vendor's own
831 Severity : enhancement
833 Description: Vanilla kernel 2.6.30 patchless client support.
838 Description: bad entry in directory xxx: inode out of bounds
839 Details : fix locking issue in the rename path which could race with any
840 other operations updating the same directory.
842 Severity : enhancement
844 Description: Make watchdog timer messages to be more clear and descriptive.
848 Description: cp -p command does not preserve the dates and timestamp
849 Details : mtime could be spoiled by a write callback
853 Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
857 Description: Allow non-root access for "lfs check".
858 Details : Added a check in obd_class_ioctl() for OBD_IOC_PING_TARGET.
860 Severity : enhancement
862 Description: quotacheck performance/scaling issues
863 Details : reduce quotacheck time on empty filesystem by skipping uninit
866 Severity : enhancement
868 Description: Enhancement for lfs(1) command to use numeric uid/gid.
870 Severity : enhancement
872 Description: Adjust locks' extents on their first enqueue, so that at the time
873 they get granted, there is no need for another pass through the
874 queues since they are already shaped into the proper forms.
878 Description: Fix mds_shrink_intent_reply()/mds_intent_policy() to pass correct
879 arguments and prevent LBUG() in lustre_shrink_reply_v2().
883 Description: Change tunefs.lustre and mkfs.lustre --mountfsoptions so that
884 exactly the specified mount options are used. Leaving off
885 any "mandatory" mount options is an error. Leaving off any
886 default mount options causes a warning, but is allowed.
887 Change errors=remount-ro from mandatory to default. Sanitize
888 the mount string before storing it. Update man pages accordingly.
892 Description: mds_getattr() should return 0, even if mds_fid2entry() fails with
893 -ENOENT. Also fix in ptlrpc_expire_one_request() to print signed
896 Severity : enhancement
898 Description: Remove set_info(KEY_UNLINKED) from MDS/OSC
900 Severity : enhancement
902 Description: Clients can replay thousands of unused locks during recovery
903 Details : Don't replay unused locks (only read locks for now) during
904 recovery. This feature is disabled by default and can be
905 enabled by running the following command on the clients:
906 lctl get_param ldlm.cancel_unused_locks_before_replay
910 Description: can't stat file in some situation.
911 Details : improve initialize osc date when target is added to mds and
912 ability to resend too big getattr request is client isn't have info
917 Description: Prevent inconsistences between linux and lustre mount structures.
918 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
919 Make the sleep interruptible.
921 Severity : enhancement
923 Description: Communicate OST degraded/readonly state via statfs to MDS
924 Details : Flags in the statfs returned from OSTs indicate whether the
925 OST is in a degraded RAID state, or if the filesystem has
926 turned read-only after a filesystem error is detected.
931 Description: don't panic if EPROTO was hit when reading symlink
932 Details : correctly handling request reference in error cases.
937 Description: open sometimes returns ENOENT instead of EACCES
938 Details : checking permission should be part of open part of mds_open, not
939 lookup part. so server should be set DISP_OPEN_OPEN disposition
940 before starting permission check.
941 Also not need revalidate dentry if client already have LOOKUP lock.
945 Description: enable client interface failover
946 Frequency : on servers with multiple network interfaces
947 Details : When a child reconnects from another NID, properly update export
948 nid hash position and ldlm reverse import.
950 Severity : enhancement
952 Description: implemented direct I/O with arbitrary (nonaligned) memory
953 addresses and file offsets.
955 Severity : enhancement
957 Description: added more recovery timeout options.
959 Severity : enhancement
961 Description: added llapi_file_open, llapi_file_create, llapi_file_get_stripe
966 Description: Avoid deadlock for local client writes
967 Frequency : only on systems with clients writing to an OST on the same node
968 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
969 memory freeing context. This allows OST threads to set the
970 PF_MEMALLOC flag on task structures in order to allocate memory
971 from reserved pools and complete IO.
972 Use GFP_HIGHUSER for OST allocations for non-local client writes,
973 so that the OST threads generate memory pressure and allow
974 inactive pages to be reclaimed.
979 Descriptoin: lock ordering violation between &cli->cl_sem and _lprocfs_lock
980 Details : move ldlm namespace creation in setup phase to avoid grab
981 _lprocfs_lock with cli_sem held.
985 Frequency : only during format of test systems
986 Description: Unable to run several mkfs.lustre on loop devices at the same time
987 Details : mkfs.lustre returns error 256 on the concurrent loop devices
988 formatting. The solution is to proper handle the error.
990 Severity : enhancement
992 Description: implement async create (obd_async_create) method for osc, to avoid
993 too long waiting new ost objects with holding ldlm lock.
997 Frequency : occasionally during network problems
998 Description: client not allowed to reconnect to OST because of active request
999 Details : abort bulk requests received by the OST once the client has timed
1000 out since the client will resend the request anyway. The client
1001 also now retries to reconnect to the same server if a connect
1002 request failed with EBUSY or -EAGAIN.
1005 Frequency : rare, if used wide striped file and one ost in down.
1007 Descriptoin: don't return error if we created a subset of objects for file.
1008 Details : lov_update_create_set() uses set->set_success as index for created
1009 objects, so if some requests failed, they will have hole at end of
1010 array and we can use qos_shrink_lsm for allocate correct lsm.
1014 Description: Slow stale export processing during normal start up
1015 Details : The global mgc lock prevents OST setup to be run in parallel.
1016 Replace the global lock with a per-config_llog_data semaphore.
1020 Description: Out or order replies might be lost on replay
1021 Details : In ptlrpc_retain_replayable_request if we cannot find retained
1022 request with tid smaller then one currently being added, add it
1023 to the start, not end of the list.
1027 Description: BUG: soft lockup - CPU#1 stuck for 10s! [ll_mdt_07:4523]
1028 Details : add cond_resched() calls to avoid hogging the cpu for too long
1029 in the hash code. Make also lustre_hash_for_each_empty() more
1032 Severity : enhancement
1034 Description: Performance improvements for debug messages with D_RPCTRACE,
1035 D_LDLM, D_QUOTA options.
1038 Frequency : only with NFS export
1040 Description: (lov_merge.c:74:lov_merge_lvb())
1041 ASSERTION(spin_is_locked(&lsm->lsm_lock)) failed (SR 71691004)
1042 Details : Fix a race in the nfs export code by populating inode
1043 info while the new inode is still locked
1045 Severity : enhancement
1047 Description: add a new file in procfs called force_lbug. Writting to this
1048 file triggers a LBUG. Only for test purpose.
1052 Description: OOM killer causes node hang
1053 Details : really interrupt the sleep in osc_enter_cache on signals
1057 Description: LustreError: 9153:0:(quota_context.c:622:dqacq_completion()) LBUG
1058 Details : fix race during quota release on the slave.
1060 Severity : enhancement
1062 Description: smaller hash bucket sizes, cleanups
1063 Details : increase hash table sizes and enabled rehashing for pools, quota,
1064 uuid, nid & per-nid stats.
1066 Severity : enhancement
1068 Description: Add ldiskfs maxdirsize mount option
1069 Details : add max_dir size mount option
1073 Description: panic in ll_statahead_thread
1074 Details : prevent parent thread to be killed before its child
1077 Frequency : only with 16TB device
1079 Description: unable to perform "mount -t lustre" of 16TB OST device
1080 Details : Mounting 16TB LUNs failed due to three bugs in mkfs.lustre.
1084 Description: ASSERTION(atomic_read(&imp->imp_inflight) == 0) failed
1085 Details : unregistering should be zero if no RPC inflight.
1089 Description: hyperion: Oops during metabench
1090 Details : Correct the refcount of lov_request_set
1092 Severity : enhancement
1094 Description: Add mptlinux and nxge drivers to Lustre builds
1096 Severity : enhancement
1098 Description: Fix watchdog timer message to be more clear
1099 Details : Make watchdog timer messages more clear and descriptive.
1103 Description: LNET soft lockups in socknal_cd thread
1104 Details : don't hog CPU for active-connecting if another connd is
1105 accepting connecting-requst from the same peer
1109 Description: recovery-small test_17 hang
1110 Details : Land several AT improvements & fixes.
1114 Description: MDS panic and hanging client processes
1115 Details : Replace exp_ops_stats with exp_nid_stats->nid_stats
1119 Description: OSS stuck in recovery.
1120 Details : fix race during recovery. class_unlink_export,
1121 class_set_export_delayed and target_queue_last_replay_reply
1122 may race while increasing/decreasing obd_recoverable_clients
1123 and obd_delayed_clients, causing recovery to wait forever.
1125 Severity : enhancement
1127 Description: add cascading_rw.c to lustre/tests
1131 Description: filter_last_id() NULL deref
1132 Details : lprocfs_filter_rd_last_id() should check for the fully
1133 setup obd device, before proceeding further.
1135 Severity : enhancement
1137 Description: Loadgen improvements
1138 Details : stacksize and locking fixes for loadgen
1142 Description: Quiet CERROR("dirty %d > system dirty_max %d\n"
1143 Details : The atomic_read() allowing the atomic_inc() are not covered
1144 by a lock. Thus they may safely race and trip this CERROR()
1145 unless we add in a small fudge factor (+1).
1147 Severity : enhancement
1149 Description: shrink_slab: nr=-9223362083340912175
1150 Details : fix spurious message from shrink_slab reporing negative nr
1154 Description: Quiet bogus previously committed transno error
1155 Details : suppress the "server went back in time" error message which
1156 is always printed even in the common case after a client eviction
1158 Severity : enhancement
1160 Description: Parallel statfs() calls result in client eviction
1161 Details : cache statfs data for 1s.
1165 Description: parallel-scale test_compilebench: @@@@@@ FAIL: compilebench
1167 Details : fix serveral issues in pinger code causing clients not to ping
1168 servers for too long, resulting in evictions.
1172 Description: e2fsck should warn when MMP update interval is extended
1173 Details : print mmp_check_interval and make it possible to abort mount
1174 operation in case it takes too long.
1178 Description: mdsrate-create-large.sh, BUG: soft lockup - CPU#0 stuck for 10s!
1179 Details : fix bug in the RHEL5's jbd2 callback patch.
1183 Description: drop number of active requests when queued for recovery
1184 Details : Now that we take a reference on the original request instead of
1185 making a copy of it for recovery. We need to drop the number of
1186 active requests or the queued requests will prevent all request
1187 processing when they exceed (srv->srv_threads_running - 1).
1189 Severity : enhancement
1191 Description: refuse to invalidate operational quota files when they are in use
1192 Details : an attempt to invalidate operational quota files on the quota
1193 master is not actually permitted by VFS (returning -EPERM), but we
1194 should not depend on that and should return the error earlier.
1198 Description: Applications stuck in jbd2_log_wait_commit during exit
1199 Details : fix deadlock between kjournald2 trying to acquire the page lock
1200 owned by an ost_io thread waiting for journal commit.
1202 -------------------------------------------------------------------------------
1204 2009-10-16 Sun Microsystems, Inc.
1206 * Support for kernels:
1207 2.6.16.60-0.42.4 (SLES 10),
1208 2.6.27.29-0.1 (SLES11, i686 & x84_64 only),
1209 2.6.18-128.7.1.el5 (RHEL 5),
1210 2.6.18-128.7.1.el5 (OEL 5),
1211 * Client support for unpatched kernels:
1212 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1213 2.6.16 - 2.6.27 vanilla (kernel.org)
1214 * Recommended e2fsprogs version: 1.41.6.sun1
1215 * ext4 support for RHEL5 is experimental and thus should not be
1218 Severity : enhancement
1220 Description: Add OEL5 support.
1222 Severity : enhancement
1224 Description: Update kernel to SLES11 2.6.27.29-0.1.
1228 Description: File checksum failures with OST read cache on
1229 Details : Disable page poisoning when the bulk transfer has to be aborted
1230 because the client got evicted.
1234 Description: Don't allow make backward step on assiging osc next id.
1235 Details : race between allocation next id and ll_sync thread can be cause
1236 of set wrong osc next id and can be kill valid ost objects.
1238 Severity : enhancement
1240 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
1242 Severity : enhancement
1244 Description: Update kernel to SLES10 SP2 2.6.16.60-0.42.4.
1248 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
1252 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
1256 Description: Error handling in osc_statfs_interpret() has been improved.
1257 Details : Check in osc_statfs_interpret() for EBADR.
1261 Description: Do not send statfs() requests to OSTs disabled by administrator.
1262 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
1266 Description: Do not update ctime for the deleted inode.
1267 Details : Check in mds_reint_unlink() before calling fsfilt_setattr().
1271 Description: Increase of the size of the LDLM resource hash.
1272 Details : Bump up RES_HASH_BITS=12.
1276 Description: correctly send lsm on open replay
1277 Details : MDS is trust to LSM size on replay open, but client can set wrong size
1282 Description: Deadlock between filter_destroy() and filter_commitrw_write().
1283 Details : filter_destroy() does not hold the DLM lock over the whole
1284 operation. If the DLM lock is dropped, filter_commitrw() can go
1285 through, causing the deadlock between page lock and i_mutex.
1286 The i_alloc_sem should also be hold in filter_destroy() while
1287 truncating the file.
1291 Description: truncate starts GFP_FS allocation under transaction causing deadlock
1292 Details : ldiskfs_truncate calls grab_cache_page which may start page
1293 allocation under an open transaction. This may lead to
1294 calling prune_icache with consequent lustre reentrance.
1297 Frequency : only when down/upgrading the MDS to 1.6/1.8 while 1.8 clients are
1298 still up and when the OST pool feature is used
1300 Description: interop testing got LBUG when run dd with OST pool
1301 :LustreError: 30032:0:(llite_lib.c:1913:ll_replace_lsm()) LBUG
1302 Details : down/upgrading the MDS to a version that doesn't/does support OST
1303 pool can cause clients to crash because the lsm has changed
1308 Description: missing tree_status on 1.8.1 RPM build
1309 Details : make rpms failed due because the tree_status file is missing.
1313 Description: continuing LustreError "mds adjust qunit failed!"
1314 Details : don't print message on the console when ->adjust_qunit fails.
1318 Description: don't increase ldlm timeout if previous client was evicted
1319 Details : if a client doesn't respond to a blocking callback within the
1320 adaptive ldlm enqueue timeout, don't adjust the adaptive estimate
1321 when the lock is next granted.
1325 Description: ost is being unmounted w/o all writes to last_rcvd landing on disk.
1326 affects recovery negatively.
1327 Details : make sure all exports have been properly destroyed by the zombie
1328 thread processed before stopping the target.
1332 Description: Performance degradation with O_DIRECT between 1.6 & 1.8.1 b190
1333 Details : disable write barrier for ext4/SLES11.
1337 Description: Kernel panic - not syncing: Out of memory and no killable
1338 processes... on OSS when iozone
1339 Details : fix memory leak in the journal checksum patch.
1343 Description: group quota "too many blocks" OSS crashes
1344 Details : we should keep the same uid/gid for lquota_chkquota() and
1345 lquota_pending_commit()
1349 Description: LustreError: 9153:0:(quota_context.c:622:dqacq_completion()) LBUG
1350 Details : don't LBUG on release quota error. Just a workaround until the
1351 problem is understood.
1353 ------------------------------------------------------------------------------
1354 2009-07-31 Sun Microsystems, Inc.
1356 * Support for kernels:
1357 2.6.16.60-0.39.3 (SLES 10),
1358 2.6.27.23-0.1 (SLES11, i686 & x84_64 only),
1359 2.6.18-128.1.14.el5 (RHEL 5),
1360 * Client support for unpatched kernels:
1361 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1362 2.6.16 - 2.6.27 vanilla (kernel.org)
1363 * Recommended e2fsprogs version: 1.41.6.sun1
1364 * File join has been disabled in this release, refer to Bugzilla 16929.
1365 * NFS export disabled when stack size < 8192. Since the NFSv4 export
1366 of Lustre filesystem with 4K stack may cause a stack overflow. For
1367 more information, please refer to bugzilla 17630.
1368 * ext4 support for RHEL5 is experimental and thus should not be
1371 Severity : enhancement
1373 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
1376 Frequency : with 1.8 server and 1.6 clients
1378 Description: correctly shrink reply for avoid send too big message to client.
1379 Details : 1.8 mds is allocate to big buffer to LOV EA data and this produce
1380 some problems with sending this reply to 1.6 client.
1384 Description: Repeated atomic allocation failures.
1385 Details : Use GFP_HIGHUSER | __GFP_NOMEMALLOC flags for memory allocations
1386 to generate memory pressure and allow reclaiming of inactive pages.
1387 At the same time, do not allow to exhaust emergency pools.
1388 For local clients the use of GFP_NOFS will be introduced in 1.8.2
1390 Severity : enhancement
1391 Bugzilla : 19846, 18289
1392 Description: Update kernel to RHEL5 2.6.18-128.1.14.el5.
1394 Severity : enhancement
1395 Bugzilla : 19625, 16893, 18668, 19848
1396 Description: Add support for SLES11 2.6.27.23-0.1.
1398 Severity : enhancement
1400 Description: Update client support to vanila kernels up to 2.6.27.
1402 Severity : enhancement
1404 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
1406 Severity : enhancement
1408 Description: Compile with -Werror by default for i686 and x86_64.
1412 Description: resolve race between obd_disconnect and class_disconnect_exports
1413 Details : if obd_disconnect will be called to already disconnected export he
1414 forget release one reference and osc module can't unloaded.
1416 Severity : enhancement
1418 Description: move AT tunable parameters for more consistent usage
1419 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
1423 Descriptoin: correctly skip time estimate if in recovery
1424 Details : rq_send_state insn't bitmask so using bitwise ops is forbid.
1428 Descriptoin: OSS DeadLock
1429 Details : Use trylock to prevent deadlock when shrink icache.
1431 Severity : enhancement
1433 Description: Allow tuning service thread via /proc
1434 Details : For each service a new
1435 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
1436 created that can be used to set min/max thread counts, and get the
1437 current number of running threads.
1439 Severity : enhancement
1441 Description: Add state history info file, enhance import info file
1442 Details : Track import connection state changes in a new osc/mdc proc file;
1443 add overview-type data to the osc/mdc import proc file.
1447 Description: Reduce small size read RPC
1448 Details : Set read-ahead limite for every file and only do read-ahead when
1449 available read-ahead pages are bigger than 1M to avoid small size
1454 Description: free_entry erroneously used groups_free instead of put_group_info
1456 Severity : enhancement
1458 Description: Make read-ahead stripe size aligned.
1460 Severity : enhancement
1462 Description: MDS create should not wait for statfs RPC while holding DLM lock.
1465 Frequency : rare, connect and disconnect target at same time
1467 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
1468 Details : don't call obd_disconnect under lov_lock. this long time
1469 operation and can block ptlrpcd which answer to connect request.
1472 Frequency : start MDS on uncleanly shutdowned MDS device
1474 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
1475 Details : stay in waiting mds<>ost recovery finished produce random bugs
1476 due race between two ll_sync thread for one lov target. send
1477 ACTIVATE event only if connect realy finished and import have
1481 Frequency : start MDS on uncleanly shutdowned MDS device
1483 Descriptoin: aborting recovery hang on MDS
1484 Details : don't throttle destroy RPCs for the MDT.
1488 Description: Slow reads beyond 8Tb offsets.
1489 Details : Page index integer overflow in ll_read_ahead_page
1493 Description: Soft lockup on OSS after MDS failover
1494 Details : MSG_CONNECT_INITIAL is not set on the initial MDS->OST connect.
1495 As a conseqence, the patch from bug 18224 is not operational
1496 and the MDS export cannot be reused on the OSTs until it gets
1500 Frequency : rare, only if using MMP with Linux RAID
1502 Description: MMP doesn't work with Linux RAID
1503 Details : While using HA for Lustre servers with Linux RAID, it is possible
1504 that MMP will not detect multiple mounts. To make this work we
1505 need to unplug the device queue in RAID when the MMP block is being
1506 written. Also while reading the MMP block, we should read it from
1507 disk and not the cached one.
1510 Frequency : rare, during recovery
1512 Description: Assertion failure in ldlm_lock_put
1513 Details : Do not put cancelled locks into replay list, hold references on
1514 locks in replay list
1518 Description: 1.6.5 mdsrate performance is slower than 1.4.11/12 (MDS is not cpu bound!)
1519 Details : create_count always drops to the min value (=32) because grow_count
1520 is being changed before the precreate RPC completes.
1523 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
1526 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
1527 Details : add proper locking for creation of jbd_4k slab cache
1531 Description: MMP check in ext3_remount() fails without displaying any error
1532 Details : When multiple mount protection fails during remount, proper error
1537 Description: Rare Client crash on resend if the file was deleted.
1538 Details : When file is opened, but open reply is lost and file is
1539 subsequently deleted before resend, resend processing logic
1540 breaks trying to open the file again, should not try to open.
1544 Description: add check for >8TB ldiskfs filesystems
1545 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
1546 Don't allow >8TB ldiskfs filesystems to be mounted without
1547 force_over_8tb mount option
1551 Description: Client locked up when running multiple instances of an app. on
1552 multiple mount points
1553 Details : ll_shrink_cache() can sleep while holding the ll_sb_lock.
1554 Convert ll_sb_lock to a read/write semaphore to fix the problem.
1558 Description: Cannot acces an NFS-mounted Lustre filesystem
1559 Details : An NFS client cannot access the Lustre filesystem NFS-mounted
1560 from a Lustre-client exporting the Lustre filesystem via NFS.
1564 Description: panic in ll_statahead_thread
1565 Details : grab dentry reference in parent process.
1568 -------------------------------------------------------------------------------
1570 tbd Sun Microsystems, Inc.
1572 * Support for kernels:
1573 2.6.16.60-0.31 (SLES 10),
1574 2.6.18-128.1.6.el5 (RHEL 5),
1575 2.6.22.14 vanilla (kernel.org)
1576 * Client support for unpatched kernels:
1577 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1578 2.6.16 - 2.6.22 vanilla (kernel.org)
1579 * Recommended e2fsprogs version: 1.40.11-sun1
1580 * File join has been disabled, refer to Bugzilla 16929.
1581 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
1582 * NFS export disabled when stack size < 8192. Since the NFSv4 export of
1583 Lustre filesystem with 4K stack may cause a stack overflow. For more
1584 information, please refer to bugzilla 17630.
1586 Severity : enhancement
1588 Description: Update to RHEL5.3 kernel-2.6.18-128.1.6.el5.
1590 Severity : enhancement
1592 Description: Update OFED release to 1.4.1 RC4
1594 Severity : major, only with big OST
1596 Description: Very poor metadata performance on Infiniband lustre configuration
1597 Details : OST object precreation becomes very slow on big OSTs. This is due
1598 to the ialloc patch spending too much time scanning groups.
1601 Frequency : during recovery
1603 Description: don't mix llog inodes with normal.
1604 Details : allocate inodes for log in last inode group
1609 Description: fix lqs' reference which won't be put in some situations
1610 Details : This patch fixes:
1611 1. In quota_check_common(), this function will check quota
1612 for user and group, but only send one return via "pending".
1613 In most cases, the pendings should be same. But that is not
1615 2. If quotaoff runs between lquota_chkquota() and
1616 lquota_pending_commit(), the same thing will happen too.
1617 That is why it comes:
1618 - if (!ll_sb_any_quota_active(qctxt->lqc_sb))
1621 -------------------------------------------------------------------------------
1623 2008-12-31 Sun Microsystems, Inc.
1625 * Support for kernels:
1626 2.6.5-7.314 (SLES 9),
1627 2.6.9-67.0.22.EL (RHEL 4),
1628 2.6.16.60-0.31 (SLES 10),
1629 2.6.18-92.1.17.el5 (RHEL 5),
1630 2.6.22.14 vanilla (kernel.org)
1631 * Client support for unpatched kernels:
1632 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1633 2.6.16 - 2.6.22 vanilla (kernel.org)
1634 * Client support for unpatched kernels:
1635 we do not recommend using patchless RHEL4 clients with kernels
1636 prior to 2.6.9-55EL (RHEL4U5).
1637 * Recommended e2fsprogs version: 1.40.11-sun1
1638 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1639 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
1640 removed cwd "./" (refer to Bugzilla 14399).
1641 * A new quota file format has been introduced in 1.6.5.
1642 The format conversion from prior releases is handled transparently,
1643 but releases older than 1.4.12/1.6.5 don't understand this new
1644 format. The automatic format conversion can be avoided by running
1645 the following command on the MDS:
1646 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
1647 For more information, please refer to bugzilla 13904.
1648 * A new quota file format was introduced in 1.6.6/1.8.0 (kernels 2.6.16+).
1649 The format conversion from prior releases is handled transparently,
1650 but releases older than 1.6.6/1.8.0 don't understand this new
1651 format. The automatic format conversion can be avoided by running
1652 the following commands on the MDS and OSS servers (for
1653 pre 1.4.12-1.6.5 quota files):
1654 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV',
1655 'tunefs.lustre --param="ost.quota_type=ug1" $MDTDEV'
1656 or (for 1.4.12/1.6.5 quota files)
1657 'tunefs.lustre --param="mdt.quota_type=ug2" $MDTDEV',
1658 'tunefs.lustre --param="ost.quota_type=ug2" $MDTDEV'
1659 For more information, please refer to bugzilla 13904.
1660 * Output of lfs quota has been made less detailed by default,
1661 old (verbose) output can be obtained by using -v option.
1662 * File join has been disabled in this release, refer to Bugzilla 16929.
1663 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
1664 * NFS export disabled when stack size < 8192. Since the NFSv4 export of
1665 Lustre filesystem with 4K stack may cause a stack overflow. For more
1666 information, please refer to bugzilla 17630.
1668 Severity : enhancement
1670 Description: Caching OSS
1671 Details : introduce data caching on the OSS. The OSS now relies on the linux
1672 kernel page cache to keep recently accessed data in memory.
1673 It is worth noting that all write requests are still flushed
1674 synchronously as in lustre 1.6.
1676 Severity : enhancement
1678 Description: Version based recovery
1679 Details : introduce finer grained recovery able to detect transaction
1680 dependencies and can deal with transaction gaps caused by clients
1681 failing at the same time as the server.
1683 Severity : enhancement
1685 Description: Enable adaptive timeouts by default
1686 Details : The Lustre timeout value in /proc/sys/lustre/timeout is now
1687 managed dynamically based on server load and should not need
1688 to be tuned manually based on cluster size. This allows Lustre
1689 to work under a wider variety of system sizes and loads, without
1690 unnecessarily causing lengthy recovery times.
1692 Severity : enhancement
1694 Description: Add OST Pools support
1695 Details : File striping can now be set to use an arbitrary pool of OSTs
1697 Severity : enhancement
1699 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
1700 Details : allow skip disconnected ost for send statfs request and hide error
1704 Frequency : rare, on llog test 6
1706 Descriptoin: don't allow connect to already connected import
1707 Details : allowing connect to already connected import is hide connecting problem.
1710 Frequency : rare, on failed llog setup
1712 Descriptoin: don't leak obd reference on failed llog setup
1713 Details : for failed llog setup - mgc forget call class_destroy_import for
1714 client import, move destroy import to more generic place.
1719 Descriptoin: allow kill process which wait statahead result
1720 Details : for some reasons 'ls' can stick in waiting result from statahead,
1721 in this case need way for kill this process.
1724 Frequency : rare, at shutdown
1726 Descriptoin: panic at umount
1727 Details : llap_shrinker can be raced with killing super block from list and
1728 this produce panic with access to already freeded pointer
1733 Descriptoin: don't lose wakeup for imp_recovery_waitq
1734 Details : recover_import_no_retry or invalidate_import and import_close can
1735 both sleep on imp_recovery_waitq, but we was send only one wakeup
1741 Descriptoin: panic in mds_open
1742 Details : don't confuse mds_finish_transno() with PTR_ERR(-ENOENT)
1747 Descriptoin: stuck in cache_remove_extent() or panic with accessing to already
1749 Details : release lock refernce only after add page to pages list.
1752 Frequency : always with long access acl
1754 Descriptoin: mds can't pack reply with long acl.
1755 Details : mds don't control size of acl but they limited by reint/getattr
1760 Frequency : on remount
1762 Description: external journal device not working after the remount
1763 Details : clear dev_rdonly flag for external journal devices in
1769 Description: shutdown vs evict race
1770 Details : client_disconnect_export vs connect request race.
1771 if client will evicted at this time - we start invalidate
1772 thread without referece to import and import can be freed
1778 Description: shrink LOV EAs before replying
1779 Details : correctly adjust LOV EA buffer for reply.
1784 Description: don't skip ost target if they assigned to file
1785 Details : Drop slow OSCs if we can, but not for requested start idx.
1786 This means "if OSC is slow and it is not the requested
1787 start OST, then it can be skipped, otherwise skip it only
1788 if it is inactive/recovering/out-of-space.
1790 Severity : enhancement
1792 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
1795 Frequency : rare, need acl's on inode.
1797 Description: client can't handle ost additional correctly
1798 Details : if ost was added after client connected to mds client can have
1799 hit lnet_try_match_md ... to big messages to wide striped files.
1800 in this case need teach client to handle config events about add
1801 lov target and update client max ea size at that event.
1803 Severity : enhancement
1805 Description: Update to sles9 kernel-2.6.5-7.314.
1807 Severity : enhancement
1809 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
1812 Frequency : Create a symlink file with a very long name
1814 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
1815 Details : If there is no extra space in the request for early cancels,
1816 ldlm_req_handles_avail() returns 0 instead of a negative value.
1821 Description: mds is deadlocked
1822 Details : in rare cases, inode in catalog can have i_no less than have parent
1823 i_no, this produce wrong order for locking during open, and
1824 parallel unlink can be lock open. this need teach mds_open to grab
1825 locks in resource id order, not at parent -> child order.
1827 Severity : enhancement
1829 Description: Add /proc entry for import status
1830 Details : The mdc, osc, and mgc import directories now have
1831 an import directory that contains useful import data for debugging
1832 connection problems.
1834 Severity : enhancement
1836 Description: Re-disable certain /proc logging
1837 Details : Enable and disable client's offset_stats, extents_stats and
1838 extents_stats_per_process stats logging on the fly.
1841 Frequency : Only on FC kernels 2.6.22+
1843 Description: oops in statahead
1844 Details : Do not drop reference count for the dentry from VFS when lookup,
1845 VFS will do that by itself.
1847 Severity : enhancement
1849 Description: Generic /proc file permissions
1850 Details : Set /Proc file permissions in a more generic way to enable non-
1851 root users operate on some /proc files.
1855 Description: Hitting mdc_commit_close() ASSERTION
1856 Details : Properly handle request reference release in
1857 ll_release_openhandle().
1861 Frequency : only patchless client
1862 Description: add workaround for race between add/remove dentry from hash
1864 Severity : enhancement
1866 Description: Allow OST glimpses to return PW locks
1870 Description: LBUG when llog conf file is full
1871 Details : When llog bitmap is full, ENOSPC should be returned for plain
1876 Description: Prevent import from entering FULL state when server in recovery
1880 Description: service mount cannot take device name with ":"
1881 Details : Only when device name contains ":/" will mount treat it as
1887 Description: replace ptlrpcd with the statahead thread to interpret the async
1888 statahead RPC callback
1892 Frequency : on recovery
1893 Description: I/O failures after umount during fail back
1894 Details : if client reconnected to restarted server we need join to recovery
1895 instead of find server handler is changed and process self eviction
1896 with cancel all locks.
1898 Severity : enhancement
1900 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
1904 Description: Kernel BUG tries to release flock
1905 Details : Lustre does not destroy flock lock before last reference goes
1906 away. So always drop flock locks when client is evicted and
1907 perform unlock regardless of successfulness of speaking to MDS.
1909 Severity : enhancement
1911 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
1913 Severity : enhancement
1915 Description: Upcall on Lustre log has been dumped
1916 Details : Allow for a user mode script to be called once a Lustre log has
1917 been dumped. It passes the filename of the dumped log to the
1918 script, the location of the script can be specified via
1919 /proc/sys/lnet/debug_log_upcall.
1924 Description: avoid messages about idr_remove called for id that is not allocated
1925 Details : Move assigment s_dev for clustered nfs to end of initialization,
1926 for avoid problem with error handling.
1931 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
1932 Details : When connection is reused this not moved from CONN_UNUSED_HASH into
1933 CONN_USED_HASH and this prodice warning when put connection again
1939 Description: avoid ASSERTION(client_stat->nid_exp_ref_count == 0) failed
1940 Details : release reference to stats when client disconnected, not
1941 when export destroyed for avoid races when client destroyed
1942 after main ost export.
1946 Description: more cleanup in mds_lov
1947 Details : add workaround for get valid ost count for avoid warnings about
1948 drop too big messages, not init llog cat under semphore which
1949 can be blocked on reconnect and break normal replay, fix access
1952 Severity : enhancement
1954 Description: Export bytes_read/bytes_write count on OSC/OST.
1958 Description: Early reply size mismatch, MGC loses connection
1959 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
1960 the connect flags are properly negotiated.
1964 Description: Properly propagate oinfo flags from lov to osc for statfs
1965 Details : restore missing copy oi_flags to lov requests.
1969 Description: exports in /proc are broken
1970 Details : recreate /proc entries for clients when they reconnect.
1972 Severity : enhancement
1974 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
1975 : l_getgroups(8), lst(8), routerstat(8)
1976 Details : included man pages for llobdstat(8), llstat(8),
1977 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
1979 Severity : enhancement
1981 Description: Implement lustre ll_show_options method.
1983 Severity : enhancement
1985 Description: Update to SLES9 kernel-2.6.5-7.312.
1987 Severity : enhancement
1989 Description: Update to RHEL4 kernel-2.6.9-67.0.22.EL.
1993 Description: exports in /proc are broken
1994 Details : recreate /proc entries for clients when they reconnect.
1998 Description: don't fail open with -ERANGE
1999 Details : if client connected until mds will be know about real ost count
2000 get LOV EA can be fail because mds not allocate enougth buffer
2005 Description: Resolve device initialization race
2006 Details : Prevent proc handler from accessing devices added to the
2007 obd_devs array but yet be intialized.
2009 Severity : enhancement
2011 Description: configure's --enable-quota should check the
2012 : kernel .config for CONFIG_QUOTA
2013 Details : configure is terminated if --enable-quota is passed but
2014 : no quota support is in kernel
2016 Severity : enhancement
2018 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
2020 Severity : enhancement
2022 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
2026 Frequency : rare, on PPC clients
2027 Description: don't swab ost objects in response about directory, because
2029 Details : bug similar bug 14856, but in different function.
2031 Severity : enhancement
2033 Description: lfs quota tool enhancement
2034 Details : added units specifiers support for setquota, default to
2035 current uid/gid for quota report, short quota stats by
2036 default, nonpositional parameters for setquota, added
2037 llapi_quotactl manual page.
2039 Severity : enhancement
2041 Description: *optional* service tags registration
2042 Details : if the "service tags" package is installed on a Lustre node
2043 When the filesystem is mounted, a local-node service tag will
2044 be created. See http://inventory.sun.com/ for more information
2045 about the Service Tags asset management system.
2049 Description: Client runs out of low memory
2050 Details : Consider only lowmem when counting initial number of llap pages
2053 Frequency : occasional
2055 Description: add refcount for osc callbacks, so avoid panic on shutdown
2057 Severity : enhancement
2059 Description: Update to RHEL4 kernel-2.6.9-67.0.20.
2062 Frequency : testing only
2064 Description: sanity test 65a fails if stripecount of -1 is set
2065 Details : handle -1 striping on filesystem in ll_dirstripe_verify
2068 Frequency : only in unusual configurations
2070 Description: Kernel panic with find ost index.
2071 Details : lov_obd have panic if some OST's have sparse indexes.
2073 Severity : enhancement
2075 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
2078 Frequency : rarely, if filesystem is mounted with -o flock
2080 Description: do not process already freed flock
2081 Details : flock can possibly be freed by another thread before it reaches
2082 to ldlm_flock_completion_ast.
2085 Frequency : rarely, if filesystem is mounted with -o flock
2087 Description: LBUG during stress test
2088 Details : Need properly lock accesses the flock deadlock detection list.
2091 Frequency : rarely, if binaries are being run from Lustre
2093 Description: oops in page fault handler
2094 Details : kernel page fault handler can return two special 'pages' in
2095 error case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
2098 Frequency : rarely, during shutdown
2100 Description: timeout with invalidate import.
2101 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
2102 should be handled by ptlrpcd. This produce long age waiting and
2103 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
2109 Description: ASSERTION(CheckWriteback(page,cmd)) failed
2110 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
2114 Frequency : only with broken builds/installations
2116 Description: no LBUG if lquota.ko and fsfilt_ldiskfs.ko are different versions
2117 Details : just return an error to a user, put a console error message
2119 Severity : enhancement
2121 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
2123 Severity : enhancement
2125 Description: Update to RHEL4 kernel-2.6.9-67.0.15.
2127 Severity : enhancement
2129 Description: enable MGS and MDT services start separately
2130 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
2131 a co-located MGS without starting the MGS, which is a complement
2132 to 'nosvc' mount option.
2135 Frequency : always, on big-endian systems
2137 Description: cleanup in ptlrpc code, related to PPC platform
2138 Details : store magic in native order avoid panic's in recovery on PPC
2139 node and forbid from this error in future. Also fix posibily
2140 of twice swab data. Fix get lov striping to userpace.
2143 Frequency : rarely, if replay get lost on server
2145 Description: server incorrectly drop resent replays lead to recovery failure.
2146 Details : do not drop replay according to msg flags, instead we check the
2147 per-export recovery request queue for duplication of transno.
2150 Frequency : after recovery
2152 Description: precreate to many object's after del orphan.
2153 Details : del orphan st in oscc last_id == next_id and this triger growing
2154 count of precreated objects. Set flag LOW to skip increase count
2155 of precreated objects.
2158 Frequency : rare, on clear nid stats
2160 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
2161 Details : when clean nid stats sometimes try destroy live entry,
2162 and this produce panic in free.
2165 Frequency : occasionally since 1.6.4
2167 Description: Stack overflow during MDS log replay
2168 Details : ease stack pressure by using a thread dealing llog_process.
2171 Frequency : very rare
2173 Description: MDT cannot be unmounted, reporting "Mount still busy"
2174 Details : Mountpoint references were being leaked during open reply
2175 reconstruction after an MDS restart. Drop mountpoint reference
2176 in reconstruct_open() and free dentry reference also.
2181 Description: wait until IO finished before start new when do lock cancel.
2182 Details : VM protocol want old IO finished before start new, in this case
2183 need wait until PG_writeback is cleared until check dirty flag
2184 and call writepages in lock cancel callback.
2189 Description: mds_mfd_close() ASSERTION(rc == 0)
2190 Details : In mds_mfd_close(), we need protect inode's writecount change
2191 within its orphan write semaphore to prevent possible races.
2194 Frequency : rare, on shutdown ost
2196 Description: don't hit live lock with umount ost.
2197 Details : shrink_dcache_parent can be in long loop with destroy dentries,
2198 use shrink_dcache_sb instead.
2201 Frequency : only when echo_client is used
2203 Description: don't panic with use echo_client
2204 Details : echo client pass NULL as client nid pointer and this produce NULL
2205 pointer dereference.
2208 Frequency : Always on 32-bit PowerPC systems
2210 Description: fix build on PPC32
2211 Details : compile code with -m64 flag produce wrong object file for PPC32.
2216 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
2217 Details : In reconstruct_* functions, LASSERTs on both the data supplied
2218 by a client, and the data on disk are dangerous and incorrect.
2219 Change them with client eviction.
2221 Severity : enhancement
2223 Description: skiplist implementation simplification
2224 Details : skiplists are used to group compatible locks on granted list
2225 that was implemented as tracking first and last lock of each lock
2226 group the patch changes that to using doubly linked lists
2230 Description: delete compatibility for 32bit qdata
2231 Details : as planned, when lustre is beyond b1_8, lquota won't support 32bit
2232 qunit. That means servers of b1_4 and servers of b1_8 can't be
2233 used together if users want to use quota.
2236 Frequency : only with administrator action
2238 Description: mount failure if config log has invalid conf_param setting
2239 Details : If administrator specified an incorrect configuration parameter
2240 with "lctl conf_param" this would cause an error during future
2241 client mounts. Instead, ignore the bad configuration parameter.
2244 Frequency : blocks per group < blocksize*8 and uninit_groups is enabled
2246 Description: ldiskfs error: XXX blocks in bitmap, YYY in gd
2247 Details : If blocks per group is less than blocksize*8, set rest of the
2251 Frequency : Application do stride read on lustre
2253 Description: The read performance will drop a lot if the application does
2255 Details : Because the stride_start_offset are missing in stride read-ahead,
2256 it will cause clients read a lot of unused pages in read-ahead,
2257 then the read-performance drops.
2261 Description: more ldlm soft lockups
2262 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
2263 starve other threads from the resource lock for a long time in
2264 case of long waiting queue, so change the debug level from
2265 D_OTHER to the less frequently used D_INFO.
2267 Severity : enhancement
2269 Description: add -gid, -group, -uid, -user options to lfs find
2271 Severity : enhancement
2273 Description: ll_recover_lost_found_objs - recover objects in lost+found
2274 Details : OST corruption and subsequent e2fsck can leave objects in the
2275 lost+found directory. Using the "ll_recover_lost_found_objs"
2276 tool, these objects can be retrieved and data can be salvaged
2277 by using the object ID saved in the fid EA on each object.
2282 Description: this bug _only_ happens when inode quota limitation is very low
2283 (less than 12), so that inode quota unit is 1 at initialization.
2284 Details : if remaining quota equates 1, it is a sign to demonstate that quota
2285 is effective now. So least quota qunit should be 2.
2289 Description: Hung threads in invalidate_inode_pages2_range
2290 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
2291 one is completed. As a result, some RPCs are stuck in the queue
2296 Description: Procfs and llog threads access destoryed import sometimes.
2297 Details : Sync the import destoryed process with procfs and llog threads by
2298 the import refcount and semaphore.
2302 Description: mds fails to respond, threads stuck in ldlm_completion_ast
2303 Details : Sort source/child resource pair after updating child resource.
2308 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
2309 Details : If insertion of an extent fails, then discard the inode
2310 preallocation and free data blocks else it can lead to duplicate
2315 Description: don't always update ctime in ext3_xattr_set_handle()
2316 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle()
2317 In some cases the ctime should not be updated, for example for
2318 2.0->1.8 compatibility it is necessary to delete an xattr and it
2319 should not update the ctime.
2323 Description: add quota statistics
2324 Details : 1. sort out quota proc entries and proc code.
2325 2. add quota statistics
2330 Description: quotas are not honored with O_DIRECT
2331 Details : all writes with the flag O_DIRECT will use grants which leads to
2332 this problem. Now using OBD_BRW_SYNC to guard this.
2336 Bugzilla : 15713/16362
2337 Description: Assertion in iopen_connect_dentry in 1.6.3
2338 Details : looking up an inode via iopen with the wrong generation number can
2339 populate the dcache with a disconneced dentry while the inode
2340 number is in the process of being reallocated. This causes an
2341 assertion failure in iopen since the inode's dentry list contains
2342 both a connected and disconnected dentry.
2346 Description: assertion failure in ldlm_handle2lock()
2347 Details : fix a race between class_handle_unhash() and class_handle2object()
2348 introduced in lustre 1.6.5 by bug 13622.
2350 Severity : enhancement
2352 Description: superblock lock contention with many SMP cores on one client
2353 Details : several client filesystem locks were highly contended on SMP
2354 NUMA systems with 8 or more cores. Per-CPU datastructures
2355 and more efficient locking implemented to reduce contention.
2360 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
2361 Details : remove the limit of 256 scsi disks in the sd_iostat patch
2366 Description: oops in sd_iostats_seq_show()
2367 Details : unloading/reloading the scsi low level driver triggers a kernel
2368 bug when trying to access the sd iostat file.
2373 Description: Kernel panics during QLogic driver reload
2374 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
2375 patch, causing memory corruption.
2380 Description: journal_dev option does not work in b1_6
2381 Details : pass mount option during pre-mount.
2383 Severity : enhancement
2385 Description: Add a FIEMAP(FIle Extent MAP) ioctl for ldiskfs
2386 Details : FIEMAP ioctl will allow an application to efficiently fetch the
2387 extent information of a file. It can be used to map logical blocks
2388 in a file to physical blocks in the block device.
2391 Frequency : only with adaptive timeout enabled
2393 Description: DEBUG_REQ() bad paging request
2394 Details : ptlrpc_at_recv_early_reply() should not modify req->rq_repmsg
2395 because it can be accessed by reply_in_callback() without the
2399 Frequency : only on Cray X2
2401 Description: X2 build failures
2402 Details : fix build failures on Cray X2.
2406 Description: xid & resent requests
2407 Details : Initialize RPC XID from clock at startup (randomly if clock is
2412 Description: quota recovery deadlock during mds failover
2413 Details : This patch includes att18982, att18236, att18237 in bz14840.
2415 1. fix osts hang when mds does failover with quotaon
2416 2. prevent watchdog storm when osts threads wait for the
2421 Description: kernel panic on racer
2422 Details : Do not access dchild->d_inode when IS_ERR(dchild) is true.
2424 Severity : enhancement
2426 Description: Add lustre_start utility to start or stop multiple Lustre servers
2431 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
2432 Details : In case of memory pressure, list_del() can be called twice on
2433 req->rq_history_list, causing a kernel oops.
2437 Description: (ptllnd_peer.c:557:kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
2438 Details : fix stack overflow in the distributed lock manager by defering
2439 export eviction after a failed ast to the elt thread instead of
2440 handling it in the dlm interpret routine.
2442 Severity : enhancement
2444 Description: More exported tunables for mballoc
2445 Details : Add support for tunable preallocation window and new tunables for
2446 large/small requests
2450 Description: Detect corruption of block bitmap and checking for preallocations
2451 Details : Checks validity of on-disk block bitmap. Also it does better
2452 checking of number of applied preallocations. When corruption is
2453 found, it turns filesystem readonly to prevent further corruptions.
2457 Frequency : only for big-endian servers
2458 Description: Check for big-endian system while mounting fs with extents feature
2459 Details : Mounting a filesystem with extents feature will fail on big-endian
2460 systems since ext3-based ldiskfs is not supported on big-endian
2461 systems. Can be over-ridden with "bigendian_extents" mount option.
2465 Description: Excessive recovery window
2466 Details : With AT enabled, the recovery window can be excessively long (6000+
2467 seconds). To address this problem, we no longer use
2468 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2469 timeout no longer depends on the service time, it is set to
2470 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2471 time via pb_service_time.
2474 Descriptoin: Don't sync journal after every i/o
2475 Details : Implement write RPC replay to allow server replies for write RPCs
2476 before data is on disk. However, this feature is disabled by
2477 default since some issues leading to data corruptions have been
2478 found during recovery (e.g. bug 19128). This feature can be enabled
2479 by running the following command on the OSSs:
2480 lctl set_param obdfilter.*.sync_journal=0
2484 Description: Watchdog triggered on MDS failover
2485 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2486 the OSTs know that the MDS "UUID" can be reused for the same export
2487 from a different NID, so we do not need to wait for the export to
2492 Descriptoin: Lustre detected file system corruption with inode out of bounds
2493 Details : don't update i_size on MDS_CLOSE for directories. This causes
2494 directory corruptions on the MDT.
2496 -------------------------------------------------------------------------------
2498 2008-05-26 Sun Microsystems, Inc.
2500 * Support for kernels:
2501 2.6.5-7.311 (SLES 9),
2502 2.6.9-67.0.7.EL (RHEL 4),
2503 2.6.16.54-0.2.5 (SLES 10),
2504 2.6.18-53.1.14.el5 (RHEL 5),
2505 2.6.22.14 vanilla (kernel.org)
2506 * Client support for unpatched kernels:
2507 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2508 2.6.16 - 2.6.22 vanilla (kernel.org)
2509 * Due to problems with nested symlinks and FMODE_EXEC (bug 12652),
2510 we do not recommend using patchless RHEL4 clients with kernels
2511 prior to 2.6.9-55EL (RHEL4U5).
2512 * Recommended e2fsprogs version: 1.40.7-sun1
2513 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2514 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
2515 removed cwd "./" (refer to Bugzilla 14399).
2516 * A new quota file format has been introduced in 1.6.5.
2517 The format conversion from prior releases is handled transparently,
2518 but releases older than 1.4.12/1.6.5 will not understand this new
2519 format. The automatic format conversion can be avoided by running
2520 the following command on the MDS before upgrading:
2521 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
2522 For more information, please refer to bugzilla 13904.
2526 Description: quota performance fix
2527 Details : quota data is written in journalled mode instead of ordered to
2528 increase performance
2532 Description: lfs support for human-readable quota grace time strings
2533 Details : lfs setquota -t and lfs quota -t represent quota grace times
2534 in "XXwXXdXXhXXmXXs" format instead of large values in seconds
2537 Frequency : always with o2ib 1.3 and sles10
2539 Description: fix build with SLES10 and o2ib v3.
2540 Details : sles10 uses diffrent name for Module.symver file but configure
2541 assume this file has same name on RHEL/SLES/vanila kernels.
2544 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
2546 Description: MDS may lose file striping (and hence file data) in some cases
2547 Details : If there are additional extended attributes stored on the MDS,
2548 in particular ACLs, SELinux, or user attributes (if user_xattr
2549 is specified for the client mount options) then there is a risk
2550 of attribute loss. Additionally, the Lustre file striping
2551 needs to be larger than default (e.g. striped over all OSTs),
2552 and an additional attribute must be stored initially in the
2553 inode and then increase in size enough to be moved to the
2554 external attribute block (e.g. ACL growing in size) for file
2557 Severity : enhancement
2559 Description: add message levels for liblustreapi
2564 Description: MDT cannot be unmounted, reporting "Mount still busy"
2565 Details : Mountpoint references were being leaked during open reply
2566 reconstruction after an MDS restart. Drop mountpoint reference
2567 in reconstruct_open() and free dentry reference also.
2572 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
2573 Details : Move the 'good_osts' check before the 'total_bavail' check. This
2574 will result in an -EAGAIN and in the exit call path we call
2575 alloc_rr() which will with increasing aggressiveness attempt to
2576 aquire precreated objects on the minimum number of required OSCs.
2580 Description: Use old size assignment to avoid deadlock
2581 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
2582 the scheduling while holding a spinlock. We do not need locking
2583 for size in ll_update_inode() because size is only updated from
2584 the MDS for directories or files without objects, so there is no
2585 other place to do the update, and concurrent access to such inodes
2586 are protected by the inode lock.
2590 Description: Use __u64 instead of int for valid bits
2594 Description: resolve "_IOWR redefined" build error on SLES10
2598 Description: dump the memory debugging after all modules are unloaded to
2599 suppress false negative in conf_sanity test 39
2603 Description: the recovery timer never expires
2604 Details : for new client connect request, the recovery timer should not be
2605 reset, otherwise recovery timer will never expired, if the old
2606 client never come. Only old client connect and first connection
2607 req should trigger recovery timer reset.
2611 Description: the min numbers of lproc stats are wrong
2612 Details : adding a new constant LC_MIN_INIT and use it for initialization
2616 Frequency : always with interactive lfs
2618 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
2621 Frequency : with multiple concurrent readdir processes in same directory
2622 Bugzilla : 15406, 15169, 15175
2623 Description: misc fixes for directory readahead.
2624 Details : prevent previous statahead async RPC callback from processing the
2625 current "statahead_info", race condition between async RPC callback
2626 add dentry into dentry hash table and "ls" thread revalidate such
2627 dentry, statahead his/miss control for hidden items, and so on.
2629 Severity : enhancement
2631 Description: build kernel-ib packages for OFED 1.3 in our release cycle
2635 Description: incore types cleaning in quota code (with respect to 64-bit limits)
2636 Details : several u32 variables declarations are replaced with u64 declarations
2641 Description: fix SLES kernel versioning
2642 Details : the kernel version for our SLES 10 kernel did not include a "-"
2643 before the "smp" at the end. while this was not a problem in
2644 general, it did mean that software trying to use the kernel
2645 version to try to detect a vendor specific kernel would fail.
2646 this was most evident by the OFED build scripts.
2651 Description: Don't update lov_desc members until making sure they are valid
2652 Details : When updating lov_desc members via proc fs, need fix their
2653 validities before doing the real update.
2656 Frequency : very rare
2658 Description: don't put request into delay list while invalidate in flight.
2659 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
2660 import in flight. this produce timeout for invalidate and sometimes
2661 can cause stale data.
2663 Severity : enhancement
2665 Description: Update kernel to SLES9 2.6.5-7.311.
2667 Severity : enhancement
2669 Description: Update kernel to RHEL4 2.6.9-67.0.7.
2674 Frequency : on PPC only
2675 Description: not convert ost objects for directory because it's not exist.
2676 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
2678 Severity : enhancement
2680 Description: Fix warnings with compile liblustre at sles10/rhel5 which have
2681 __u64 as usingied long long type.
2684 Frequency : rare, on shutdown
2686 Description: race process ast vs remove callback
2687 Details : removing callback before disconnect import open race with
2688 processing callback.
2690 Severity : enhancement
2692 Description: Update kernel to SLES9 2.6.5-7.311.
2694 Severity : enhancement
2696 Description: Files open for execute are not marked busy on SLES10
2697 Details : Add FMODE_EXEC to SLES10 SP1 server kernel series.
2699 Severity : enhancement
2701 Description: Add server support for vanilla-2.6.22.14.
2704 Frequency : occasional
2706 Description: Avoid lov_create() getting stuck in obd_statfs_rqset()
2707 Details : If an OST is down the MDS will hang indefinitely in
2708 obd_statfs_rqset() waiting for the statfs data. While for
2709 MDS QOS usage of statfs, it should not stuck in waiting.
2711 Severity : enhancement
2713 Description: Disable adaptive timeouts by default
2716 Frequency : on network error
2718 Description: panic with double free request if network error
2719 Details : mdc_finish_enqueue is finish request if any network error occuring,
2720 but it's true only for synchronus enqueue, for async enqueue
2721 (via ptlrpcd) this incorrect and ptlrpcd want finish request
2725 Frequency : rare, on recovery
2727 Description: read procfs can produce deadlock in some situation
2728 Details : Holding lprocfs lock with send rpc can produce block for destroy
2729 obd objects and this also block reconnect with -EALREADY.
2730 This isn't fix all lprocfs bugs - but make it rare.
2732 Severity : enhancement
2734 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
2737 Frequency : frequent on X2 node
2739 Description: mdc_set_open_replay_data LBUG
2740 Details : Set replay data for requests that are eligible for replay.
2745 Description: lustre_mgs: operation 101 on unconnected MGS
2746 Details : When MGC is disconnected from MGS long enough, MGS will evict the
2747 MGC, and late on MGC cannot successfully connect to MGS and a lot
2748 of the error messages complaining that MGS is not connected.
2751 Frequency : rare, depends on device drivers and load
2753 Description: MDS or OSS nodes crash due to stack overflow
2754 Details : Code changes in 1.6.4 increased the stack usage of some functions.
2755 In some cases, in conjunction with device drivers that use a lot
2756 of stack, the MDS (or possibly OSS) service threads could overflow
2757 the stack. One change which was identified to consume additional
2758 stack has been reworked to avoid the extra stack usage.
2760 Severity : enhancement
2762 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
2764 Severity : enhancement
2766 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
2768 Severity : enhancement
2770 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
2772 Serverity : enhancement
2774 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
2777 Frequency : rare on shutdown OST
2779 Description: Don't allow skipping OSTs if index has been specified.
2780 Details : Don't allow skipping OSTs if index has been specified, make locking
2781 in internal create lots better.
2786 Description: ASSERTION(!PageDirty(page)) failed
2787 Details : Wrong check could lead to an assertion failure under specific
2793 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
2794 Details : access to bitfield in structure is always rounded to long
2795 and this produce problem with not atomic change any bit.
2800 Description: Lustre make rpms failed.
2801 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
2802 builds Lustre rpms from tarball.
2804 Severity : enhancement
2806 Description: Update to SLES9 SP4 kernel-2.6.5-7.308.
2809 Frequency : rare on shutdown OST
2811 Description: If llog cancel was not send before clean_exports phase, this can
2812 produce deadlock in llog code.
2813 Details : If llog thread has last reference to obd and call class_import_put
2814 this produce deadlock because llog_cleanup_commit_master wait when
2815 last llog_commit_thread exited, but this never success because was
2816 called from llog_commit_thread.
2819 Frequency : only if OST index is skipped
2821 Description: NULL lov_tgts causing MDS oops
2822 Details : more safe checks for NULL lov_tgts for avoid oops.
2824 Severity : enhancement
2826 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
2828 Severity : enhancement
2830 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
2835 Description: make mgs_setparam() handle fsname containing dash
2836 Details : fsname containing a dash does not work with lctl conf_param
2838 Severity : enhancement
2840 Description: Update to RHEL4 Update-6 kernel-2.6.9-67.EL.
2843 Frequency : rare, in recovery and (or) destroy lovobjid file.
2845 Description: rewrite lov objid code.
2846 Details : Cleanup for lov objid code, remove scability problems and wrong
2847 locking. Fix sending last_id into ost.
2849 Severity : enhancement
2851 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
2853 Severity : enhancement
2855 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
2856 Details : Use d_move_locked instead of __d_move.
2859 Frequency : rare, at shutdown
2861 Description: access already free / zero obd_namespace.
2862 Details : if client_disconnect_export was called without force flag set,
2863 and exist connect request in flight, this can produce access to
2864 NULL pointer (or already free pointer) when connect_interpret
2865 store ocd flags in obd_namespace.
2868 Frequency : only at startup
2870 Description: not alloc memory with spinlock held.
2871 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
2872 if any spinlock held.
2877 Description: lfs find does not continue on file error
2878 Details : Continue other files processing when a file/dir is absent.
2882 Description: Inconsistent usage of lustre_pack_reply()
2883 Details : Standardize the usage of lustre_pack_reply() such that it
2884 always generate a CERROR on failure.
2887 Frequency : very rare
2889 Description: Fix replay if there is an un-replied request and open
2890 Details : In some cases, older replay request will revert the
2891 mcd->mcd_last_xid on MDS which is used to record the client's
2892 latest sent request.
2894 Severity : enhancement
2896 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
2898 Severity : enhancement
2900 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
2902 Severity : enhancement
2904 Description: Update to SLES9 kernel-2.6.5-7.287.3.
2906 Severity : enhancement
2908 Description: Update to RHEL4 kernel-2.6.9-55.0.12.EL.
2910 Severity : enhancement
2912 Description: Build SLES10 patchless client fails
2913 Details : The configure was broken by run ./configure with
2914 --with-linux-obj=.... argument for patchless client. When the
2915 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
2916 can't search header adequately. Use absolute path such as
2917 -I($LINUX)/include instead.
2919 Severity : enhancement
2921 Description: Lustre Page Accounting
2922 Details : New macros for page alloc and free which enable accounting
2923 of page allocation of Lustre. Use percpu counters to store memory
2924 and page statistics.
2927 Frequency : only if debugging is disabled
2929 Description: LASSERT_{REQ,REP}SWAB macros are buggy
2930 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
2931 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
2932 these macros and replace them with their definitions instead.
2937 Description: interrupt oig_wait produce painc on resend.
2938 Details : brw_redo_request can be used for resend requests from ptlrpcd and
2939 private set, and this produce situation when rq_ptlrpcd_data not
2940 copyed to new allocated request and triggered LBUG on assert
2941 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
2942 ptlrpcd set if request is changed and can be safety changed to use
2945 Severity : enhancement
2947 Description: organize the server-side client stats on per-nid basis
2948 Details : Change the structure of stats under obdfilter and mds to
2956 The "uuid"s file would list the uuids of _active_ exports.
2957 And the clear entry is to clear all stats and stale nids.
2962 Description: Processes looping in ll_readdir() on Lustre clients finally causing
2963 a full node pseudo-hang
2964 Details : Concurrent access to the same directory from multiple clients with
2965 intensive file creation/removal can cause a client node to spin in
2966 ll_readdir(). i_version must be increased every time the lock is
2967 cancelled to ensure a revalidate is done.
2972 Description: touch file failed when fs is not full
2973 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
2974 otherwise we can get ENOSP while fs is not full.
2977 Frequency : only for Cray XT3
2978 Bugzilla : 12829/13455
2979 Description: Changing primary group doesn't change the group lustre assigns to
2981 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
2982 overridden with the primary group provided by the group upcall,
2983 whereas the supplied fsgid can be trusted if it is in the list of
2984 supplementary groups returned by the group upcall.
2986 Severity : enhancement
2988 Description: Root Squash Functionality
2989 Details : Implementation of NFS-like root squash capability. Specifically,
2990 don't allow someone with root access on a client node to be able
2991 to manipulate files owned by root on a server node.
2993 Severity : enhancement
2995 Description: Slow trucate/writes to huge files at high offsets.
2996 Details : Directly associate cached pages to lock that protect those pages,
2997 this allows us to quickly find what pages to write and remove
2998 once lock callback is received.
3003 Description: Too many locks accumulating on client during NFS usage
3004 Details : mds_open improperly used accmode to find out access mode to a
3005 file. Also mdc_intent_lock logic to find out if we already have
3006 lock similar to just received was flawed since introduction of
3007 skiplists - locks are now added to the front of the granted
3012 Description: Hit ASSERTION(obd->obd_stopping == 1) failed in some setup failed
3014 Details : In obd setup failure handler, obd_stopping will not necessarily to
3015 be 1, and obd_set_up should also be checked to make sure whether
3016 obd is completely setup.
3018 Severity : enhancement
3020 Description: Allow masking D_WARNING, D_ERROR messages from console
3021 Details : Console messages can now be disabled via lnet.printk.
3026 Description: User code with malformed file open parameter crashes client node
3027 Details : Before packing join_file req, all the related reference should be
3028 checked carefully in case some malformed flags cause fake join_file
3034 Description: shrink/enlarge qunit size when needed; fix the problem of coarse
3035 grain of quota doing harm to quota's accuracy
3036 Details : qunit size will be changed when quota limitation is too low/high;
3037 record the pending quota write in order to get more accureate
3038 quota; delete the patch for bug12588, which is unnecessary when
3039 this patch is landed. This bug also contains fixes for bug 14526,
3040 14299, 14601 and 13794.
3044 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
3045 Details : ldlm_completion_ast() assumes that a lock is granted when the req
3046 mode is equal to the granted mode. However, it should also check
3047 that LDLM_FL_CP_REQD is not set.
3051 Description: Heavy nfs access might result in deadlocks
3052 Details : After ELC code landed, it is now improper to enqueue any mds
3053 locks under och_sem, because enqueue might want to decide to
3054 cancel open locks for same inode we are holding och_sem for.
3058 Description: 35% write performance drop with ldiskfs2 when quotas are on
3059 Details : Enable ext3 journalled quota by default to improve performance
3060 when quotas are turned on.
3064 Description: Client eviction while running blogbench
3065 Details : A lot of unlink operations with concurrent I/O can lead to a
3066 deadlock causing evictions. To address the problem, the number of
3067 oustanding OST_DESTROY requests is now throttled to
3068 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
3069 callbacks are processed in priority.
3072 Frequency : RHEL4 only
3074 Description: mkfs is very slow on IA64/RHEL4
3075 Details : A performance regression has been discovered in the MPT Fusion
3076 driver between versions 3.02.73rh and 3.02.99.00rh. As a
3077 consequence, we have downgraded the MPT Fusion driver in the RHEL4
3078 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
3081 Frequency : PPC/PPC64 only
3083 Description: conflicts between asm-ppc64/types.h and lustre_types.h
3084 Details : fix duplicated definitions between asm-ppc64/types.h and
3085 lustre_types.h on PPC.
3088 Frequency : PPC/PPC64 only
3090 Description: asm-ppc/segment.h does not exist
3091 Details : fix compile issue on PPC.
3095 Description: data checksumming impacts single node performance
3096 Details : add support for several checksum algorithms. Currently, CRC32 and
3097 Adler-32 are supported. The checksum type can be changed on the fly
3098 through /proc/fs/lustre/osc/*/checksum_type.
3102 Description: use adler32 for page checksums
3103 Details : when available, use the Adler-32 algorithm instead of CRC32 for
3108 Description: better handle error messages in extents code
3110 Severity : enhancement
3112 Description: SNMP support enhancement
3113 Details : Adding total number of sampled request for an MDS node in snmp
3116 Severity : enhancement
3118 Description: Optimize ldlm waiting list processing for PR extent locks
3119 Details : When processing waiting list for read extent lock and meeting read
3120 lock that is same or wider to it that is not contended, skip
3121 processing rest of the list and immediatelly return current
3122 status of conflictness, since we are guaranteed there are no
3123 conflicting locks in the rest of the list.
3127 Description: Time out and refuse to reconnect
3128 Details : When the failover node is the primary node, it is possible
3129 to have two identical connections in imp_conn_list. We must
3130 compare not conn's pointers but NIDs, otherwise we
3131 can defeat connection throttling.
3135 Description: Client not clear own cache if answer to reconnect is lost.
3136 Details : Client gets evicted from server. Now client also thinks it is
3137 disconnected (or gets ENOTCONN on its operation) and decides to
3138 reconnect. Server receives reconnect message, but cannot find
3139 export. New export is created that is fully valid (new cookie!),
3140 but reply is lost and not reported to client. Client reconnects
3141 again and gets back a just-created connection, but it is not new
3142 so client thinks it was not evicted and does not do recovery.
3146 Description: Detect stride IO mode in read-ahead
3147 Details : When a client does stride read, read-ahead should detect that and
3148 read-ahead pages according to the detected stride pattern.
3152 Description: build for x2 fails
3153 Details : fix compile issue on Cray systems.
3155 Severity : enhancement
3157 Description: implement readv/writev APIs(aio_read/aio_writes in newer kernels)
3158 Details : This greatly improves speed of NFS writes on 2.6 kernels.
3161 Frequency : only on PPC/SLES10
3163 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
3164 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
3165 be defined. Add a hack in mkfs_lustre.c to work around this compile
3170 Description: LASSERT on MDS when client holding flock lock dies
3171 Details : ldlm pool logic depends on number of granted locks equal to
3172 number of released locks which is not true for flock locks, so
3173 just exclude such locks from consideration.
3177 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
3178 Details : Use fsfilt_sync() for both the whole filesystem sync and
3179 individual file sync to eliminate dangerous inode locking
3180 with I_LOCK that can lead to a deadlock.
3184 Description: Update an obsolete wirecheck.c generator
3185 Details : Update wirecheck.c/wirehdr.c and regenerate wiretest.c
3189 Description: Client can panic on open sometimes
3190 Details : It is possible that we try to free already freed request in
3191 ll_file_open in some error cases when we send request from
3196 Description: performance in 1.6.3
3197 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
3198 sure that 1MB requests are not fragmented by the block layer.
3202 Description: LDLM soft lockups - improvement
3203 Details : It is be possible to send the lock handle along with each read
3204 or write request because the client is already doing a lock match
3205 itself so there isn't any reason the OST should have to re-do that
3211 Description: lfs quota fails with deactivated OSTS
3212 Details : With this patch, three improvements are included:
3213 1. detete the softlimit in mds and osts when use "lfs quota".
3214 2. display the inaccurate data in the output of "lfs quota".
3215 3. try to get quota info when "lfs quota" is executed.
3220 Description: Extent locks not granted with no conflicts sometimes.
3221 Details : When race occurs in glimpse handler and nothing is returned,
3222 we do not reprocess the queue after lock cancel, and that leads
3223 to a stall until next activity on a resource
3226 Frequency : failover with quotaon
3228 Description: during mds failovers with quota on, OSTs got into deadlock state
3229 and causing dumpstack.
3230 Details : for every quota slave, at any time, there is only one quota req
3231 is sent to quota master for every uid/gid. Before that quota req
3232 returns, all the thread relative to the same uid/gid will wait.
3233 So if the quota req is lost because mds failovers or any other
3234 reasons, this bug will be hit. Now, dqacq_interpret() will handle
3235 quota reqs who time out.
3237 Severity : enhancement
3240 Description: when quota slave checks if quota is enough, there is an unnecessary
3242 Details : place this wait on necessary place instead of always waiting.
3244 --------------------------------------------------------------------------------
3246 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
3248 * Support for kernels:
3249 2.6.5-7.286 (SLES 9),
3250 2.6.9-55.0.9.EL (RHEL 4),
3251 2.6.16.53-0.8 (SLES 10),
3252 2.6.18-8.1.14.el5 (RHEL 5),
3253 2.6.18.8 vanilla (kernel.org)
3254 * Client support for unpatched kernels:
3255 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3256 2.6.16 - 2.6.22 vanilla (kernel.org)
3257 * Due to recently discovered recovery problems, we do not recommend
3258 using patchless RHEL 4 clients with this or any earlier release.
3259 * Recommended e2fsprogs version: 1.40.2-cfs1
3260 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3263 Frequency : occasional
3265 Description: MDS hang or stay in waiting lock
3266 Details : If client receive lock with CBPENDING flag ldlm need send lock
3267 cancel as separate rpc, to avoid situation when cancel request
3268 can't processed due all i/o threads stay in wait lock.
3271 Frequency : occasional
3273 Description: Do not fail import if osc_interpret_create gets -EAGAIN
3274 Details : If osc_interpret_create got -EAGAIN it immediately exits and
3275 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
3276 oscc_has_objects and see OSC has no objests and call
3277 oscc_internal_create to resend create request.
3279 Severity : enhancement
3281 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
3282 Details : Update which_patch & target file for SLES10 latest kernel.
3284 Severity : enhancement
3286 Description: add --type and --size parameters to lfs find
3287 Details : Enhance lfs find by adding filetype and filesize parameters. Also
3288 multiple OBDs can now be specified for the --obd option.
3290 Severity : enhancement
3292 Description: eliminate client locks in face of contention
3293 Details : file contention detection and lockless i/o implementation
3294 for contended files.
3296 Severity : enhancement
3298 Description: Remove client patches from SLES 10 kernel.
3299 Details : This causes SLES 10 clients to behave as patchless clients
3300 even on a Lustre-patched (server) kernel.
3302 Severity : enhancement
3304 Description: use i_size_read and i_size_write in 2.6 port
3305 Details : replace inode->i_size access with i_size_read/write()
3308 Frequency : when removing large files
3310 Description: scheduling issue during removal of large Lustre files
3311 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
3312 It causes scheduling issues when removing large files (17TB in the
3318 Description: 1.4.11 Can't handle directories with stripe set and extended ACLs
3319 Details : Impossible (EPROTO is returned) to access a directory that has a
3320 non-default striping and ACLs.
3323 Frequency : only on ppc
3325 Description: /proc/fs/lustre/devices broken on ppc
3326 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
3327 We should make sure the type of 'index' is loff_t and then cast
3328 explicitly as needed below. Do not assign an explicitly cast
3332 Frequency : only for rhel5
3334 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
3335 Details : Modify the target file & which_kernel.
3338 Frequency : if the uninit_groups feature is enabled on ldiskfs
3340 Description: e2fsck reports "invalid unused inodes count"
3341 Details : If a new ldiskfs filesystem is created with the "uninit_groups"
3342 feature and only a single inode is created in a group then the
3343 "bg_unused_inodes" count is incorrectly updated. Creating a
3344 second inode in that group would update it correctly.
3349 Description: buffer overruns could theoretically occur
3350 Details : llapi_semantic_traverse() modifies the "path" argument by
3351 appending values to the end of the origin string, and a buffer
3352 overrun may occur. Adding buffer overrun check in liblustreapi.
3354 Severity : enhancement
3356 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla.
3359 Frequency : only if filesystem is inconsistent
3361 Description: handle "serious error: objid * already exists" more gracefully
3362 Details : If LAST_ID value on disk is smaller than the objects existing in
3363 the O/0/d* directories, it indicates disk corruption and causes an
3364 LBUG(). If the object is 0-length, then we should use the existing
3365 object. This will help to avoid a full fsck in most cases.
3367 Severity : enhancement
3369 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
3370 Details : Modify vm-tunables-rhel4.patch.
3372 Severity : enhancement
3374 Description: Kernel config for 2.6.18-vanilla.
3375 Details : Modify targets/2.6-vanilla.target.in.
3376 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
3377 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
3378 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
3379 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
3382 Frequency : occasional
3384 Description: improve handling recoverable errors
3385 Details : If request processed with error which can be recoverable on server
3386 request should be resend, otherwise page released from cache and
3392 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
3393 Details : Modify target file & which_patch.
3394 A flaw was found in the IA32 system call emulation provided
3395 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
3396 value could be stored in the %RAX register, which could trigger an
3397 out-of-bounds system call table access. An untrusted local user
3398 could exploit this flaw to run code in the kernel
3399 (ie a root privilege escalation). (CVE-2007-4573).
3403 Description: change order of libsysio includes
3404 Details : '#include sysio.h' should always come before '#include xtio.h'
3406 Severity : enhancement
3408 Description: adapt the lustre_config script to support the upgrade case
3409 Details : Add "-u" option for lustre_config script to support upgrading 1.4
3410 server targets to 1.6 in parallel.
3415 Description: To avoid grant space > avaible space when the disk is almost
3416 full. Without this patch you might see the error "grant XXXX >
3417 available" or some LBUG about grant, when the disk is almost
3419 Details : In filter_check_grant, for non_grant cache write, we should
3420 check the left space by if (*left > ungranted + bytes), instead
3421 of (*left > ungranted), because only we are sure the left space
3422 is enough for another "bytes", then the ungrant space should be
3423 increase. In client, we should update cl_avail_grant only there is
3424 OBD_MD_FLGRANT in the reply.
3429 Description: Update RHEL 4 kernel to fix local root privilege escalation.
3430 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
3431 described in CVE-2007-4573. This problem could allow untrusted
3432 local users to gain root access.
3435 Frequency : when using O_DIRECT and quotas
3437 Description: Incorrect file ownership on O_DIRECT output files
3438 Details : block usage reported by 'lfs quota' does not take into account
3439 files that have been written with O_DIRECT.
3443 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
3444 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
3445 window to 0. If it does hit cache miss, and read-pattern is still
3446 stride-io mode, does not reset the stride window, but also does not
3447 increase the stride window length in this case.
3449 --------------------------------------------------------------------------------
3451 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
3453 * Support for kernels:
3454 2.6.5-7.286 (SLES 9),
3455 2.6.9-55.0.2.EL (RHEL 4),
3456 2.6.16.46-0.14 (SLES 10),
3457 2.6.18-8.1.8.el5 (RHEL 5),
3458 2.6.18.8 vanilla (kernel.org)
3459 * Client support for unpatched kernels:
3460 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3461 2.6.16 - 2.6.21 vanilla (kernel.org)
3462 * Due to recently discovered recovery problems, we do not recommend
3463 using patchless RHEL 4 clients with this or any earlier release.
3464 * Recommended e2fsprogs version: 1.40.2-cfs1
3465 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3469 Description: Fix errors in lfs documentation
3470 Details : Fixes man pages
3472 Severity : enhancement
3474 Description: Adaptive timeouts
3475 Details : RPC timeouts adapt to changing server load and network
3476 conditions to reduce resend attempts and improve recovery time.
3478 Severity : enhancement
3480 Description: llapi_file_create() does not allow some changes
3481 Details : add llapi_file_open() that allows specifying the file creation
3482 mode and open flags, and also returns an open file handle.
3484 Severity : enhancement
3486 Description: df doesn't work properly if diskfs blocksize != 4K
3487 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
3490 Frequency : always on directories with default striping set
3492 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
3493 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
3495 Severity : enhancement
3497 Description: merge and cleanup kernel patches.
3498 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
3500 Severity : enhancement
3502 Description: RedHat Update kernel for RHEL5
3503 Details : Kernel config file for RHEL5.
3505 Severity : enhancement
3507 Description: OSS needs mutliple precreate threads
3508 Details : Add ability to start more than one create thread per OSS.
3511 Frequency : only with quota on the root user
3513 Description: mds_obd_create error creating tmp object
3514 Details : When the user sets quota on root, llog will be affected and can't
3515 create files and write files.
3518 Frequency : Always on ia64 patchless client, and possibly others.
3520 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
3521 Details : This allows the patchless client to be loaded on architectures
3522 without this export.
3527 Description: disorder of journal start and llog_add cause deadlock.
3528 Details : in llog_origin_connect, journal start should happen before
3529 llog_add keep the same order as other functions to avoid
3532 Severity : enhancement
3534 Description: RedHat Update kernel for RHEL5
3535 Details : Modify the kernel config file more closer RHEL5.
3537 Severity : enhancement
3539 Description: Build failure against Centos5 (RHEL5)
3540 Details : Define PAGE_SIZE when it isn't present.
3543 Frequency : occasionally when using NFS
3545 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
3546 Details : with NFS, the anon dentry's parent was set to itself in
3547 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
3548 corresponding dentry other than use rec->ur_name.
3550 Severity : enhancement
3552 Description: enable data checksumming by default
3553 Details : enable checksum by default, allow --disable-checksum
3554 configure option and "-o nochecksum" mount option. Checksums
3555 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
3558 Frequency : Occasionally with failover
3560 Description: Client eviction due to failover config
3561 Details : after a connection loss, the lustre client should attempt to
3562 reconnect to the last active server first before trying the
3563 other potential connections.
3565 Severity : enhancement
3567 Description: client-side metadata stat-ahead during readdir(directory readahead)
3568 Details : perform client-side metadata stat-ahead when the client detects
3569 readdir and sequential stat of dir entries therein
3574 Description: when mds and osts use different quota unit(32bit and 64bit),
3575 quota will be released repeatly.
3576 Details : void sending multiple quota reqs to mds, which will keep the
3577 status between the reqs.
3580 Frequency : only with liblustre clients on XT3
3582 Description: evictions taking too long
3583 Details : allow llrd to evict clients directly on OSTs
3587 Description: osts not allocated evenly to files
3588 Details : change the condition to increase offset_idx
3590 Severity : enhancement
3592 Description: self-adjustable client's lru lists
3593 Details : use adaptive algorithm for managing client cached locks lru
3594 lists according to current server load, other client's work
3595 pattern, memory activities, etc. Both, server and client
3596 side namespaces provide number of proc tunables for controlling
3601 Description: rewrite ext2-derived code in llite/dir.c and obdclass/uuid.c
3602 Details : rewrite inherited code (uuid parsing code from ext2 utils and
3603 readdir code from ext3) from scratch preserving functionality.
3607 Description: Only those disconnect error should be returned by rq_status.
3608 Details : In open/enqueue processs, Some errors, which will cause client
3609 disconnected, should be returned by rq_status, while other
3610 errors should still be returned by intent, then mdc or llite will
3613 Severity : enhancement
3615 Description: Tune the kernel for good SCSI performance.
3616 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
3617 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
3621 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
3622 Bugzilla : 13547, 13627
3623 Description: Data corruption for OSTs that are formatted larger than 2TB
3625 Details : When generating the bio request for lustre file writes the
3626 sector number would overflow a temporary variable before being
3627 used for the IO. The data reads correctly from Lustre (which
3628 will overflow in a similar manner) but other file data or
3629 filesystem metadata may be corrupted in some cases.
3633 Description: "lfs find -obd UUID" prints directories
3634 Details : "lfs find -obd UUID" will return all directory names instead
3635 of just file names. It is incorrect because the directories
3636 do not reside on the OSTs.
3639 Bugzilla : 13596, 13614
3640 Description: MDS hang after unclean shutdown of lots of clients
3641 Details : Never resend AST requests.
3645 Frequency : Always, for kernels after 2.6.16
3646 Description: Fix warning idr_remove called for id=.. which is not allocated.
3647 Details : Last kernels save old s_dev before kill super and not allow
3648 to restore from callback - restore it before call kill_anon_super.
3652 Description: Fix error on 'ls .' at the top of the Lustre mount.
3653 Details : Don't revalidate dentry if it is a root dentry.
3655 --------------------------------------------------------------------------------
3657 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
3659 * Support for kernels:
3660 2.6.5-7.286 (SLES 9),
3661 2.6.9-55.0.2.EL (RHEL 4),
3662 2.6.16.46-0.14 (SLES 10),
3663 2.6.18-8.1.8.el5 (RHEL 5),
3664 2.6.18.8 vanilla (kernel.org)
3665 * Client support for unpatched kernels:
3666 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3667 2.6.16 - 2.6.22 vanilla (kernel.org)
3668 * Due to recently discovered recovery problems, we do not recommend
3669 using patchless RHEL 4 clients with this or any earlier release.
3670 * Recommended e2fsprogs version: 1.39.cfs8
3671 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3676 Description: block reactivating mgc import until all deactivates complete
3677 Details : Fix race when failing back MDT/MGS to itself (testing)
3679 Severity : enhancement
3681 Description: lfs setstripe enhancement
3682 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
3685 Frequency : mds/oss recovery
3687 Description: llog ctxt is refrenced after it has been freed.
3688 Details : llog ctxt refcount was added to avoide the race between ctxt free
3689 and llog recovery process. Each llog user must hold ctxt refcount
3690 before it access the llog. And the llog ctxt can only be freed
3691 when its refcount is zero.
3693 Severity : enhancement
3695 Description: randomly memory allocation failure util
3696 Details : Make lustre randomly failed allocating memory for testing purpose.
3699 Frequency : only for SLES10
3701 Description: Update kernel patch for SLES10 SP1
3702 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
3704 Severity : enhancement
3706 Description: omit set fsid for export NFS
3707 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
3708 as FS_REQUIRES_DEV for avoid problems with generate fsid.
3711 Frequency : after network failures
3713 Description: Add sync option to mount_lustre.c
3714 Details : Client loses data written to lustre after a network interruption.
3716 Severity : enhancement
3718 Description: Error message improvement.
3719 Details : Merging of two LCONSOLE_ERROR_MSG into one.
3721 Severity : enhancement
3723 Description: don't use GFP_* in generic Lustre code.
3724 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
3726 Severity : enhancement
3728 Description: obdclass is limited by single OBD_ALLOC(idarray)
3729 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
3731 Severity : enhancement
3733 Description: updated patchess for new RHEL4 kernel
3734 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
3736 Severity : enhancement
3738 Description: warnings with build patchless client with vanila 2.6.19 and up
3739 Details : change old ctl_table style and replace ctl_table/ctl_table_header
3740 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
3742 Severity : enhancement
3744 Description: O_DIRECT bypasses client statistics.
3745 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
3746 read_ahead_stats were updated. Copied stats section from
3747 osc_send_oap_rpc() into async_internal().
3750 Frequency : only for Cray XT3
3752 Description: peer credits not enough on many OST per OSS systems.
3753 Details : Use new lnet way to add credits as we need those for pings and ASTs
3756 Frequency : only with liblustre
3758 Description: Liblustre is not releasing flock locks on file close.
3759 Details : Release flock locks on file close.
3762 Frequency : only for RHEL4
3764 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
3765 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
3766 Update target file and kernel config.
3770 Description: lustre support for RHEL5
3771 Details : Add support for RHEL5.
3776 Description: ASSERTION(export != NULL) failed in target_handle_connect
3777 Details : Assetion hit is result of rare race between disconnect and connet
3778 to same nid. target_handle_connect found old connect cockie and
3779 tried to reconnect, but can't find export for this cockie.
3784 Description: umount blocks forever on error
3785 Details : In result of wrong using obd_no_recov and obd_force flags client
3786 can hand if cancel or some other requests is lost.
3789 Frequency : Only for SLES
3791 Description: sanity_quota fail test_1
3792 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
3793 file, which makes TSTID[2] inunique.
3795 Severity : enhancement
3797 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
3798 Details : Update target/ChangeLog/which_patch .
3800 Severity : enhancement
3802 Description: jbd statistics
3803 Details : Port older jbd statistics patch for sles10
3808 Description: lvbo_init failed for resource with missing objects.
3809 Details : Fix returning error if we do stat for file with missing/corrupted
3810 objects and i_size set to all sum of size of all avaible objects.
3811 if we to truncate/write to missing object - it is recreated.
3816 Description: Oops in read and write path when failing to allocate lock.
3817 Details : Check if lock allocation failed and return error back.
3820 Frequency : When flocks are used.
3822 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
3823 Details : Flock locks might destroy just granted lock if it could be merged
3824 with another existing flock, this is done in completion handler,
3825 so teach ldlm_cli_enquque_fini that this is a valid case for
3831 Description: reply_lock_interpret crash due to race with it and lock cancel.
3832 Details : Do not replay locks that are being cancelled. Do not reference
3833 locks by their address during replay, just by their handle.
3836 Frequency : only with deactivated OSTs
3838 Description: lstripe command fails for valid OST index
3839 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
3840 instead of lov->desc.ld_active_tgt_count.
3842 Severity : enhancement
3844 Description: Build failure against Centos5 (RHEL5)
3845 Details : Use getpagesize() instead of PAGE_SIZE.
3847 --------------------------------------------------------------------------------
3849 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
3851 * Support for kernels:
3852 2.6.5-7.283 (SLES 9),
3853 2.6.9-55.EL (RHEL 4),
3854 2.6.16.46-0.14 (SLES 10),
3855 2.6.18.8 vanilla (kernel.org)
3856 * Client support for unpatched kernels:
3857 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3858 2.6.16 - 2.6.22 vanilla (kernel.org)
3859 * Due to recently discovered recovery problems, we do not recommend
3860 using patchless RHEL 4 clients with this or any earlier release.
3861 * Recommended e2fsprogs version: 1.39.cfs8
3862 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3863 * Starting with this release, the ldiskfs backing filesystem required
3864 by Lustre is now in its own package, lustre-ldiskfs. This package
3865 should be installed. It is versioned separately from Lustre and
3866 may be released separately in future.
3871 Description: server LBUG when shutting down
3872 Details : Block umount forever until the mount refcount is zero rather
3873 than giving up after an arbitrary timeout.
3875 Severity : enhancement
3877 Description: add optional extra BUILD_VERSION info
3878 Details : add a new environment variable (namely LUSTRE_VERS) which allows
3879 to override the lustre version.
3882 Frequency : 2.6.18 servers only
3884 Description: ll_kern_mount() doesn't release the module reference
3885 Details : The ldiskfs module reference count never drops down to 0
3886 because ll_kern_mount() doesn't release the module reference.
3891 Description: server LBUG when using old ost_num_threads parameter
3892 Details : Accept the old ost_num_threads parameter but warn that it
3893 is deprecated, and fix an off-by-one error that caused an LBUG.
3898 Description: Transient SCSI error results in persistent IO issue
3899 Details : iobuf->dr_error is not reinitialized to 0 between two
3903 Frequency : sometimes when underlying device returns I/O errors
3905 Description: OSTs not going read-only during write failures
3906 Details : OSTs are not remounted read-only when the journal commit threads
3907 get I/O errors because fsfilt_ext3 calls journal_start/stop()
3908 instead of the ext3 wrappers.
3911 Frequency : only with 10000 clients or more
3913 Description: poor connect scaling with increasing client count
3914 Details : Don't run filter_grant_sanity_check for more than 100 exports
3915 to improve scaling for large numbers of clients.
3918 Frequency : SLES10 only
3920 Description: sanity-quota.sh quotacheck failed: rc = -22
3921 Details : Quotas cannot be enabled on SLES10.
3924 Frequency : liblustre clients only
3926 Description: getdirentries does not give error when run on compute nodes
3927 Details : getdirentries does not fail when the size specified as an argument
3928 is too small to contain at least one entry
3930 Severity : enhancement
3932 Description: Add LNET router traceability for debug purposes
3933 Details : If a checksum failure occurs with a router as part of the
3934 IO path, the NID of the last router that forwarded the bulk data
3935 is printed so it can be identified.
3940 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
3941 Details : Due to a race condition, liblustre clients were occasionally
3942 evicted incorrectly.
3944 Severity : enhancement
3946 Description: lfs setstripe use optional parameters instead of postional
3949 Severity : enhancement
3951 Description: Nanosecond timestamp support for ldiskfs
3952 Details : The on-disk ldiskfs filesystem has added support for nanosecond
3953 resolution timestamps. There is not yet support for this at
3954 the Lustre filesystem level.
3957 Frequency : during server recovery
3959 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
3960 Details : request with rq_no_resend flag not awake l_wait_event if they get a
3964 Frequency : nfs export on patchless client
3966 Description: connectathon hang when test nfs export over patchless client
3967 Details : Disconnected dentry cannot be found with lookup, so we do not need
3968 to unhash it or make it invalid
3971 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
3972 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
3973 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
3978 Description: open req refcounting wrong on reconnect
3979 Details : If reconnect happened between getting open reply from server and
3980 call to mdc_set_replay_data in ll_file_open, we will schedule
3981 replay for unreferenced request that we are about to free.
3982 Subsequent close will crash in variety of ways.
3983 Check that request is still eligible for replay in
3984 mdc_set_replay_data().
3989 Description: disable writes to filesystem when reading health_check file
3990 Details : the default for reading the health_check proc file has changed
3991 to NOT do a journal transaction and write to disk, because this
3992 can cause reads of the /proc file to hang and block HA state
3993 checking on a healthy but otherwise heavily loaded system. It
3994 is possible to return to the previous behaviour during configure
3995 with --enable-health-write.
3997 Severity : enhancement
3999 Description: 64-bit inode version
4000 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
4001 to the inode. This will be required for version-based recovery.
4006 Description: MDS fails to start if a duplicate client export is detected
4007 Details : in some rare cases it was possible for a client to connect to
4008 an MDS multiple times. Upon recovery the MDS would detect this
4009 and fail during startup. Handle this more gracefully.
4011 Severity : enhancement
4013 Description: Add -o localflock option to simulate old noflock behaviour.
4014 Details : This will achieve local-only flock/fcntl locks coherentness.
4019 Description: log_commit_thread vs filter_destroy race leads to crash
4020 Details : Take import reference before releasing llog record semaphore
4025 Description: Wrong request locking in request set processing
4026 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
4027 imp_delayed_list, in this place should be used imp_lock.
4030 Frequency : when reconnecting
4032 Description: Grant leak when OSC reconnect to OST
4033 Details : When osc reconnect ost, OST(filter) should check whether it
4034 should grant more space to client by comparing fed_grant and
4035 cl_avail_grant, and return the granted space to client instead
4036 of "new granted" space, because client will call osc_init_grant
4037 to update the client grant space info.
4040 Frequency : when client reconnects to OST
4042 Description: Grant leak when OSC does a resend and replays bulk write
4043 Details : When osc reconnect to OST, OST(filter) should clear grant info of
4044 bulk write request, because the grant info while be sync between
4045 OSC and OST when reconnect, and we should ignore the grant info
4046 these of resend/replay write req.
4051 Description: Grant space more than avaiable space sometimes.
4052 Details : When then OST is about to be full, if two bulk writing from
4053 different clients came to OST. Accord the avaliable space of the
4054 OST, the first req should be permitted, and the second one
4055 should be denied by ENOSPC. But if the seconde arrived before
4056 the first one is commited. The OST might wrongly permit second
4057 writing, which will cause grant space > avaiable space.
4060 Frequency : when client is evicted
4062 Description: Grant might be wrongly erased when osc is evicted by OST
4063 Details : when the import is evicted by server, it will fork another
4064 thread ptlrpc_invalidate_import_thread to invalidate the
4065 import, where the grant will be set to 0. While the original
4066 thread will update the grant it got when connecting. So if
4067 the former happened latter, the grant will be wrongly errased
4068 because of this race.
4073 Description: Checking Stale with correct fid
4074 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
4075 to check whether it is stale, because sometimes, we want the
4076 enqueue happened anyway, and op_data.fid2 will not be initialized.
4078 Severity : enhancement
4080 Description: update patchless client
4081 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
4084 Frequency : only with 2.4 kernel
4086 Description: random memory corruption
4087 Details : size of struct ll_inode_info is to big for union inode.u and this
4088 can be cause of random memory corruption.
4093 Description: Memory leak in recovery
4094 Details : Lov_mds_md was not free in an error handler in mds_create_object.
4095 It should also check obd_fail before fsfilt_start, otherwise if
4096 fsfilt_start return -EROFS,(failover mds during mds recovery).
4097 then the req will return with repmsg->transno = 0 and rc = EROFS.
4098 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
4099 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
4100 be freed no matter whether fsfilt_commit success or not.
4103 Frequency : only with large numbers of cores on a single node
4105 Description: Prevents from taking the superblock lock in llap_from_page for
4107 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
4108 for llap_from_page call in ll_removepage() prevents from taking
4109 the superblock lock for a soon died page.
4114 Description: Not check open intent error before release open handle
4115 Details : in some rare cases, the open intent error is not checked before
4116 release open handle, which may cause
4117 ASSERTION(open_req->rq_transno != 0), because it tries to release
4118 the failed open handle.
4123 Description: Set cat log bitmap only after create log success.
4124 Details : in some rare cases, the cat log bitmap is set too early. and it
4125 should be set only after create log success.
4129 Description: Accessing a block bevice can re-enable I/O when Lustre is
4130 tearing down a device.
4131 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
4136 Description: service threads may hog cpus when there are a lot of requests
4137 Details : Insert cond_resched to give other threads a chance to use some CPU
4142 Description: the cat log was not initialized in recovery
4143 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
4144 unlink log on mds will not be initialized until mds post
4145 recovery. And also in mds post recovery, the unlink log will
4146 initialization will be done asynchronausly, so there will be race
4147 between add unlink log and unlink log initialization.
4151 Description: brw_stats were being printed incorrectly
4152 Details : brw_stats were being printed as log2 but all of them were not
4153 recorded as log2. Also remove some code duplication arising from
4154 filter_tally_{read,write}.
4158 Frequency : rare, only in recovery.
4159 Description: ASSERTION(req->rq_type != LI_POISON) failed
4160 Details : imp_lock should be held while iterating over imp_sending_list for
4161 prevent destroy request after get timeout in ptlrpc_queue_wait.
4165 Description: replay-single.sh test 52 fails
4166 Details : A lock's skiplist need to be cleanup when it being unlinked
4167 from its resource list.
4172 Description: Short directio read returns full requested size rather than
4174 Details : Direct I/O operations should return actual amount of bytes
4175 transferred rather than requested size.
4177 Severity : enhancement
4179 Description: metadata RPC reduction (e.g. for rm performance)
4180 Details : decrease the amount of synchronous RPC between clients and servers
4181 by canceling conflicing lock before the operation on the client
4182 and packing thier handles into the main operation RPC to server.
4184 Severity : enhancement
4186 Description: add #ifdef HAVE_KERNEL_CONFIG_H
4187 Details : kernels from 2.6.19 not need include linux/config.h, but add
4188 include linux/autoconf.h in commpiler command line.
4190 Severity : enhancement
4192 Description: patchless client support for 2.6.22 kernel
4193 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
4194 removed. In this case we need drop using os depended interface to
4195 kmem_cache and use cfs_mem_cache API.
4200 Description: fix mal-formatted messages
4201 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
4205 Frequency : always in liblustre
4206 Description: wrong IS_ERR implementation in liblustre.h
4207 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
4212 Description: Correct condition for output debug message.
4213 Details : inode i_nlink equal zero is not enough for output message about
4214 disk corruption, i_ctime and i_mode should be also checked.
4218 Frequency : always in patchless client
4219 Description: add configure check for truncate_complete_page
4220 Details : improve checks for exported symbols. This allow run check without
4221 sources, but with Module.symvers shipped with kernel distribution.
4222 add check for truncate_complete_page used by patchless client.
4226 Description: sanity.sh test_77h fails with "test_77h file compare failed"
4227 Details : test_77h uses a file which was messed by other test case.
4231 Description: Not Check whether lov_tgts is NULL in some lov functions
4232 Details : Checking whether lov_tgts is NULL in some functions.
4236 Description: replace obdo_alloc() with OBDO_ALLOC macro
4237 Details : nothing special is done in obdo_alloc() function, and for
4238 debugging purpose, it needs to be replaced with macros.
4242 Description: bad return value and errno from fcntl call
4243 Details : In liblustre API, errno should be a negative value if error
4248 Description: ptlrpc_check_set() LBUG
4249 Details : In case of positive reply from server and failed client bulk
4250 callback after bulk transfer shouldn't LBUG, but process this
4251 request as erroneous.
4253 Severity : enhancement
4255 Description: VFS operations stats tool.
4256 Details : Tool which collects stats by tracking value written in pid,
4257 ppid, gid and uses llstat to generate output to plot graph using
4259 Updated lustre/utils/Makefile.am
4260 Added lustre/utils/ltrack_stats.c
4262 Severity : enhancement
4264 Description: 2.6.18 server support (lustre 1.6.1)
4265 Details : Support for 2.6.18 kernels on the server side.
4270 Description: ASSERTION(imp->imp_conn_current) failed
4271 Details : an assertion failure is hit if a client node boots and attempts to
4272 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
4275 Frequency : only for i686
4277 Description: 1.4.11 RC1 build fails for RHEL 4, i686
4278 Details : Fixed config variable for build.
4283 Description: Updated patchess for new RHEL4 kernel
4284 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
4285 Updated patch jbd-stats-2.6.9.patch
4286 Updated patch qsnet-rhel4-2.6.patch
4287 Updated patch quota-deadlock-on-pagelock-core.patch
4288 Updated patch vfs_intent-2.6-rhel4.patch
4289 Updated patch vfs_races-2.6-rhel4.patch
4290 Updated series file 2.6-rhel4-titech.series
4291 Updated series file 2.6-rhel4.series
4292 Updated kernel config files
4297 Description: lquota slave complains LBUG when reconnecting with mds
4299 Details : quota slave depends qctxt->lqc_import to send its quota request.
4300 This pointer will be invalid if mds did failover or broke its
4301 connect to osts, which leads to LBUG.
4304 Frequency : when qunit size is too small(less than 20M)
4306 Description: write is stopped by improper -EDQUOT
4307 Details : If the master is busy and qunit size is small enough(let's say 1M),
4308 the slave can not get quota from master on time, which will lead
4309 slave to trigger a -EQUOTA to client.
4314 Description: Deadlock during metadata tests
4315 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
4316 called with the per-dentry lock held.
4319 Frequency : SLES9 only
4321 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
4322 Details : set CONFIG_NR_CPUS to 128 instead of 8.
4324 Severity : enhancement
4326 Description: remove fs_prep_san_write operation and related patches
4327 Details : remove the ext3-san-jdike patches which are no longer useful.
4332 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
4333 Details : a disk corruption can cause the mballoc code to assert on a
4334 double free or other extent corruptions. Handle these with
4335 ext3_error() instead of assertions.
4338 Frequency : only with mballoc3 code and deep extent trees
4340 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
4341 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
4342 can cause the filesystem to be remounted read-only.
4347 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
4348 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
4349 few external users of this ioctl number is lfsck's call to
4350 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
4351 Add the old ioctl number to the handler so both old and new
4356 Description: parallel lock callbacks
4357 Details : Instead of sending blocking and completion callbacks as separated
4358 requests, adding them to a set and sending in parallel.
4362 Description: Disable most debugging by default
4363 Details : To improve performance, disable most logging (for debug purposes)
4364 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
4365 default, and HA includes fewer messages.
4369 Frequency : only run on patchless client.
4370 Description: use do_facet on sanity.sh for test handling recoverables errors
4371 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
4375 Description: hash tables for lists of nids, connections and uuids
4376 Details : Hash tables noticeably help when a lot of clients connect to a
4377 server, to faster identify duplicate connections or reconnects,
4378 also to faster find export to evict in manual eviction case.
4382 Description: Sometimes, when the server evict a client, and the client will
4383 not be evicted as soon as possible.
4384 Details : In enqueue req, the error was returned by intent, instead of
4385 rq_status which make ptlrpc layer not detect this error, and
4386 does not evict the client. So enqueue error should be returned
4390 Frequency : only at startup
4392 Description: Delay client connections to MDT until fist MDT->OST connect
4393 Details : If a client tried to create a new file before the MDT had
4394 connected to any OSTs, the create would return EIO. Now
4395 the client will simply block until the MDT connects to the
4396 first OST and the create can succeed.
4399 Frequency : only for SLES9
4401 Description: Routinely utilize latest Quadrics drivers in CFS releases
4402 Details : Update patch qsnet-suse-2.6.patch.
4405 Frequency : only for sles10
4407 Description: Update patches for SLES 10 SP1 kernel.
4408 Details : Update the patch vfs_intent-2.6-sles10.patch.
4413 Description: Routinely utilize latest Quadrics drivers in CFS releases
4414 Details : Update patch qsnet-rhel4-2.6.patch.
4417 Frequency : at statup only
4419 Description: mds_lov_synchronize race leads to various problems
4420 Details : simultaneous MDT->OST connections at startup can cause the
4421 sync to abort, leaving the OSC in a bad state.
4426 Description: Using wrong pointer in osc_brw_prep_request
4427 Details : Access to array[-1] can produce panic if kernel compiled with
4428 CONFIG_PAGE_ALLOC enabled
4430 Severity : enhancement
4432 Description: Async OSC create to avoid the blocking unnecessarily.
4433 Details : If a OST has no remain object, system will block on the creating
4434 when need to create a new object on this OST. Now, ways use
4435 pre-created objects when available, instead of blocking on an
4436 empty osc while others are not empty. If we must block, we block
4437 for the shortest possible period of time.
4441 Frequency : only in recovery
4442 Description: Mark OST as early accessible if his start SYNC.
4443 Details : osc_precreate return flag early accessible if oscc marked as
4444 OSCC_FLAG_SYNC_IN_PROGRESS.
4449 Description: Sometimes precreate code can triger create object on wrong ost
4450 Details : Wrong protected or not not restored variables aftre precreate loop
4451 can produce creation object on wrong ost.
4454 Frequency : oss recovery
4456 Description: llog_commit_thread cleanup should sync with llog_commit_thread
4458 Details : llog_commit_thread_count should be synced between llog_commit
4459 start and cleanup, so new llog_commit thread should not be started
4460 when llog_commit threads being stopped to avoid accessing some
4463 Severity : enhancement
4465 Description: Add printing inode info into message about error in writepage.
4467 --------------------------------------------------------------------------------
4469 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
4474 Frequency : on some architectures
4476 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
4477 Details : Uninitialized flags sometimes cause configuration commands to
4481 Frequency : patchless clients only
4483 Description: missing __iget() symbol export
4484 Details : The __iget() symbol export is missing. To avoid the need for
4485 this on patchless clients the deathrow inode reaper is turned
4486 off, and we depend on the VM to clean up old inodes. This
4487 dependency was during via the fix for bug 12181.
4492 Description: sanity.sh fail: test_52b
4493 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
4494 incorrect inode's flags to client.
4496 --------------------------------------------------------------------------------
4498 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
4500 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
4501 INTEROPERATE with older versions automatically. In many cases a
4502 special upgrade step is needed. Please read the
4503 user documentation before upgrading any part of a 1.4.x system.
4504 * WARNING: Lustre configuration and startup changes are required with
4505 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
4507 * Support for kernels:
4508 2.4.21-47.0.1.EL (RHEL 3),
4509 2.6.5-7.283 (SLES 9),
4510 2.6.9-42.0.10.EL (RHEL 4),
4511 2.6.12.6 vanilla (kernel.org),
4512 2.6.16.27-0.9 (SLES10)
4513 * Client support for unpatched kernels:
4514 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
4515 2.6.16 - 2.6.19 vanilla (kernel.org),
4516 2.6.9-42.0.8EL (RHEL 4)
4517 * Recommended e2fsprogs version: 1.39.cfs6
4518 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
4521 Severity : enhancement
4523 Description: Async OSC create to avoid the blocking unnecessarily.
4524 Details : If a OST has no remain object, system will block on the creating
4525 when need to create a new object on this OST. Now, ways use
4526 pre-created objects when available, instead of blocking on an
4527 empty osc while others are not empty. If we must block, we block
4528 for the shortest possible period of time.
4530 Severity : enhancement
4532 Description: MountConf
4533 Details : Lustre configuration is now managed via mkfs and mount
4534 commands instead of lmc and lconf. New obd types (MGS, MGC)
4535 are added for dynamic configuration management. See
4536 https://mail.clusterfs.com/wikis/lustre/MountConf for
4539 Severity : enhancement
4541 Description: dynamic OST addition
4542 Details : OSTs can now be added to a live filesystem
4544 Severity : enhancement
4546 Description: startup order invariance
4547 Details : MDTs and OSTs can be started in any order. Clients only
4548 require the MDT to complete startup.
4550 Severity : enhancement
4552 Description: parallel, asynchronous orphan cleanup
4553 Details : orphan cleanup is now performed in separate threads for each
4554 OST, allowing parallel non-blocking operation.
4556 Severity : enhancement
4558 Description: optimized stripe assignment
4559 Details : stripe assignments are now made based on ost space available,
4560 ost previous usage, and OSS previous usage, in order to try
4561 to optimize storage space and networking resources.
4563 Severity : enhancement
4565 Description: Permanently set tunables
4566 Details : All writable /proc/fs/lustre tunables can now be permanently
4567 set on a per-server basis, at mkfs time or on a live
4570 Severity : enhancement
4572 Description: Lustre message v2
4573 Details : Add lustre message format v2.
4575 Severity : enhancement
4577 Description: client OST exclusion list
4578 Details : Clients can be started with a list of OSTs that should be
4579 declared "inactive" for known non-responsive OSTs.
4583 Description: ENOENT returned for valid filehandle during dbench.
4584 Details : Check if a directory has children when invalidating dentries
4585 associated with an inode during lock cancellation. This fixes
4586 an incorrect ENOENT sometimes seen for valid filehandles during
4587 testing with dbench.
4590 Frequency : SFS test only (otherwise harmless)
4592 Description: SPEC SFS validation failure on NFS v2 over lustre.
4593 Details : Changes the blocksize for regular files to be 2x RPC size,
4594 and not depend on stripe size.
4596 Severity : enhancement
4598 Description: fine-grained SMP locking inside DLM
4599 Details : Improve DLM performance on SMP systems by removing the single
4600 per-namespace lock and replace it with per-resource locks.
4602 Severity : enhancement
4604 Description: don't hold multiple extent locks at one time
4605 Details : To avoid client eviction during large writes, locks are not
4606 held on multiple stripes at one time or for very large writes.
4607 Otherwise, clients can block waiting for a lock on a failed OST
4608 while holding locks on other OSTs and be evicted.
4610 Severity : enhancement
4612 Description: Multiple MD RPCs in flight.
4613 Details : Further unserialise some read-only MDT RPCs - learn about intents.
4614 To avoid overly-overloading MDT, introduce a limit on number of
4615 MDT RPCs in flight for a single client and add /proc controls
4616 to adjust this limit.
4618 Severity : enhancement
4620 Description: client read/write statistics
4621 Details : Add client read/write call usage stats for performance
4622 analysis of user processes.
4623 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
4624 file access. extents_stats shows chunk size distribution.
4625 extents_stats_per_process show chunk size distribution per
4628 Severity : enhancement
4630 Description: per-client statistics on server
4631 Details : Add ldlm and operations statistics for each client in
4632 /proc/fs/lustre/mds|obdfilter/*/exports/
4634 Severity : enhancement
4636 Description: improved MDT statistics
4637 Details : Add detailed MDT operations statistics in
4638 /proc/fs/lustre/mds/*/stats
4640 Severity : enhancement
4642 Description: VFS operations stats
4643 Details : Add client VFS call stats, trackable by pid, ppid, or gid
4644 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
4649 Description: Fix client-side osc byte counters
4650 Details : The osc read/write byte counters in
4651 /proc/fs/lustre/osc/*/stats are now working
4654 Frequency : always as root on SLES
4656 Description: Failure of copying files with lustre special EAs.
4657 Details : Client side always return success for setxattr call for lustre
4658 special xattr (currently only "trusted.lov").
4663 Description: Refcount LNET uuids
4664 Details : The global LNET uuid list grew linearly with every startup;
4665 refcount repeated list entries instead of always adding to
4668 Severity : enhancement
4670 Description: Dynamic service threads
4671 Details : Within a small range, start extra service threads
4672 automatically when the request queue builds up.
4675 Frequency : mixed-endian client/server environments
4677 Description: mixed-endian crashes
4678 Details : The new msg_v2 system had some failures in mixed-endian
4681 Severity : enhancement
4683 Description: Easy OST removal
4684 Details : OSTs can be permanently deactivated with e.g. 'lctl
4685 conf_param lustre-OST0001.osc.active=0'
4687 Severity : enhancement
4689 Description: MGS proc entries
4690 Details : Added basic proc entries for the MGS showing what filesystems
4693 Severity : enhancement
4695 Description: provide MGS failover
4696 Details : Added config lock reacquisition after MGS server failover.
4698 Severity : enhancement
4700 Description: add Linux 2.4 support
4701 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
4705 Description: a large application tries to do I/O to the same resource and dies
4706 in the middle of it.
4707 Details : Check the req->rq_arrival time after the call to
4708 ost_brw_lock_get(), but before we do anything about
4709 processing it & sending the BULK transfer request. This
4710 should help move old stale pending locks off the queue as
4711 quickly as obd_timeout.
4714 Frequency : when an incorrect nid is specified during startup
4716 Description: ptlrpc connect to non-existant node causes kernel crash
4717 Details : LNET can't be re-entered from an event callback, which
4718 happened when we expire a message after the export has been
4719 cleaned up. Instead, hand the zombie cleanup off to another
4722 Severity : enhancement
4724 Description: plain/inodebits lock performance improvement
4725 Details : Grouping plain/inodebits in granted list by their request modes
4726 and bits policy, thus improving the performance of search through
4730 Frequency : only if OST filesystem is corrupted
4732 Description: client incorrectly hits assertion in ptlrpc_replay_req()
4733 Details : for a short time RPCs with bulk IO are in the replay list,
4734 but replay of bulk IOs is unimplemented. If the OST filesystem
4735 is corrupted due to disk cache incoherency and then replay is
4736 started it is possible to trip an assertion. Avoid putting
4737 committed RPCs into the replay list at all to avoid this issue.
4740 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
4742 Description: System hang on startup
4743 Details : This bug allowed the liblustre (e.g. catamount) client to
4744 return to the app before handling all startup RPCs. This
4745 could leave the node unresponsive to lustre network traffic
4746 and manifested as a server ptllnd timeout.
4748 Severity : enhancement
4750 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
4751 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
4752 boolean to control whether to print peer debug info when a
4753 client's RPC times out.
4756 Frequency : only for kernels with patches from Lustre below 1.4.3
4758 Description: Remove old rdonly API
4759 Details : Remove old rdonly API which unused from at least lustre 1.4.3
4762 Frequency : only for devices with external journals
4764 Description: Set external device read-only also
4765 Details : During a commanded failover stop, we set the disk device
4766 read-only while the server shuts down. We now also set any
4767 external journal device read-only at the same time.
4770 Frequency : when upgrading from 1.4 while trying to change parameters
4772 Description: The wrong (new) MDC name was used when setting parameters for
4773 upgraded MDT's. Also allows changing of OSC (and MDC)
4774 parameters if --writeconf is specified at tunefs upgrade time.
4777 Frequency : when setting specific ost indicies
4779 Description: QOS code breaks on skipped indicies
4780 Details : Add checks for missing OST indicies in the QOS code, so OSTs
4781 created with --index need not be sequential.
4783 Severity : enhancement
4785 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
4786 Details : The uninit_groups feature works in conjunction with the kernel
4787 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
4788 up the pass1 processing of e2fsck. This is a read-only feature
4789 in ldiskfs2 only, so older kernels and current ldiskfs cannot
4790 mount filesystems that have had this feature enabled.
4792 Severity : enhancement
4794 Description: Improve multi-block allocation algorithm to avoid fragmentation
4795 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
4796 allocation locality and avoid filesystem fragmentation.
4798 ------------------------------------------------------------------------------
4800 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
4802 * Support for kernels:
4803 2.4.21-47.0.1.EL (RHEL 3)
4804 2.6.5-7.283 (SLES 9)
4805 2.6.9-42.0.10.EL (RHEL 4)
4806 2.6.12.6 vanilla (kernel.org)
4807 2.6.16.27-0.9 (SLES 10)
4808 * Recommended e2fsprogs version: 1.39.cfs5
4810 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
4814 Frequency : occasional, depends on client load and configuration
4815 Bugzilla : 12181, 12203
4816 Description: data loss for recently-modified files
4818 Details : In some cases it is possible that recently written or created
4819 files may not be written to disk in a timely manner (this should
4820 normally be within 30s unless client IO load is very high).
4821 The problem appears as zero-length files or files that are a
4822 multiple of 1MB in size after a client crash or client eviction
4823 that are missing data at the end of the file.
4825 This problem is more likely to be hit on clients where files are
4826 repeatedly created and unlinked in the same directory, clients
4827 have a large amount of RAM, have many CPUs, the filesystem has
4828 many OSTs, the clients are rebooted frequently, and/or the files
4829 are not accessed by other nodes after being written.
4831 The presence of the problem can be detected by looking at
4832 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
4833 smaller than the second (nr_unused) then dirty files will not
4834 be flushed automatically to disk. "sync; sleep 10" should be
4835 run several times on the node before unmounting it to update
4836 Lustre (this is also safe to run on nodes without this problem).
4838 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
4839 that can cause this same problem, so customers using that kernel
4840 also need to update the kernel in addition to Lustre. In order
4841 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
4843 It is normal that files written just before a client crash (less
4844 than 30s) may not yet have been flushed to disk, even for local
4848 Frequency : frequent on thin XT3 nodes
4850 Description: UUID collision on thin XT3 Linux nodes
4851 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
4852 randomly, since we relied on an insufficiently-seeded PRNG.
4857 Description: OSS hangs after "All ost request buffers busy"
4858 Details : A deadlock between quota and journal operations caused OSS
4859 hangs after printing "All ost request buffers busy."
4862 Frequency : always on liblustre builds
4864 Description: Cleanup compiler warnings on liblustre
4867 Frequency : always on liblustre builds on XT3
4869 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
4873 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
4874 > 32000 subdirectories
4875 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
4876 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
4877 subdirectory count crosses 32000. This will aid e2fsck to
4878 correctly handle more than 32000 subdirectories.
4881 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
4883 Description: System hang on startup
4884 Details : This bug allowed the liblustre (e.g. catamount) client to
4885 return to the app before handling all startup RPCs. This
4886 could leave the node unresponsive to lustre network traffic
4887 and manifested as a server ptllnd timeout.
4889 Severity : enhancement
4891 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
4892 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
4893 boolean to control whether to print peer debug info when a
4894 client's RPC times out.
4899 Description: make O_SYNC working on 2.6 kernels
4900 Details : 2.6 kernels use different method for mark pages for write,
4901 so need add a code to lustre for O_SYNC work.
4906 Description: Failure to close file and release space on NFS
4907 Details : Put inode details into lock acquired in ll_intent_file_open.
4908 Use mdc_intent_lock in ll_intent_open to properly
4909 detect all kind of errors unhandled by mdc_enqueue.
4914 Description: proc file read during shutdown sometimes raced obd removal,
4916 Details : Add lock to prevent obd access after proc file removal.
4919 Frequency : Only for files larger than 4GB on 32-bit clients.
4921 Description: improperly doing page alignment of locks
4922 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
4923 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
4928 Description: RPCs being resent when they shouldn't be
4929 Details : Some RPCs that should not be resent are being resent. This
4930 can cause inconsistencies in the RPC state machine. Do not
4931 resend such requests.
4934 Frequency : rare, only with NFS export
4936 Description: Crash on NFS re-export node
4937 Details : under very unusual load conditions an assertion is hit in
4938 ll_intent_file_open()
4941 Frequency : only if OST filesystem is corrupted
4943 Description: client incorrectly hits assertion in ptlrpc_replay_req()
4944 Details : for a short time RPCs with bulk IO are in the replay list,
4945 but replay of bulk IOs is unimplemented. If the OST filesystem
4946 is corrupted due to disk cache incoherency and then replay is
4947 started it is possible to trip an assertion. Avoid putting
4948 committed RPCs into the replay list at all to avoid this issue.
4953 Description: large O_DIRECT requests fail under memory pressure/fragmentation
4954 Details : Large single O_DIRECT read and write calls can fail to allocate
4955 a sufficiently large buffer to process the request. In case of
4956 allocation failure the allocation is retried with a smaller
4957 buffer and broken into smaller requests.
4959 Severity : enhancement
4961 Description: Add -o localflock option to simulate old noflock behaviour.
4962 Details : This will achieve local-only flock/fcntl locks coherentness.
4967 Description: versioning check is incomplete
4968 Details : Checking the version difference of client vs. server, report
4969 error if the gap is too big.
4974 Description: add support PG_writeback bit
4975 Details : add support for PG_writeback bit for Lustre, for more carefull
4976 work with page cache in 2.6 kernel. This also fix some deadlocks
4977 and remove hack for work O_SYNC with 2.6 kernel.
4979 Severity : enhancement
4981 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
4982 Details : The uninit_groups feature works in conjunction with the kernel
4983 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
4984 up the pass1 processing of e2fsck. This is a read-only feature
4985 in ldiskfs2 only, so older kernels and current ldiskfs cannot
4986 mount filesystems that have had this feature enabled.
4988 Severity : enhancement
4990 Description: Improve multi-block allocation algorithm to avoid fragmentation
4991 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
4992 allocation locality and avoid filesystem fragmentation.
4994 ------------------------------------------------------------------------------
4996 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
4998 * Support for kernels:
4999 2.6.9-42.0.3.EL (RHEL 4)
5000 2.6.5-7.276 (SLES 9)
5001 2.4.21-47.0.1.EL (RHEL 3)
5002 2.6.12.6 vanilla (kernel.org)
5003 2.6.16.21-0.8 (SLES10)
5004 * Recommended e2fsprogs version: 1.39.cfs2-0
5006 * The backwards-compatible /proc/sys/portals symlink has been removed
5007 in this release. Before upgrading, please ensure that you change
5008 any configuration scripts or /etc/sysctl.conf files that access
5009 /proc/sys/portals/* or sysctl portals.* to use the corresponding
5010 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
5011 in advance of the upgrade on any system running Lustre 1.4.6 or
5012 newer, since /proc/sys/lnet was added in that version.
5013 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
5017 Frequency : only when quota is used
5019 Description: avoid scanning export list for quota master
5020 Details : Change the algorithms to avoid scanning export list in order
5021 to improve the efficiency.
5024 Frequency : MDS failover only, very rarely
5026 Description: "went back in time" messages on mds failover
5027 Details : The greatest transno may be lost when the current operation
5028 finishes with an error (transno==0) and the client's last_rcvd
5029 record is over-written. Save the greatest transno in the
5030 mds_last_transno for this case.
5033 Frequency : always for specific kernels and striping counts
5035 Description: client may get "Matching packet too big" without ACL support
5036 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
5037 when trying to access files in certain configurations. The
5038 clients should in fact be denied when mounting because they do
5039 not understand ACLs.
5042 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
5044 Description: many clients connecting with IO in progress causes connect timeouts
5045 Details : Avoid synchronous journal commits to avoid delays caused by many
5046 clients connecting/disconnecting when bulk IO is in progress.
5047 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
5048 OST_IO_PORTAL to avoid delays behind potentially many pending
5052 Frequency : occasionally with multiple writers to a single file
5054 Description: shared writes to file may result in wrong size reported by stat()
5055 Details : Allow growing of kms when extent lock is cancelled
5058 Frequency : always with random mmap IO to multi-striped file
5060 Description: mmap write might be lost if we are writing to a 'hole' in stripe
5061 Details : Only if the hole is at the end of OST object so that kms is too
5062 small. Fix is to increase kms accordingly in ll_nopage.
5065 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
5067 Description: writes to a missing object would leak memory on the OST
5068 Details : If there is an inconsistency between the MDS and OST filesystems,
5069 such that the MDS references an object that doesn't exist, writes
5070 to that object will leak memory due to incorrect cleanup in the
5071 error handling path, eventually running out of memory on the OST.
5076 Description: Creating too long symlink causes lustre errors
5077 Details : Check symlink and name lengths before sending requests to MDS.
5080 Frequency : only if flock is enabled (not on by default)
5082 Description: posix locks not released on fd closure on 2.6.9+
5083 Details : We failed to add posix locks to list of inode locks on 2.6.9+
5084 kernels, this caused such locks not to be released on fd close and
5085 then assertions on fs unmount about still used locks.
5088 Frequency : MDS failover only, very rarely
5090 Description: clients may get ASSERTION(granted_lock != NULL)
5091 Details : When request was taking a long time, and a client was resending
5092 a getattr by name lock request. The were multiple lock requests
5093 with the same client lock handle and
5094 mds_getattr_name->fixup_handle_for_resent_request found one of the
5095 lock handles but later failed with ASSERTION(granted_lock != NULL).
5100 Description: handle->h_buffer_credits > 0, assertion failure
5101 Details : h_buffer_credits is zero after truncate, causing assertion
5102 failure. This patch extends the transaction or creates a new
5106 Frequency : NFS re-export or patchless client
5107 Bugzilla : 11179, 10796
5108 Description: Crash on NFS re-export node (__d_move)
5109 Details : We do not want to hash the dentry if we don't have a lock.
5110 But if this dentry is later used in d_move, we'd hit uninitialised
5111 list head d_hash, so we just do this to init d_hash field but
5112 leave dentry unhashed.
5115 Frequency : NFS re-export or patchless client
5117 Description: NFS exports has problem with symbolic link
5118 Details : lustre client didn't properly install dentry when re-exported
5119 to NFS or running patchless client.
5122 Frequency : NFS re-export or patchless client
5124 Description: Various nfs/patchless fixes.
5125 Details : fixes reuse disconected alias for lookup process - this fixes
5126 warning "find_exported_dentry: npd != pd",
5127 fix permission error with open files at nfs.
5128 fix apply umask when do revalidate.
5131 Frequency : occasional
5133 Description: Crash on NFS re-export node
5134 Details : calling clear_page() on the wrong pointer triggered oops in
5135 generic_mapping_read().
5138 Frequency : rarely, using O_DIRECT IO
5140 Description: unaligned directio crashes client with LASSERT
5141 Details : check for unaligned buffers before trying any requests.
5144 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
5146 Description: stale data returned from RAID cache
5147 Details : If only a small amount of IO is done to the RAID device before
5148 reading it again it is possible to get stale data from the RAID
5149 cache instead of reading it from disk.
5152 Frequency : always for sles10 kernel
5154 Description: sles10 support
5155 Details : ll_follow_link: compile fixes and using of nd_set_link
5156 under newer kernels.
5159 Frequency : depends on arch, kernel and compiler version, always on sles10
5162 Description: recursive or deep enough symlinks cause stack overflow
5163 Details : getting rid of large stack-allocated variable in
5167 Frequency : depends on hardware
5169 Description: lustre write performance loss in the SLES10 kernel
5170 Details : the performance loss is caused by using of write barriers in the
5171 ext3 code. The SLES10 kernel turns barrier support on by
5172 default. The fix is to undo that change for ldiskfs.
5174 ------------------------------------------------------------------------------
5176 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
5178 * Support for kernels:
5179 2.6.9-42.0.3EL (RHEL 4)
5180 2.6.5-7.276 (SLES 9)
5181 2.4.21-47.0.1.EL (RHEL 3)
5182 2.6.12.6 vanilla (kernel.org)
5186 Frequency : quota enabled and large files being deleted
5188 Description: releasing more than 4GB of quota at once hangs OST
5189 Details : If a user deletes more than 4GB of files on a single OST it
5190 will cause the OST to spin in an infinite loop. Release
5191 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
5196 Description: statfs data retrieved from /proc may be stale or zero
5197 Details : When reading per-device statfs data from /proc, in the
5198 {kbytes,files}_{total,free,avail} files, it may appear
5199 as zero or be out of date.
5202 Frequency : systems with MD RAID1 external journal devices
5204 Description: lconf's call to blkid is confused by RAID1 journal devices
5205 Details : Use the "blkid -l" flag to locate the MD RAID device instead
5206 of returning all block devices that match the journal UUID.
5209 Frequency : always, for aggregate stripe size over 4GB
5211 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
5212 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
5213 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
5216 Frequency : always if "lfs find" used on a local file/directory
5218 Description: "lfs find" segfaults if used on a local file/directory
5219 Details : The case where a directory component was not specified wasn't
5220 handled correctly. Handle this properly.
5223 Frequency : always on ppc64
5225 Description: the write to an ext3 filesystem mounted with mballoc got stuck
5226 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
5227 perform endianness conversion.
5230 Frequency : rarely (truncate to non-zero file size after write under load)
5231 Bugzilla : 10730, 10687
5232 Description: Files padded with zeros to next 4K multiple
5233 Details : With filesystems mounted using the "extents" option (2.6 kernels)
5234 it is possible that files that are truncated to a non-zero size
5235 immediately after being written are filled with zero bytes beyond
5236 the truncated size. No file data is lost.
5238 Severity : enhancement
5240 Description: Allow recovery/failover for liblustre clients.
5241 Details : liblustre clients were unaware of failover configurations until
5244 Severity : enhancement
5246 Description: user file locks should fail when not mounting with flock option
5247 Details : Set up an error-returning stub in ll_file_operations.lock field
5248 to prevent incorrect behaviour when client is mounted without
5249 flock option. Also, set up properly f_op->flock field for
5253 Frequency : always on ia64
5255 Description: "lfs df" loops on printing out MDS statfs information
5256 Details : The obd_ioctl_data was not initialized and in some systems
5257 this caused a failure during the ioctl that did not return
5258 an error. Initialize the struct and return an error on failure.
5261 Frequency : SLES 9 only
5263 Description: Error of copying files with lustre special EAs as root
5264 Details : Client side always return success for setxattr call for lustre
5265 special xattr (currently only "trusted.lov").
5268 Frequency : rarely on clusters with both ia64+i386 clients
5270 Description: ia64+i686 clients doing shared IO on the same file may LBUG
5271 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
5272 clients are doing concurrent writes to the same file it is
5273 possible that the ia64 clients may LASSERT because the OST
5274 extent locks are not PAGE_SIZE aligned. Ensure that grown
5275 locks are always aligned on the request boundary.
5278 Frequency : specific use, occasional
5280 Description: Overwriting in use executable truncates on-disk binary image
5281 Details : If one node attempts to overwrite an executable in use by
5282 another node, we now correctly return ETXTBSY instead of
5283 truncating the file.
5285 Severity : enhancement
5287 Description: Async OSC create to avoid the blocking unnecessarily.
5288 Details : If a OST has no remain object, system will block on the creating
5289 when need to create a new object on this OST. Now, ways use
5290 pre-created objects when available, instead of blocking on an
5291 empty osc while others are not empty. If we must block, we block
5292 for the shortest possible period of time.
5297 Description: chmod on Lustre root is propagated to other clients
5298 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
5299 invalid by the follow_mount time.
5302 Frequency : liblustre clients only
5304 Description: Race in 'instant cancel' lock handling could lead to such locks
5305 never to be granted in case of SMP MDS
5306 Details : Do not destroy not yet granted but cbpending locks in
5310 Frequency : replay/resend of open
5312 Description: non null lock assetion failure in mds_intent_policy
5313 Details : Trying to replay/resend lockless open requests resulted in
5314 mds_open() returning 0 with no lock. Now it sets a flag if
5315 a lock is going to be returned.
5317 Severity : enhancement
5319 Description: Checksum enhancements
5320 Details : New checksum enhancements allow for resending RPCs that failed
5323 Severity : enhancement
5325 Description: Tunables on number of dirty pages in cacche
5326 Details : Allow to set limit on number of dirty pages cached.
5331 Description: client crash on unmount - lock still has references
5332 Details : In some error handling cases it was possible to leak a lock
5333 reference on a client while accessing a file. This was not
5334 harmful to the client during operation, but would cause the
5335 client to crash when the filesystem is unmounted.
5338 Frequency : specific case, rare
5340 Description: ETXTBSY on mds though file not in use by client
5341 Details : ETXTBSY is no longer incorrectly returned when attempting to
5342 chmod or chown a directory that the user previously tried to
5343 execute or a currently-executing binary.
5346 Frequency : extremely rare except on liblustre-based clients
5348 Description: Lustre space not freed when files are deleted
5349 Details : Clean up open-unlinked files after client eviction. Previously
5350 the unlink was skipped and the files remained as orphans.
5355 Description: OST failure "would be an LBUG" in waiting_locks_callback()
5356 Details : In some cases it was possible to send a blocking callback to
5357 a client doing a glimpse, even though that client didn't get
5358 a lock granted. When the glimpse lock is cancelled on the OST
5359 the freed lock is left on the waiting list and corrupted the list.
5362 Frequency : all core dumps
5364 Description: Broke core dumps to lustre
5365 Details : Negative dentry may be unhashed if parent does not have UPDATE
5366 lock, but some callers, e.g. do_coredump, expect dentry to be
5367 hashed after successful create, hash it in ll_create_it.
5369 ------------------------------------------------------------------------------
5371 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
5373 * Support for kernels:
5374 2.6.9-42.0.2.EL (RHEL 4)
5375 2.6.5-7.276 (SLES 9)
5376 2.4.21-40.EL (RHEL 3)
5377 2.6.12.6 vanilla (kernel.org)
5381 Frequency : always on RHEL 3
5383 Description: Number of open files grows over time
5384 Details : The number of open files grows over time, whether or not
5385 Lustre is started. This was due to a filp leak introduced
5386 by one of our kernel patches.
5388 ------------------------------------------------------------------------------
5390 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
5392 * Support for kernels:
5393 2.6.9-42.EL (RHEL 4)
5394 2.6.5-7.267 (SLES 9)
5395 2.4.21-40.EL (RHEL 3)
5396 2.6.12.6 vanilla (kernel.org)
5401 Bugzilla : 5719, 9635, 9792, 9684
5402 Description: OST (or MDS) trips assertions in (re)connection under heavy load
5403 Details : If a server is under heavy load and cannot reply to new
5404 connection requests before the client resends the (re)connect,
5405 the connection handling code can behave badly if two service
5406 threads are concurrently handing separate (re)connections from
5407 the same client. Add better locking to the connection handling
5408 code, and ensure that only a single connection will be processed
5409 for a given client UUID, even if the lock is dropped.
5411 Severity : enhancement
5413 Description: add TCP zero-copy support to kernel
5414 Details : Add support to the kernel TCP stack to allow zero-copy bulk
5415 sends if the hardware supports scatter-gather and checksumming.
5416 This allows socklnd to do client-write and server-read more
5417 efficiently and reduce CPU utilization from skbuf copying.
5420 Frequency : only if NFS exporting from client
5422 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
5423 Details : If setting attributes on a file created under NFS that had
5424 never been opened it would be possible to oops the client
5425 if the file had no objects.
5429 Bugzilla : 9326, 10402, 10897
5430 Description: client crash in ptlrpcd_wake() thread when sending async RPC
5431 Details : It is possible that ptlrpcd_wake() dereferences a freed async
5432 RPC. In rare cases the ptlrpcd thread alread processed the RPC
5433 before ptlrpcd_wake() was called and the request was freed.
5436 Frequency : always for liblustre
5438 Description: liblustre client does MDS+OSTs setattr RPC for each write
5439 Details : When doing a write from a liblustre client, the client
5440 incorrectly issued an RPC to the MDS and each OST the file was
5441 striped over in order to update the timestamps. When writing
5442 with small chunks and many clients this could overwhelm the MDS
5443 with RPCs. In all cases it would slow down the write because
5444 these RPCs are unnecessary.
5446 Severity : enhancement
5448 Description: allow number of MDS service threads to be changed at module load
5449 Details : It is now possible to change the number of MDS service threads
5450 running. Adding "options mds mds_num_threads={N}" to the MDS's
5451 /etc/modprobe.conf will set the number of threads for the next
5452 time Lustre is restarted (assuming the "mds" module is also
5453 reloaded at that time). The default number of threads will
5454 stay the same, 32 for most systems.
5459 Description: OST crash if filesystem is unformatted or corrupt
5460 Details : If an OST is started on a device that has never been formatted
5461 or if the filesystem is corrupt and cannot even mount then the
5462 error handling cleanup routines would dereference a NULL pointer.
5467 Description: NULL pointer deref in llap_from_page.
5468 Details : get_cache_page_nowait can return a page with NULL (or otherwise
5469 incorrect) mapping if the page was truncated/reclaimed while it was
5470 searched for. Check for this condition and skip such pages when
5471 doing readahead. Introduce extra check to llap_from_page() to
5472 verify page->mapping->host is non-NULL (so page is not anonymous).
5475 Frequency : Sometimes when using sys_sendfile
5477 Description: "page not covered by a lock" warnings from ll_readpage
5478 Details : sendfile called ll_readpage without right page locks present.
5479 Now we introduced ll_file_sendfile that does necessary locking
5480 around call to generic_file_sendfile() much like we do in
5484 Frequency : with certain MDS communication failures at client mount time
5486 Description: NULL pointer deref after failed client mount
5487 Details : a client connection request may delayed by the network layer
5488 and not be sent until after the PTLRPC layer has timed out the
5489 request. If the client fails the mount immediately it will try
5490 to clean up before the network times out the request. Add a
5491 reference from the request import to the obd device and delay
5492 the cleanup until the network drops the request.
5495 Frequency : occasionally during client (re)connect
5497 Description: assertion failure during client (re)connect
5498 Details : processing a client connection request may be delayed by the
5499 client or server longer than the client connect timeout. This
5500 causes the client to resend the connection request. If the
5501 original connection request is replied in this interval, the
5502 client may trip an assertion failure in ptlrpc_connect_interpret()
5503 which thought it would be the only running connect process.
5506 Frequency : only with obd_echo servers and clients that are rebooted
5508 Description: kernel BUG accessing uninitialized data structure
5509 Details : When running an obd_echo server it did not start the ping_evictor
5510 thread, and when a client was evicted an uninitialized data
5511 structure was accessed. Start the ping_evictor in the RPC
5512 service startup instead of the OBD startup.
5514 Severity : enhancement
5515 Bugzilla : 10193 (patchless)
5516 Description: Remove dependency on various unexported kernel interfaces.
5517 Details : No longer need reparent_to_init, exit_mm, exit_files,
5518 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
5521 Frequency : rare (only users of deprecated and unsupported LDAP config)
5523 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
5524 Details : LDAP apparently contains 'lustreName' attributes instead of
5525 'name'. A simple remapping of the name is sufficient.
5528 Frequency : rare (only with non-default dump_on_timeout debug enabled)
5530 Description: waiting_locks_callback trips kernel BUG if client is evicted
5531 Details : Running with the dump_on_timeout debug flag turned on makes
5532 it possible that the waiting_locks_callback() can try to dump
5533 the Lustre kernel debug logs from an interrupt handler. Defer
5534 this log dumping to the expired_lock_main() thread.
5536 Severity : enhancement
5538 Description: Support NFS exporting on 2.6 kernels.
5539 Details : Implement non-rawops metadata methods for NFS server to use without
5540 changing NFS server code.
5543 Frequency : very rare (synthetic metadata workload only)
5545 Description: two racing renames might cause an MDS thread to deadlock
5546 Details : Running the "racer" program may cause one MDS thread to rename
5547 a file from being the source of a rename to being the target of
5548 a rename at exactly the same time that another thread is doing
5549 so, and the second thread has already enqueued these locks after
5550 doing a lookup of the target and is trying to relock them in
5551 order. Ensure that we don't try to re-lock the same resource.
5554 Frequency : only very large systems with liblustre clients
5556 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
5557 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
5558 all OSTs in parallel. This allows the eviction of stale liblustre
5559 clients to proceed much faster than if they were done in series,
5560 and also offers similar improvements for other set_info RPCs.
5565 Description: excessive CPU usage during initial read phase on client
5566 Details : During the initial read phase on a client, it would agressively
5567 retry readahead on the file, consuming too much CPU and impacting
5568 performance (since 1.4.5.8). Improve the readahead algorithm
5569 to avoid this, and also improve some other common cases (read
5570 of small files in particular, where "small" is files smaller than
5571 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
5576 Description: MDS crash when receiving packet with unknown intent.
5577 Details : Do not LBUG in unknown intent case, just return -EFAULT
5579 Severity : enhancement
5580 Bugzilla : 9293, 9385
5581 Description: MDS RPCs are serialised on client. This is unnecessary for some.
5582 Details : Do not serialize getattr (non-intent version) and statfs.
5585 Frequency : occasional, when OST network is overloaded/intermittent
5587 Description: client evicted by OST after bulk IO timeout
5588 Details : If a client sends a bulk IO request (read or write) the OST
5589 may evict the client if it is unresposive to its data GET/PUT
5590 request. This is incorrect if the network is overloaded (takes
5591 too long to transfer the RPC data) or dropped the OST GET/PUT
5592 request. There is no need to evict the client at all, since
5593 the pinger and/or lock callbacks will handle this, and the
5594 client can restart the bulk request.
5597 Frequency : Always when mmapping file with no objects
5599 Description: client crashes when mmapping file with no objects
5600 Details : Check that we actually have objects in a file before doing any
5601 operations on objects in ll_vm_open, ll_vm_close and
5607 Description: Request leak when working with deleted CWD
5608 Details : Introduce advanced request refcount tracking for requests
5609 referenced from lustre intent.
5611 Severity : Enhancement
5613 Description: Cache open file handles on client.
5614 Details : MDS now will return special lock along with openhandle, if
5615 requested and client is allowed to hold openhandle, even if unused,
5616 until such a lock is revoked. Helps NFS a lot, since NFS is opening
5617 closing files for every read/write openration.
5619 Severity : Enhancement
5621 Description: Cache open negative dentries on client when possible.
5622 Details : Guard negative dentries with UPDATE lock on parent dir, drop
5623 negative dentries on lock revocation.
5628 Description: Remounting a client read-only wasn't possible with a zconf mount
5629 Details : It wasn't possible to remount a client read-only with llmount.
5631 Severity : enhancement
5632 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
5633 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
5634 MPICH is included to add Lustre-specific ADIO interfaces.
5635 This is based closely on the UFS ADIO layer and only differs
5636 in file creation, in order to allow the OST striping to be set.
5637 This is user-contributed code and not supported by CFS.
5642 Description: extended inode attributes (immutable, append-only) work improperly
5643 when 2.4 and 2.6 kernels are used on client/server or vice versa
5644 Details : Introduce kernel-independent values for these flags.
5646 Severity : enhancement
5649 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
5650 Details : Many of the /proc/ tunables can only be tuned at a megabyte
5651 granularity. Now, Fractional MB granularity is be supported,
5652 this is very useful for low memory system.
5654 Severity : enhancement
5656 Description: Getattr by fid
5657 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
5658 locks, avoids extra getattr rpc requests to MDS, allows '/' to
5659 have locks and avoids getattr rpc requests for it on every stat.
5662 Frequency : Always, for filesystems larger than 2TB
5664 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
5665 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
5666 filesystems larger than 2TB to be mounted with mballoc enabled.
5669 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
5671 Description: filesystem corruption for non-standard kernels and very large OSTs
5672 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
5673 filesystems larger than 2TB is mounted then the kernel will
5674 silently corrupt the start of the filesystem. CONFIG_LBD is
5675 enabled for all CFS-supported kernels, but the possibility of
5676 this happening with a modified kernel config exists.
5678 Severity : enhancement
5680 Description: add client O_DIRECT support for 2.6 kernels
5681 Details : It is now possible to do O_DIRECT reads and writes to files
5682 in the Lustre client mountpoint on 2.6 kernel clients.
5684 Severity : enhancement
5686 Description: parallel glimpse, setattr, statfs, punch, destroy requests
5687 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
5688 parallel, not waiting for response from every OST before sending
5689 a rpc to the next OST.
5694 Description: setattr vs write race when updating file timestamps
5695 Details : Client processes that update a file timestamp into the past
5696 right after writing to the file (e.g. tar) it is possible that
5697 the updated file modification time can be reset to the current
5698 time due to a race between processing the setattr and write RPC.
5700 Severity : enhancement
5702 Description: Bring 'lfs find' closer in line with regular Linux find.
5703 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
5704 -print0 options and obtains all the needed info through the lustre
5707 Severity : enhancement
5709 Description: support up to 1024 configured devices on one node
5710 Details : change obd_dev array from statically allocated to dynamically
5711 allocated structs as they are first used to reduce memory usage
5716 Description: Flush dirty partially truncated pages during truncate
5717 Details : Immediatelly flush partially truncated pages in filter_setattr,
5718 this way we completely avoid having any pages in page cache on OST
5719 and can retire ugly workarounds during writes to flush such pages.
5724 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
5725 Details : protect inode from truncation within vfs_unlink() context
5726 just take a reference before calling vfs_unlink() and release it
5727 when parent's i_sem is free.
5730 Frequency : always, if extents are used on OSTs
5732 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
5733 Details : OSTs using the extents format would not zero the high 16 bits of
5734 the index physical block number. This is not a problem for any
5735 OST filesystems smaller than 16TB, and no kernels support ext3
5736 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
5737 new/modified files) and can be fixed for existing filesystems
5738 with e2fsprogs-1.39-cfs1.
5743 Description: import connection selection may be incorrect if timer wraps
5744 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
5745 import connections to be ignored if the 32-bit jiffies counter
5746 wraps. Use a 64-bit jiffies counter.
5749 Frequency : during server recovery
5751 Description: crash after server is denying duplicate export
5752 Details : If clients are resending connect requests to the server, the
5753 server refuses to allow a client to connect multiple times.
5754 Fixed a bug in the handling of this case.
5757 Frequency : very large clusters immediately after boot
5759 Description: LNET request buffers exhausted under heavy short-term load
5760 Details : If a large number of client requests are generated on a service
5761 that has previously never seen so many requests it is possible
5762 that the request buffer growth cannot keep up with the spike in
5763 demand. Instead of dropping incoming requests, they are held in
5764 the LND until the RPC service can accept more requests.
5767 Frequency : Sometimes during replay
5769 Description: Assertion failure in ll_local_open after replay.
5770 Details : If replay happened on an open request reply before we were able
5771 to set replay handler, reply will become not swabbed tripping the
5772 assertion in ll_local_open. Now we set the handler right after
5773 recognising of open request
5776 Frequency : very rare
5778 Description: kernel reports "badness in vsnprintf"
5779 Details : Reading from the "recovery_status" /proc file in small chunks
5780 may cause a negative length in lprocfs_obd_rd_recovery_status()
5781 call to vsnprintf() (which is otherwise harmless). Exit early
5782 if there is no more space in the output buffer.
5784 Severity : enhancement
5786 Description: clear OBD RPC statistics by writing to them
5787 Details : It is now possible to clear the OBD RPC statistics by writing
5788 to the "stats" file.
5793 Description: Client mtime is not the same on different clients after utimes
5794 Details : In some cases, the client was using the utimes() syscall on
5795 a file cached on another node. The clients now validate the
5796 ctime from the MDS + OSTs to determine which one is right.
5801 Description: Inability to activate failout mode
5802 Details : lconf script incorrectly assumed that in python string's numeric
5803 value is used in comparisons.
5806 Frequency : always with multiple stripes per file
5808 Description: Inefficient object allocation for mutli-stripe files
5809 Details : When selecting which OSTs to stripe files over, for files with
5810 a stripe count that divides evenly into the number of OSTs,
5811 the MDS is always picking the same starting OST for each file.
5812 Return the OST selection heuristic to the original design.
5817 Description: mount failures may take full timeout to return an error
5818 Details : Under some heavy load conditions it is possible that a
5819 failed mount can wait for the full obd_timeout interval,
5820 possibly several minutes, before reporting an error.
5821 Instead return an error as soon as the status is known.
5823 ------------------------------------------------------------------------------
5825 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
5827 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
5828 INTEROPERATE with older versions automatically. Please read the
5829 user documentation before upgrading any part of a live system.
5830 * WARNING: Lustre networking configuration changes are required with
5831 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
5834 * Support for kernels:
5835 2.6.9-22.0.2.EL (RHEL 4)
5836 2.6.5-7.244 (SLES 9)
5837 2.6.12.6 vanilla (kernel.org)
5840 Severity : enhancement
5841 Bugzilla : 7981/8208
5842 Description: Introduced Lustre Networking (LNET)
5843 Details : LNET is new networking infrastructure for Lustre, it includes
5844 a reorganized network configuration mode (see the user
5845 documentation for full details) as well as support for routing
5846 between different network fabrics. Lustre Networking Devices
5847 (LNDs) for the supported network fabrics have also been
5848 created for this new infrastructure.
5850 Severity : enhancement
5851 Description: Introduced Access control lists
5852 Details : clients can set ACLs on files and directories in order to have
5853 more fine-grained permissions than the standard Unix UGO+RWX.
5854 The MDS must be started with the "-o acl" mount option.
5856 Severity : enhancement
5857 Description: Introduced filesystem quotas
5858 Details : Administrators may now establish per-user quotas on the
5861 Severity : enhancement
5863 Description: Configuration change for the XT3
5864 The PTLLND is now used to run Lustre over Portals on the XT3
5865 The configure option(s) --with-cray-portals are no longer used.
5866 Rather --with-portals=<path-to-portals-includes> is used to
5867 enable building on the XT3. In addition to enable XT3 specific
5868 features the option --enable-cray-xt3 must be used.
5873 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
5874 Details : A race between allocating a new llcd and re-getting the llcd_lock
5875 allowed another thread to grab newly-allocated llcd.
5877 Severity : enhancement
5879 Description: 2.6 OST async journal commit and locking fix to improve performance
5880 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
5881 2.6 kernels are now async as they already were in 2.4 kernels so
5882 that they can commit concurrently with the network bulk transfer.
5883 For block-allocated files the filter allocation semaphore is held
5884 to avoid filesystem fragmentation during allocation. BKL lock
5885 removed for 2.6 xattr operations where it is no longer needed.
5890 Description: lconf incorrectly determined whether two IP networks could talk
5891 Details : In some more complicated routing and multiple-network
5892 configurations, lconf will avoid trying to make a network
5893 connection to a disjoint part of the IP space. It was doing the
5894 math incorrectly for one set of cases.
5899 Description: Fix for potential infinite loop processing records in an llog.
5900 Details : If an llog record is corrupted/zeroed, it is possible to loop
5901 forever in llog_process(). Validate the llog record length
5902 and skip the remainder of the block on error.
5905 Frequency : occasional (liblustre only)
5907 Description: liblustre could not open files whose last component is a symlink
5908 Details : sysio_path_walk() would incorrectly pass the open intent to
5909 intermediate path components.
5912 Frequency : rare (liblustre only with non-standard tuning)
5913 Bugzilla : 7201 (7350)
5914 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
5915 Details : llu_lookup_finish_locks() tries to set lock data on a lock
5916 after it has been released, only do this for referenced locks
5918 Severity : enhancement
5920 Description: specifying an (invalid) directory default stripe_size of -1
5921 would reset the directory default striping
5922 Details : stripe_size -1 was used internally to signal directory stripe
5923 removal, now use "all default" to signal dir stripe removal
5924 as a directory striping of "all default" is not useful
5927 Frequency : common for large clusters running liblustre clients
5929 Description: doing an ls when liblustre clients are running is slow
5930 Details : sending a glimpse AST to a liblustre client waits for every AST
5931 to time out, as liblustre clients will not respond. Since they
5932 cannot cache data we refresh the OST lock LVB from disk instead.
5934 Severity : enhancement
5936 Description: doing an ls at the same time as file IO can be slow
5937 Details : enqueue and other "small" requests can be blocked behind many
5938 large IO requests. Create a new OST IO portal for non-IO
5939 requests so they can be processed faster.
5942 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
5944 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
5945 created file with mode 000
5946 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
5947 allow an ftruncate() as a non-root user to a file with mode 000.
5948 Lustre now respects this flag to disable mode checks when
5949 truncating a file owned by the user
5952 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
5955 Description: Revoking locks from clients that went dead or catatonic might take
5957 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
5958 cancellation of such locks instant on servers without waiting for
5959 any reply from clients. Clients drops these locks when cancel
5960 notification from server is received without replying.
5963 Frequency : liblustre-only, when liblustre client dies or becomes busy
5965 Description: Doing ls on Linux clients can take a long time with active
5967 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
5968 granting such locks to it in the first place if possible. Locks
5969 are taken by proxy on the OST during the read or write and
5970 dropped immediately afterward. Add connect flags handling, do
5971 not grant locks to liblustre clients for glimpse ASTs.
5973 Severity : enhancement
5975 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
5976 Details : Existing read-ahead algorithm is tuned for the case of streamlined
5977 sequential reads and behaves badly with applications doing random
5978 reads. Improve it by reading ahead at least read region, and
5979 avoiding excessive large RPC for small reads.
5981 Severity : enhancement
5983 Description: Creating more than 1000 files for a single job may cause a load
5984 imbalance on the OSTs if there are also a large number of OSTs.
5985 Details : qos_prep_create() uses an OST index reseed value that is an
5986 even multiple of the number of available OSTs so that if the
5987 reseed happens in the middle of the object allocation it will
5988 still utilize the OSTs as uniformly as possible.
5993 Description: OST or MDS may oops in ping_evictor_main()
5994 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
5995 but doesn't restart at beginning of obd_exports_timed list
5998 Severity : enhancement
6000 Description: improve by-nid export eviction on the MDS and OST
6001 Details : allow multiple exports with the same NID to be evicted at one
6002 time without re-searching the exports list.
6005 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
6007 Description: MDS may oops in groups_free()
6008 Details : in rare race conditions a newly allocated group_info struct is
6009 freed again, and this can be NULL. The 2.4 compatibility code
6010 for groups_free() checked for a NULL pointer, but 2.6 did not.
6013 Frequency : common for liblustre clients doing little filesystem IO
6014 Bugzilla : 9352, 7313
6015 Description: server may evict liblustre clients accessing contended locks
6016 Details : if a client is granted a lock or receives a completion AST
6017 with a blocking AST already set it would not reply to the AST
6018 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
6021 Frequency : lfs setstripe, only systems with more than 160 OSTs
6023 Description: unable to set striping with a starting offset beyond OST 160
6024 Details : llapi_create_file() incorrectly limited the starting stripe
6025 index to the maximum single-file stripe count.
6028 Frequency : LDAP users only
6030 Description: lconf did not handle in-kernel recovery with LDAP properly
6031 Details : lconf/LustreDB get_refs() is searching the wrong namespace
6033 Severity : enhancement
6035 Description: bind OST threads to NUMA nodes to improve performance
6036 Details : all OST threads are uniformly bound to CPUs on a single NUMA
6037 node and do their allocations there to localize memory access
6039 Severity : enhancement
6041 Description: llmount can determine client NID directly from Myrinet (GM)
6042 Details : the client NID code from gmnalnid was moved directly into
6043 llmount, removing the need to use this or specifying the
6044 client NID explicitly when mounting GM clients with zeroconf
6047 Frequency : if client is started with down MDS
6049 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
6050 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
6051 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
6052 ptlrpc_import_delayed_req() because we didn't check if the
6053 request was interrupted, and we also didn't break out of the
6054 event loop if there was no timeout
6059 Description: data loss during non-page-aligned writes to a single file from
6060 both multiple nodes and multiple threads on one node at same time
6061 Details : updates to KMS and lsm weren't protected by common lock. Resulting
6062 inconsistency led to false short-reads, that were cached and later
6063 used by ->prepare_write() to fill in partially written page,
6064 leading to data loss.
6067 Frequency : always, if lconf --abort_recovery used
6069 Description: lconf --abort_recovery fails with 'Operation not supported'
6070 Details : lconf was attempting to abort recovery on the MDT device and not
6073 Severity : enhancement
6075 Description: remove cleanup logs
6076 Details : replace lconf-generated cleanup logs with lustre internal
6077 cleanup routines. Eliminates the need for client-cleanup and
6080 Severity : enhancement
6082 Description: add support for EAs (user and system) on lustre filesystems
6083 Details : it is now possible to store extended attributes in the Lustre
6084 client filesystem, and with the user_xattr mount option it
6085 is possible to allow users to store EAs on their files also
6087 Severity : enhancement
6089 Description: Add possibility (config option) to show minimal available OST free
6091 Details : When compiled with --enable-mindf configure option, statfs(2)
6092 (and so, df) will return least minimal free space available from
6093 all OSTs as amount of free space on FS, instead of summary of
6094 free spaces of all OSTs.
6096 Severity : enhancement
6098 Description: do not expand extent locks acquired on OST-side
6099 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
6100 by server: they are not cached anyway.
6103 Frequency : when mmap is used/binaries executed from Lustre
6105 Description: Unmmap pages before throwing them away from read cache.
6106 Details : llap_shrink cache now attempts to unmap pages before discarding
6107 them (if unmapping failed - do not discard). SLES9 kernel has
6108 extra checks that trigger if this unmapping is not done first.
6113 Description: lconf didn't resolve symlinks before checking to see whether a
6114 given mountpoint was already in use
6117 Frequency : when migrating failover services
6118 Bugzilla : 6395, 9514
6119 Description: When migrating a subset of services from a node (e.g. failback
6120 from a failover service node) the remaining services would
6121 time out and evict clients.
6122 Details : lconf --force (implied by --failover) sets the global obd_timeout
6123 to 5 seconds in order to quickly disconnect, but this caused
6124 other RPCs to time out too quickly. Do not change the global
6125 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
6127 Severity : enhancement
6128 Frequency : if MDS is started with down OST
6129 Bugzilla : 9439,5706
6130 Description: Allow startup/shutdown of an MDS without depending on the
6131 availability of the OSTs.
6132 Details : Asynchronously call mds_lov_synchronize during MDS startup.
6133 Add appropriate locking and lov-osc refcounts for safe
6134 cleaning. Add osc abort_inflight calls in case the
6135 synchronize never started.
6138 Frequency : occasional (Cray XT3 only)
6140 Description: root not authorized to access files in CRAY_PORTALS environment
6141 Details : The client process capabilities were not honoured on the MDS in
6142 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
6143 been accessed by an authorized user then root was able to access
6144 the file on the local client also. The root user capabilities
6145 are now allowed on the MDS, as this environment has secure UID.
6148 Frequency : occasional
6150 Description: ldiskfs "too long searching" message happens too often
6151 Details : A debugging message (otherwise harmless) prints too often on
6152 the OST console. This has been reduced to only happen when
6153 there are fragmentation problems on the filesystem.
6158 Description: Division by zero in statfs when all OSCs are inactive
6159 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
6160 lov_statfs divides by value returned by lov_get_stripecnt().
6164 Bugzilla : 9489, 3273
6165 Description: First write from each client to each OST was only 4kB in size,
6166 to initialize client writeback cache, which caused sub-optimal
6167 RPCs and poor layout on disk for the first writen file.
6168 Details : Clients now request an initial cache grant at (re)connect time
6169 and so that they can start streaming writes to the cache right
6170 away and always do full-sized RPCs if there is enough data.
6171 If the OST is rebooted the client also re-establishes its grant
6172 so that client cached writes will be honoured under the grant.
6177 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
6178 Details : Now I/O RPCs go to different portal number and (presumably) fast
6179 lock requests (and glimses) and other RPCs get their own service
6180 threads pool that should be able to service those RPCs
6183 Severity : enhancement
6185 Description: Ability to exchange lustre version between client and servers and
6186 issue warnings at client side if client is too old. Also for
6187 liblustre clients there is ability to refuse connection of too old
6189 Details : New 'version' field is added to connect data structure that is
6190 filled with version info. That info is later checked by server and
6194 Frequency : rare, liblustre only.
6195 Bugzilla : 9296, 9581
6196 Description: Two simultaneous writes from liblustre at offset within same page
6197 might proceed at the same time overwriting eachother with stale
6199 Details : I/O lock withing llu_file_prwv was released too early, before data
6200 actually was hitting the wire. Extended lock-holding time until
6201 server acknowledges receiving data.
6204 Frequency : extremely rare. Never observed in practice.
6206 Description: avoid generating lustre_handle cookie of 0.
6207 Details : class_handle_hash() generates handle cookies by incrementing
6208 global counter, and can hit 0 occasionaly (this is unlikely, but
6209 not impossible, because initial value of cookie counter is
6210 selected randonly). Value of 0 is used as a sentinel meaning
6211 "unassigned handle" --- avoid it. Also coalesce two critical
6212 sections in this function into one.
6214 Severity : enhancement
6216 Description: allow liblustre clients to delegate truncate locking to OST
6217 Details : To avoid overhead of locking, liblustre client instructs OST to
6218 take extent lock in ost_punch() on client's behalf. New connection
6219 flag is added to handle backward compatibility.
6221 Severity : enhancement
6222 Bugzilla : 4928, 7341, 9758
6223 Description: allow number of OST service threads to be specified
6224 Details : a module parameter allows the number of OST service threads
6225 to be specified via "options ost ost_num_threads={N}" in the
6226 OSS's /etc/modules.conf or /etc/modprobe.conf.
6230 Bugzilla : 6146, 9635, 9895
6231 Description: servers crash with bad pointer in target_handle_connect()
6232 Details : In rare cases when a client is reconnecting it was possible that
6233 the connection request was the last reference for that export.
6234 We would temporarily drop the export reference and get a new
6235 one, but this may have been the last reference and the export
6236 was just destroyed. Get new reference before dropping old one.
6238 Severity : enhancement
6239 Frequency : if client is started with failover MDS
6241 Description: Allow multiple MDS hostnames in the mount command
6242 Details : Try to read the configuration from all specified MDS
6243 hostnames during a client mount in case the "primary"
6246 Severity : enhancement
6248 Description: Stop sending data to evicted clients as soon as possible.
6249 Details : Check if the client we are about to send or are sending data to
6250 was evicted already. (Check is done every second of waiting,
6251 for which l_wait_event interface was extended to allow checking
6252 of exit condition at specified intervals).
6255 Frequency : rare, normally only when NFS exporting is done from client
6257 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
6259 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
6262 Frequency : always, if srand() is called before liblustre initialization
6264 Description: Liblustre uses system PRNG disturbing its usage by user application
6265 Details : Introduce internal to lustre fast and high-quality PRNG for
6266 lustre usage and make liblustre and some other places in generic
6267 lustre code to use it.
6269 Severity : enhancement
6270 Bugzilla : 9477, 9557, 9870
6271 Description: Verify that the MDS configuration logs are updated when xml is
6272 Details : Check if the .xml configuration logs are newer than the config
6273 logs stored on the MDS and report an error if this is the case.
6274 Request --write-conf, or allow starting with --old_conf.
6276 Severity : enhancement
6278 Description: Handle symlinks in the path when checking if Lustre is mounted.
6279 Details : Resolve intermediate symlinks when checking if a client has
6280 mounted a filesystem to avoid duplicate client mounts.
6285 Description: lconf can hit an error exception but still return success.
6286 Details : The lconf command catches the Command error exception at the top
6287 level script context and will exit with the associated exit
6288 status, but doesn't ensure that this exit status is non-zero.
6293 Description: failure of ptlrpc thread startup can cause oops
6294 Details : Starting a ptlrpc service thread can fail if there are a large
6295 number of threads or the server memory is very fragmented.
6296 Handle this without oopsing.
6299 Frequency : always, only if liblustre and non-default acceptor port was used
6301 Description: liblustre cannot connect to servers with non-default acceptor port
6302 Details : tcpnal_set_default_params() was not called and was therefore
6303 ignoring the environment varaible TCPNAL_PORT, as well as other
6304 TCPNAL_ environment variables
6309 Description: two objects could be created on the same OST for a single file
6310 Details : If an OST is down, in some cases it was possible to create two
6311 objects on a single OST for a single file. No problems other
6312 than potential performance impact and spurious error messages.
6316 Bugzilla : 5681, 9562
6317 Description: Client may oops in ll_unhash_aliases
6318 Details : Client dcache may become inconsistent in race condition.
6319 In some cases "getcwd" can fail if the current directory is
6325 Description: Inode refcounting problems in NFS export code
6326 Details : link_raw functions used to call d_instantiate without obtaining
6327 extra inode reference first.
6331 Bugzilla : 9942, 9903
6332 Description: Referencing freed requests leading to crash, memleaks with NFS.
6333 Details : We used to require that call to ll_revalidate_it was always
6334 followed by ll_lookup_it. Also with revalidate_special() it is
6335 possible to call ll_revalidate_it() twice for the same dentry
6336 even if first occurence returned success. This fix changes semantic
6337 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
6338 reference on a request referred from the intent.
6339 ll_intent_release() then releases such a request.
6342 Frequency : rare, normally benchmark loads only
6344 Description: unlinked inodes were kept in memory on the client
6345 Details : If a client is repeatedly creating and unlinking files it
6346 can accumulate a lot of stale inodes in the inode slab cache.
6347 If there is no other client load running this can cause the
6348 client node to run out of memory. Instead flush old inodes
6349 from client cache that have the same inode number as a new inode.
6352 Frequency : SLES9 2.6.5 kernel and long filenames only
6353 Bugzilla : 9969, 10379
6354 Description: utime reports stale NFS file handle
6355 Details : SLES9 uses out-of-dentry names in some cases, which confused
6356 the lustre dentry revalidation. Change it to always use the
6360 Frequency : rare, unless heavy write-truncate concurrency is continuous
6361 Bugzilla : 4180, 6984, 7171, 9963, 9331
6362 Description: OST becomes very slow and/or deadlocked during object unlink
6363 Details : filter_destroy() was holding onto the parent directory lock
6364 while truncating+unlinking objects. For very large objects this
6365 may block other threads for a long time and slow overall OST
6366 responsiveness. It may also be possible to get a lock ordering
6367 deadlock in this case, or run out of journal credits because of
6368 the combined truncate+unlink. Solution is to do object truncate
6369 first in one transaction without parent lock, and then do the
6370 final unlink in a new transaction with the parent lock. This
6371 reduces the lock hold time dramatically.
6374 Frequency : rare, 2.4 kernels only
6376 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
6377 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
6378 before unmounting their filesystem in order to avoid deadlock.
6379 The kernel_locked() function in 2.4 kernels only checks whether
6380 the kernel lock is held, not whether it is this process that is
6381 holding it as 2.6 kernels do.
6386 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
6387 Details : The client ptlrpc code may be trying to reconnect to a down
6388 server before a previous connection attempt has timed out.
6389 Increase the reconnect interval to be longer than the connection
6390 timeout interval to avoid sending duplicate connections to
6394 Frequency : echo_client brw_test command
6396 Description: fix echo_client to work with OST preallocated code
6397 Details : OST preallocation code (5137) didn't take echo_client IO path
6398 into account: echo_client calls filter methods outside of any
6399 OST thread and, hence, there is no per-thread preallocated
6400 pages and buffers to use. Solution: hijack pga pages for IO. As
6401 a byproduct, this avoids unnecessary data copying.
6405 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
6406 Description: Client can oops in mdc_commit_close() after open replay
6407 Details : It was possible for the MDS to return an open request with no
6408 transaction number in mds_finish_transno() if the client was
6409 evicted, but without actually returning an error. Clients
6410 would later try to replay that open and may trip an assertion
6411 Simplify the client close codepath, and always return an error
6412 from the MDS in case the open is not successful.
6415 Frequency : rare, 2.6 OSTs only
6417 Description: OST may deadlock under high load on fragmented files
6418 Details : If there was a heavy load and highly-fragmented OST filesystems
6419 it was possible to have all the OST threads deadlock waiting on
6420 allocation of biovecs, because the biovecs were not released
6421 until the entire RPC IO was completed. Instead, release biovecs
6422 as soon as they are complete to ensure forward IO progress.
6424 Severity : enhancement
6426 Description: Support for specifying external journal device at mount
6427 Details : If an OST or MDS device is formatted with an external journal
6428 device, this device major/minor is stored in the ext3 superblock
6429 and may not be valid for failover. Allow detecting and
6430 specifying the external journal at mount time.
6435 Description: Mounting an MDS with pending unlinked files may cause oops
6436 Details : target_finish_recovery() calls mds_postrecov() which returned
6437 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
6438 considers this an error and immediately begins cleaning up the
6439 lov, just after starting the mds_lov process
6441 Severity : enhancement
6443 Description: Implement 'lfs df' to report actual free space on per-OST basis
6444 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
6445 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
6446 usage of objects; '-h' to report in human readable format.
6448 ------------------------------------------------------------------------------
6450 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
6457 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
6458 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
6459 at startup during orphan unlinking. Instead, skip buddy update
6460 before setup, as it will be regenerated after recovery is complete.
6465 Description: If an OST is inactive, its locks might reference stale inodes.
6466 Details : lov_change_cbdata() must iterate over all namespaces, even if
6467 they are inactive to clear inode references from the lock.
6469 Severity : enhancement
6470 Frequency : occasional, if non-standard max_dirty_mb used
6472 Description: Client will block write RPCs if not enough grant
6473 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
6474 then the client will block writes while waiting for another RPC
6475 to complete instead of consuming its dirty limit. With change
6476 we get improved performance when max_dirty_mb is small.
6478 Severity : enhancement
6479 Bugzilla : 3389, 6253
6480 Description: Add support for supplementary groups on the MDS.
6481 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
6482 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
6483 lookups for user supplementary groups into a cache.
6487 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
6488 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
6489 for liblustre O_EXCL is enforced by MDS.
6493 Description: readdir never returns NULL in liblustre.
6494 Details : Corrected llu_iop_getdirentries logic, to return offset of next
6495 dentry in struct dirent.
6499 Frequency : liblustre only, depends on application IO pattern
6500 Description: liblustre clients evicted if not contacting servers
6501 Details : Don't put liblustre clients into the ping_evictor list, so
6502 they will not be evicted by the pinger ever.
6504 Severity : enhancement
6506 Description: Add ability to evict clients by NID from MDS.
6507 Details : By echoing "nid:$NID" string into
6508 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
6509 $NID would be instantly evicted from this MDS and from all active
6510 OSTs connected to it.
6514 Description: Do not query file size twice, somewhat slowing stat(2) calls.
6515 Details : lookup_it_finish() used to query file size from OSTs that was not
6520 Description: service threads change working directory to that of init
6521 Details : Starting lustre service threads may pin the working directory
6522 of the parent thread, making that filesystem busy. Threads
6523 now change to the working directory of init to avoid this.
6527 Frequency : during shutdown only
6528 Description: shutdown with a failed MDS or OST can cause unmount to hang
6529 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
6534 Frequency : 2.6 only
6535 Description: chmod/chown may include an extra supplementary group
6536 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
6537 supplementary group and if none is specified this is used.
6540 Bugzilla : 5479 (6816)
6542 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
6543 Details : If lookup is in progress on a file that is unlinked we might try
6544 to revalidate the inode and fail in revalidate after lookup is
6545 complete and ll_file_open() enqueues the open again but
6546 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
6549 Frequency : always, if lconf --abort_recovery used
6551 Description: lconf --abort_recovery fails with 'Operation not supported'
6552 Details : lconf was attempting to abort recovery on the MDT device and not
6555 ------------------------------------------------------------------------------
6557 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
6562 Frequency : rare (only unsupported configurations with a node running as an
6564 Bugzilla : 6514, 5137
6565 Description: Mounting a Lustre file system on a node running as an OST could
6567 Details : OSTs now preallocates memory needed to write out data at
6568 startup, instead of when needed, to avoid having to
6569 allocate memory in possibly low memory situations.
6570 Specifically, if the file system is mounted on on OST,
6571 memory pressure could force it to try to write out data,
6572 which it needed to allocate memory to do. Due to the low
6573 memory, it would be unable to do so and the node would
6574 become unresponsive.
6576 Severity : enhancement
6578 Description: Addition of lconf --service command line option
6579 Details : lconf now accepts a '--service <arg>' option, which is
6580 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
6582 Severity : enhancement
6584 Description: Failover mode is now the default for OSTs.
6585 Details : By default, OSTs will now run in failover mode. To return to
6586 the old behaviour, add '--failout' to the lmc line for OSTs.
6588 Severity : enhancement
6590 Description: Health checks are now provided for MDS and OSTs
6591 Details : Additional detailed health check information on MSD and OSTs
6592 is now provided through the procfs health_check value.
6595 Frequency : occasional, depends on IO load
6597 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
6598 after numerous create/delete cycles
6599 Details : The ext3 inode allocation policy would not allocate new inodes
6600 very well on the OSTs because there are no new directories
6601 being created. Instead we look for groups with free space if
6602 the parent directories are nearly full.
6607 Description: Network or server problems during mount may cause partially
6608 mounted clients instead of returning an error.
6609 Details : The config llog parsing code may overwrite the error return
6610 code during mount error handling, returning success instead
6616 Description: MDS can fail to allocate large reply buffers
6617 Details : After long uptimes the MDS can fail to allocate large reply
6618 buffers (e.g. zconf client mount config records) due to memory
6619 fragmentation or consumption by the buffer cache. Preallocate
6620 some large reply buffers so that these replies can be sent even
6621 under memory pressure.
6625 Frequency : rare (liblustre)
6626 Description: fsx running with liblustre complained that using truncate() to
6627 extend the file doesn't work. This patch corrects that issue.
6628 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
6629 ATTR_SIZE and lsm use in llu_setattr_raw.
6633 Frequency : rare, only 2.6 kernels
6634 Description: Unusual file access patterns on the MDS may result in inode
6635 data being lost in very rare circumstances.
6636 Details : Bad interaction between the ea-in-inode patch and the "no-read"
6637 code in the 2.6 kernel caused the inode and/or EA data not to
6638 be read from disk, causing single-file corruption.
6642 Frequency : rare, only 2.6 filesystems using extents
6643 Description: Heavy concurrent write and delete load may cause data corruption.
6644 Details : It was possible under high-load situations to have an extent
6645 metadata block in the block device cache from a just-unlinked
6646 file overwrite a newly-allocated data block. We now unmap any
6647 metadata buffers that alias just-allocated data blocks.
6651 Frequency : filesystems with default stripe_count larger than 77
6652 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
6653 Details : lconf specifies an inode size of 4096 bytes when the default
6654 stripe_count is larger than 77. This conflicts with the default
6655 inode density of 1 per 4096 bytes. Allocate smaller inodes in
6656 this case to avoid pinning too much memory for large EAs.
6658 ------------------------------------------------------------------------------
6660 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
6665 Frequency : rare (extremely heavy IO load with hundreds of clients)
6667 Description: Client is evicted, gets IO error writing to file
6668 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
6669 caused clients to be evicted for AST timeouts. The fixes in
6670 bug 5192 mean we no longer need to have such short AST timeouts
6671 so ldlm_timeout has been increased.
6674 Frequency : occasional during --force or --failover shutdown under load
6675 Bugzilla : 5949, 4834
6676 Description: Server oops/LBUG if stopped with --force or --failover under load
6677 Details : a collection of import/export refcount and cleanup ordering
6678 issues fixed for safer force cleanup
6681 Frequency : only filesystems larger than 120 OSTs
6682 Bugzilla : 5990, 6223
6683 Description: lfs getstripe would oops on a very large filesystem
6684 Details : lov_getconfig used kfree on vmalloc'd memory
6687 Frequency : only filesystems exporting via NFS to Solaris 10 clients
6688 Bugzilla : 6242, 6243
6689 Description: reading from files that had been truncated to a non-zero size
6690 but never opened returned no data
6691 Details : ll_file_read() reads zeros from no-object files to EOF
6696 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
6697 Details : There's little harm in aborting MDS/OSS recovery and letting it
6698 try again, so I removed the LASSERT and return an error instead.
6700 Severity : enhancement
6702 Description: New debugging infrastructure for tracking down data corruption
6703 Details : The I/O checksum code was replaced to: (a) control it at runtime,
6704 (b) cover more of the client-side code path, and (c) try to narrow
6705 down where problems occurred
6709 Bugzilla : 3819, 4364, 4397, 6313
6710 Description: Racing close and eviction MDS could cause assertion in mds_close
6711 Details : It was possible to get multiple mfd references during close and
6712 client eviction, leading to one thread referencing a freed mfd.
6714 Severity: : enhancement
6715 Bugzilla : 3262, 6359
6716 Description: Attempts to reconnect to servers are now more aggressive.
6717 Details : This builds on the enhanced upcall-less recovery that was added
6718 in 1.4.2. When trying to reconnect to servers, clients will
6719 now try each server in the failover group every 10 seconds. By
6720 default, clients would previously try one server every 25 seconds.
6725 Description: After recovery, certain operations trigger a failed
6726 assertion on a client.
6727 Details : Failing over an mds, using lconf -d --failover, while a
6728 client was doing a readdir() call would cause the client to
6729 LBUG after recovery completed and the readdir() was resent.
6731 Severity : enhancement
6733 Description: Default groups are now added by lconf
6734 Details : You can now run lconf --group <servicename> without having to
6735 manually add groups with lmc.
6738 Frequency : occasional
6740 Description: Nodes with an elan id of 0 trigger a failed assertion
6743 Frequency : always when accessing e.g. tty/console device nodes
6745 Description: tty and some other devices nodes cannot be used on lustre
6746 Details : file's private_data field is used by device data and lustre
6747 values in there got lost. New field was added to struct file to
6748 store fs-specific private data.
6751 Frequency : when exporting Lustre via NFS
6753 Description: NFSD failed occasionally when looking up a path component
6754 Details : NFSD is looking up ".." which was broken in ext3 directories
6755 that had grown large enough to become hashed.
6758 Frequency : Clusters with multiple interfaces not on the same subnet
6760 Description: Nodes will repeatedly try to reconnect to an interface which it
6761 cannot reach and report an error to the log.
6762 Details : Extra peer list entries will be created by lconf with some peers
6763 unreachable. lconf now validates the peer before adding it.
6766 Frequency : Only if a default stripe is set on the filesystem root.
6768 Description: Setting a default stripe on the filesystem root prevented the
6769 filesystem from being remounted.
6770 Details : The client was sending extra request flags in the root getattr
6771 request and did not allocate a reply buffer for the dir EA.
6774 Frequency : occasional, higher if lots of files are accessed by one client
6775 Bugzilla : 6159, 6097
6776 Description: Client trips assertion regarding lsm mismatch/magic
6777 Details : While revalidating inodes the VFS looks up inodes with ifind()
6778 and in rare cases can find an inode that is being freed.
6779 The ll_test_inode() code will free the lsm during ifind()
6780 when it finds an existing inode and then the VFS later attaches
6781 this free lsm to a new inode.
6785 Bugzilla : 6422, 7030
6786 Description: MDS deadlock between mkdir and client eviction
6787 Details : Creating a new file via mkdir or mknod (starting a transaction
6788 and getting the ns lock) can deadlock with client eviction
6789 (gets ns lock and trying to finish a synchronous transaction).
6792 Frequency : occasional
6793 Description: While starting a server, the fsfilt_ext3 module could not be
6795 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
6796 kernels. Previously, lconf would still use the ext3 name
6797 when trying to load modules. Now, it will correctly use
6798 ext3 on 2.4 and ldiskfs on 2.6.
6800 Severity : enhancement
6801 Description: The default stripe count has been changed to 1
6802 Details : The interpretation of the default stripe count (0, to lfs
6803 or lmc) has been changed to mean striping across a single
6804 OST, rather than all available. For general usage we have
6805 found a stripe count of 1 or 2 works best.
6807 Severity : enhancement
6808 Description: Add support for compiling against Cray portals.
6809 Details : Conditional compiling for some areas that are different
6813 Frequency : occasional
6814 Bugzilla : 6409, 6834
6815 Description: Creating files with an explicit stripe count may lead to
6816 a failed assertion on the MDS
6817 Details : If some OSTs are full or unavailable, creating files may
6818 trigger a failed assertion on the MDS. Now, Lustre will
6819 try to use other servers or return an error to the
6823 Frequency : occasional
6825 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
6826 caused lock timeout during readahead (since 1.4.2).
6827 Details : Processes doing readahead might match a lock that hasn't been
6828 granted yet if there are overlapping and conflicting lock
6829 requests. The readahead process waits on ungranted lock
6830 (original lock is CBPENDING), while OST waits for that process
6831 to cancel CBPENDING read lock and eventually evicts client.
6833 Severity : enhancement
6835 Description: Initial enabling of flock support for clients
6836 Details : Implements fcntl advisory locking and file status functions.
6837 This feature is provided as an optional mount flag (default
6838 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
6839 locking are implemented yet, and those that are are not guaranteed
6840 to be completely correct in production environments.
6841 mount -t lustre -o [flock|noflock] ...
6844 Frequency : occasional
6846 Description: OSTs running 2.4 kernels but with extents enabled might trip an
6847 assertion in the ext3 JBD (journaling) layer.
6848 Details : The b_committed_data struct is protected by the big kernel lock
6849 in 2.4 kernels, serializing journal_commit_transaction() and
6850 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
6851 access to this struct. In 2.6 kernels there is finer grained
6852 locking to improve SMP performance of the JBD layer.
6856 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
6858 -----------------------------------------------------------------------------
6860 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
6862 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
6863 versions of Lustre. Please update all servers and clients to
6864 version 1.4.2 or later at the same time. You must also run
6865 "lconf --write-conf {config}.xml" on the MDS while it is stopped
6866 to update the configuration logs.
6868 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
6869 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
6870 - swab llog records of type '0' so we get proper header size/idx (5861)
6871 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
6872 - fix rename of one directory over another leaking an inode (5953)
6873 - avoid SetPageDirty on 2.6 (5981)
6874 - don't re-add just-being-destroyed locks to the waiting list (5653)
6875 - when creating new directories, inherit the parent's custom
6876 striping settings if present parent (3048)
6877 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
6878 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
6879 - don't hold client locks on temporary worklist from l_lru (5666)
6880 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
6881 - automatically evict dead clients (5921)
6882 - Update file size properly in create+truncate+fstat case (6196)
6883 - Do not unhash mountpoint dentries, do not allow removal of
6885 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
6886 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
6887 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
6888 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
6889 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
6890 - Correctly handle reads of files with no objects (6243)
6891 - lctl recover will also mark a device active if deactivate used (5933)
6893 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
6894 - allow --write-conf on an MDS with different nettype than client (5619)
6895 - don't write config llogs to MDS for mounts not from that MDS (5617)
6896 - lconf should create multiple TCP connections from a client (5201)
6897 - init scripts are now turned off by default; run chkconfig --on
6898 lustre and chkconfig --on lustrefs to use them
6899 - upcalls are no longer needed for clients to recover to failover
6901 - add --abort-recovery option to lconf to abort recovery on device
6903 - add support for an arbitrary number of OSTs (3026)
6904 - Quota support protocol changes.
6905 - forward compatibility changes to wire structs (6007)
6906 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
6907 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
6908 - improved "lustre status" script
6909 - initialize blocksize for non-regular files (6062)
6910 - added --disable-server and --disable-client configure options (5782)
6911 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
6912 - Vanilla 2.4.29 support
6913 - increase maximum number of obd devices to 520 (6242)
6914 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
6915 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
6916 SLES 9 2.6.5 kernels
6917 - limit stripes per file to 160 (the maximum EA size) (6093)
6919 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
6922 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
6923 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
6924 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
6925 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
6926 - small ext3 extents cleanups and fixes (5733)
6927 - improved mballoc code, several small races and bugs fixed (5733, 5638)
6928 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
6929 - avoid needless client->OST connect, fix handle mismatch (5317)
6930 - fix DLM error path that led to out-of-sync client, long delays (5779)
6931 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
6932 - fix several locking issues related to i_size (5492,5624,5654,5672)
6933 - don't move pending lock onto export if it is already evicted (5683)
6934 - fix kernel oops when creating .foo in unlinked directory (5548)
6935 - fix deadlock in obdfilter statistics vs. object create (5811)
6936 - use time_{before,after} to avoid timer jiffies wrap (5882)
6937 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
6938 - Do not leak request if server was not able to process it (5154)
6939 - If mds_open unable to find parent dir, make that negative lookup(5154)
6940 - don't create new directories with extent-mapping (5909, 5936)
6942 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
6943 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
6944 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
6945 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
6946 - support for RHEL4 kernel on i686 (5773)
6947 - provide error messages when incompatible logs are encountered (5898)
6949 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
6950 * version 1.4.0.10 (1.4.1 release candidate 1)
6952 - don't keep a lock reference when lock is not granted (4238)
6953 - unsafe list practices (rarely) led to infinite eviction loop (4908)
6954 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
6955 - drop import inflight refcount on signal_completed_replay error (5255)
6956 - unlock page after async write error during send (3677)
6957 - handle missing objects in filter_preprw_read properly (5265)
6958 - no transno return for symlink open, don't save no-trasno open (3440)
6959 - don't try to complete elan receive that already failed (4012)
6960 - free RPC server reply state on error (5406)
6961 - clean up thread from ptlrpc_start_thread() on error (5160)
6962 - readahead could read extra page into cache that wasn't ejected (5388)
6963 - prevent races in class_attach/setup/cleanup/detach (5260)
6964 - don't dereference de->d_inode after l_dput of de (5458)
6965 - use "int" for stripe value returned from lock_to_stripe (5544)
6966 - mballoc allocation and error-checking fixes in 2.6 (5504)
6967 - block device patches to fix I/O request sizes in 2.6 (5482)
6968 - look up hostnames for IB nals (5602)
6969 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
6970 - don't start multiple acceptors for the same port (5277)
6971 - fix incorrect LASSERT in mds_getattr_name (5635)
6972 - export a proc file for general "ping" checking (5628)
6973 - fix "lfs check" to not block when the MDS is down (5628)
6975 - service request history (4965)
6976 - put {ll,lov,osc}_async_page structs in a single slab (4699)
6977 - create an "evict_client" /proc entry on OSTs, like the MDS has
6978 - fix mount usage message, return errors per mount(8) (5168)
6979 - change grep [] to grep "[]" in tests so they work in more UMLs
6980 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
6981 - remove ancient LOV_MAGIC_V0 EA support (5047)
6982 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
6983 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
6984 - allow readahead allocations to fail when low on memory (5383)
6985 - mmap locking landed again, after considerable improvement (2828)
6986 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
6988 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
6991 - send OST transaction number in read/write reply to free req (4966)
6992 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
6993 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
6994 - the watchdog thread now runs as interruptible (5246)
6995 - flock/lockf fixes (but it's still disabled, pending 5135)
6996 - don't use EXT3 constants in llite code (5094)
6997 - memory shortage at startup could cause assertion (5176)
6999 - reorganization of lov code
7000 - single portals codebase
7002 - add extents/mballoc support (5025)
7003 - direct I/O reads in the obdfilter (4048)
7004 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
7006 tbd Cluster File Systems, Inc. <info@clusterfs.com>
7009 - send OST transaction number in read/write reply to free req (4966)
7010 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
7011 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
7012 - the watchdog thread now runs as interruptible (5246)
7013 - handle missing objects in filter_preprw_read properly (5265)
7014 - unsafe list practices (rarely) led to infinite eviction loop (4908)
7015 - drop import inflight refcount on signal_completed_replay error (5255)
7016 - unlock page after async write error during send (3677)
7017 - return original error code on reconstructed replies (3761)
7018 - no transno return for symlink open, don't save no-trasno open (3440)
7020 - add pid to ldlm debugging output (4922)
7021 - bump the watchdog timeouts -- we can't handle 30sec yet
7022 - extra debugging for orphan dentry/inode bug (5259)
7024 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
7027 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
7028 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
7029 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
7030 - fix NULL dereference in /proc/sys/portals/routes (4827)
7031 - allow failed mdc_close() operations to be interrupted (4561)
7032 - stop precreate on OST before MDS would time out on it (4778)
7033 - don't send partial-page writes before EOF from client (4410)
7034 - discard client grant for sub-page writes on large-page clients (4520)
7035 - don't free dentries not owned by NFS code, check generation (4806)
7036 - fix lsm leak if mds_create_objects() fails (4801)
7037 - limit debug_daemon file size, always print CERROR messages (4789)
7038 - use transno after validating reply (3892)
7039 - process timed out requests if import state changes (3754)
7040 - update mtime on OST during writes, return in glimpse (4829)
7041 - add mkfsoptions to LDAP (4679)
7042 - use ->max_readahead method instead of zapping global ra (5039)
7043 - don't interrupt __l_wait_event() during strace
7045 - add software watchdogs to catch hung threads quickly (4941)
7046 - make lustrefs init script start after nfs is mounted
7047 - fix CWARN/ERROR duplication (4930)
7048 - return async write errors to application if possible (2248)
7049 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
7050 - print NAL number in %x format (4645)
7051 - update barely-supported suse-2.4.21-171 series (4842)
7052 - support for sles 9 %post scripts
7053 - support for building 2.6 kernel-source packages
7054 - support for sles km_* packages
7056 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
7059 - ignore -ENOENT errors in osc_destroy (3639)
7060 - notify osc create thread that OSC is being cleaned up (4600)
7061 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
7062 - reconstruct ost_handle() like mds_handle() (4657)
7063 - create a new thread to do import eviction to avoid deadlock (3969)
7064 - let lconf resolve symlinked-to devices (4629)
7065 - don't unlink "objects" from directory with default EA (4554)
7066 - hold socknal file ref over connect in case target is down (4394)
7067 - allow more than 32000 subdirectories in a single directory (3244)
7068 - fix blocks count for O_DIRECT writes (3751)
7069 - OST returns ENOSPC from object create when no space left (4539)
7070 - don't send truncate RPC if file size isn't changing (4410)
7071 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
7072 - bind to privileged port in socknal and tcpnal (3689)
7074 - rate limit CERROR/CWARN console message to avoid overload (4519)
7075 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
7076 - basic mmap support (3918)
7077 - kernel patch series update from b1_4 (4711)
7079 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
7082 - avoid crash during MDS cleanup with OST shut down (2775)
7083 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
7084 - don't use bad inodes on the MDS (3744)
7085 - dynamic object preallocation to improve recovery speed (4236)
7086 - don't hold spinlock over lock dumping or change debug flags (4401)
7087 - don't zero obd_dev when it is force cleaned (3651)
7088 - print grants to console if they go negative (4431)
7089 - "lctl deactivate" will stop automatic recovery attempts (3406)
7090 - look for existing locks in ldlm_handle_enqueue() (3764)
7091 - don't resolve lock handle twice in recovery avoiding race (4401)
7092 - revalidate should check working dir is a directory (4134)
7094 - don't always mark "slow" obdfilter messages as errors (4418)
7096 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
7099 - don't close LustreDB during write_conf until it is done (3860)
7100 - fix typo in lconf for_each_profile (3821)
7101 - allow dumping logs from multiple threads at one time (3820)
7102 - don't allow multiple threads in OSC recovery (3812)
7103 - fix debug_size parameters (3864)
7104 - fix mds_postrecov to initialize import for llog ctxt (3121)
7105 - replace config semaphore with spinlock (3306)
7106 - be sure to send a reply for a CANCEL rpc with bad export (3863)
7107 - don't allow enqueue to complete on a destroyed export (3822)
7108 - down write_lock before checking llog header bitmap (3825)
7109 - recover from lock replay timeout (3764)
7110 - up llog sem before sending rpc (3652)
7111 - reduce ns lock hold times when setting kms (3267)
7112 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
7113 - fix NULL deref and obd_dev leak on setup error (3312)
7114 - replace some LBUG about llog ops with error handling (3841)
7115 - don't match INVALID dentries from d_lookup and spin (3784)
7116 - hold dcache_lock while marking dentries INVALID and hashing (4255)
7117 - fix invalid assertion in ptlrpc_set_wait (3880)
7119 - add libwrap support for the TCP acceptor (3996)
7120 - add /proc/sys/portals/routes for non-root route listing (3994)
7121 - allow setting MDS UUID in .xml (2580)
7122 - print the stack of a process that LBUGs (4228)
7124 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
7127 - don't cleanup request in ll_file_open() on failed MDS open (3430)
7128 - make sure to unset replay flag from failed open requests (3440)
7129 - if default stripe count is 0, use OST count for inode size (3636)
7130 - update parent mtime/ctime on client for create/unlink (2611)
7131 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
7132 - free recovery state on server during a forced cleanup (3571)
7133 - unregister_reply for resent reqs (3063)
7134 - loop back devices mounting and status check on 2.6 (3563)
7135 - fix resource-creation race that can provoke i_size == 0 (3513)
7136 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
7137 - more debugging for page-accounting assertion (3746)
7138 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
7139 - avoid deadlock after precreation failure (3758)
7140 - fix race and lock order deadlock in orphan handling (3450, 3750)
7141 - add validity checks when grabbing inodes from l_ast_data (3599)
7143 - add /proc/.../recovery_status to obdfilter (3428)
7144 - lightweight CDEBUG infrastructure, debug daemon (3668)
7145 - change default OSC RPC parameters to be better on small clusters
7146 - turn off OST read cache for files smaller than 32MB
7147 - install man pages and include them in rpms (3100)
7148 - add new init script for (un)mounting lustre filesystems (2593)
7149 - run chkconfig in %post for init scripts (3701)
7150 - drop scimac NAL (unmaintained)
7152 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
7155 - clean kiobufs before and after use (3485)
7156 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
7157 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
7158 - change init script to fail more clearly if not run as root (1528)
7159 - allow clients to reconnect during replay (1742)
7160 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
7161 - don't do DNS lookups on NIDs too small for IP addresses (3442)
7162 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
7163 - fix cond_resched (3554)
7164 - only evict unfinished clients after recovery (3515)
7165 - allow bulk resend, prevent data loss (3570)
7166 - dynamic ptlrpc request buffer allocation (2102)
7167 - don't allow unlinking open directory if it isn't empty (2904)
7168 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
7169 - remove extraneous obd dereference causing LASSERT failure (3334)
7170 - don't use get_cycles() when creating temp. files on the mds (3156)
7171 - hold i_sem when setting i_size in ll_extent_lock() (3564)
7172 - handle EEXIST for set-stripe, set proper directory name (3336)
7174 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
7175 - fix ksocknal_fmb_callback() error messages (2918)
7177 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
7180 - don't copy lvb into (possibly NULL) reply on error (2983)
7181 - don't deref dentry after dput, don't free lvb on error (2922)
7182 - use the kms to determine writeback rpc length (2947)
7183 - increment oti_logcookies when osc is inactive (2948)
7184 - update client's i_blocks count via lvb messages (2543)
7185 - handle intent open/close of special files properly (1557)
7186 - mount MDS with errors=remount-ro, like obdfilter (2009)
7187 - initialize lock handle to avoid ASSERT on error cleanup (3057)
7188 - don't use cancelling-locks' kms values (2947)
7189 - use highest lock extent for kms, not last one (2925)
7190 - don't dereference ERR_PTR() dentry in error handling path (3107)
7191 - fix thread race in portals_debug_dumplog() (3122)
7192 - create lprocfs device entries at setup instead of at attach (1519)
7193 - common AST error handler, don't evict client on completion race (3145)
7194 - zero nameidata in detach_mnt in 2.6 (3118)
7195 - verify d_inode after revalidate_special is valid in 2.6 (3116)
7196 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
7197 - initialize RPC timeout timer earlier for 2.6 (3219)
7198 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
7199 - print nal/nid for unknown nid (3258)
7200 - additional checks for oscc recovery before doing precreate (3284)
7201 - fix ll_extent_lock() error return code for 64-bit systems (3043)
7202 - don't crash in mdc_close for bad permissions on open (3285)
7203 - zero i_rdev for non-device files (3147)
7204 - clear page->private before handing to FS, better assertion (3119)
7205 - tune the read pipeline (3236)
7206 - fix incorrect decref of invalidated dentry (2350)
7207 - provide read-ahead stats and refine rpc in flight stats (3328)
7208 - don't hold journal transaction open across create RPC (3313)
7209 - update atime on MDS at close time (3265)
7210 - close LDAP connection when recovering to avoid server load (3315)
7211 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
7212 - don't leak open file on MDS after open resend (3325)
7213 - serialize filter_precreate and filter_destroy_precreated (3329)
7214 - loop device shouldn't call sync_dev() for nul device (3092)
7215 - clear page cache after eviction (2766)
7216 - resynchronize MDS->OST in background (2824)
7217 - refuse to mount the same filesystem twice on same mountpoint (3394)
7218 - allow llmount to create routes for mounting behind routers (3320)
7219 - push lock cancellation to blocking thread for glimpse ASTs (3409)
7220 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
7221 - fix rare problem with rename on htree directories (3417)
7223 - allow default OST striping configuration per directory (1414)
7224 - fix compilation for qswnal for 2.6 kernels (3125)
7225 - increase maximum number of MDS request buffers for large systems
7226 - change liblustreapi to be useful for external progs like lfsck (3098)
7227 - increase local configuration timeout for slow disks (3353)
7228 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
7230 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
7233 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
7234 - don't overwrite extent policy data in reply if lock was blocked (2901)
7235 - drop filter export grants atomically with removal from device (2663)
7236 - del obd_self_export from work_list in class_disconnect_exports (2908)
7237 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
7238 - swab reply message in mdc_close, other PPC fixes (2464)
7239 - fix destroying of named logs (2325)
7240 - overwrite old logs when running lconf --write_conf (2264)
7241 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
7242 - fix race in target_handle_connect (2898)
7243 - mds_reint_create() should take same inode create lock (2926)
7244 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
7245 - don't close files for self_export to avoid uninitialized obd (2936)
7246 - allow MDS with the same name as client node (2939)
7247 - hold dentry reference for closed log files for unlink (2325)
7248 - reserve space for all logs during transactions (2059)
7249 - don't evict page beyond end of stripe extent (2925)
7250 - don't oops on a deleted current working directory (2399)
7251 - handle hard links to targets without a parent properly (2517)
7252 - don't dereference NULL lock when racing during eviction (2867)
7253 - don't grow lock extents when lots of conflicting locks (2919)
7255 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
7258 - account for cache space usage on clients to avoid data loss (974)
7259 - lfsck support in lustre kernel code (2349)
7260 - reduce journal credits needed for BRW writes (2370)
7261 - orphan handling to avoid losing space on client/server crashes
7262 - ptlrpcd can be blocked, stopping ALL progress (2477)
7263 - use lock value blocks to assist in proper KMS, faster stat (1021)
7264 - takes i_sem instead of DLM locks internally on obdfilter (2720)
7265 - recovery for initial connections (2355)
7266 - fixes for mds_cleanup_orphans (1934)
7267 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
7268 - block all file creations until orphan recovery completes (1901)
7269 - client remove rq_connection from request struct (2423)
7270 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
7271 - recovery timer race (2670)
7272 - mdc_close recovey bug (2532)
7273 - ptlrpc cleanup bug (2710)
7274 - mds timeout on local locks (2588)
7275 - namespace lock held during RPCs (2431)
7276 - handle interrupted sync write properly (2503)
7277 - don't try to handle a message that hasn't been replied to (2699)
7278 - client assert failure during cleanup after abort recovery (2701)
7279 - leak mdc device after failed mount (2712)
7280 - ptlrpc_check_set allows timedout requests to complete (2714)
7281 - wait for inflight reqs when ptlrpcd finishes (2710)
7282 - make sure unregistered services are removed from the srv_list
7283 - reset bulk XID's when resending them (caught by 1138 test)
7284 - unregister_bulk after timeout
7285 - fix lconf error (2694)
7286 - handle write after unfinished setstripe, stripe-only getstripe (2388)
7287 - readahead locks pages, leaves pending causing memory pressure (2673)
7288 - increase OST request buffers to 4096 on large machines (2729)
7289 - fix up permission of existing directories in simple_mkdir (2661)
7290 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
7291 - don't assign transno to errored transactions (2742)
7292 - don't delete objects on OST if given a bogus objid from MDS (2751)
7293 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
7294 - if rq_no_resend, then timeout request after recovery (2432)
7295 - fix MDS llog_logid record size, 64-bit array alignment (2733)
7296 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
7297 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
7298 - MDS recovery shouldn't delete objects that it has given out (2730)
7299 - if enqueue arrives after completion, don't clobber LVB (2819)
7300 - don't unlock pages twice when trigger_group_io returns error (2814)
7301 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
7302 - don't write pages to disk if there was an error (1450)
7303 - don't ping imports that have recovery disabled (2676)
7304 - take buffered bytes into account when balancing socknal conn (2817)
7305 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
7306 - reconnect unlink llog connection after MDS reconnects to OST (2816)
7307 - remove little-endian swabbing of llog records (1987)
7308 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
7309 - retry reposting request buffers if they fail (1191)
7310 - grow extent at grant time to avoid granting a revoked lock (2809)
7311 - lock revoke doesn't evict page if covered by a second lock (2765)
7312 - disable VM readahead to avoid reading outside lock extents (2805)
7314 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
7315 - updated kernel patches for hp-2.4.20 kernel (2681)
7317 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
7320 - fix truncated write corruption (2366)
7321 - fix for failed assertion in iopen_connect_dentry (1792,2517)
7323 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
7324 - ensure len(uuid) < 37 in lmc (1171)
7325 - fix ia64 OOPS in llog_test (2255)
7326 - zero end of page at obdfilter for partial page writes (2648)
7327 - don't leave stale dentries around after renames (bug 2428)
7328 - fix timeouts when evicting a client with a single lock held (2642)
7329 - set deadline for the initial HELLO message to drain (2634)
7330 - print out dotted-quad IP addresses in the socknal (2302)
7332 - additional debugging for MDS client eviction problem (2443)
7333 - fix mkfsoptions support for osts (2603, 2604)
7335 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
7338 - add series for the vanilla 2.6.0 kernel
7339 - add series for the vanilla 2.4.24 kernel
7340 - add series for a cray x86/64 UL kernel drop
7341 - fix xattr patches for the vanilla 2.4.19 series
7343 - generate true UUIDs in lmc (1171)
7344 - have portals stack dumping break in UML (2466)
7345 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
7346 - allocate with _NOFS in ldlm to avoid deadlock (1933)
7347 - wake callback waiting threads on client eviction (2460)
7348 - Add --ptldebug and --subsystem to lmc (1719)
7349 - update assertion to allow safe interrupt allocation
7350 - set rq_no_resend for cancel requests (2432)
7351 - recalculate ptlrpcd timeout after resend (2494)
7352 - call vfs_rmdir when removing pending directories (2368)
7353 - fix renaming a file to itself (2429)
7354 - lmc creates a default one-stripe lov (2454)
7355 - expand procfs space to handle large clusters (2326)
7356 - increase UML stack to avoid overflow
7357 - update lconf's list of debug and subsystem masks
7358 - fix lfs find --obd (2510)
7359 - /proc tunable for disabling filter read caching (2591)
7360 - stop rpm packages from altering slapd.conf (2301)
7361 - disable nagle in the socknal under 0conf (2578)
7362 - choose mds inode size based on stripe count (2572)
7363 - fix kernel-source rpm problems (2516)
7365 - add --disable-doc to avoid pdf generation (2421)
7366 - update documentation, tests, type-os, comments
7367 - avoid format warnings on ia64
7368 - remove the TOE NAL
7369 - tiny code cleanups by removing unused fields
7371 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
7374 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
7375 - remove the most blatant lies from BUILDING file (2371)
7376 - change default debug level to reasonable production setting
7377 - reduce client side cache size to reduce cache flush time
7378 - reduce max RPCs in flight to avoid unnecessary file fragmentation
7379 - make TCP zerocopy and pinger support enabled by default (2476)
7380 - sync writes completed after process exits caused crashes (2319)
7381 - maintain correct mount count on the MDS (2356)
7382 - backout 1557, because 2316 wasn't really fixed
7383 - better file I/O statistics gathering in /proc
7384 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
7385 - another kernel patch to fix zero-copy TCP function export
7386 - don't take duplicate lock when processing re-sent getattr (2420)
7387 - lctl uses obd_self_export instead of creating new conn (2353)
7388 - MDS/OST recovery case which requires object creation asserted (2425)
7389 - move lfs from /usr/sbin to /usr/bin in packages
7390 - fix race between mds_client_add and mds_client_free (2417)
7391 - use kmalloc instead of slabs in portals (2430)
7392 - don't create duplicate records when a failover MDS is present (2442)
7393 - remove unnecessary mount age check (2332)
7394 - don't remove directory inodes from locks prematurely (2451)
7395 - don't break if MDS service name is the same as hostname (2103)
7396 - fix races in client write RPC generation when cache full (2482)
7398 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
7401 - remove now-unused request->rq_obd (278)
7402 - if an allocation fails, print out how much memory we've used (1933)
7403 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
7404 - add the "configurable stack size" patch to most series files (1256)
7405 - ability to write large log records, for 100+ OST configs (2306)
7406 - fix NULL deref when filter_prep fails (2314)
7407 - fix operator precedence error in filter_sync
7408 - dynamic allocation of socknal TX descriptors (2315)
7409 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
7410 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
7411 - ensure that CWARN messages in Portals always get to the syslog
7412 - __init/__exit are not for prototype decls (ldlm_init/exit)
7413 - x86-64 compile warning fixes
7414 - fix gateway LMC keyword conflict (2318)
7415 - fix MDS lock inversions in getattr/reint paths (1844)
7416 - fix a rare lock re-ordering bug, which caused deadlock (2322)
7417 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
7418 - DLM race condition prevented some lock evictions (2328)
7419 - ENOMEM detection and retry on socknal sends (2230)
7420 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
7421 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
7422 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
7423 - better error messages when a client is rejected during recovery (1505)
7424 - avoid cancelling locks which were never granted, after failure (2330)
7425 - fix i_sem/journal inversion in mds_client_add (2333)
7426 - fix truncate/getattr lock cycle deadlock (2334)
7427 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
7428 - fix two rare exit paths which could leak an l_lock() ref (2321)
7429 - fencepost error in MDS/OST orphan recovery (2226)
7430 - make log record alignment 8 bytes (1988)
7431 - lstripe now fails when requested offset > ost_count (2237)
7432 - ensure that all kernel series have a complete list.h (1607)
7433 - fix crashes in special-file operations (2316)
7434 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
7436 - allow configurable automake binary, for testing new versions
7437 - small update to the lfs documentation
7439 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
7441 * fix negative export reference count in fsfilt_sync (2312)
7443 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
7444 * release candidate 0.9.1
7446 - orphans are moved into the PENDING directory for possible recovery
7447 - replayed opens now open by fid for orphan/rename safety (1042)
7448 - last close of an orphan inode generates a transno (683)
7449 - chdir() and mount() now pin the directory entry (1020)
7450 - avoid CERROR in normal ll_setattr_raw() error case (1500)
7451 - discard very old requests without processing them (1502)
7452 - remove some common, well-understood CERRORs (1505)
7453 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
7454 - clear "grant" flags in OST replies until OST grant code lands (1644)
7455 - fix read performance by not clobbering i_blksize on client (1598)
7456 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
7457 - make LRU size a /proc tunable, clears locks when reduced (707)
7458 - fix some lprocfs rot that prevented ptlbd from loading (1732)
7459 - server locks take references on exports now (1558)
7460 - build fixes for 2.4.20-rh trees (1663)
7461 - return an error from lov_create if all OSCs are inactive (1751)
7462 - fix import levels when a reconnect happens without a timeout (1597)
7463 - exit early from mds_open if we get a lookup error (1749)
7464 - partial page read at EOF wouldn't wait for disk before sending (1642)
7465 - avoid NULL deref in obdfilter when reading page past EOF (1592)
7466 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
7467 - fix LBUG in ll_it_open_error with rc = -2 (1861)
7468 - write/truncate lock inversion (1639)
7469 - Don't auto-load obdclass, portals modules during cleanup (1495)
7470 - fix timestamps from jumping to "now" (1763)
7471 - extra journal assertions (1648)
7472 - add an extra multiunlink test (1771)
7473 - fix read_record/write_record API (1776)
7474 - fix leak of offset_extent, possible incorrect i_size later (1772)
7475 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
7476 - Debugging for the kqswnal_get_idle_tx problems (1820)
7477 - Allow recovery to be attempted multiple times (1536)
7478 - Write out MDS last_rcvd file after it is first created (1600)
7479 - Fix tx_descriptor leak in failed transmit situations (1827)
7480 - ext3 journaling fixes for assertion failure after IO error (1871)
7481 - class_export_put() on freed export after completion AST error (1896)
7482 - Fix revalidate looping in VFS (1322)
7483 - Don't access a freed export during MDS_REINT timeout (1521)
7484 - Add open-unlink recovery support on the MDS (1673,1764)
7485 - Return an error if no MDS data was read from last_rcvd (1946)
7486 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
7487 - Don't setup a disk device that doesn't match exported UUID (317)
7488 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
7489 - avoid committing NULL handle in force close
7490 - local.sh is now a one-stripe LOV configuration
7491 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
7492 - fix ext3 htree duplicate directory entry corruption (1516)
7493 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
7494 - update to new LOV EA format (2097)
7495 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
7497 2003-06-15 Phil Schwan <phil@clusterfs.com>
7500 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
7501 - per-import/export recovery handling (958, 931, 959)
7502 - multiple last-rcvd slots, for serving multiple FSes (949)
7503 - connections are again shared between multiple imp/exports (963, 964)
7504 - "umount -f" would hang if any requests needed to be sent (393, 978)
7505 - avoid pinning large req buffer by copying for queued messages (989)
7506 - add "uuid" to "lctl device" command to help upcalls (991)
7507 - "open" RPCs with transnos would confuse recovery counters (1037)
7508 - do proper endian conversion of all wire messages (288, 340, 891)
7509 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
7510 - call waiting locks callback from LDLM recovery thread (1127, 1151)
7511 - fix ptlrpc_connection leak in target_handle_connect (1174)
7512 - fix import refcounting bug in OST and MDS cleanup (1134)
7513 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
7514 - fix very unlikely obd_types race condition (501)
7515 - remove osc_open hack for echo_client (1187)
7516 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
7517 - a failure in read_inode2 leads to deadlock (1139)
7518 - cancel ack-locks as soon as transaction is committed (1072)
7519 - fix major leaks and crashes in the bulk I/O path (937, 1057)
7520 - make sure to commitrw after any preprw to avoid deadlock (1162)
7521 - failing to execute a file in a lustre FS would lock inode (1203)
7522 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
7523 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
7524 - obd_uuid2tgt would walk off the end of the list (1255)
7525 - on IA64 the osc would give portals incorrect bulk size (1258)
7526 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
7527 - fix lock inversion caused by new llite matching code (1282)
7528 - limit the number of dirty pages on a client to 10MB (1286)
7529 - timed out locks were not being corrected cancelled (1289)
7530 - fix O_DIRECT above 4GB on IA-32 (1292)
7531 * major user-visible changes
7532 - fail out/fail over policy now controlled by the upcall (993)
7534 - add OBD_PING to check server availability and failure (954)
7535 - lustre messages are now sent in sending host order (288, 340, 891)
7536 - add eadatalen to MDS getattr reply (340)
7537 - OST read replies may contain second buffer, with per-page status (593)
7539 2003-03-11 Phil Schwan <phil@clusterfs.com>
7542 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
7543 - failed open()s could cause deadlock; fixed (867, 869)
7544 - stop cancelling OST locks when files are closed (481)
7545 - overlapping XID spaces caused network corruption (851, 853)
7546 - fix unsafe fsfilt counter arithmetic; change to atomic_t
7547 - setattr_raw added, to do single-RPC, server-side setattrs
7548 - lmc/lconf syntax change for OST UUIDs
7549 - fix crashy race condition between ptlrpc_free_req and osc_close
7550 - don't use request in mdc_enqueue if we hit a timeout (889)
7551 - don't set the inode i_size for regular files from the MDS (896)
7552 - handle out of order completion AST (842)
7553 - don't LBUG if a lock request times out after receiving AST (913)
7554 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
7555 - if a bad lock AST arrives, send an error instead of dropping entirely
7556 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
7557 - fix leak in bulk IO when only partially completed (899, 900, 926)
7558 - fix O_DIRECT for ia64 (55)
7559 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
7560 - C-z after timeout could hang a process forever; fixed (977)
7562 - client-side I/O cache (678, 924, 929, 941, 970)
7564 - READPAGE and SETATTRs which don't take server-side locks get
7567 2003-02-11 Phil Schwan <phil@clusterfs.com>
7570 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
7571 - Fix fsfilt_extN_readpage() to read a full page of directory
7572 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
7573 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
7574 - Fix imbalanced LOV object allocation and out-of-bound access (469)
7575 - Most intent operations were removed, in favour of a new RPC mode
7576 that does a single RPC to the server and bypasses most of the VFS
7577 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
7578 - Aggressively cancel local locks on DLM servers
7579 - mds_reint_unlink sends EA to the client if it's the last nlink.
7580 client uses that EA to unlink OST objects.
7581 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
7582 - recursive symlinks were fixed (439)
7583 - fixed NULL deref in DEBUG_REQ
7584 - filter_update_lastobjid no longer calls sync, which annoyed extN
7585 - fixed multi-client small-writes to a single file problem (445)
7586 - fixed mtime updates during file writes (607)
7587 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
7588 - fixed bug in obd_brw_read/write() (under guise of testing 367)
7589 - fixed Linux OST size reporting problem (444, 656)
7590 - OST now updates object mtime with writes or setattr (607, 619)
7591 - client verifies file size before zeroing page past EOF (445)
7592 - OST now writes last allocated objid to disk with allocation (108)
7593 - LOV on echo now works (409)
7595 - mds_reint_unlink sends a new buffer, with the EA included. this
7596 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
7597 set if a regular file was being unlinked, and it was the last link
7598 - use PtlGet from the target for bulk writes (315)
7599 - OST now updates object mtime with writes or setattr (607, 619)
7600 - LDLM now has a grant-time callback to revalidate locked items, if
7602 - Many MDS operations were reorganized to combat race conditions
7604 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
7605 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
7607 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
7610 - Fully reactivate OST imports after reconnection (512, others)
7611 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
7612 - More graceful error handling for truncating on dead OST (515)
7613 - Don't error out unless we're actually accessing dead stripes (474)
7614 - Fix garbage sizes when stripes are missing (410)
7615 - LRU counters were broken, causing constant lock purge (433, 432)
7616 - garbage on read from stripes with failed OSTs (441)
7617 - mark OSCs as active before reconnecting during recovery (438)
7618 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
7619 - lfind did not preserve OST order in output (443)
7620 - symlinks cause hung clients, incorrect data (439)
7621 - stop dereferencing request after dropping refcount (457)
7622 - don't LASSERT(spin_is_locked) on non-SMP (455)
7623 - fixes for many rename() bugs
7624 - fstat didn't correctly synchronize attributes (399)
7625 - server must handle lock cancellation during blocking AST prep (487)
7626 - bulk descriptors were free()d too soon (511)
7627 - fix paths in lconf, which would load incorrect modules (451, 507)
7628 - fix confusing lconf 'host not found' error message (386)
7629 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
7630 - fix race condition in mdc_blocking_ast() for inode access (526)
7631 - fix lov_unpackmd() unpacking wrong number of stripes (537)
7632 - fix lov_set_osc_active() marking wrong OSC inactive (440)
7633 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
7634 - fix multiple writes of stripe MD to MDS (358, maybe 519)
7635 - fix lstripe in several ways (kernel side) (527)
7636 - fix request leak in ldlm_cli_enqueue (262)
7637 - incorrect OSC was marked inactive after OST failure
7638 - call mds_fs_cleanup before unmounting filesystem (524)
7639 - fix races between taking ns_lock and ldlm_lock_change_resource
7640 - fix races updating LOV export open file list
7641 - fix lov_enqueue error path, avoid decref-ing bad lock handle
7642 - fix recovery NULL deref in ldlm_cli_cancel_unused
7643 - fix some DLM races by using new hash table for lock handles (419)
7644 - permit the client to specify desired inodes, at replay
7645 - duplicate requests when we queue them for replay reintegration
7646 - fix last_rcvd offset calculation
7647 - sync after each recovered transaction, so we always make progress
7648 - never, not always, ERESTART requests without transnos
7649 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
7651 - skip replay if the MDS says that the client is already connected
7652 - don't check for a recovery-enabled export to match lctl's UUID
7653 - don't INC_USE_COUNT for phantom exports
7654 - don't crash when cleaning up phantom exports (567)
7655 - don't double-finish or set replay data for errored mdc_open requests
7656 - abort requests when they time out, so we don't get old replies
7657 - send/receive replies for AST messages again
7658 - if the client says that it doesn't have the lock, cancel it on the
7660 - if we timeout during I/O, don't try to cancel an in-use lock; instead
7661 mark it as destroyed, it will all work out when decref is called
7662 - fix module use counts (22, 581)
7664 - ASTs now expect a reply (server cancels lock on error reply)
7666 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
7669 - fix many simultaneous client startup (392)
7670 - fix dentry->d_it clobbering
7671 - credentials weren't being shipped for readdir/getattr operations
7672 - remove invalid assertions triggered during some concurrent MD
7674 - proper Lustre versions added (336, 389)
7675 - fix memory leak for create error case (398)
7676 - fix LOV locking bug that would get cli/srv out of sync
7677 - fix echo client over LOV (409)
7678 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
7679 - fix double-O_EXCL intent crash (424)
7680 - avoid sending multiple lock CANCELs (352)
7682 - MDS can do multi-client recovery (modulo bugs in new code)
7684 - many updates, edits, cleanups
7686 2002-11-18 Phil Schwan <phil@clusterfs.com>
7689 - fix null d_it dereference (346)
7690 - fix full OST/dbench hang (333)
7691 - fix permission problem with file removal (286)
7692 - fix removal of OSCs from LOV when they fail
7693 - fix NULL deref during bulk timeout (214)
7694 - fix problems related to multiple filesystems on one MDS (241)
7695 - fixed serious subtle metadata locking bugs
7696 - free locks on clients when inodes are removed due to memory
7698 - fix inode pointer in lock data (285)
7699 - partial support for multiple MDS on a single host (241)
7700 - data locks weren't cancelled at clear_inode time (290, 311)
7701 - intent locks could lead to unbounded lock growth (205)
7702 - added a maximum lock count, an LRU list, and a flusher
7703 - fix multiple rename (365)
7704 - properly abstracted the echo client
7705 - OSC locked 1 byte too many; fixed
7706 - rewrote brw callback code:
7707 - fixed recovery bugs related to LOVs (306)
7708 - fixed too-many-pages-in-one-write crash (191)
7709 - fixed (again) crash in sync_io_timeout (214)
7710 - probably fixed callback-related race (385)
7712 - Add capability to MDS protocol
7713 - LDLM cancellations and callbacks on different portals
7715 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
7718 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
7719 - defer open object destruction to close time (601981, 138)
7720 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
7721 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
7722 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
7723 - don't call obd_setattr in ll_file_release for destroyed objects
7724 * protocol change to lustre_msg: move |version| and add |flags|
7725 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
7726 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
7727 * added replay of create, unlink, link and rename operations during
7728 MDS failover; recovery should be much more robust now
7729 * remove failed OSCs from LOVs (only lov_create uses this so far)
7730 * the lustre-HOWTO was brought (more) up to date (582544)
7732 2002-10-23 Phil Schwan <phil@clusterfs.com>
7735 - in-use dentries weren't being reused properly (617851)
7736 - prevent multiple LDLM setup (599178)
7737 - fix LOV size calculations for truncate (617853)
7738 - fix client handling of MDS intent errors (POSIX)
7739 - fix permission bug in lovstripe.c test (624321)
7740 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
7741 - truncate past end of file could corrupt data
7742 - proper cleanup after timeouts, crashes, etc (592524, 550815)
7743 - a race in recovery could return ETIMEDOUT to apps (623947)
7744 - building outside the source directory was fixed
7745 * the lustre-HOWTO was brought (more) up to date (582544)
7746 * major progress was made on recovery functionality
7748 2002-10-10 Phil Schwan <phil@clusterfs.com>
7751 - recovery deadlock fix
7752 - rm -rf causes LBUG fix (617817)
7753 - file open by multiple tasks fix (618962)
7754 - directory permissions bugs (602707 and 620007)
7755 - journal_stop fixed with locking (611313)
7756 - O_APPEND failures resolved (618273, perhaps 614459)
7757 - lconf PATH fix (619770)
7758 - IA64 build fix (621450)
7759 - RPC buffer sizes scale with amount of memory
7761 2002-10-01 Phil Schwan <phil@clusterfs.com>
7764 - locks would be cancelled without throwing away data pages,
7765 resulting in inconsistent data (605627)
7766 - inode attributes were not always being refreshed (605627, 612449)
7767 - lconf now continues to cleanup after lctl reports an error
7768 - MDS now enforces user permissions (602707)
7769 - lprocfs cleanup fixed, but not yet enabled (614157)
7770 - fixed infinite server hang, should a client not respond to an AST
7771 - avoid going into recovery if user calls readlink() with a buffer
7772 that's too small (613941)
7773 - AST RPCs no longer require replies (614867) -- this may be changed
7774 - don't crash server if client sends an IOV that's too big (611336)
7775 - fixed lock conversion deadlock (611892)
7776 - fixed the following of symlinks (614622)
7777 * recovery: the server can remove locks from a client that dies, other
7778 clients can make progress
7779 * more extN patch fixes
7780 * compile-time configurable ptlrpc buffer allocations
7782 - collaborative read cache document
7783 - Lustre Lite Performance CDR document-in-progress
7785 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
7788 - fix typo in patch-2.4.18
7790 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
7793 - clear ptlrpc request each time in handle_incoming_request()
7794 - unlink of files now destroys the object on the OST
7796 2002-09-19 Peter Braam <braam@clusterfs.com>
7798 * add hard link support
7799 * change obdfile creation method
7800 * kernel patch changed
7802 2002-09-19 Peter Braam <braam@clusterfs.com>
7805 - stack overflow bug in extN fixed
7807 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
7809 * documentation updates
7810 - add man pages for config tools
7811 - update tests/README to describe testing with new config tools
7812 - finish metadata API descriptions
7813 * bug fixes and cleanups
7814 - statfs workaround for 16TB limit
7815 - LOV stripe allocation improved, can stripe on subset of OSTs
7816 - LOV file size/IO offset was wrong for files > 4GB in size
7817 - object EA data was being dropped, caused files to be unreadable
7818 - memory overflow with non-LOV OST caused memory corruption
7819 - fixed regression tests to work with new config tools, obdfilter
7820 - fixed bug when directory size became larger than 1 block
7821 - fixed bug (for single client case) when PWD was deleted
7822 - invalidate local directory pages when doing intent-based ops
7823 - avoid LDLM oops when lock callback contained bad data
7825 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
7827 * documentation updates
7828 * bug fixes and cleanups
7829 - configuration tools
7832 - 64-bit compile warnings
7833 - 64-bit internal statfs data
7835 * test_brw on persistent OST devices
7837 * lprocfs (disabled)
7839 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
7841 * documentation updates
7842 * bug fixes and cleanups
7843 * configuration tools
7845 2002-08-30 Peter J. Braam <braam@clusterfs.com>
7848 * many small fixes to 0_5_4
7849 * io/network handling
7850 * thinkos in MDS operations
7852 2002-08-24 Peter J. Braam <braam@clusterfs.com>
7855 * crucial basic fixes to 0.5.3
7856 * IOR, Iozone work over Elan
7859 2002-08-07 Phil Schwan <phil@clusterfs.com>
7860 * version 0_5_3, our first alpha
7861 * we use the new Portals iovs
7862 * documentation updates
7863 * bug fixes and cleanups
7864 * small changes in the DLM wire protocol
7866 2002-07-25 Peter J. Braam <braam@clusterfs.com>
7867 * version 0_5_1 with some initial stability,
7868 * locking on MD and file I/O.
7869 * documentation updates
7870 * several bug fixes since 0.5.0
7871 * small changes in wire protocol
7873 2002-07-18 Phil Schwan <phil@clusterfs.com>
7875 * delivered as Lustre Light Alpha
7876 * fixed a crash after handling invalid MDS requests
7877 * fixed directory pages for architectures with non-4k pages sizes
7879 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
7880 * release version v0_4_4
7881 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
7883 * Fixes a number of interruption problems with OST operations.
7884 * Update documentation for portals header changes
7885 * Move all wire protocol structs/defines to lustre_idl.h
7886 * Fixes symlink length bug.
7887 * Add tcpdump to repository.
7889 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
7890 * release version v0_4_3
7891 * Fixes statfs for inodes on extN.
7892 * Fixes bug in runtests which would delete /etc/hosts.
7893 * Use 64-bit object IDs wherever possible (not into VFS though)
7894 Remove ost_get_info, which is unused by lustre, and out of date.
7896 2002-07-03 Peter Braam <braam@clusterfs.com>
7897 * release version v0_4_2 Fixes a lookup error (type not passed)
7898 * move forward to head of Portals
7899 * move forward to latest Lustre kernel
7901 2002-06-25 Peter Braam <braam@clusterfs.com>
7902 * release version v0_4_1. Hopefully stable on single node use.