1 2012-xx-xx Whamcloud, Inc.
4 2.6.18-308.4.1.el5 (RHEL 5)
5 2.6.32-220.13.1.el6 (RHEL 6) - patchless client only
6 * Recommended e2fsprogs version: 1.41.90.wc3
8 See http://wiki.whamcloud.com/display/PUB/Changelog+1.8 for more information.
10 Severity : enhancement
12 Description: Kernel update [RHEL5.8 2.6.18-308.4.1.el5]
14 Severity : enhancement
16 Description: Update RHEL6.2 patchless client to 2.6.32-220.13.1.
18 Severity : enhancement
20 Description: Kernel update [RHEL5.8 2.6.18-308.1.1.el5]
22 Severity : enhancement
24 Description: Update RHEL6.2 kernel to 2.6.32-220.7.1.
26 Severity : enhancement
28 Description: Add patchless client support for RHEL6.2.
30 -------------------------------------------------------------------------------
32 2011-10-31 Whamcloud, Inc.
34 * Support for kernels:
35 2.6.18-274.3.1.el5 (RHEL 5)
36 2.6.32-131.12.1.el6 (RHEL 6) - patchless client only
37 * Recommended e2fsprogs version: 1.41.90.wc3
39 See http://wiki.whamcloud.com/display/PUB/Changelog+1.8 for more information.
41 -------------------------------------------------------------------------------
43 2011-06-30 Whamcloud, Inc.
45 * Support for kernels:
46 2.6.18-238.12.1.el5 (RHEL 5)
47 2.6.32-131.2.1.el6 (RHEL 6) - patchless client only
48 * Recommended e2fsprogs version: 1.41.90.wc3
49 * Use ext4-based ldiskfs as default for RHEL5
50 * Add support for 24TB LUN (RHEL5 server only)
52 Severity : enhancement
54 Description: Add support for RHEL5.6 (2.6.18-238.12.1.el5)
56 Severity : enhancement
57 Jira : LU-62, LU-73, LU-402, LU-130
58 Description: Add RHEL6 client support
62 Description: Fix client crash introduced in 1.8.5
63 Details : The patch from bugzilla ticket 18213 landed for 1.8.5 can cause
64 a crash in mdc_exit_request() when the process is interrupted.
68 Description: ls -l reports wrong file size
69 Details : Always update LVB from disk when a glimpse callback returns an
72 Severity : major (2.6.32-based kernels only)
74 Description: Add workaround for a race causing an assertion failure in
75 clear_inode() on lustre clients.
79 Description: Fix interoperability issue between 1.8 clients and 2.x servers
80 Details : The ldlm hash handling in Lustre 2.x is incompatible with the one
81 used in Lustre 1.8. Add support for 64-bit dir name hash to the
86 Description: slow IO with read-intense application
87 Details : Align the readahead extent by 1M after when it is trimed by
92 Description: Skip quotacheck on administratively disabled OSTs
96 Description: Some autogen improvements
98 Severity : enhancement
99 Jira : LU-123, LU-296, LU-230
100 Description: Add support for yaml data logging and auster script.
102 Severity : enhancement
104 Description: Allow OSTs to be created with no primary node
105 Details : Add a --servicenode parameter for mkfs.lustre to treat all
106 service nodes equally.
108 -------------------------------------------------------------------------------
110 2011-05-12 Oracle, Inc.
112 * Support for kernels:
113 2.6.16.60-0.42.8 (SLES 10),
114 2.6.27.39-0.3.1 (SLES 11),
115 2.6.18-194.3.1.el5 (RHEL 5)
116 2.6.18-194.3.1.0.1.el5 (OEL 5)
117 * Client support for unpatched kernels:
118 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
119 2.6.16 - 2.6.32 vanilla (kernel.org)
120 * Recommended e2fsprogs version: 1.41.10-sun2
121 * The async journal commit feature (bug 19128) and the cancel
122 lock before replay feature (bug 16774) are disabled by default.
126 Description: llap pages are moved to head of LRU list instead of tail
130 Description: disable statahead by default due to important races found in the
131 statahead implementation.
135 Description: release request ref through ptlrpc_server_drop_request()
136 Details : make sure the request is protected by rq_refcount while
137 being processed in ptlrpc_server_handle_req_in().
141 Description: correct check for transno value in filter_finish_transno()
145 Description: Set body->eadatasize in mdc_getattr_pack()
149 Description: Handle unsent requests with rq_net_err in ptlrpc_check_set()
153 Description: apps stuck in ptlrpc_check_set() during direct I/O
154 Details : start async bulk unregistering at the same time at reply unlink
158 Description: lfs setstripe --pool broken in 1.8.4
159 Details : fix llapi_search_fsname() to handle relative pathnames
163 Description: Only force the mode change if we're changing the size as well
164 Details : Fix regression introduced in 1.8.0. The offending code was added
165 by commit 77ba4b2141d04180211efa8a75c11ab0abf7fafb to remove
166 setgid/setuid bits when do_truncate() is called on the file.
167 We should only force the change when that occurs, similarly to
168 ll_setattr() in lustre/llite/llite_lib.c
170 Severity : enhancement
172 Description: add procfs tunable to enable/disable lockless direct I/O
173 Details : llite.lustre-*.lockless_direct_io=0 will disable default semantics
174 of direct I/O that forces it to be lockless. lockless_direct_io
175 value, however, will be ignored if per-file LL_FILE_LOCKED_DIRECTIO
180 Description: Lustre servers crashing with NULL pointer errors
181 Details : Make sure the request is protected by rq_refcount while being
182 processed in ptlrpc_server_handle_req_in().
186 Description: kernel BUG at drivers/scsi/sd.c
187 Details : Remove sd iostats patch from sles11 patch series.
191 Description: Modified value of at_min is not taken into account
195 Description: lustre grants flock exclusive locks to two fd's in the same process
199 Description: LBUG in the mfs_verify_child()
203 Description: ASSERTION(atomic_read(&client_stat->nid_exp_ref_count) == 0)
204 Details : Reconnecting from a different nid causes old per-nid stats not to be
209 Description: create proper macro check for bdi interface.
211 -------------------------------------------------------------------------------
213 2010-10-29 Oracle, Inc.
215 * Support for kernels:
216 2.6.16.60-0.69.1 (SLES 10),
217 2.6.32.19-0.2.1 (SLES11),
218 2.6.18-194.17.1.el5 (RHEL 5)
219 2.6.18-194.17.1.0.1.el5 (OEL 5)
220 * Client support for unpatched kernels:
221 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
222 2.6.16 - 2.6.32 vanilla (kernel.org)
223 * Recommended e2fsprogs version: 1.41.10-sun2
224 * The async journal commit feature (bug 19128) and the cancel
225 lock before replay feature (bug 16774) are disabled by default.
229 Description: atime is not properly updated on an MDS
231 Severity : enhancement
233 Description: Update to RHEL5.5 kernel 2.6.19-194.17.1.el5.
234 Update to OEL5.5 kernel 2.6.19-194.17.1.0.1.el5.
236 Severity : enhancement
238 Description: Update to SLES10 SP3 kernel 2.6.16.60-0.69.1.
241 Frequency : only with SLES10
243 Description: Use OFED "KMP" provided by Novell
244 Details : SLES10 SP3 ships with OFED in a separate "KMP" package.
245 Lustre is now built against this package. That means you need to
246 install the ofed-kmp package from Novell for the patchless client
247 and from our download site for the server. Note that the ofed-kmp
248 that Novell ships may not exactly match the kernel version but
249 should still be compatible.
251 Severity : enhancement
253 Description: Update SLES11 SP1 kernel to 2.6.32.19-0.2.1.
257 Description: Enabling quotas fails with non-consecutive OST numbering.
261 Description: Fix kernel warning due to lookup_on_len() called without i_mutex
266 Description: Account direct i/o inflight rpcs separately from non-direct i/o so
267 that direct i/o, which is limited by max_rpcs_in_flight, should not
268 block non-direct i/o, which is not limited by max_rpcs_in_flight.
272 Description: Fix per-NID reporting on outstanding writes
276 Description: Reduce stack pressure by uninlining some mds and ptlrpc functions.
280 Description: Remove LASSERT in lprocfs_rd_conn_uuid() since conn == NULL is a
285 Description: fix obdo leak issue in ll_setattr_raw()
289 Description: limit MMP interval
291 Severity : enhancement
293 Description: add several lfs ost enhancements
297 Description: Too many default ACLs break directory access on new directories
301 Description: Lustre inode size is not coherent across nodes.
302 Details : Update lvbo from disk when AST fails with EINVAL. Lvbo is updated
303 on EINVAL error in ldlm_handle_ast_error(). The updates in
304 filter_intent_policy() and ldlm_cb_interpret() have been removed as
309 Description: Oops at __percpu_counter_add+0x1b
310 Details : Use bdi_init()/bdi_destroy() to proper initialize backing_dev_info
315 Description: add mount option to generate 32bit ino, this can be used for 32bit
316 application compatibility.
320 Description: keep reference count for "lli_sai" to prevent it to be released
321 when "statahead_enter()"
325 Description: allow quotacheck over OSTs with sparse indices
329 Description: Objects not getting deleted for files which have been removed
330 Details : ll_have_md_lock() should differentiate between CR and CW OPEN
335 Description: pin object's inode in memory to avoid certain timeouts
339 Description: fix LBUG when obdfilter-survey is interrupted.
341 -------------------------------------------------------------------------------
343 2010-07-31 Oracle, Inc.
345 * Support for kernels:
346 2.6.16.60-0.42.8 (SLES 10),
347 2.6.27.39-0.3.1 (SLES11),
348 2.6.18-194.3.1.el5 (RHEL 5)
349 2.6.18-194.3.1.0.1.el5 (OEL 5)
350 * Client support for unpatched kernels:
351 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
352 2.6.16 - 2.6.30 vanilla (kernel.org)
353 * Recommended e2fsprogs version: 1.41.10-sun2
354 * The async journal commit feature (bug 19128) and the cancel
355 lock before replay feature (bug 16774) are disabled by default.
357 Severity : enhancement
359 Description: Update RHEL5.5 kernel to 2.6.18-194.8.1.el5 and OEL5.5 kernel to
360 2.6.18-194.8.1.0.1.el5.
362 Severity : enhancement
364 Description: using inkernel OFED stack for rhel5 & oel5.
366 Severity : enhancement
368 Description: Add "lfs_migrate" script from manual into lustre/scripts and RPMs
369 Details : lfs_migrate does a "poor man's" migration of files from their
370 current OST layout to a new OST layout as chosen by the MDS.
374 Description: mds_orphan_add_link()) error linking orphan to PENDING
375 Details : quota limits might disallow linking orphans to PENDING
376 when unlinking a file - temporary raise threads' privileges
377 when processing unlinks.
379 Severity : enhancement
381 Description: add conf-param -d option to remove permanent settings.
382 Details : Add the ability to remove permanent lctl conf_param settings.
383 (Previously conf_param settings could only be changed, not
384 removed.) This also provides a method to change failover
385 nid locations. Improve lctl man page.
387 Severity : enhancement
389 Description: add list_param to b1_8 and add "-R" option to list params
392 Severity : enhancement
394 Description: lfs quota output is not very convenient for awk/sed-parsing
395 Details : Some positions in lfs quota output table could be empty or
396 non-empty which made it hard to parse it with scripts, now a dash
397 is put instead of space where there is not supposed to be any data.
399 Severity : enhancement
401 Description: fix obdfilter-survey script to work properly with remote oss-s
403 Severity : enhancement
405 Description: add new OBDFILTER_SURVEY test suite
407 Severity : enhancement
409 Description: add new multiple mount protection (MMP) test suite
411 Severity : enhancement
413 Description: add support for async journal commit in echo client
415 Severity : enhancement
417 Description: allow userland programs to include <lustre/lustre_idl.h>
418 from stardard include directories
420 Severity : enhancement
422 Description: The prune-icache-use-trylock is no longer needed now that
423 the patch from bug 20008 is landed.
427 Description: The shrink grant feature is still active on the client although the
428 connect flag is not set.
432 Description: Don't leak grant space if the write failed with quota exceeded.
436 Description: Don't consume grant space twice on recoverable resent.
440 Description: a race condition could lead to SIGBUS being sent to an
441 application using mmap-ped files from Lustre
442 Details : truncate_complete_page implementation for the patchless
443 client could arbitrarily unset PG_Uptodate flag for a
444 page being kicked from the page cache, an uptodate check
445 right after a readpage call in filemap_fault could fail
446 because of that as though the page read had been unsuccessful.
450 Description: dlm lock slab shrinking is not efficient
451 Details : The dlm_locks slab can grow significantly and consumes a lot of
452 memory on the server. Set a hardlimit to grant_plan.
456 Description: Lustre does not do 1MB IOs to HW RAID
457 Details : Bump MAX_PHYS/HW_SEGMENTS and SG_ALL to 256 in the RHEL5 kernel.
458 This is what we do already for SLES kernels.
462 Description: bump maximum number of phys/hw segments in the SLES11 kernel
463 until s/g chaining works properly.
468 Description: LSI Fusion MPT driver hacks to improve performance
469 Details : Set CONFIG_FUSION_MAX_SGE to 256 for RHEL5
471 Severity : enhancement
473 Description: increase default md stripe_cache_size to 16k
476 Bugzilla : 15587/21439
477 Description: don't handle security.capability xattr
478 Details : CONFIG_SECURITY_FILE_CAPABILITIES is enabled by default on SLES11.
479 This results in additional getxattr calls, causing VBR test
480 failures as well as a preformance drop when writing.
484 Description: obdfilter-survey is no longer working
485 Details : revert patch from bug 20355 to resolve an issue with lctl --threads
486 not working correctly with $(PTHREAD_LIBS) being linked to lctl.
490 Description: ll_shrink_cache does not handle __GFP_FS properly
494 Description: lfs getstripe shows wrong info for directories
495 Details : Set correct LOVEA default values for filesystem-wide.
499 Description: FSX checksum false positves due to mmap IO
500 Details : Use OBD_FL_MMAP flag for IOs on a memory mapped file. Do not print
501 checksum errors, if the flag is set on a request.
505 Description: file operations after eviction have successful return values
506 Details : use vfs ->flush callback to return any pending async errors
511 Description: mdsrate fails to write after 1.3+M files opened
512 Details : decrease memory usage on clients by recycling dentries and
517 Description: obdfilter-survey gives unreasonably high numbers
518 Details : Wait for all threads to complete when running test_brw.
522 Description: do not set lustre read_only device when server umount and keep
523 client records for recoverable ones
527 Description: move sync_on_lock_cancel tunable to the obdfilter layer
528 Details : move the tunable to trigger a journal flush on lock cancel from
529 the ost layer to the obdfilter layer. This tunable is useful
530 when using the async journal commit feature.
534 Description: exp->exp_nid_stats == NULL in filter_tally()
535 Details : fix race with per-nid stats by delaying procfs cleanup until
540 Description: extent lock cancellation on client can keep the cpu busy for too
545 Description: Do not fail OST activation when a llog is not found, just
546 issue an error message.
550 Description: Don't enable extents by default for MDT.
554 Description: Protect bitfield access to ptlrpc_request's rq_flags, since
555 the AT code can access it concurrently while sending early
560 Description: Disable lockless truncate by default since it is sometimes flawed
561 and causes the write_disjoint test to fail.
565 Description: OSSs which don't have the patch from bug 20278 can trigger an
568 Severity : enhancement
570 Description: don't print message to the console when we have not managed to
575 Description: The MDS fails to synchronize OSTs which registered with the MGS
576 after the MDT. The problem is that OBD_NOTIFY_CREATE events are
577 raised too early and thus discarded by the MDT stack.
578 The fix consists of issuing OBD_NOTIFY_CREATE event in the lov
583 Description: Fix race when the ping evictor and a service thread execute
584 target_recovery_check_and_stop() concurrently.
588 Description: quota broadcast can trigger a LBUG on the MDT if there are
593 Description: Resetting the lov_objid values to last_id reported by the OST
594 during orphan recovery is incorrect and can cause the same
595 objects to be allocated twice.
597 Severity : enhancement
599 Description: "weak-modules" support
600 Details : Implement "weak-modules" support which enables kernel modules
601 to be used with any kernel that implements the same kABI. In
602 order to achieve this modules are now installed in
603 /lib/modules/$(uname -r)/updates/kernel on all distributions.
605 Severity : enhancement
607 Description: add writeconf as mount option
609 Severity : enhancement
611 Description: produce debuginfo packages for SLES.
613 Severity : enhancement
615 Description: add failover nidlist to the import proc file.
617 Severity : enhancement
619 Description: fix LUSTRE_SEQ_MAX_WIDTH for interoperability between 1.8
620 clients and 2.0 servers.
622 Severity : enhancement
624 Description: lfs find -s does not work correctly because of a bug in
629 Description: ll_read_ahead_page() must validate the dlm lock before using
634 Description: Prevent failover nids from registering with MGS first.
638 Description: fix lock inversion in ll_setattr_raw().
642 Description: object allocation is not balanced across OSTs.
643 Details : osc_precreate() should return 0, if there are enough objects left.
645 -------------------------------------------------------------------------------
647 2010-04-30 Oracle, Inc.
649 * Support for kernels:
650 2.6.16.60-0.42.8 (SLES 10),
651 2.6.27.39-0.3.1 (SLES11),
652 2.6.18-164.11.1.el5 (RHEL 5)
653 2.6.18-164.11.1.0.1.el5 (OEL 5)
654 * Client support for unpatched kernels:
655 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
656 2.6.16 - 2.6.30 vanilla (kernel.org)
657 * Recommended e2fsprogs version: 1.41.10-sun2
658 * The async journal commit feature (bug 19128) and the cancel
659 lock before replay feature (bug 16774) are disabled by default.
663 Description: open-unlinked directories trigger MDS LBUG
664 Details : Fix regression introduced by the patch from bug 19640.
665 ext3_inc_count() can reset nlink to 1 when the directory
666 is indexed and inode->i_nlink == 2. Work around the problem
667 by incrementing nlink by 2 instead of 1.
671 Description: Reconnects are not throttled
672 Details : Don't wake up pinger on reconnect failures and rely on regular
673 pings to trigger the next reconnection. Please note that the
674 pinger already uses a smaller interval if the import is
678 Frequency : only with NFS export
680 Description: Console flooded with error message from ll_inode_from_lock()
682 Details : in mds_open, initialize the child_res_id before enqueuing
683 the OPEN lock for the child inode, then to avoid senting
684 wrong ldlm_res_id to client.
688 Description: allow multiple instances of the same nid in NID hash
689 Details : Case of multiple separate clients from the same NID (as
690 with liblustre) is legitimate and so we should allow
691 multiple instances of the same NID in nid hash.
695 Description: the readahead code can sleep on a semaphore while holding a
697 Details : in ras_update, "lov_get_info" could be called during increasing
698 readahead windows, which tries to get the mutex lock "lov_lock"
699 while holding the spin_lock "ras_lock", then causes system
704 Description: ASSERTION(cli->cl_avail_grant >= 0) failed
705 Details : fix assertion failure in the grant code.
709 Description: Use CNETERR (which is rate limited) in specific places in
710 the portal's LNET driver to avoid flooding the console.
714 Description: include last created object in precreate slow case
718 Description: don't do rep-ack if not created anything
719 Details : mds_open currently always put a lock into a rep-ack regardless
720 if something was created or not. This is pointless and only
721 creates needless contention. In fact the entire idea was to do
722 this for real creates as a recovery protection.
726 Description: Spurious error messages from smp_processor_id() on preemptible
728 Details : Disable a preemption by grabbing the lock in fs_trace_get_tcd()
729 first. The function fs_trace_get_tcd() was moved up.
733 Description: interval_erase() fix
734 Details : interval_erase() calls update_maxhigh() properly when child
739 Description: "lfs df" does not print stats for all mountpoints
740 Details : Print all mounted lustre filesystems with "lfs df"
744 Description: lfs setstripe -p not longer work with a relative pathname
745 Details : Use realpath() to provide absolute pathname.
749 Description: fix for truncated reply buffer
750 Details : reply buffer could be referred by reply_in_callback after released
755 Description: lustre.lov error when backing up symlinks with extended attributes
756 Details : Improved logic in ll_listxattr()
760 Description: properly handle null value for setattr -n lustre.lov
761 Details : Running "setfattr -n trusted.lov ." causes a NULL dereference
762 in ll_setxattr() due to no checking if "value" is NULL.
763 This command now resets to the default striping when executed
768 Description: stack overflow on lock cancellation due to fsync call
769 Details : sync_on_lock_cancel is needed for recovery when async journal
770 is enabled, but we actually just need to make sure that
771 metadata blocks have hit the journal, so doing a fs sync
772 should be enough and should consume less stack (just create an
773 empty handle and commmit it).
777 Description: using current->journal_info to store per-thread data leads
778 to problem under memory pressure
779 Details : disable the per-thread data (current->journal_info) containing
780 the lock info during I/O to work around the issue for short tem
784 Description: control DCACHE_LUSTRE_INVALID flag with MDS_INODELOCK_LOOKUP lock
785 Details : DCACHE_LUSTRE_INVALID is controlled by MDS_INODELOCK_LOOKUP
786 lock which is corresponding to "IT_LOOKUP", do not skip invalidate
791 Description: Cannot send after transport shutdown
792 Details : Clear imp_vbr_failed flag upon eviction
796 Description: soft lock in request set code during recovery
797 Details : during recovery, uses req->rq_set itself to replay the request
798 instead of ptlrpcd_recovery_pc
802 Description: Use CFS_ALLOC_IO instead of _STD in llap_from_page_with_lockh
803 Details : During an ll_readahead under ll_readpage, we have seen the the
804 OBD_SLAB_ALLOC hang under ldlm_pools_shrink when trying to lock
805 a page that is already locked by the readahead code.
810 Description: stop waitting for next replay transno if shutdown
811 Details : if the system is shutting down, wake up service thread blocked
812 to wait for next replay transno during recovery, then all the
813 references held by queued requests can be dropped and device
818 Description: lov_merge_lvb()) ASSERTION(spin_is_locked(&lsm->lsm_lock)) failed
819 Details : Protect lli->lli_smd pointer updates with lli->lli_lock.
823 Description: per-nid stats should not access lustre-hash internal structures
828 Description: mount.lustre fails to pass some options to mount()
833 Description: ext4 extent allocation is slower than in ext3
834 Details : Increase the default value of MB_DEFAULT_ORDER2_REQS to 8,
835 enlarge ext4 preallocation table for 2048 4K blocks extents
840 Description: incorrect triggering of synchronous IO
841 Details : The OSC can mistakenly fall back to synchronous IO when the
842 max_dirty_mb limit is reached and no write requests have yet
843 been issued. This can occur when the dirty pages are spread
844 over many files all of which are below the optimal request size.
848 Description: Optimize quota_ctl operations by sending requests in parallel
849 Details : Send MDS->OST quota_ctl requests in parallel, do not resend.
850 Compiled from two attachments in the ticket.
854 Description: ordering issue between transaction start & i_mutex
855 Details : start the transaction earlier in llog_lvfs_destroy to get
856 transaction start and inode mutex lock nested properly.
860 Description: lru resize SLV can get stuck
861 Details : calculate SLV with a greater precision to not lose small
862 changes due to interger math truncation; round up SLV only
863 if the amount of granted locks less than the limit to not
864 get stuck with this SLV
868 Description: avoid divide-by-zero in lprocfs_rd_import()
872 Description: lfs quota failed when OSTs are down
873 Details : really return approximate block/inode usage when OSTs are down
877 Description: llobdstat fix and enhancement
878 Details : add a counter to set a limit to how many samples will be returned
879 fix a wildcard in the path to limit to obdfilter stats only
883 Description: recovery-small 51 hang - MDT can not stop, Mount still busy with
885 Details : abort lock enqueue processing sooner when we umount is in progress.
887 Severity : enhancement
889 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and OEL5.4 kernel to
890 2.6.18-164.11.1.0.1.el5.
892 Severity : enhancement
894 Description: error message improvements
895 Details : Use INFO/WARN instead of WARN/ERROR for the slow messages.
896 Simplify MDT/OST service start message.
897 Suppress "changing the import ..." warning.
900 Bugzilla : 21961/17914
901 Description: ignore trailing -mdc when determining index number
904 Frequency : only with SLES11
906 Description: fix for a race condition in linux quotas implementation
907 Details : dq_flags(struct dquot) access is not properly locked which could
908 lead to certain inconsistencies when accessing it using non-atomic
909 bit operations like __set_bit in do_set_dqblk.
913 Description: return any pending async errors in close(2) using flush callback
915 -------------------------------------------------------------------------------
917 2010-01-29 Sun Microsystems, Inc.
919 * Support for kernels:
920 2.6.16.60-0.42.8 (SLES 10),
921 2.6.27.39-0.3.1 (SLES11),
922 2.6.18-164.11.1.el5 (RHEL 5)
923 2.6.18-164.6.1.0.1.el5 (OEL 5)
924 * Client support for unpatched kernels:
925 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
926 2.6.16 - 2.6.30 vanilla (kernel.org)
927 * Recommended e2fsprogs version: 1.41.6.sun1
928 * The async journal commit feature (bug 19128) and the cancel
929 lock before replay feature (bug 16774) are disabled by default.
931 Severity : enhancement
933 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and
934 OEL5.4 kernel to 2.6.18-164.11.1.0.1.el5.
936 Severity : enhancement
937 Bugzilla : 21511/19848
938 Description: Update SLES11 kernel to 2.6.27.39-0.3.1.
940 Severity : enhancement
942 Description: Update supported SLES10 kernel to 2.6.16.60-0.42.8.
944 Severity : enhancement
946 Description: Update kernel to RHEL5.4 2.6.18-164.6.1.el5 and
947 OEL5 2.6.18-164.6.1.0.1.el5(Both in-kernel OFED enabled).
949 Severity : enhancement
951 Description: Build kernels (RHEL5, OEL5 and SLES10/11) using the vendor's own
954 Severity : enhancement
956 Description: Vanilla kernel 2.6.30 patchless client support.
961 Description: bad entry in directory xxx: inode out of bounds
962 Details : fix locking issue in the rename path which could race with any
963 other operations updating the same directory.
965 Severity : enhancement
967 Description: Make watchdog timer messages to be more clear and descriptive.
971 Description: cp -p command does not preserve the dates and timestamp
972 Details : mtime could be spoiled by a write callback
976 Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
980 Description: Allow non-root access for "lfs check".
981 Details : Added a check in obd_class_ioctl() for OBD_IOC_PING_TARGET.
983 Severity : enhancement
985 Description: quotacheck performance/scaling issues
986 Details : reduce quotacheck time on empty filesystem by skipping uninit
989 Severity : enhancement
991 Description: Enhancement for lfs(1) command to use numeric uid/gid.
993 Severity : enhancement
995 Description: Adjust locks' extents on their first enqueue, so that at the time
996 they get granted, there is no need for another pass through the
997 queues since they are already shaped into the proper forms.
1001 Description: Fix mds_shrink_intent_reply()/mds_intent_policy() to pass correct
1002 arguments and prevent LBUG() in lustre_shrink_reply_v2().
1006 Description: Change tunefs.lustre and mkfs.lustre --mountfsoptions so that
1007 exactly the specified mount options are used. Leaving off
1008 any "mandatory" mount options is an error. Leaving off any
1009 default mount options causes a warning, but is allowed.
1010 Change errors=remount-ro from mandatory to default. Sanitize
1011 the mount string before storing it. Update man pages accordingly.
1015 Description: mds_getattr() should return 0, even if mds_fid2entry() fails with
1016 -ENOENT. Also fix in ptlrpc_expire_one_request() to print signed
1019 Severity : enhancement
1021 Description: Remove set_info(KEY_UNLINKED) from MDS/OSC
1023 Severity : enhancement
1025 Description: Clients can replay thousands of unused locks during recovery
1026 Details : Don't replay unused locks (only read locks for now) during
1027 recovery. This feature is disabled by default and can be
1028 enabled by running the following command on the clients:
1029 lctl get_param ldlm.cancel_unused_locks_before_replay
1033 Description: can't stat file in some situation.
1034 Details : improve initialize osc date when target is added to mds and
1035 ability to resend too big getattr request is client isn't have info
1040 Description: Prevent inconsistences between linux and lustre mount structures.
1041 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
1042 Make the sleep interruptible.
1044 Severity : enhancement
1046 Description: Communicate OST degraded/readonly state via statfs to MDS
1047 Details : Flags in the statfs returned from OSTs indicate whether the
1048 OST is in a degraded RAID state, or if the filesystem has
1049 turned read-only after a filesystem error is detected.
1054 Description: don't panic if EPROTO was hit when reading symlink
1055 Details : correctly handling request reference in error cases.
1060 Description: open sometimes returns ENOENT instead of EACCES
1061 Details : checking permission should be part of open part of mds_open, not
1062 lookup part. so server should be set DISP_OPEN_OPEN disposition
1063 before starting permission check.
1064 Also not need revalidate dentry if client already have LOOKUP lock.
1068 Description: enable client interface failover
1069 Frequency : on servers with multiple network interfaces
1070 Details : When a child reconnects from another NID, properly update export
1071 nid hash position and ldlm reverse import.
1073 Severity : enhancement
1075 Description: implemented direct I/O with arbitrary (nonaligned) memory
1076 addresses and file offsets.
1078 Severity : enhancement
1080 Description: added more recovery timeout options.
1082 Severity : enhancement
1084 Description: added llapi_file_open, llapi_file_create, llapi_file_get_stripe
1089 Description: Avoid deadlock for local client writes
1090 Frequency : only on systems with clients writing to an OST on the same node
1091 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
1092 memory freeing context. This allows OST threads to set the
1093 PF_MEMALLOC flag on task structures in order to allocate memory
1094 from reserved pools and complete IO.
1095 Use GFP_HIGHUSER for OST allocations for non-local client writes,
1096 so that the OST threads generate memory pressure and allow
1097 inactive pages to be reclaimed.
1102 Descriptoin: lock ordering violation between &cli->cl_sem and _lprocfs_lock
1103 Details : move ldlm namespace creation in setup phase to avoid grab
1104 _lprocfs_lock with cli_sem held.
1108 Frequency : only during format of test systems
1109 Description: Unable to run several mkfs.lustre on loop devices at the same time
1110 Details : mkfs.lustre returns error 256 on the concurrent loop devices
1111 formatting. The solution is to proper handle the error.
1113 Severity : enhancement
1115 Description: implement async create (obd_async_create) method for osc, to avoid
1116 too long waiting new ost objects with holding ldlm lock.
1120 Frequency : occasionally during network problems
1121 Description: client not allowed to reconnect to OST because of active request
1122 Details : abort bulk requests received by the OST once the client has timed
1123 out since the client will resend the request anyway. The client
1124 also now retries to reconnect to the same server if a connect
1125 request failed with EBUSY or -EAGAIN.
1128 Frequency : rare, if used wide striped file and one ost in down.
1130 Descriptoin: don't return error if we created a subset of objects for file.
1131 Details : lov_update_create_set() uses set->set_success as index for created
1132 objects, so if some requests failed, they will have hole at end of
1133 array and we can use qos_shrink_lsm for allocate correct lsm.
1137 Description: Slow stale export processing during normal start up
1138 Details : The global mgc lock prevents OST setup to be run in parallel.
1139 Replace the global lock with a per-config_llog_data semaphore.
1143 Description: Out or order replies might be lost on replay
1144 Details : In ptlrpc_retain_replayable_request if we cannot find retained
1145 request with tid smaller then one currently being added, add it
1146 to the start, not end of the list.
1150 Description: BUG: soft lockup - CPU#1 stuck for 10s! [ll_mdt_07:4523]
1151 Details : add cond_resched() calls to avoid hogging the cpu for too long
1152 in the hash code. Make also lustre_hash_for_each_empty() more
1155 Severity : enhancement
1157 Description: Performance improvements for debug messages with D_RPCTRACE,
1158 D_LDLM, D_QUOTA options.
1161 Frequency : only with NFS export
1163 Description: (lov_merge.c:74:lov_merge_lvb())
1164 ASSERTION(spin_is_locked(&lsm->lsm_lock)) failed (SR 71691004)
1165 Details : Fix a race in the nfs export code by populating inode
1166 info while the new inode is still locked
1168 Severity : enhancement
1170 Description: add a new file in procfs called force_lbug. Writting to this
1171 file triggers a LBUG. Only for test purpose.
1175 Description: OOM killer causes node hang
1176 Details : really interrupt the sleep in osc_enter_cache on signals
1180 Description: LustreError: 9153:0:(quota_context.c:622:dqacq_completion()) LBUG
1181 Details : fix race during quota release on the slave.
1183 Severity : enhancement
1185 Description: smaller hash bucket sizes, cleanups
1186 Details : increase hash table sizes and enabled rehashing for pools, quota,
1187 uuid, nid & per-nid stats.
1189 Severity : enhancement
1191 Description: Add ldiskfs maxdirsize mount option
1192 Details : add max_dir size mount option
1196 Description: panic in ll_statahead_thread
1197 Details : prevent parent thread to be killed before its child
1200 Frequency : only with 16TB device
1202 Description: unable to perform "mount -t lustre" of 16TB OST device
1203 Details : Mounting 16TB LUNs failed due to three bugs in mkfs.lustre.
1207 Description: ASSERTION(atomic_read(&imp->imp_inflight) == 0) failed
1208 Details : unregistering should be zero if no RPC inflight.
1212 Description: hyperion: Oops during metabench
1213 Details : Correct the refcount of lov_request_set
1215 Severity : enhancement
1217 Description: Add mptlinux and nxge drivers to Lustre builds
1219 Severity : enhancement
1221 Description: Fix watchdog timer message to be more clear
1222 Details : Make watchdog timer messages more clear and descriptive.
1226 Description: LNET soft lockups in socknal_cd thread
1227 Details : don't hog CPU for active-connecting if another connd is
1228 accepting connecting-requst from the same peer
1232 Description: recovery-small test_17 hang
1233 Details : Land several AT improvements & fixes.
1237 Description: MDS panic and hanging client processes
1238 Details : Replace exp_ops_stats with exp_nid_stats->nid_stats
1242 Description: OSS stuck in recovery.
1243 Details : fix race during recovery. class_unlink_export,
1244 class_set_export_delayed and target_queue_last_replay_reply
1245 may race while increasing/decreasing obd_recoverable_clients
1246 and obd_delayed_clients, causing recovery to wait forever.
1248 Severity : enhancement
1250 Description: add cascading_rw.c to lustre/tests
1254 Description: filter_last_id() NULL deref
1255 Details : lprocfs_filter_rd_last_id() should check for the fully
1256 setup obd device, before proceeding further.
1258 Severity : enhancement
1260 Description: Loadgen improvements
1261 Details : stacksize and locking fixes for loadgen
1265 Description: Quiet CERROR("dirty %d > system dirty_max %d\n"
1266 Details : The atomic_read() allowing the atomic_inc() are not covered
1267 by a lock. Thus they may safely race and trip this CERROR()
1268 unless we add in a small fudge factor (+1).
1270 Severity : enhancement
1272 Description: shrink_slab: nr=-9223362083340912175
1273 Details : fix spurious message from shrink_slab reporing negative nr
1277 Description: Quiet bogus previously committed transno error
1278 Details : suppress the "server went back in time" error message which
1279 is always printed even in the common case after a client eviction
1281 Severity : enhancement
1283 Description: Parallel statfs() calls result in client eviction
1284 Details : cache statfs data for 1s.
1288 Description: parallel-scale test_compilebench: @@@@@@ FAIL: compilebench
1290 Details : fix serveral issues in pinger code causing clients not to ping
1291 servers for too long, resulting in evictions.
1295 Description: e2fsck should warn when MMP update interval is extended
1296 Details : print mmp_check_interval and make it possible to abort mount
1297 operation in case it takes too long.
1301 Description: mdsrate-create-large.sh, BUG: soft lockup - CPU#0 stuck for 10s!
1302 Details : fix bug in the RHEL5's jbd2 callback patch.
1306 Description: drop number of active requests when queued for recovery
1307 Details : Now that we take a reference on the original request instead of
1308 making a copy of it for recovery. We need to drop the number of
1309 active requests or the queued requests will prevent all request
1310 processing when they exceed (srv->srv_threads_running - 1).
1312 Severity : enhancement
1314 Description: refuse to invalidate operational quota files when they are in use
1315 Details : an attempt to invalidate operational quota files on the quota
1316 master is not actually permitted by VFS (returning -EPERM), but we
1317 should not depend on that and should return the error earlier.
1321 Description: Applications stuck in jbd2_log_wait_commit during exit
1322 Details : fix deadlock between kjournald2 trying to acquire the page lock
1323 owned by an ost_io thread waiting for journal commit.
1325 -------------------------------------------------------------------------------
1327 2009-10-16 Sun Microsystems, Inc.
1329 * Support for kernels:
1330 2.6.16.60-0.42.4 (SLES 10),
1331 2.6.27.29-0.1 (SLES11, i686 & x84_64 only),
1332 2.6.18-128.7.1.el5 (RHEL 5),
1333 2.6.18-128.7.1.el5 (OEL 5),
1334 * Client support for unpatched kernels:
1335 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1336 2.6.16 - 2.6.27 vanilla (kernel.org)
1337 * Recommended e2fsprogs version: 1.41.6.sun1
1338 * ext4 support for RHEL5 is experimental and thus should not be
1341 Severity : enhancement
1343 Description: Add OEL5 support.
1345 Severity : enhancement
1347 Description: Update kernel to SLES11 2.6.27.29-0.1.
1351 Description: File checksum failures with OST read cache on
1352 Details : Disable page poisoning when the bulk transfer has to be aborted
1353 because the client got evicted.
1357 Description: Don't allow make backward step on assiging osc next id.
1358 Details : race between allocation next id and ll_sync thread can be cause
1359 of set wrong osc next id and can be kill valid ost objects.
1361 Severity : enhancement
1363 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
1365 Severity : enhancement
1367 Description: Update kernel to SLES10 SP2 2.6.16.60-0.42.4.
1371 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
1375 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
1379 Description: Error handling in osc_statfs_interpret() has been improved.
1380 Details : Check in osc_statfs_interpret() for EBADR.
1384 Description: Do not send statfs() requests to OSTs disabled by administrator.
1385 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
1389 Description: Do not update ctime for the deleted inode.
1390 Details : Check in mds_reint_unlink() before calling fsfilt_setattr().
1394 Description: Increase of the size of the LDLM resource hash.
1395 Details : Bump up RES_HASH_BITS=12.
1399 Description: correctly send lsm on open replay
1400 Details : MDS is trust to LSM size on replay open, but client can set wrong size
1405 Description: Deadlock between filter_destroy() and filter_commitrw_write().
1406 Details : filter_destroy() does not hold the DLM lock over the whole
1407 operation. If the DLM lock is dropped, filter_commitrw() can go
1408 through, causing the deadlock between page lock and i_mutex.
1409 The i_alloc_sem should also be hold in filter_destroy() while
1410 truncating the file.
1414 Description: truncate starts GFP_FS allocation under transaction causing deadlock
1415 Details : ldiskfs_truncate calls grab_cache_page which may start page
1416 allocation under an open transaction. This may lead to
1417 calling prune_icache with consequent lustre reentrance.
1420 Frequency : only when down/upgrading the MDS to 1.6/1.8 while 1.8 clients are
1421 still up and when the OST pool feature is used
1423 Description: interop testing got LBUG when run dd with OST pool
1424 :LustreError: 30032:0:(llite_lib.c:1913:ll_replace_lsm()) LBUG
1425 Details : down/upgrading the MDS to a version that doesn't/does support OST
1426 pool can cause clients to crash because the lsm has changed
1431 Description: missing tree_status on 1.8.1 RPM build
1432 Details : make rpms failed due because the tree_status file is missing.
1436 Description: continuing LustreError "mds adjust qunit failed!"
1437 Details : don't print message on the console when ->adjust_qunit fails.
1441 Description: don't increase ldlm timeout if previous client was evicted
1442 Details : if a client doesn't respond to a blocking callback within the
1443 adaptive ldlm enqueue timeout, don't adjust the adaptive estimate
1444 when the lock is next granted.
1448 Description: ost is being unmounted w/o all writes to last_rcvd landing on disk.
1449 affects recovery negatively.
1450 Details : make sure all exports have been properly destroyed by the zombie
1451 thread processed before stopping the target.
1455 Description: Performance degradation with O_DIRECT between 1.6 & 1.8.1 b190
1456 Details : disable write barrier for ext4/SLES11.
1460 Description: Kernel panic - not syncing: Out of memory and no killable
1461 processes... on OSS when iozone
1462 Details : fix memory leak in the journal checksum patch.
1466 Description: group quota "too many blocks" OSS crashes
1467 Details : we should keep the same uid/gid for lquota_chkquota() and
1468 lquota_pending_commit()
1472 Description: LustreError: 9153:0:(quota_context.c:622:dqacq_completion()) LBUG
1473 Details : don't LBUG on release quota error. Just a workaround until the
1474 problem is understood.
1476 ------------------------------------------------------------------------------
1477 2009-07-31 Sun Microsystems, Inc.
1479 * Support for kernels:
1480 2.6.16.60-0.39.3 (SLES 10),
1481 2.6.27.23-0.1 (SLES11, i686 & x84_64 only),
1482 2.6.18-128.1.14.el5 (RHEL 5),
1483 * Client support for unpatched kernels:
1484 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1485 2.6.16 - 2.6.27 vanilla (kernel.org)
1486 * Recommended e2fsprogs version: 1.41.6.sun1
1487 * File join has been disabled in this release, refer to Bugzilla 16929.
1488 * NFS export disabled when stack size < 8192. Since the NFSv4 export
1489 of Lustre filesystem with 4K stack may cause a stack overflow. For
1490 more information, please refer to bugzilla 17630.
1491 * ext4 support for RHEL5 is experimental and thus should not be
1494 Severity : enhancement
1496 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
1499 Frequency : with 1.8 server and 1.6 clients
1501 Description: correctly shrink reply for avoid send too big message to client.
1502 Details : 1.8 mds is allocate to big buffer to LOV EA data and this produce
1503 some problems with sending this reply to 1.6 client.
1507 Description: Repeated atomic allocation failures.
1508 Details : Use GFP_HIGHUSER | __GFP_NOMEMALLOC flags for memory allocations
1509 to generate memory pressure and allow reclaiming of inactive pages.
1510 At the same time, do not allow to exhaust emergency pools.
1511 For local clients the use of GFP_NOFS will be introduced in 1.8.2
1513 Severity : enhancement
1514 Bugzilla : 19846, 18289
1515 Description: Update kernel to RHEL5 2.6.18-128.1.14.el5.
1517 Severity : enhancement
1518 Bugzilla : 19625, 16893, 18668, 19848
1519 Description: Add support for SLES11 2.6.27.23-0.1.
1521 Severity : enhancement
1523 Description: Update client support to vanila kernels up to 2.6.27.
1525 Severity : enhancement
1527 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
1529 Severity : enhancement
1531 Description: Compile with -Werror by default for i686 and x86_64.
1535 Description: resolve race between obd_disconnect and class_disconnect_exports
1536 Details : if obd_disconnect will be called to already disconnected export he
1537 forget release one reference and osc module can't unloaded.
1539 Severity : enhancement
1541 Description: move AT tunable parameters for more consistent usage
1542 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
1546 Descriptoin: correctly skip time estimate if in recovery
1547 Details : rq_send_state insn't bitmask so using bitwise ops is forbid.
1551 Descriptoin: OSS DeadLock
1552 Details : Use trylock to prevent deadlock when shrink icache.
1554 Severity : enhancement
1556 Description: Allow tuning service thread via /proc
1557 Details : For each service a new
1558 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
1559 created that can be used to set min/max thread counts, and get the
1560 current number of running threads.
1562 Severity : enhancement
1564 Description: Add state history info file, enhance import info file
1565 Details : Track import connection state changes in a new osc/mdc proc file;
1566 add overview-type data to the osc/mdc import proc file.
1570 Description: Reduce small size read RPC
1571 Details : Set read-ahead limite for every file and only do read-ahead when
1572 available read-ahead pages are bigger than 1M to avoid small size
1577 Description: free_entry erroneously used groups_free instead of put_group_info
1579 Severity : enhancement
1581 Description: Make read-ahead stripe size aligned.
1583 Severity : enhancement
1585 Description: MDS create should not wait for statfs RPC while holding DLM lock.
1588 Frequency : rare, connect and disconnect target at same time
1590 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
1591 Details : don't call obd_disconnect under lov_lock. this long time
1592 operation and can block ptlrpcd which answer to connect request.
1595 Frequency : start MDS on uncleanly shutdowned MDS device
1597 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
1598 Details : stay in waiting mds<>ost recovery finished produce random bugs
1599 due race between two ll_sync thread for one lov target. send
1600 ACTIVATE event only if connect realy finished and import have
1604 Frequency : start MDS on uncleanly shutdowned MDS device
1606 Descriptoin: aborting recovery hang on MDS
1607 Details : don't throttle destroy RPCs for the MDT.
1611 Description: Slow reads beyond 8Tb offsets.
1612 Details : Page index integer overflow in ll_read_ahead_page
1616 Description: Soft lockup on OSS after MDS failover
1617 Details : MSG_CONNECT_INITIAL is not set on the initial MDS->OST connect.
1618 As a conseqence, the patch from bug 18224 is not operational
1619 and the MDS export cannot be reused on the OSTs until it gets
1623 Frequency : rare, only if using MMP with Linux RAID
1625 Description: MMP doesn't work with Linux RAID
1626 Details : While using HA for Lustre servers with Linux RAID, it is possible
1627 that MMP will not detect multiple mounts. To make this work we
1628 need to unplug the device queue in RAID when the MMP block is being
1629 written. Also while reading the MMP block, we should read it from
1630 disk and not the cached one.
1633 Frequency : rare, during recovery
1635 Description: Assertion failure in ldlm_lock_put
1636 Details : Do not put cancelled locks into replay list, hold references on
1637 locks in replay list
1641 Description: 1.6.5 mdsrate performance is slower than 1.4.11/12 (MDS is not cpu bound!)
1642 Details : create_count always drops to the min value (=32) because grow_count
1643 is being changed before the precreate RPC completes.
1646 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
1649 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
1650 Details : add proper locking for creation of jbd_4k slab cache
1654 Description: MMP check in ext3_remount() fails without displaying any error
1655 Details : When multiple mount protection fails during remount, proper error
1660 Description: Rare Client crash on resend if the file was deleted.
1661 Details : When file is opened, but open reply is lost and file is
1662 subsequently deleted before resend, resend processing logic
1663 breaks trying to open the file again, should not try to open.
1667 Description: add check for >8TB ldiskfs filesystems
1668 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
1669 Don't allow >8TB ldiskfs filesystems to be mounted without
1670 force_over_8tb mount option
1674 Description: Client locked up when running multiple instances of an app. on
1675 multiple mount points
1676 Details : ll_shrink_cache() can sleep while holding the ll_sb_lock.
1677 Convert ll_sb_lock to a read/write semaphore to fix the problem.
1681 Description: Cannot acces an NFS-mounted Lustre filesystem
1682 Details : An NFS client cannot access the Lustre filesystem NFS-mounted
1683 from a Lustre-client exporting the Lustre filesystem via NFS.
1687 Description: panic in ll_statahead_thread
1688 Details : grab dentry reference in parent process.
1691 -------------------------------------------------------------------------------
1693 tbd Sun Microsystems, Inc.
1695 * Support for kernels:
1696 2.6.16.60-0.31 (SLES 10),
1697 2.6.18-128.1.6.el5 (RHEL 5),
1698 2.6.22.14 vanilla (kernel.org)
1699 * Client support for unpatched kernels:
1700 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1701 2.6.16 - 2.6.22 vanilla (kernel.org)
1702 * Recommended e2fsprogs version: 1.40.11-sun1
1703 * File join has been disabled, refer to Bugzilla 16929.
1704 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
1705 * NFS export disabled when stack size < 8192. Since the NFSv4 export of
1706 Lustre filesystem with 4K stack may cause a stack overflow. For more
1707 information, please refer to bugzilla 17630.
1709 Severity : enhancement
1711 Description: Update to RHEL5.3 kernel-2.6.18-128.1.6.el5.
1713 Severity : enhancement
1715 Description: Update OFED release to 1.4.1 RC4
1717 Severity : major, only with big OST
1719 Description: Very poor metadata performance on Infiniband lustre configuration
1720 Details : OST object precreation becomes very slow on big OSTs. This is due
1721 to the ialloc patch spending too much time scanning groups.
1724 Frequency : during recovery
1726 Description: don't mix llog inodes with normal.
1727 Details : allocate inodes for log in last inode group
1732 Description: fix lqs' reference which won't be put in some situations
1733 Details : This patch fixes:
1734 1. In quota_check_common(), this function will check quota
1735 for user and group, but only send one return via "pending".
1736 In most cases, the pendings should be same. But that is not
1738 2. If quotaoff runs between lquota_chkquota() and
1739 lquota_pending_commit(), the same thing will happen too.
1740 That is why it comes:
1741 - if (!ll_sb_any_quota_active(qctxt->lqc_sb))
1744 -------------------------------------------------------------------------------
1746 2008-12-31 Sun Microsystems, Inc.
1748 * Support for kernels:
1749 2.6.5-7.314 (SLES 9),
1750 2.6.9-67.0.22.EL (RHEL 4),
1751 2.6.16.60-0.31 (SLES 10),
1752 2.6.18-92.1.17.el5 (RHEL 5),
1753 2.6.22.14 vanilla (kernel.org)
1754 * Client support for unpatched kernels:
1755 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1756 2.6.16 - 2.6.22 vanilla (kernel.org)
1757 * Client support for unpatched kernels:
1758 we do not recommend using patchless RHEL4 clients with kernels
1759 prior to 2.6.9-55EL (RHEL4U5).
1760 * Recommended e2fsprogs version: 1.40.11-sun1
1761 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1762 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
1763 removed cwd "./" (refer to Bugzilla 14399).
1764 * A new quota file format has been introduced in 1.6.5.
1765 The format conversion from prior releases is handled transparently,
1766 but releases older than 1.4.12/1.6.5 don't understand this new
1767 format. The automatic format conversion can be avoided by running
1768 the following command on the MDS:
1769 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
1770 For more information, please refer to bugzilla 13904.
1771 * A new quota file format was introduced in 1.6.6/1.8.0 (kernels 2.6.16+).
1772 The format conversion from prior releases is handled transparently,
1773 but releases older than 1.6.6/1.8.0 don't understand this new
1774 format. The automatic format conversion can be avoided by running
1775 the following commands on the MDS and OSS servers (for
1776 pre 1.4.12-1.6.5 quota files):
1777 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV',
1778 'tunefs.lustre --param="ost.quota_type=ug1" $MDTDEV'
1779 or (for 1.4.12/1.6.5 quota files)
1780 'tunefs.lustre --param="mdt.quota_type=ug2" $MDTDEV',
1781 'tunefs.lustre --param="ost.quota_type=ug2" $MDTDEV'
1782 For more information, please refer to bugzilla 13904.
1783 * Output of lfs quota has been made less detailed by default,
1784 old (verbose) output can be obtained by using -v option.
1785 * File join has been disabled in this release, refer to Bugzilla 16929.
1786 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
1787 * NFS export disabled when stack size < 8192. Since the NFSv4 export of
1788 Lustre filesystem with 4K stack may cause a stack overflow. For more
1789 information, please refer to bugzilla 17630.
1791 Severity : enhancement
1793 Description: Caching OSS
1794 Details : introduce data caching on the OSS. The OSS now relies on the linux
1795 kernel page cache to keep recently accessed data in memory.
1796 It is worth noting that all write requests are still flushed
1797 synchronously as in lustre 1.6.
1799 Severity : enhancement
1801 Description: Version based recovery
1802 Details : introduce finer grained recovery able to detect transaction
1803 dependencies and can deal with transaction gaps caused by clients
1804 failing at the same time as the server.
1806 Severity : enhancement
1808 Description: Enable adaptive timeouts by default
1809 Details : The Lustre timeout value in /proc/sys/lustre/timeout is now
1810 managed dynamically based on server load and should not need
1811 to be tuned manually based on cluster size. This allows Lustre
1812 to work under a wider variety of system sizes and loads, without
1813 unnecessarily causing lengthy recovery times.
1815 Severity : enhancement
1817 Description: Add OST Pools support
1818 Details : File striping can now be set to use an arbitrary pool of OSTs
1820 Severity : enhancement
1822 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
1823 Details : allow skip disconnected ost for send statfs request and hide error
1827 Frequency : rare, on llog test 6
1829 Descriptoin: don't allow connect to already connected import
1830 Details : allowing connect to already connected import is hide connecting problem.
1833 Frequency : rare, on failed llog setup
1835 Descriptoin: don't leak obd reference on failed llog setup
1836 Details : for failed llog setup - mgc forget call class_destroy_import for
1837 client import, move destroy import to more generic place.
1842 Descriptoin: allow kill process which wait statahead result
1843 Details : for some reasons 'ls' can stick in waiting result from statahead,
1844 in this case need way for kill this process.
1847 Frequency : rare, at shutdown
1849 Descriptoin: panic at umount
1850 Details : llap_shrinker can be raced with killing super block from list and
1851 this produce panic with access to already freeded pointer
1856 Descriptoin: don't lose wakeup for imp_recovery_waitq
1857 Details : recover_import_no_retry or invalidate_import and import_close can
1858 both sleep on imp_recovery_waitq, but we was send only one wakeup
1864 Descriptoin: panic in mds_open
1865 Details : don't confuse mds_finish_transno() with PTR_ERR(-ENOENT)
1870 Descriptoin: stuck in cache_remove_extent() or panic with accessing to already
1872 Details : release lock refernce only after add page to pages list.
1875 Frequency : always with long access acl
1877 Descriptoin: mds can't pack reply with long acl.
1878 Details : mds don't control size of acl but they limited by reint/getattr
1883 Frequency : on remount
1885 Description: external journal device not working after the remount
1886 Details : clear dev_rdonly flag for external journal devices in
1892 Description: shutdown vs evict race
1893 Details : client_disconnect_export vs connect request race.
1894 if client will evicted at this time - we start invalidate
1895 thread without referece to import and import can be freed
1901 Description: shrink LOV EAs before replying
1902 Details : correctly adjust LOV EA buffer for reply.
1907 Description: don't skip ost target if they assigned to file
1908 Details : Drop slow OSCs if we can, but not for requested start idx.
1909 This means "if OSC is slow and it is not the requested
1910 start OST, then it can be skipped, otherwise skip it only
1911 if it is inactive/recovering/out-of-space.
1913 Severity : enhancement
1915 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
1918 Frequency : rare, need acl's on inode.
1920 Description: client can't handle ost additional correctly
1921 Details : if ost was added after client connected to mds client can have
1922 hit lnet_try_match_md ... to big messages to wide striped files.
1923 in this case need teach client to handle config events about add
1924 lov target and update client max ea size at that event.
1926 Severity : enhancement
1928 Description: Update to sles9 kernel-2.6.5-7.314.
1930 Severity : enhancement
1932 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
1935 Frequency : Create a symlink file with a very long name
1937 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
1938 Details : If there is no extra space in the request for early cancels,
1939 ldlm_req_handles_avail() returns 0 instead of a negative value.
1944 Description: mds is deadlocked
1945 Details : in rare cases, inode in catalog can have i_no less than have parent
1946 i_no, this produce wrong order for locking during open, and
1947 parallel unlink can be lock open. this need teach mds_open to grab
1948 locks in resource id order, not at parent -> child order.
1950 Severity : enhancement
1952 Description: Add /proc entry for import status
1953 Details : The mdc, osc, and mgc import directories now have
1954 an import directory that contains useful import data for debugging
1955 connection problems.
1957 Severity : enhancement
1959 Description: Re-disable certain /proc logging
1960 Details : Enable and disable client's offset_stats, extents_stats and
1961 extents_stats_per_process stats logging on the fly.
1964 Frequency : Only on FC kernels 2.6.22+
1966 Description: oops in statahead
1967 Details : Do not drop reference count for the dentry from VFS when lookup,
1968 VFS will do that by itself.
1970 Severity : enhancement
1972 Description: Generic /proc file permissions
1973 Details : Set /Proc file permissions in a more generic way to enable non-
1974 root users operate on some /proc files.
1978 Description: Hitting mdc_commit_close() ASSERTION
1979 Details : Properly handle request reference release in
1980 ll_release_openhandle().
1984 Frequency : only patchless client
1985 Description: add workaround for race between add/remove dentry from hash
1987 Severity : enhancement
1989 Description: Allow OST glimpses to return PW locks
1993 Description: LBUG when llog conf file is full
1994 Details : When llog bitmap is full, ENOSPC should be returned for plain
1999 Description: Prevent import from entering FULL state when server in recovery
2003 Description: service mount cannot take device name with ":"
2004 Details : Only when device name contains ":/" will mount treat it as
2010 Description: replace ptlrpcd with the statahead thread to interpret the async
2011 statahead RPC callback
2015 Frequency : on recovery
2016 Description: I/O failures after umount during fail back
2017 Details : if client reconnected to restarted server we need join to recovery
2018 instead of find server handler is changed and process self eviction
2019 with cancel all locks.
2021 Severity : enhancement
2023 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
2027 Description: Kernel BUG tries to release flock
2028 Details : Lustre does not destroy flock lock before last reference goes
2029 away. So always drop flock locks when client is evicted and
2030 perform unlock regardless of successfulness of speaking to MDS.
2032 Severity : enhancement
2034 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
2036 Severity : enhancement
2038 Description: Upcall on Lustre log has been dumped
2039 Details : Allow for a user mode script to be called once a Lustre log has
2040 been dumped. It passes the filename of the dumped log to the
2041 script, the location of the script can be specified via
2042 /proc/sys/lnet/debug_log_upcall.
2047 Description: avoid messages about idr_remove called for id that is not allocated
2048 Details : Move assigment s_dev for clustered nfs to end of initialization,
2049 for avoid problem with error handling.
2054 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
2055 Details : When connection is reused this not moved from CONN_UNUSED_HASH into
2056 CONN_USED_HASH and this prodice warning when put connection again
2062 Description: avoid ASSERTION(client_stat->nid_exp_ref_count == 0) failed
2063 Details : release reference to stats when client disconnected, not
2064 when export destroyed for avoid races when client destroyed
2065 after main ost export.
2069 Description: more cleanup in mds_lov
2070 Details : add workaround for get valid ost count for avoid warnings about
2071 drop too big messages, not init llog cat under semphore which
2072 can be blocked on reconnect and break normal replay, fix access
2075 Severity : enhancement
2077 Description: Export bytes_read/bytes_write count on OSC/OST.
2081 Description: Early reply size mismatch, MGC loses connection
2082 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
2083 the connect flags are properly negotiated.
2087 Description: Properly propagate oinfo flags from lov to osc for statfs
2088 Details : restore missing copy oi_flags to lov requests.
2092 Description: exports in /proc are broken
2093 Details : recreate /proc entries for clients when they reconnect.
2095 Severity : enhancement
2097 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
2098 : l_getgroups(8), lst(8), routerstat(8)
2099 Details : included man pages for llobdstat(8), llstat(8),
2100 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
2102 Severity : enhancement
2104 Description: Implement lustre ll_show_options method.
2106 Severity : enhancement
2108 Description: Update to SLES9 kernel-2.6.5-7.312.
2110 Severity : enhancement
2112 Description: Update to RHEL4 kernel-2.6.9-67.0.22.EL.
2116 Description: exports in /proc are broken
2117 Details : recreate /proc entries for clients when they reconnect.
2121 Description: don't fail open with -ERANGE
2122 Details : if client connected until mds will be know about real ost count
2123 get LOV EA can be fail because mds not allocate enougth buffer
2128 Description: Resolve device initialization race
2129 Details : Prevent proc handler from accessing devices added to the
2130 obd_devs array but yet be intialized.
2132 Severity : enhancement
2134 Description: configure's --enable-quota should check the
2135 : kernel .config for CONFIG_QUOTA
2136 Details : configure is terminated if --enable-quota is passed but
2137 : no quota support is in kernel
2139 Severity : enhancement
2141 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
2143 Severity : enhancement
2145 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
2149 Frequency : rare, on PPC clients
2150 Description: don't swab ost objects in response about directory, because
2152 Details : bug similar bug 14856, but in different function.
2154 Severity : enhancement
2156 Description: lfs quota tool enhancement
2157 Details : added units specifiers support for setquota, default to
2158 current uid/gid for quota report, short quota stats by
2159 default, nonpositional parameters for setquota, added
2160 llapi_quotactl manual page.
2162 Severity : enhancement
2164 Description: *optional* service tags registration
2165 Details : if the "service tags" package is installed on a Lustre node
2166 When the filesystem is mounted, a local-node service tag will
2167 be created. See http://inventory.sun.com/ for more information
2168 about the Service Tags asset management system.
2172 Description: Client runs out of low memory
2173 Details : Consider only lowmem when counting initial number of llap pages
2176 Frequency : occasional
2178 Description: add refcount for osc callbacks, so avoid panic on shutdown
2180 Severity : enhancement
2182 Description: Update to RHEL4 kernel-2.6.9-67.0.20.
2185 Frequency : testing only
2187 Description: sanity test 65a fails if stripecount of -1 is set
2188 Details : handle -1 striping on filesystem in ll_dirstripe_verify
2191 Frequency : only in unusual configurations
2193 Description: Kernel panic with find ost index.
2194 Details : lov_obd have panic if some OST's have sparse indexes.
2196 Severity : enhancement
2198 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
2201 Frequency : rarely, if filesystem is mounted with -o flock
2203 Description: do not process already freed flock
2204 Details : flock can possibly be freed by another thread before it reaches
2205 to ldlm_flock_completion_ast.
2208 Frequency : rarely, if filesystem is mounted with -o flock
2210 Description: LBUG during stress test
2211 Details : Need properly lock accesses the flock deadlock detection list.
2214 Frequency : rarely, if binaries are being run from Lustre
2216 Description: oops in page fault handler
2217 Details : kernel page fault handler can return two special 'pages' in
2218 error case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
2221 Frequency : rarely, during shutdown
2223 Description: timeout with invalidate import.
2224 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
2225 should be handled by ptlrpcd. This produce long age waiting and
2226 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
2232 Description: ASSERTION(CheckWriteback(page,cmd)) failed
2233 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
2237 Frequency : only with broken builds/installations
2239 Description: no LBUG if lquota.ko and fsfilt_ldiskfs.ko are different versions
2240 Details : just return an error to a user, put a console error message
2242 Severity : enhancement
2244 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
2246 Severity : enhancement
2248 Description: Update to RHEL4 kernel-2.6.9-67.0.15.
2250 Severity : enhancement
2252 Description: enable MGS and MDT services start separately
2253 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
2254 a co-located MGS without starting the MGS, which is a complement
2255 to 'nosvc' mount option.
2258 Frequency : always, on big-endian systems
2260 Description: cleanup in ptlrpc code, related to PPC platform
2261 Details : store magic in native order avoid panic's in recovery on PPC
2262 node and forbid from this error in future. Also fix posibily
2263 of twice swab data. Fix get lov striping to userpace.
2266 Frequency : rarely, if replay get lost on server
2268 Description: server incorrectly drop resent replays lead to recovery failure.
2269 Details : do not drop replay according to msg flags, instead we check the
2270 per-export recovery request queue for duplication of transno.
2273 Frequency : after recovery
2275 Description: precreate to many object's after del orphan.
2276 Details : del orphan st in oscc last_id == next_id and this triger growing
2277 count of precreated objects. Set flag LOW to skip increase count
2278 of precreated objects.
2281 Frequency : rare, on clear nid stats
2283 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
2284 Details : when clean nid stats sometimes try destroy live entry,
2285 and this produce panic in free.
2288 Frequency : occasionally since 1.6.4
2290 Description: Stack overflow during MDS log replay
2291 Details : ease stack pressure by using a thread dealing llog_process.
2294 Frequency : very rare
2296 Description: MDT cannot be unmounted, reporting "Mount still busy"
2297 Details : Mountpoint references were being leaked during open reply
2298 reconstruction after an MDS restart. Drop mountpoint reference
2299 in reconstruct_open() and free dentry reference also.
2304 Description: wait until IO finished before start new when do lock cancel.
2305 Details : VM protocol want old IO finished before start new, in this case
2306 need wait until PG_writeback is cleared until check dirty flag
2307 and call writepages in lock cancel callback.
2312 Description: mds_mfd_close() ASSERTION(rc == 0)
2313 Details : In mds_mfd_close(), we need protect inode's writecount change
2314 within its orphan write semaphore to prevent possible races.
2317 Frequency : rare, on shutdown ost
2319 Description: don't hit live lock with umount ost.
2320 Details : shrink_dcache_parent can be in long loop with destroy dentries,
2321 use shrink_dcache_sb instead.
2324 Frequency : only when echo_client is used
2326 Description: don't panic with use echo_client
2327 Details : echo client pass NULL as client nid pointer and this produce NULL
2328 pointer dereference.
2331 Frequency : Always on 32-bit PowerPC systems
2333 Description: fix build on PPC32
2334 Details : compile code with -m64 flag produce wrong object file for PPC32.
2339 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
2340 Details : In reconstruct_* functions, LASSERTs on both the data supplied
2341 by a client, and the data on disk are dangerous and incorrect.
2342 Change them with client eviction.
2344 Severity : enhancement
2346 Description: skiplist implementation simplification
2347 Details : skiplists are used to group compatible locks on granted list
2348 that was implemented as tracking first and last lock of each lock
2349 group the patch changes that to using doubly linked lists
2353 Description: delete compatibility for 32bit qdata
2354 Details : as planned, when lustre is beyond b1_8, lquota won't support 32bit
2355 qunit. That means servers of b1_4 and servers of b1_8 can't be
2356 used together if users want to use quota.
2359 Frequency : only with administrator action
2361 Description: mount failure if config log has invalid conf_param setting
2362 Details : If administrator specified an incorrect configuration parameter
2363 with "lctl conf_param" this would cause an error during future
2364 client mounts. Instead, ignore the bad configuration parameter.
2367 Frequency : blocks per group < blocksize*8 and uninit_groups is enabled
2369 Description: ldiskfs error: XXX blocks in bitmap, YYY in gd
2370 Details : If blocks per group is less than blocksize*8, set rest of the
2374 Frequency : Application do stride read on lustre
2376 Description: The read performance will drop a lot if the application does
2378 Details : Because the stride_start_offset are missing in stride read-ahead,
2379 it will cause clients read a lot of unused pages in read-ahead,
2380 then the read-performance drops.
2384 Description: more ldlm soft lockups
2385 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
2386 starve other threads from the resource lock for a long time in
2387 case of long waiting queue, so change the debug level from
2388 D_OTHER to the less frequently used D_INFO.
2390 Severity : enhancement
2392 Description: add -gid, -group, -uid, -user options to lfs find
2394 Severity : enhancement
2396 Description: ll_recover_lost_found_objs - recover objects in lost+found
2397 Details : OST corruption and subsequent e2fsck can leave objects in the
2398 lost+found directory. Using the "ll_recover_lost_found_objs"
2399 tool, these objects can be retrieved and data can be salvaged
2400 by using the object ID saved in the fid EA on each object.
2405 Description: this bug _only_ happens when inode quota limitation is very low
2406 (less than 12), so that inode quota unit is 1 at initialization.
2407 Details : if remaining quota equates 1, it is a sign to demonstate that quota
2408 is effective now. So least quota qunit should be 2.
2412 Description: Hung threads in invalidate_inode_pages2_range
2413 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
2414 one is completed. As a result, some RPCs are stuck in the queue
2419 Description: Procfs and llog threads access destoryed import sometimes.
2420 Details : Sync the import destoryed process with procfs and llog threads by
2421 the import refcount and semaphore.
2425 Description: mds fails to respond, threads stuck in ldlm_completion_ast
2426 Details : Sort source/child resource pair after updating child resource.
2431 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
2432 Details : If insertion of an extent fails, then discard the inode
2433 preallocation and free data blocks else it can lead to duplicate
2438 Description: don't always update ctime in ext3_xattr_set_handle()
2439 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle()
2440 In some cases the ctime should not be updated, for example for
2441 2.0->1.8 compatibility it is necessary to delete an xattr and it
2442 should not update the ctime.
2446 Description: add quota statistics
2447 Details : 1. sort out quota proc entries and proc code.
2448 2. add quota statistics
2453 Description: quotas are not honored with O_DIRECT
2454 Details : all writes with the flag O_DIRECT will use grants which leads to
2455 this problem. Now using OBD_BRW_SYNC to guard this.
2459 Bugzilla : 15713/16362
2460 Description: Assertion in iopen_connect_dentry in 1.6.3
2461 Details : looking up an inode via iopen with the wrong generation number can
2462 populate the dcache with a disconneced dentry while the inode
2463 number is in the process of being reallocated. This causes an
2464 assertion failure in iopen since the inode's dentry list contains
2465 both a connected and disconnected dentry.
2469 Description: assertion failure in ldlm_handle2lock()
2470 Details : fix a race between class_handle_unhash() and class_handle2object()
2471 introduced in lustre 1.6.5 by bug 13622.
2473 Severity : enhancement
2475 Description: superblock lock contention with many SMP cores on one client
2476 Details : several client filesystem locks were highly contended on SMP
2477 NUMA systems with 8 or more cores. Per-CPU datastructures
2478 and more efficient locking implemented to reduce contention.
2483 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
2484 Details : remove the limit of 256 scsi disks in the sd_iostat patch
2489 Description: oops in sd_iostats_seq_show()
2490 Details : unloading/reloading the scsi low level driver triggers a kernel
2491 bug when trying to access the sd iostat file.
2496 Description: Kernel panics during QLogic driver reload
2497 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
2498 patch, causing memory corruption.
2503 Description: journal_dev option does not work in b1_6
2504 Details : pass mount option during pre-mount.
2506 Severity : enhancement
2508 Description: Add a FIEMAP(FIle Extent MAP) ioctl for ldiskfs
2509 Details : FIEMAP ioctl will allow an application to efficiently fetch the
2510 extent information of a file. It can be used to map logical blocks
2511 in a file to physical blocks in the block device.
2514 Frequency : only with adaptive timeout enabled
2516 Description: DEBUG_REQ() bad paging request
2517 Details : ptlrpc_at_recv_early_reply() should not modify req->rq_repmsg
2518 because it can be accessed by reply_in_callback() without the
2522 Frequency : only on Cray X2
2524 Description: X2 build failures
2525 Details : fix build failures on Cray X2.
2529 Description: xid & resent requests
2530 Details : Initialize RPC XID from clock at startup (randomly if clock is
2535 Description: quota recovery deadlock during mds failover
2536 Details : This patch includes att18982, att18236, att18237 in bz14840.
2538 1. fix osts hang when mds does failover with quotaon
2539 2. prevent watchdog storm when osts threads wait for the
2544 Description: kernel panic on racer
2545 Details : Do not access dchild->d_inode when IS_ERR(dchild) is true.
2547 Severity : enhancement
2549 Description: Add lustre_start utility to start or stop multiple Lustre servers
2554 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
2555 Details : In case of memory pressure, list_del() can be called twice on
2556 req->rq_history_list, causing a kernel oops.
2560 Description: (ptllnd_peer.c:557:kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
2561 Details : fix stack overflow in the distributed lock manager by defering
2562 export eviction after a failed ast to the elt thread instead of
2563 handling it in the dlm interpret routine.
2565 Severity : enhancement
2567 Description: More exported tunables for mballoc
2568 Details : Add support for tunable preallocation window and new tunables for
2569 large/small requests
2573 Description: Detect corruption of block bitmap and checking for preallocations
2574 Details : Checks validity of on-disk block bitmap. Also it does better
2575 checking of number of applied preallocations. When corruption is
2576 found, it turns filesystem readonly to prevent further corruptions.
2580 Frequency : only for big-endian servers
2581 Description: Check for big-endian system while mounting fs with extents feature
2582 Details : Mounting a filesystem with extents feature will fail on big-endian
2583 systems since ext3-based ldiskfs is not supported on big-endian
2584 systems. Can be over-ridden with "bigendian_extents" mount option.
2588 Description: Excessive recovery window
2589 Details : With AT enabled, the recovery window can be excessively long (6000+
2590 seconds). To address this problem, we no longer use
2591 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2592 timeout no longer depends on the service time, it is set to
2593 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2594 time via pb_service_time.
2597 Descriptoin: Don't sync journal after every i/o
2598 Details : Implement write RPC replay to allow server replies for write RPCs
2599 before data is on disk. However, this feature is disabled by
2600 default since some issues leading to data corruptions have been
2601 found during recovery (e.g. bug 19128). This feature can be enabled
2602 by running the following command on the OSSs:
2603 lctl set_param obdfilter.*.sync_journal=0
2607 Description: Watchdog triggered on MDS failover
2608 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2609 the OSTs know that the MDS "UUID" can be reused for the same export
2610 from a different NID, so we do not need to wait for the export to
2615 Descriptoin: Lustre detected file system corruption with inode out of bounds
2616 Details : don't update i_size on MDS_CLOSE for directories. This causes
2617 directory corruptions on the MDT.
2619 -------------------------------------------------------------------------------
2621 2008-05-26 Sun Microsystems, Inc.
2623 * Support for kernels:
2624 2.6.5-7.311 (SLES 9),
2625 2.6.9-67.0.7.EL (RHEL 4),
2626 2.6.16.54-0.2.5 (SLES 10),
2627 2.6.18-53.1.14.el5 (RHEL 5),
2628 2.6.22.14 vanilla (kernel.org)
2629 * Client support for unpatched kernels:
2630 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2631 2.6.16 - 2.6.22 vanilla (kernel.org)
2632 * Due to problems with nested symlinks and FMODE_EXEC (bug 12652),
2633 we do not recommend using patchless RHEL4 clients with kernels
2634 prior to 2.6.9-55EL (RHEL4U5).
2635 * Recommended e2fsprogs version: 1.40.7-sun1
2636 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2637 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
2638 removed cwd "./" (refer to Bugzilla 14399).
2639 * A new quota file format has been introduced in 1.6.5.
2640 The format conversion from prior releases is handled transparently,
2641 but releases older than 1.4.12/1.6.5 will not understand this new
2642 format. The automatic format conversion can be avoided by running
2643 the following command on the MDS before upgrading:
2644 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
2645 For more information, please refer to bugzilla 13904.
2649 Description: quota performance fix
2650 Details : quota data is written in journalled mode instead of ordered to
2651 increase performance
2655 Description: lfs support for human-readable quota grace time strings
2656 Details : lfs setquota -t and lfs quota -t represent quota grace times
2657 in "XXwXXdXXhXXmXXs" format instead of large values in seconds
2660 Frequency : always with o2ib 1.3 and sles10
2662 Description: fix build with SLES10 and o2ib v3.
2663 Details : sles10 uses diffrent name for Module.symver file but configure
2664 assume this file has same name on RHEL/SLES/vanila kernels.
2667 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
2669 Description: MDS may lose file striping (and hence file data) in some cases
2670 Details : If there are additional extended attributes stored on the MDS,
2671 in particular ACLs, SELinux, or user attributes (if user_xattr
2672 is specified for the client mount options) then there is a risk
2673 of attribute loss. Additionally, the Lustre file striping
2674 needs to be larger than default (e.g. striped over all OSTs),
2675 and an additional attribute must be stored initially in the
2676 inode and then increase in size enough to be moved to the
2677 external attribute block (e.g. ACL growing in size) for file
2680 Severity : enhancement
2682 Description: add message levels for liblustreapi
2687 Description: MDT cannot be unmounted, reporting "Mount still busy"
2688 Details : Mountpoint references were being leaked during open reply
2689 reconstruction after an MDS restart. Drop mountpoint reference
2690 in reconstruct_open() and free dentry reference also.
2695 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
2696 Details : Move the 'good_osts' check before the 'total_bavail' check. This
2697 will result in an -EAGAIN and in the exit call path we call
2698 alloc_rr() which will with increasing aggressiveness attempt to
2699 aquire precreated objects on the minimum number of required OSCs.
2703 Description: Use old size assignment to avoid deadlock
2704 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
2705 the scheduling while holding a spinlock. We do not need locking
2706 for size in ll_update_inode() because size is only updated from
2707 the MDS for directories or files without objects, so there is no
2708 other place to do the update, and concurrent access to such inodes
2709 are protected by the inode lock.
2713 Description: Use __u64 instead of int for valid bits
2717 Description: resolve "_IOWR redefined" build error on SLES10
2721 Description: dump the memory debugging after all modules are unloaded to
2722 suppress false negative in conf_sanity test 39
2726 Description: the recovery timer never expires
2727 Details : for new client connect request, the recovery timer should not be
2728 reset, otherwise recovery timer will never expired, if the old
2729 client never come. Only old client connect and first connection
2730 req should trigger recovery timer reset.
2734 Description: the min numbers of lproc stats are wrong
2735 Details : adding a new constant LC_MIN_INIT and use it for initialization
2739 Frequency : always with interactive lfs
2741 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
2744 Frequency : with multiple concurrent readdir processes in same directory
2745 Bugzilla : 15406, 15169, 15175
2746 Description: misc fixes for directory readahead.
2747 Details : prevent previous statahead async RPC callback from processing the
2748 current "statahead_info", race condition between async RPC callback
2749 add dentry into dentry hash table and "ls" thread revalidate such
2750 dentry, statahead his/miss control for hidden items, and so on.
2752 Severity : enhancement
2754 Description: build kernel-ib packages for OFED 1.3 in our release cycle
2758 Description: incore types cleaning in quota code (with respect to 64-bit limits)
2759 Details : several u32 variables declarations are replaced with u64 declarations
2764 Description: fix SLES kernel versioning
2765 Details : the kernel version for our SLES 10 kernel did not include a "-"
2766 before the "smp" at the end. while this was not a problem in
2767 general, it did mean that software trying to use the kernel
2768 version to try to detect a vendor specific kernel would fail.
2769 this was most evident by the OFED build scripts.
2774 Description: Don't update lov_desc members until making sure they are valid
2775 Details : When updating lov_desc members via proc fs, need fix their
2776 validities before doing the real update.
2779 Frequency : very rare
2781 Description: don't put request into delay list while invalidate in flight.
2782 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
2783 import in flight. this produce timeout for invalidate and sometimes
2784 can cause stale data.
2786 Severity : enhancement
2788 Description: Update kernel to SLES9 2.6.5-7.311.
2790 Severity : enhancement
2792 Description: Update kernel to RHEL4 2.6.9-67.0.7.
2797 Frequency : on PPC only
2798 Description: not convert ost objects for directory because it's not exist.
2799 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
2801 Severity : enhancement
2803 Description: Fix warnings with compile liblustre at sles10/rhel5 which have
2804 __u64 as usingied long long type.
2807 Frequency : rare, on shutdown
2809 Description: race process ast vs remove callback
2810 Details : removing callback before disconnect import open race with
2811 processing callback.
2813 Severity : enhancement
2815 Description: Update kernel to SLES9 2.6.5-7.311.
2817 Severity : enhancement
2819 Description: Files open for execute are not marked busy on SLES10
2820 Details : Add FMODE_EXEC to SLES10 SP1 server kernel series.
2822 Severity : enhancement
2824 Description: Add server support for vanilla-2.6.22.14.
2827 Frequency : occasional
2829 Description: Avoid lov_create() getting stuck in obd_statfs_rqset()
2830 Details : If an OST is down the MDS will hang indefinitely in
2831 obd_statfs_rqset() waiting for the statfs data. While for
2832 MDS QOS usage of statfs, it should not stuck in waiting.
2834 Severity : enhancement
2836 Description: Disable adaptive timeouts by default
2839 Frequency : on network error
2841 Description: panic with double free request if network error
2842 Details : mdc_finish_enqueue is finish request if any network error occuring,
2843 but it's true only for synchronus enqueue, for async enqueue
2844 (via ptlrpcd) this incorrect and ptlrpcd want finish request
2848 Frequency : rare, on recovery
2850 Description: read procfs can produce deadlock in some situation
2851 Details : Holding lprocfs lock with send rpc can produce block for destroy
2852 obd objects and this also block reconnect with -EALREADY.
2853 This isn't fix all lprocfs bugs - but make it rare.
2855 Severity : enhancement
2857 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
2860 Frequency : frequent on X2 node
2862 Description: mdc_set_open_replay_data LBUG
2863 Details : Set replay data for requests that are eligible for replay.
2868 Description: lustre_mgs: operation 101 on unconnected MGS
2869 Details : When MGC is disconnected from MGS long enough, MGS will evict the
2870 MGC, and late on MGC cannot successfully connect to MGS and a lot
2871 of the error messages complaining that MGS is not connected.
2874 Frequency : rare, depends on device drivers and load
2876 Description: MDS or OSS nodes crash due to stack overflow
2877 Details : Code changes in 1.6.4 increased the stack usage of some functions.
2878 In some cases, in conjunction with device drivers that use a lot
2879 of stack, the MDS (or possibly OSS) service threads could overflow
2880 the stack. One change which was identified to consume additional
2881 stack has been reworked to avoid the extra stack usage.
2883 Severity : enhancement
2885 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
2887 Severity : enhancement
2889 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
2891 Severity : enhancement
2893 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
2895 Serverity : enhancement
2897 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
2900 Frequency : rare on shutdown OST
2902 Description: Don't allow skipping OSTs if index has been specified.
2903 Details : Don't allow skipping OSTs if index has been specified, make locking
2904 in internal create lots better.
2909 Description: ASSERTION(!PageDirty(page)) failed
2910 Details : Wrong check could lead to an assertion failure under specific
2916 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
2917 Details : access to bitfield in structure is always rounded to long
2918 and this produce problem with not atomic change any bit.
2923 Description: Lustre make rpms failed.
2924 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
2925 builds Lustre rpms from tarball.
2927 Severity : enhancement
2929 Description: Update to SLES9 SP4 kernel-2.6.5-7.308.
2932 Frequency : rare on shutdown OST
2934 Description: If llog cancel was not send before clean_exports phase, this can
2935 produce deadlock in llog code.
2936 Details : If llog thread has last reference to obd and call class_import_put
2937 this produce deadlock because llog_cleanup_commit_master wait when
2938 last llog_commit_thread exited, but this never success because was
2939 called from llog_commit_thread.
2942 Frequency : only if OST index is skipped
2944 Description: NULL lov_tgts causing MDS oops
2945 Details : more safe checks for NULL lov_tgts for avoid oops.
2947 Severity : enhancement
2949 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
2951 Severity : enhancement
2953 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
2958 Description: make mgs_setparam() handle fsname containing dash
2959 Details : fsname containing a dash does not work with lctl conf_param
2961 Severity : enhancement
2963 Description: Update to RHEL4 Update-6 kernel-2.6.9-67.EL.
2966 Frequency : rare, in recovery and (or) destroy lovobjid file.
2968 Description: rewrite lov objid code.
2969 Details : Cleanup for lov objid code, remove scability problems and wrong
2970 locking. Fix sending last_id into ost.
2972 Severity : enhancement
2974 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
2976 Severity : enhancement
2978 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
2979 Details : Use d_move_locked instead of __d_move.
2982 Frequency : rare, at shutdown
2984 Description: access already free / zero obd_namespace.
2985 Details : if client_disconnect_export was called without force flag set,
2986 and exist connect request in flight, this can produce access to
2987 NULL pointer (or already free pointer) when connect_interpret
2988 store ocd flags in obd_namespace.
2991 Frequency : only at startup
2993 Description: not alloc memory with spinlock held.
2994 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
2995 if any spinlock held.
3000 Description: lfs find does not continue on file error
3001 Details : Continue other files processing when a file/dir is absent.
3005 Description: Inconsistent usage of lustre_pack_reply()
3006 Details : Standardize the usage of lustre_pack_reply() such that it
3007 always generate a CERROR on failure.
3010 Frequency : very rare
3012 Description: Fix replay if there is an un-replied request and open
3013 Details : In some cases, older replay request will revert the
3014 mcd->mcd_last_xid on MDS which is used to record the client's
3015 latest sent request.
3017 Severity : enhancement
3019 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
3021 Severity : enhancement
3023 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
3025 Severity : enhancement
3027 Description: Update to SLES9 kernel-2.6.5-7.287.3.
3029 Severity : enhancement
3031 Description: Update to RHEL4 kernel-2.6.9-55.0.12.EL.
3033 Severity : enhancement
3035 Description: Build SLES10 patchless client fails
3036 Details : The configure was broken by run ./configure with
3037 --with-linux-obj=.... argument for patchless client. When the
3038 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
3039 can't search header adequately. Use absolute path such as
3040 -I($LINUX)/include instead.
3042 Severity : enhancement
3044 Description: Lustre Page Accounting
3045 Details : New macros for page alloc and free which enable accounting
3046 of page allocation of Lustre. Use percpu counters to store memory
3047 and page statistics.
3050 Frequency : only if debugging is disabled
3052 Description: LASSERT_{REQ,REP}SWAB macros are buggy
3053 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
3054 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
3055 these macros and replace them with their definitions instead.
3060 Description: interrupt oig_wait produce painc on resend.
3061 Details : brw_redo_request can be used for resend requests from ptlrpcd and
3062 private set, and this produce situation when rq_ptlrpcd_data not
3063 copyed to new allocated request and triggered LBUG on assert
3064 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
3065 ptlrpcd set if request is changed and can be safety changed to use
3068 Severity : enhancement
3070 Description: organize the server-side client stats on per-nid basis
3071 Details : Change the structure of stats under obdfilter and mds to
3079 The "uuid"s file would list the uuids of _active_ exports.
3080 And the clear entry is to clear all stats and stale nids.
3085 Description: Processes looping in ll_readdir() on Lustre clients finally causing
3086 a full node pseudo-hang
3087 Details : Concurrent access to the same directory from multiple clients with
3088 intensive file creation/removal can cause a client node to spin in
3089 ll_readdir(). i_version must be increased every time the lock is
3090 cancelled to ensure a revalidate is done.
3095 Description: touch file failed when fs is not full
3096 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
3097 otherwise we can get ENOSP while fs is not full.
3100 Frequency : only for Cray XT3
3101 Bugzilla : 12829/13455
3102 Description: Changing primary group doesn't change the group lustre assigns to
3104 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
3105 overridden with the primary group provided by the group upcall,
3106 whereas the supplied fsgid can be trusted if it is in the list of
3107 supplementary groups returned by the group upcall.
3109 Severity : enhancement
3111 Description: Root Squash Functionality
3112 Details : Implementation of NFS-like root squash capability. Specifically,
3113 don't allow someone with root access on a client node to be able
3114 to manipulate files owned by root on a server node.
3116 Severity : enhancement
3118 Description: Slow trucate/writes to huge files at high offsets.
3119 Details : Directly associate cached pages to lock that protect those pages,
3120 this allows us to quickly find what pages to write and remove
3121 once lock callback is received.
3126 Description: Too many locks accumulating on client during NFS usage
3127 Details : mds_open improperly used accmode to find out access mode to a
3128 file. Also mdc_intent_lock logic to find out if we already have
3129 lock similar to just received was flawed since introduction of
3130 skiplists - locks are now added to the front of the granted
3135 Description: Hit ASSERTION(obd->obd_stopping == 1) failed in some setup failed
3137 Details : In obd setup failure handler, obd_stopping will not necessarily to
3138 be 1, and obd_set_up should also be checked to make sure whether
3139 obd is completely setup.
3141 Severity : enhancement
3143 Description: Allow masking D_WARNING, D_ERROR messages from console
3144 Details : Console messages can now be disabled via lnet.printk.
3149 Description: User code with malformed file open parameter crashes client node
3150 Details : Before packing join_file req, all the related reference should be
3151 checked carefully in case some malformed flags cause fake join_file
3157 Description: shrink/enlarge qunit size when needed; fix the problem of coarse
3158 grain of quota doing harm to quota's accuracy
3159 Details : qunit size will be changed when quota limitation is too low/high;
3160 record the pending quota write in order to get more accureate
3161 quota; delete the patch for bug12588, which is unnecessary when
3162 this patch is landed. This bug also contains fixes for bug 14526,
3163 14299, 14601 and 13794.
3167 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
3168 Details : ldlm_completion_ast() assumes that a lock is granted when the req
3169 mode is equal to the granted mode. However, it should also check
3170 that LDLM_FL_CP_REQD is not set.
3174 Description: Heavy nfs access might result in deadlocks
3175 Details : After ELC code landed, it is now improper to enqueue any mds
3176 locks under och_sem, because enqueue might want to decide to
3177 cancel open locks for same inode we are holding och_sem for.
3181 Description: 35% write performance drop with ldiskfs2 when quotas are on
3182 Details : Enable ext3 journalled quota by default to improve performance
3183 when quotas are turned on.
3187 Description: Client eviction while running blogbench
3188 Details : A lot of unlink operations with concurrent I/O can lead to a
3189 deadlock causing evictions. To address the problem, the number of
3190 oustanding OST_DESTROY requests is now throttled to
3191 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
3192 callbacks are processed in priority.
3195 Frequency : RHEL4 only
3197 Description: mkfs is very slow on IA64/RHEL4
3198 Details : A performance regression has been discovered in the MPT Fusion
3199 driver between versions 3.02.73rh and 3.02.99.00rh. As a
3200 consequence, we have downgraded the MPT Fusion driver in the RHEL4
3201 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
3204 Frequency : PPC/PPC64 only
3206 Description: conflicts between asm-ppc64/types.h and lustre_types.h
3207 Details : fix duplicated definitions between asm-ppc64/types.h and
3208 lustre_types.h on PPC.
3211 Frequency : PPC/PPC64 only
3213 Description: asm-ppc/segment.h does not exist
3214 Details : fix compile issue on PPC.
3218 Description: data checksumming impacts single node performance
3219 Details : add support for several checksum algorithms. Currently, CRC32 and
3220 Adler-32 are supported. The checksum type can be changed on the fly
3221 through /proc/fs/lustre/osc/*/checksum_type.
3225 Description: use adler32 for page checksums
3226 Details : when available, use the Adler-32 algorithm instead of CRC32 for
3231 Description: better handle error messages in extents code
3233 Severity : enhancement
3235 Description: SNMP support enhancement
3236 Details : Adding total number of sampled request for an MDS node in snmp
3239 Severity : enhancement
3241 Description: Optimize ldlm waiting list processing for PR extent locks
3242 Details : When processing waiting list for read extent lock and meeting read
3243 lock that is same or wider to it that is not contended, skip
3244 processing rest of the list and immediatelly return current
3245 status of conflictness, since we are guaranteed there are no
3246 conflicting locks in the rest of the list.
3250 Description: Time out and refuse to reconnect
3251 Details : When the failover node is the primary node, it is possible
3252 to have two identical connections in imp_conn_list. We must
3253 compare not conn's pointers but NIDs, otherwise we
3254 can defeat connection throttling.
3258 Description: Client not clear own cache if answer to reconnect is lost.
3259 Details : Client gets evicted from server. Now client also thinks it is
3260 disconnected (or gets ENOTCONN on its operation) and decides to
3261 reconnect. Server receives reconnect message, but cannot find
3262 export. New export is created that is fully valid (new cookie!),
3263 but reply is lost and not reported to client. Client reconnects
3264 again and gets back a just-created connection, but it is not new
3265 so client thinks it was not evicted and does not do recovery.
3269 Description: Detect stride IO mode in read-ahead
3270 Details : When a client does stride read, read-ahead should detect that and
3271 read-ahead pages according to the detected stride pattern.
3275 Description: build for x2 fails
3276 Details : fix compile issue on Cray systems.
3278 Severity : enhancement
3280 Description: implement readv/writev APIs(aio_read/aio_writes in newer kernels)
3281 Details : This greatly improves speed of NFS writes on 2.6 kernels.
3284 Frequency : only on PPC/SLES10
3286 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
3287 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
3288 be defined. Add a hack in mkfs_lustre.c to work around this compile
3293 Description: LASSERT on MDS when client holding flock lock dies
3294 Details : ldlm pool logic depends on number of granted locks equal to
3295 number of released locks which is not true for flock locks, so
3296 just exclude such locks from consideration.
3300 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
3301 Details : Use fsfilt_sync() for both the whole filesystem sync and
3302 individual file sync to eliminate dangerous inode locking
3303 with I_LOCK that can lead to a deadlock.
3307 Description: Update an obsolete wirecheck.c generator
3308 Details : Update wirecheck.c/wirehdr.c and regenerate wiretest.c
3312 Description: Client can panic on open sometimes
3313 Details : It is possible that we try to free already freed request in
3314 ll_file_open in some error cases when we send request from
3319 Description: performance in 1.6.3
3320 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
3321 sure that 1MB requests are not fragmented by the block layer.
3325 Description: LDLM soft lockups - improvement
3326 Details : It is be possible to send the lock handle along with each read
3327 or write request because the client is already doing a lock match
3328 itself so there isn't any reason the OST should have to re-do that
3334 Description: lfs quota fails with deactivated OSTS
3335 Details : With this patch, three improvements are included:
3336 1. detete the softlimit in mds and osts when use "lfs quota".
3337 2. display the inaccurate data in the output of "lfs quota".
3338 3. try to get quota info when "lfs quota" is executed.
3343 Description: Extent locks not granted with no conflicts sometimes.
3344 Details : When race occurs in glimpse handler and nothing is returned,
3345 we do not reprocess the queue after lock cancel, and that leads
3346 to a stall until next activity on a resource
3349 Frequency : failover with quotaon
3351 Description: during mds failovers with quota on, OSTs got into deadlock state
3352 and causing dumpstack.
3353 Details : for every quota slave, at any time, there is only one quota req
3354 is sent to quota master for every uid/gid. Before that quota req
3355 returns, all the thread relative to the same uid/gid will wait.
3356 So if the quota req is lost because mds failovers or any other
3357 reasons, this bug will be hit. Now, dqacq_interpret() will handle
3358 quota reqs who time out.
3360 Severity : enhancement
3363 Description: when quota slave checks if quota is enough, there is an unnecessary
3365 Details : place this wait on necessary place instead of always waiting.
3367 --------------------------------------------------------------------------------
3369 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
3371 * Support for kernels:
3372 2.6.5-7.286 (SLES 9),
3373 2.6.9-55.0.9.EL (RHEL 4),
3374 2.6.16.53-0.8 (SLES 10),
3375 2.6.18-8.1.14.el5 (RHEL 5),
3376 2.6.18.8 vanilla (kernel.org)
3377 * Client support for unpatched kernels:
3378 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3379 2.6.16 - 2.6.22 vanilla (kernel.org)
3380 * Due to recently discovered recovery problems, we do not recommend
3381 using patchless RHEL 4 clients with this or any earlier release.
3382 * Recommended e2fsprogs version: 1.40.2-cfs1
3383 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3386 Frequency : occasional
3388 Description: MDS hang or stay in waiting lock
3389 Details : If client receive lock with CBPENDING flag ldlm need send lock
3390 cancel as separate rpc, to avoid situation when cancel request
3391 can't processed due all i/o threads stay in wait lock.
3394 Frequency : occasional
3396 Description: Do not fail import if osc_interpret_create gets -EAGAIN
3397 Details : If osc_interpret_create got -EAGAIN it immediately exits and
3398 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
3399 oscc_has_objects and see OSC has no objests and call
3400 oscc_internal_create to resend create request.
3402 Severity : enhancement
3404 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
3405 Details : Update which_patch & target file for SLES10 latest kernel.
3407 Severity : enhancement
3409 Description: add --type and --size parameters to lfs find
3410 Details : Enhance lfs find by adding filetype and filesize parameters. Also
3411 multiple OBDs can now be specified for the --obd option.
3413 Severity : enhancement
3415 Description: eliminate client locks in face of contention
3416 Details : file contention detection and lockless i/o implementation
3417 for contended files.
3419 Severity : enhancement
3421 Description: Remove client patches from SLES 10 kernel.
3422 Details : This causes SLES 10 clients to behave as patchless clients
3423 even on a Lustre-patched (server) kernel.
3425 Severity : enhancement
3427 Description: use i_size_read and i_size_write in 2.6 port
3428 Details : replace inode->i_size access with i_size_read/write()
3431 Frequency : when removing large files
3433 Description: scheduling issue during removal of large Lustre files
3434 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
3435 It causes scheduling issues when removing large files (17TB in the
3441 Description: 1.4.11 Can't handle directories with stripe set and extended ACLs
3442 Details : Impossible (EPROTO is returned) to access a directory that has a
3443 non-default striping and ACLs.
3446 Frequency : only on ppc
3448 Description: /proc/fs/lustre/devices broken on ppc
3449 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
3450 We should make sure the type of 'index' is loff_t and then cast
3451 explicitly as needed below. Do not assign an explicitly cast
3455 Frequency : only for rhel5
3457 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
3458 Details : Modify the target file & which_kernel.
3461 Frequency : if the uninit_groups feature is enabled on ldiskfs
3463 Description: e2fsck reports "invalid unused inodes count"
3464 Details : If a new ldiskfs filesystem is created with the "uninit_groups"
3465 feature and only a single inode is created in a group then the
3466 "bg_unused_inodes" count is incorrectly updated. Creating a
3467 second inode in that group would update it correctly.
3472 Description: buffer overruns could theoretically occur
3473 Details : llapi_semantic_traverse() modifies the "path" argument by
3474 appending values to the end of the origin string, and a buffer
3475 overrun may occur. Adding buffer overrun check in liblustreapi.
3477 Severity : enhancement
3479 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla.
3482 Frequency : only if filesystem is inconsistent
3484 Description: handle "serious error: objid * already exists" more gracefully
3485 Details : If LAST_ID value on disk is smaller than the objects existing in
3486 the O/0/d* directories, it indicates disk corruption and causes an
3487 LBUG(). If the object is 0-length, then we should use the existing
3488 object. This will help to avoid a full fsck in most cases.
3490 Severity : enhancement
3492 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
3493 Details : Modify vm-tunables-rhel4.patch.
3495 Severity : enhancement
3497 Description: Kernel config for 2.6.18-vanilla.
3498 Details : Modify targets/2.6-vanilla.target.in.
3499 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
3500 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
3501 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
3502 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
3505 Frequency : occasional
3507 Description: improve handling recoverable errors
3508 Details : If request processed with error which can be recoverable on server
3509 request should be resend, otherwise page released from cache and
3515 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
3516 Details : Modify target file & which_patch.
3517 A flaw was found in the IA32 system call emulation provided
3518 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
3519 value could be stored in the %RAX register, which could trigger an
3520 out-of-bounds system call table access. An untrusted local user
3521 could exploit this flaw to run code in the kernel
3522 (ie a root privilege escalation). (CVE-2007-4573).
3526 Description: change order of libsysio includes
3527 Details : '#include sysio.h' should always come before '#include xtio.h'
3529 Severity : enhancement
3531 Description: adapt the lustre_config script to support the upgrade case
3532 Details : Add "-u" option for lustre_config script to support upgrading 1.4
3533 server targets to 1.6 in parallel.
3538 Description: To avoid grant space > avaible space when the disk is almost
3539 full. Without this patch you might see the error "grant XXXX >
3540 available" or some LBUG about grant, when the disk is almost
3542 Details : In filter_check_grant, for non_grant cache write, we should
3543 check the left space by if (*left > ungranted + bytes), instead
3544 of (*left > ungranted), because only we are sure the left space
3545 is enough for another "bytes", then the ungrant space should be
3546 increase. In client, we should update cl_avail_grant only there is
3547 OBD_MD_FLGRANT in the reply.
3552 Description: Update RHEL 4 kernel to fix local root privilege escalation.
3553 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
3554 described in CVE-2007-4573. This problem could allow untrusted
3555 local users to gain root access.
3558 Frequency : when using O_DIRECT and quotas
3560 Description: Incorrect file ownership on O_DIRECT output files
3561 Details : block usage reported by 'lfs quota' does not take into account
3562 files that have been written with O_DIRECT.
3566 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
3567 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
3568 window to 0. If it does hit cache miss, and read-pattern is still
3569 stride-io mode, does not reset the stride window, but also does not
3570 increase the stride window length in this case.
3572 --------------------------------------------------------------------------------
3574 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
3576 * Support for kernels:
3577 2.6.5-7.286 (SLES 9),
3578 2.6.9-55.0.2.EL (RHEL 4),
3579 2.6.16.46-0.14 (SLES 10),
3580 2.6.18-8.1.8.el5 (RHEL 5),
3581 2.6.18.8 vanilla (kernel.org)
3582 * Client support for unpatched kernels:
3583 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3584 2.6.16 - 2.6.21 vanilla (kernel.org)
3585 * Due to recently discovered recovery problems, we do not recommend
3586 using patchless RHEL 4 clients with this or any earlier release.
3587 * Recommended e2fsprogs version: 1.40.2-cfs1
3588 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3592 Description: Fix errors in lfs documentation
3593 Details : Fixes man pages
3595 Severity : enhancement
3597 Description: Adaptive timeouts
3598 Details : RPC timeouts adapt to changing server load and network
3599 conditions to reduce resend attempts and improve recovery time.
3601 Severity : enhancement
3603 Description: llapi_file_create() does not allow some changes
3604 Details : add llapi_file_open() that allows specifying the file creation
3605 mode and open flags, and also returns an open file handle.
3607 Severity : enhancement
3609 Description: df doesn't work properly if diskfs blocksize != 4K
3610 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
3613 Frequency : always on directories with default striping set
3615 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
3616 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
3618 Severity : enhancement
3620 Description: merge and cleanup kernel patches.
3621 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
3623 Severity : enhancement
3625 Description: RedHat Update kernel for RHEL5
3626 Details : Kernel config file for RHEL5.
3628 Severity : enhancement
3630 Description: OSS needs mutliple precreate threads
3631 Details : Add ability to start more than one create thread per OSS.
3634 Frequency : only with quota on the root user
3636 Description: mds_obd_create error creating tmp object
3637 Details : When the user sets quota on root, llog will be affected and can't
3638 create files and write files.
3641 Frequency : Always on ia64 patchless client, and possibly others.
3643 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
3644 Details : This allows the patchless client to be loaded on architectures
3645 without this export.
3650 Description: disorder of journal start and llog_add cause deadlock.
3651 Details : in llog_origin_connect, journal start should happen before
3652 llog_add keep the same order as other functions to avoid
3655 Severity : enhancement
3657 Description: RedHat Update kernel for RHEL5
3658 Details : Modify the kernel config file more closer RHEL5.
3660 Severity : enhancement
3662 Description: Build failure against Centos5 (RHEL5)
3663 Details : Define PAGE_SIZE when it isn't present.
3666 Frequency : occasionally when using NFS
3668 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
3669 Details : with NFS, the anon dentry's parent was set to itself in
3670 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
3671 corresponding dentry other than use rec->ur_name.
3673 Severity : enhancement
3675 Description: enable data checksumming by default
3676 Details : enable checksum by default, allow --disable-checksum
3677 configure option and "-o nochecksum" mount option. Checksums
3678 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
3681 Frequency : Occasionally with failover
3683 Description: Client eviction due to failover config
3684 Details : after a connection loss, the lustre client should attempt to
3685 reconnect to the last active server first before trying the
3686 other potential connections.
3688 Severity : enhancement
3690 Description: client-side metadata stat-ahead during readdir(directory readahead)
3691 Details : perform client-side metadata stat-ahead when the client detects
3692 readdir and sequential stat of dir entries therein
3697 Description: when mds and osts use different quota unit(32bit and 64bit),
3698 quota will be released repeatly.
3699 Details : void sending multiple quota reqs to mds, which will keep the
3700 status between the reqs.
3703 Frequency : only with liblustre clients on XT3
3705 Description: evictions taking too long
3706 Details : allow llrd to evict clients directly on OSTs
3710 Description: osts not allocated evenly to files
3711 Details : change the condition to increase offset_idx
3713 Severity : enhancement
3715 Description: self-adjustable client's lru lists
3716 Details : use adaptive algorithm for managing client cached locks lru
3717 lists according to current server load, other client's work
3718 pattern, memory activities, etc. Both, server and client
3719 side namespaces provide number of proc tunables for controlling
3724 Description: rewrite ext2-derived code in llite/dir.c and obdclass/uuid.c
3725 Details : rewrite inherited code (uuid parsing code from ext2 utils and
3726 readdir code from ext3) from scratch preserving functionality.
3730 Description: Only those disconnect error should be returned by rq_status.
3731 Details : In open/enqueue processs, Some errors, which will cause client
3732 disconnected, should be returned by rq_status, while other
3733 errors should still be returned by intent, then mdc or llite will
3736 Severity : enhancement
3738 Description: Tune the kernel for good SCSI performance.
3739 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
3740 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
3744 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
3745 Bugzilla : 13547, 13627
3746 Description: Data corruption for OSTs that are formatted larger than 2TB
3748 Details : When generating the bio request for lustre file writes the
3749 sector number would overflow a temporary variable before being
3750 used for the IO. The data reads correctly from Lustre (which
3751 will overflow in a similar manner) but other file data or
3752 filesystem metadata may be corrupted in some cases.
3756 Description: "lfs find -obd UUID" prints directories
3757 Details : "lfs find -obd UUID" will return all directory names instead
3758 of just file names. It is incorrect because the directories
3759 do not reside on the OSTs.
3762 Bugzilla : 13596, 13614
3763 Description: MDS hang after unclean shutdown of lots of clients
3764 Details : Never resend AST requests.
3768 Frequency : Always, for kernels after 2.6.16
3769 Description: Fix warning idr_remove called for id=.. which is not allocated.
3770 Details : Last kernels save old s_dev before kill super and not allow
3771 to restore from callback - restore it before call kill_anon_super.
3775 Description: Fix error on 'ls .' at the top of the Lustre mount.
3776 Details : Don't revalidate dentry if it is a root dentry.
3778 --------------------------------------------------------------------------------
3780 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
3782 * Support for kernels:
3783 2.6.5-7.286 (SLES 9),
3784 2.6.9-55.0.2.EL (RHEL 4),
3785 2.6.16.46-0.14 (SLES 10),
3786 2.6.18-8.1.8.el5 (RHEL 5),
3787 2.6.18.8 vanilla (kernel.org)
3788 * Client support for unpatched kernels:
3789 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3790 2.6.16 - 2.6.22 vanilla (kernel.org)
3791 * Due to recently discovered recovery problems, we do not recommend
3792 using patchless RHEL 4 clients with this or any earlier release.
3793 * Recommended e2fsprogs version: 1.39.cfs8
3794 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3799 Description: block reactivating mgc import until all deactivates complete
3800 Details : Fix race when failing back MDT/MGS to itself (testing)
3802 Severity : enhancement
3804 Description: lfs setstripe enhancement
3805 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
3808 Frequency : mds/oss recovery
3810 Description: llog ctxt is refrenced after it has been freed.
3811 Details : llog ctxt refcount was added to avoide the race between ctxt free
3812 and llog recovery process. Each llog user must hold ctxt refcount
3813 before it access the llog. And the llog ctxt can only be freed
3814 when its refcount is zero.
3816 Severity : enhancement
3818 Description: randomly memory allocation failure util
3819 Details : Make lustre randomly failed allocating memory for testing purpose.
3822 Frequency : only for SLES10
3824 Description: Update kernel patch for SLES10 SP1
3825 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
3827 Severity : enhancement
3829 Description: omit set fsid for export NFS
3830 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
3831 as FS_REQUIRES_DEV for avoid problems with generate fsid.
3834 Frequency : after network failures
3836 Description: Add sync option to mount_lustre.c
3837 Details : Client loses data written to lustre after a network interruption.
3839 Severity : enhancement
3841 Description: Error message improvement.
3842 Details : Merging of two LCONSOLE_ERROR_MSG into one.
3844 Severity : enhancement
3846 Description: don't use GFP_* in generic Lustre code.
3847 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
3849 Severity : enhancement
3851 Description: obdclass is limited by single OBD_ALLOC(idarray)
3852 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
3854 Severity : enhancement
3856 Description: updated patchess for new RHEL4 kernel
3857 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
3859 Severity : enhancement
3861 Description: warnings with build patchless client with vanila 2.6.19 and up
3862 Details : change old ctl_table style and replace ctl_table/ctl_table_header
3863 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
3865 Severity : enhancement
3867 Description: O_DIRECT bypasses client statistics.
3868 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
3869 read_ahead_stats were updated. Copied stats section from
3870 osc_send_oap_rpc() into async_internal().
3873 Frequency : only for Cray XT3
3875 Description: peer credits not enough on many OST per OSS systems.
3876 Details : Use new lnet way to add credits as we need those for pings and ASTs
3879 Frequency : only with liblustre
3881 Description: Liblustre is not releasing flock locks on file close.
3882 Details : Release flock locks on file close.
3885 Frequency : only for RHEL4
3887 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
3888 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
3889 Update target file and kernel config.
3893 Description: lustre support for RHEL5
3894 Details : Add support for RHEL5.
3899 Description: ASSERTION(export != NULL) failed in target_handle_connect
3900 Details : Assetion hit is result of rare race between disconnect and connet
3901 to same nid. target_handle_connect found old connect cockie and
3902 tried to reconnect, but can't find export for this cockie.
3907 Description: umount blocks forever on error
3908 Details : In result of wrong using obd_no_recov and obd_force flags client
3909 can hand if cancel or some other requests is lost.
3912 Frequency : Only for SLES
3914 Description: sanity_quota fail test_1
3915 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
3916 file, which makes TSTID[2] inunique.
3918 Severity : enhancement
3920 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
3921 Details : Update target/ChangeLog/which_patch .
3923 Severity : enhancement
3925 Description: jbd statistics
3926 Details : Port older jbd statistics patch for sles10
3931 Description: lvbo_init failed for resource with missing objects.
3932 Details : Fix returning error if we do stat for file with missing/corrupted
3933 objects and i_size set to all sum of size of all avaible objects.
3934 if we to truncate/write to missing object - it is recreated.
3939 Description: Oops in read and write path when failing to allocate lock.
3940 Details : Check if lock allocation failed and return error back.
3943 Frequency : When flocks are used.
3945 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
3946 Details : Flock locks might destroy just granted lock if it could be merged
3947 with another existing flock, this is done in completion handler,
3948 so teach ldlm_cli_enquque_fini that this is a valid case for
3954 Description: reply_lock_interpret crash due to race with it and lock cancel.
3955 Details : Do not replay locks that are being cancelled. Do not reference
3956 locks by their address during replay, just by their handle.
3959 Frequency : only with deactivated OSTs
3961 Description: lstripe command fails for valid OST index
3962 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
3963 instead of lov->desc.ld_active_tgt_count.
3965 Severity : enhancement
3967 Description: Build failure against Centos5 (RHEL5)
3968 Details : Use getpagesize() instead of PAGE_SIZE.
3970 --------------------------------------------------------------------------------
3972 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
3974 * Support for kernels:
3975 2.6.5-7.283 (SLES 9),
3976 2.6.9-55.EL (RHEL 4),
3977 2.6.16.46-0.14 (SLES 10),
3978 2.6.18.8 vanilla (kernel.org)
3979 * Client support for unpatched kernels:
3980 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3981 2.6.16 - 2.6.22 vanilla (kernel.org)
3982 * Due to recently discovered recovery problems, we do not recommend
3983 using patchless RHEL 4 clients with this or any earlier release.
3984 * Recommended e2fsprogs version: 1.39.cfs8
3985 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3986 * Starting with this release, the ldiskfs backing filesystem required
3987 by Lustre is now in its own package, lustre-ldiskfs. This package
3988 should be installed. It is versioned separately from Lustre and
3989 may be released separately in future.
3994 Description: server LBUG when shutting down
3995 Details : Block umount forever until the mount refcount is zero rather
3996 than giving up after an arbitrary timeout.
3998 Severity : enhancement
4000 Description: add optional extra BUILD_VERSION info
4001 Details : add a new environment variable (namely LUSTRE_VERS) which allows
4002 to override the lustre version.
4005 Frequency : 2.6.18 servers only
4007 Description: ll_kern_mount() doesn't release the module reference
4008 Details : The ldiskfs module reference count never drops down to 0
4009 because ll_kern_mount() doesn't release the module reference.
4014 Description: server LBUG when using old ost_num_threads parameter
4015 Details : Accept the old ost_num_threads parameter but warn that it
4016 is deprecated, and fix an off-by-one error that caused an LBUG.
4021 Description: Transient SCSI error results in persistent IO issue
4022 Details : iobuf->dr_error is not reinitialized to 0 between two
4026 Frequency : sometimes when underlying device returns I/O errors
4028 Description: OSTs not going read-only during write failures
4029 Details : OSTs are not remounted read-only when the journal commit threads
4030 get I/O errors because fsfilt_ext3 calls journal_start/stop()
4031 instead of the ext3 wrappers.
4034 Frequency : only with 10000 clients or more
4036 Description: poor connect scaling with increasing client count
4037 Details : Don't run filter_grant_sanity_check for more than 100 exports
4038 to improve scaling for large numbers of clients.
4041 Frequency : SLES10 only
4043 Description: sanity-quota.sh quotacheck failed: rc = -22
4044 Details : Quotas cannot be enabled on SLES10.
4047 Frequency : liblustre clients only
4049 Description: getdirentries does not give error when run on compute nodes
4050 Details : getdirentries does not fail when the size specified as an argument
4051 is too small to contain at least one entry
4053 Severity : enhancement
4055 Description: Add LNET router traceability for debug purposes
4056 Details : If a checksum failure occurs with a router as part of the
4057 IO path, the NID of the last router that forwarded the bulk data
4058 is printed so it can be identified.
4063 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
4064 Details : Due to a race condition, liblustre clients were occasionally
4065 evicted incorrectly.
4067 Severity : enhancement
4069 Description: lfs setstripe use optional parameters instead of postional
4072 Severity : enhancement
4074 Description: Nanosecond timestamp support for ldiskfs
4075 Details : The on-disk ldiskfs filesystem has added support for nanosecond
4076 resolution timestamps. There is not yet support for this at
4077 the Lustre filesystem level.
4080 Frequency : during server recovery
4082 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
4083 Details : request with rq_no_resend flag not awake l_wait_event if they get a
4087 Frequency : nfs export on patchless client
4089 Description: connectathon hang when test nfs export over patchless client
4090 Details : Disconnected dentry cannot be found with lookup, so we do not need
4091 to unhash it or make it invalid
4094 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
4095 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
4096 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
4101 Description: open req refcounting wrong on reconnect
4102 Details : If reconnect happened between getting open reply from server and
4103 call to mdc_set_replay_data in ll_file_open, we will schedule
4104 replay for unreferenced request that we are about to free.
4105 Subsequent close will crash in variety of ways.
4106 Check that request is still eligible for replay in
4107 mdc_set_replay_data().
4112 Description: disable writes to filesystem when reading health_check file
4113 Details : the default for reading the health_check proc file has changed
4114 to NOT do a journal transaction and write to disk, because this
4115 can cause reads of the /proc file to hang and block HA state
4116 checking on a healthy but otherwise heavily loaded system. It
4117 is possible to return to the previous behaviour during configure
4118 with --enable-health-write.
4120 Severity : enhancement
4122 Description: 64-bit inode version
4123 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
4124 to the inode. This will be required for version-based recovery.
4129 Description: MDS fails to start if a duplicate client export is detected
4130 Details : in some rare cases it was possible for a client to connect to
4131 an MDS multiple times. Upon recovery the MDS would detect this
4132 and fail during startup. Handle this more gracefully.
4134 Severity : enhancement
4136 Description: Add -o localflock option to simulate old noflock behaviour.
4137 Details : This will achieve local-only flock/fcntl locks coherentness.
4142 Description: log_commit_thread vs filter_destroy race leads to crash
4143 Details : Take import reference before releasing llog record semaphore
4148 Description: Wrong request locking in request set processing
4149 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
4150 imp_delayed_list, in this place should be used imp_lock.
4153 Frequency : when reconnecting
4155 Description: Grant leak when OSC reconnect to OST
4156 Details : When osc reconnect ost, OST(filter) should check whether it
4157 should grant more space to client by comparing fed_grant and
4158 cl_avail_grant, and return the granted space to client instead
4159 of "new granted" space, because client will call osc_init_grant
4160 to update the client grant space info.
4163 Frequency : when client reconnects to OST
4165 Description: Grant leak when OSC does a resend and replays bulk write
4166 Details : When osc reconnect to OST, OST(filter) should clear grant info of
4167 bulk write request, because the grant info while be sync between
4168 OSC and OST when reconnect, and we should ignore the grant info
4169 these of resend/replay write req.
4174 Description: Grant space more than avaiable space sometimes.
4175 Details : When then OST is about to be full, if two bulk writing from
4176 different clients came to OST. Accord the avaliable space of the
4177 OST, the first req should be permitted, and the second one
4178 should be denied by ENOSPC. But if the seconde arrived before
4179 the first one is commited. The OST might wrongly permit second
4180 writing, which will cause grant space > avaiable space.
4183 Frequency : when client is evicted
4185 Description: Grant might be wrongly erased when osc is evicted by OST
4186 Details : when the import is evicted by server, it will fork another
4187 thread ptlrpc_invalidate_import_thread to invalidate the
4188 import, where the grant will be set to 0. While the original
4189 thread will update the grant it got when connecting. So if
4190 the former happened latter, the grant will be wrongly errased
4191 because of this race.
4196 Description: Checking Stale with correct fid
4197 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
4198 to check whether it is stale, because sometimes, we want the
4199 enqueue happened anyway, and op_data.fid2 will not be initialized.
4201 Severity : enhancement
4203 Description: update patchless client
4204 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
4207 Frequency : only with 2.4 kernel
4209 Description: random memory corruption
4210 Details : size of struct ll_inode_info is to big for union inode.u and this
4211 can be cause of random memory corruption.
4216 Description: Memory leak in recovery
4217 Details : Lov_mds_md was not free in an error handler in mds_create_object.
4218 It should also check obd_fail before fsfilt_start, otherwise if
4219 fsfilt_start return -EROFS,(failover mds during mds recovery).
4220 then the req will return with repmsg->transno = 0 and rc = EROFS.
4221 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
4222 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
4223 be freed no matter whether fsfilt_commit success or not.
4226 Frequency : only with large numbers of cores on a single node
4228 Description: Prevents from taking the superblock lock in llap_from_page for
4230 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
4231 for llap_from_page call in ll_removepage() prevents from taking
4232 the superblock lock for a soon died page.
4237 Description: Not check open intent error before release open handle
4238 Details : in some rare cases, the open intent error is not checked before
4239 release open handle, which may cause
4240 ASSERTION(open_req->rq_transno != 0), because it tries to release
4241 the failed open handle.
4246 Description: Set cat log bitmap only after create log success.
4247 Details : in some rare cases, the cat log bitmap is set too early. and it
4248 should be set only after create log success.
4252 Description: Accessing a block bevice can re-enable I/O when Lustre is
4253 tearing down a device.
4254 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
4259 Description: service threads may hog cpus when there are a lot of requests
4260 Details : Insert cond_resched to give other threads a chance to use some CPU
4265 Description: the cat log was not initialized in recovery
4266 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
4267 unlink log on mds will not be initialized until mds post
4268 recovery. And also in mds post recovery, the unlink log will
4269 initialization will be done asynchronausly, so there will be race
4270 between add unlink log and unlink log initialization.
4274 Description: brw_stats were being printed incorrectly
4275 Details : brw_stats were being printed as log2 but all of them were not
4276 recorded as log2. Also remove some code duplication arising from
4277 filter_tally_{read,write}.
4281 Frequency : rare, only in recovery.
4282 Description: ASSERTION(req->rq_type != LI_POISON) failed
4283 Details : imp_lock should be held while iterating over imp_sending_list for
4284 prevent destroy request after get timeout in ptlrpc_queue_wait.
4288 Description: replay-single.sh test 52 fails
4289 Details : A lock's skiplist need to be cleanup when it being unlinked
4290 from its resource list.
4295 Description: Short directio read returns full requested size rather than
4297 Details : Direct I/O operations should return actual amount of bytes
4298 transferred rather than requested size.
4300 Severity : enhancement
4302 Description: metadata RPC reduction (e.g. for rm performance)
4303 Details : decrease the amount of synchronous RPC between clients and servers
4304 by canceling conflicing lock before the operation on the client
4305 and packing thier handles into the main operation RPC to server.
4307 Severity : enhancement
4309 Description: add #ifdef HAVE_KERNEL_CONFIG_H
4310 Details : kernels from 2.6.19 not need include linux/config.h, but add
4311 include linux/autoconf.h in commpiler command line.
4313 Severity : enhancement
4315 Description: patchless client support for 2.6.22 kernel
4316 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
4317 removed. In this case we need drop using os depended interface to
4318 kmem_cache and use cfs_mem_cache API.
4323 Description: fix mal-formatted messages
4324 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
4328 Frequency : always in liblustre
4329 Description: wrong IS_ERR implementation in liblustre.h
4330 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
4335 Description: Correct condition for output debug message.
4336 Details : inode i_nlink equal zero is not enough for output message about
4337 disk corruption, i_ctime and i_mode should be also checked.
4341 Frequency : always in patchless client
4342 Description: add configure check for truncate_complete_page
4343 Details : improve checks for exported symbols. This allow run check without
4344 sources, but with Module.symvers shipped with kernel distribution.
4345 add check for truncate_complete_page used by patchless client.
4349 Description: sanity.sh test_77h fails with "test_77h file compare failed"
4350 Details : test_77h uses a file which was messed by other test case.
4354 Description: Not Check whether lov_tgts is NULL in some lov functions
4355 Details : Checking whether lov_tgts is NULL in some functions.
4359 Description: replace obdo_alloc() with OBDO_ALLOC macro
4360 Details : nothing special is done in obdo_alloc() function, and for
4361 debugging purpose, it needs to be replaced with macros.
4365 Description: bad return value and errno from fcntl call
4366 Details : In liblustre API, errno should be a negative value if error
4371 Description: ptlrpc_check_set() LBUG
4372 Details : In case of positive reply from server and failed client bulk
4373 callback after bulk transfer shouldn't LBUG, but process this
4374 request as erroneous.
4376 Severity : enhancement
4378 Description: VFS operations stats tool.
4379 Details : Tool which collects stats by tracking value written in pid,
4380 ppid, gid and uses llstat to generate output to plot graph using
4382 Updated lustre/utils/Makefile.am
4383 Added lustre/utils/ltrack_stats.c
4385 Severity : enhancement
4387 Description: 2.6.18 server support (lustre 1.6.1)
4388 Details : Support for 2.6.18 kernels on the server side.
4393 Description: ASSERTION(imp->imp_conn_current) failed
4394 Details : an assertion failure is hit if a client node boots and attempts to
4395 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
4398 Frequency : only for i686
4400 Description: 1.4.11 RC1 build fails for RHEL 4, i686
4401 Details : Fixed config variable for build.
4406 Description: Updated patchess for new RHEL4 kernel
4407 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
4408 Updated patch jbd-stats-2.6.9.patch
4409 Updated patch qsnet-rhel4-2.6.patch
4410 Updated patch quota-deadlock-on-pagelock-core.patch
4411 Updated patch vfs_intent-2.6-rhel4.patch
4412 Updated patch vfs_races-2.6-rhel4.patch
4413 Updated series file 2.6-rhel4-titech.series
4414 Updated series file 2.6-rhel4.series
4415 Updated kernel config files
4420 Description: lquota slave complains LBUG when reconnecting with mds
4422 Details : quota slave depends qctxt->lqc_import to send its quota request.
4423 This pointer will be invalid if mds did failover or broke its
4424 connect to osts, which leads to LBUG.
4427 Frequency : when qunit size is too small(less than 20M)
4429 Description: write is stopped by improper -EDQUOT
4430 Details : If the master is busy and qunit size is small enough(let's say 1M),
4431 the slave can not get quota from master on time, which will lead
4432 slave to trigger a -EQUOTA to client.
4437 Description: Deadlock during metadata tests
4438 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
4439 called with the per-dentry lock held.
4442 Frequency : SLES9 only
4444 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
4445 Details : set CONFIG_NR_CPUS to 128 instead of 8.
4447 Severity : enhancement
4449 Description: remove fs_prep_san_write operation and related patches
4450 Details : remove the ext3-san-jdike patches which are no longer useful.
4455 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
4456 Details : a disk corruption can cause the mballoc code to assert on a
4457 double free or other extent corruptions. Handle these with
4458 ext3_error() instead of assertions.
4461 Frequency : only with mballoc3 code and deep extent trees
4463 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
4464 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
4465 can cause the filesystem to be remounted read-only.
4470 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
4471 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
4472 few external users of this ioctl number is lfsck's call to
4473 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
4474 Add the old ioctl number to the handler so both old and new
4479 Description: parallel lock callbacks
4480 Details : Instead of sending blocking and completion callbacks as separated
4481 requests, adding them to a set and sending in parallel.
4485 Description: Disable most debugging by default
4486 Details : To improve performance, disable most logging (for debug purposes)
4487 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
4488 default, and HA includes fewer messages.
4492 Frequency : only run on patchless client.
4493 Description: use do_facet on sanity.sh for test handling recoverables errors
4494 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
4498 Description: hash tables for lists of nids, connections and uuids
4499 Details : Hash tables noticeably help when a lot of clients connect to a
4500 server, to faster identify duplicate connections or reconnects,
4501 also to faster find export to evict in manual eviction case.
4505 Description: Sometimes, when the server evict a client, and the client will
4506 not be evicted as soon as possible.
4507 Details : In enqueue req, the error was returned by intent, instead of
4508 rq_status which make ptlrpc layer not detect this error, and
4509 does not evict the client. So enqueue error should be returned
4513 Frequency : only at startup
4515 Description: Delay client connections to MDT until fist MDT->OST connect
4516 Details : If a client tried to create a new file before the MDT had
4517 connected to any OSTs, the create would return EIO. Now
4518 the client will simply block until the MDT connects to the
4519 first OST and the create can succeed.
4522 Frequency : only for SLES9
4524 Description: Routinely utilize latest Quadrics drivers in CFS releases
4525 Details : Update patch qsnet-suse-2.6.patch.
4528 Frequency : only for sles10
4530 Description: Update patches for SLES 10 SP1 kernel.
4531 Details : Update the patch vfs_intent-2.6-sles10.patch.
4536 Description: Routinely utilize latest Quadrics drivers in CFS releases
4537 Details : Update patch qsnet-rhel4-2.6.patch.
4540 Frequency : at statup only
4542 Description: mds_lov_synchronize race leads to various problems
4543 Details : simultaneous MDT->OST connections at startup can cause the
4544 sync to abort, leaving the OSC in a bad state.
4549 Description: Using wrong pointer in osc_brw_prep_request
4550 Details : Access to array[-1] can produce panic if kernel compiled with
4551 CONFIG_PAGE_ALLOC enabled
4553 Severity : enhancement
4555 Description: Async OSC create to avoid the blocking unnecessarily.
4556 Details : If a OST has no remain object, system will block on the creating
4557 when need to create a new object on this OST. Now, ways use
4558 pre-created objects when available, instead of blocking on an
4559 empty osc while others are not empty. If we must block, we block
4560 for the shortest possible period of time.
4564 Frequency : only in recovery
4565 Description: Mark OST as early accessible if his start SYNC.
4566 Details : osc_precreate return flag early accessible if oscc marked as
4567 OSCC_FLAG_SYNC_IN_PROGRESS.
4572 Description: Sometimes precreate code can triger create object on wrong ost
4573 Details : Wrong protected or not not restored variables aftre precreate loop
4574 can produce creation object on wrong ost.
4577 Frequency : oss recovery
4579 Description: llog_commit_thread cleanup should sync with llog_commit_thread
4581 Details : llog_commit_thread_count should be synced between llog_commit
4582 start and cleanup, so new llog_commit thread should not be started
4583 when llog_commit threads being stopped to avoid accessing some
4586 Severity : enhancement
4588 Description: Add printing inode info into message about error in writepage.
4590 --------------------------------------------------------------------------------
4592 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
4597 Frequency : on some architectures
4599 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
4600 Details : Uninitialized flags sometimes cause configuration commands to
4604 Frequency : patchless clients only
4606 Description: missing __iget() symbol export
4607 Details : The __iget() symbol export is missing. To avoid the need for
4608 this on patchless clients the deathrow inode reaper is turned
4609 off, and we depend on the VM to clean up old inodes. This
4610 dependency was during via the fix for bug 12181.
4615 Description: sanity.sh fail: test_52b
4616 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
4617 incorrect inode's flags to client.
4619 --------------------------------------------------------------------------------
4621 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
4623 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
4624 INTEROPERATE with older versions automatically. In many cases a
4625 special upgrade step is needed. Please read the
4626 user documentation before upgrading any part of a 1.4.x system.
4627 * WARNING: Lustre configuration and startup changes are required with
4628 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
4630 * Support for kernels:
4631 2.4.21-47.0.1.EL (RHEL 3),
4632 2.6.5-7.283 (SLES 9),
4633 2.6.9-42.0.10.EL (RHEL 4),
4634 2.6.12.6 vanilla (kernel.org),
4635 2.6.16.27-0.9 (SLES10)
4636 * Client support for unpatched kernels:
4637 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
4638 2.6.16 - 2.6.19 vanilla (kernel.org),
4639 2.6.9-42.0.8EL (RHEL 4)
4640 * Recommended e2fsprogs version: 1.39.cfs6
4641 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
4644 Severity : enhancement
4646 Description: Async OSC create to avoid the blocking unnecessarily.
4647 Details : If a OST has no remain object, system will block on the creating
4648 when need to create a new object on this OST. Now, ways use
4649 pre-created objects when available, instead of blocking on an
4650 empty osc while others are not empty. If we must block, we block
4651 for the shortest possible period of time.
4653 Severity : enhancement
4655 Description: MountConf
4656 Details : Lustre configuration is now managed via mkfs and mount
4657 commands instead of lmc and lconf. New obd types (MGS, MGC)
4658 are added for dynamic configuration management. See
4659 https://mail.clusterfs.com/wikis/lustre/MountConf for
4662 Severity : enhancement
4664 Description: dynamic OST addition
4665 Details : OSTs can now be added to a live filesystem
4667 Severity : enhancement
4669 Description: startup order invariance
4670 Details : MDTs and OSTs can be started in any order. Clients only
4671 require the MDT to complete startup.
4673 Severity : enhancement
4675 Description: parallel, asynchronous orphan cleanup
4676 Details : orphan cleanup is now performed in separate threads for each
4677 OST, allowing parallel non-blocking operation.
4679 Severity : enhancement
4681 Description: optimized stripe assignment
4682 Details : stripe assignments are now made based on ost space available,
4683 ost previous usage, and OSS previous usage, in order to try
4684 to optimize storage space and networking resources.
4686 Severity : enhancement
4688 Description: Permanently set tunables
4689 Details : All writable /proc/fs/lustre tunables can now be permanently
4690 set on a per-server basis, at mkfs time or on a live
4693 Severity : enhancement
4695 Description: Lustre message v2
4696 Details : Add lustre message format v2.
4698 Severity : enhancement
4700 Description: client OST exclusion list
4701 Details : Clients can be started with a list of OSTs that should be
4702 declared "inactive" for known non-responsive OSTs.
4706 Description: ENOENT returned for valid filehandle during dbench.
4707 Details : Check if a directory has children when invalidating dentries
4708 associated with an inode during lock cancellation. This fixes
4709 an incorrect ENOENT sometimes seen for valid filehandles during
4710 testing with dbench.
4713 Frequency : SFS test only (otherwise harmless)
4715 Description: SPEC SFS validation failure on NFS v2 over lustre.
4716 Details : Changes the blocksize for regular files to be 2x RPC size,
4717 and not depend on stripe size.
4719 Severity : enhancement
4721 Description: fine-grained SMP locking inside DLM
4722 Details : Improve DLM performance on SMP systems by removing the single
4723 per-namespace lock and replace it with per-resource locks.
4725 Severity : enhancement
4727 Description: don't hold multiple extent locks at one time
4728 Details : To avoid client eviction during large writes, locks are not
4729 held on multiple stripes at one time or for very large writes.
4730 Otherwise, clients can block waiting for a lock on a failed OST
4731 while holding locks on other OSTs and be evicted.
4733 Severity : enhancement
4735 Description: Multiple MD RPCs in flight.
4736 Details : Further unserialise some read-only MDT RPCs - learn about intents.
4737 To avoid overly-overloading MDT, introduce a limit on number of
4738 MDT RPCs in flight for a single client and add /proc controls
4739 to adjust this limit.
4741 Severity : enhancement
4743 Description: client read/write statistics
4744 Details : Add client read/write call usage stats for performance
4745 analysis of user processes.
4746 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
4747 file access. extents_stats shows chunk size distribution.
4748 extents_stats_per_process show chunk size distribution per
4751 Severity : enhancement
4753 Description: per-client statistics on server
4754 Details : Add ldlm and operations statistics for each client in
4755 /proc/fs/lustre/mds|obdfilter/*/exports/
4757 Severity : enhancement
4759 Description: improved MDT statistics
4760 Details : Add detailed MDT operations statistics in
4761 /proc/fs/lustre/mds/*/stats
4763 Severity : enhancement
4765 Description: VFS operations stats
4766 Details : Add client VFS call stats, trackable by pid, ppid, or gid
4767 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
4772 Description: Fix client-side osc byte counters
4773 Details : The osc read/write byte counters in
4774 /proc/fs/lustre/osc/*/stats are now working
4777 Frequency : always as root on SLES
4779 Description: Failure of copying files with lustre special EAs.
4780 Details : Client side always return success for setxattr call for lustre
4781 special xattr (currently only "trusted.lov").
4786 Description: Refcount LNET uuids
4787 Details : The global LNET uuid list grew linearly with every startup;
4788 refcount repeated list entries instead of always adding to
4791 Severity : enhancement
4793 Description: Dynamic service threads
4794 Details : Within a small range, start extra service threads
4795 automatically when the request queue builds up.
4798 Frequency : mixed-endian client/server environments
4800 Description: mixed-endian crashes
4801 Details : The new msg_v2 system had some failures in mixed-endian
4804 Severity : enhancement
4806 Description: Easy OST removal
4807 Details : OSTs can be permanently deactivated with e.g. 'lctl
4808 conf_param lustre-OST0001.osc.active=0'
4810 Severity : enhancement
4812 Description: MGS proc entries
4813 Details : Added basic proc entries for the MGS showing what filesystems
4816 Severity : enhancement
4818 Description: provide MGS failover
4819 Details : Added config lock reacquisition after MGS server failover.
4821 Severity : enhancement
4823 Description: add Linux 2.4 support
4824 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
4828 Description: a large application tries to do I/O to the same resource and dies
4829 in the middle of it.
4830 Details : Check the req->rq_arrival time after the call to
4831 ost_brw_lock_get(), but before we do anything about
4832 processing it & sending the BULK transfer request. This
4833 should help move old stale pending locks off the queue as
4834 quickly as obd_timeout.
4837 Frequency : when an incorrect nid is specified during startup
4839 Description: ptlrpc connect to non-existant node causes kernel crash
4840 Details : LNET can't be re-entered from an event callback, which
4841 happened when we expire a message after the export has been
4842 cleaned up. Instead, hand the zombie cleanup off to another
4845 Severity : enhancement
4847 Description: plain/inodebits lock performance improvement
4848 Details : Grouping plain/inodebits in granted list by their request modes
4849 and bits policy, thus improving the performance of search through
4853 Frequency : only if OST filesystem is corrupted
4855 Description: client incorrectly hits assertion in ptlrpc_replay_req()
4856 Details : for a short time RPCs with bulk IO are in the replay list,
4857 but replay of bulk IOs is unimplemented. If the OST filesystem
4858 is corrupted due to disk cache incoherency and then replay is
4859 started it is possible to trip an assertion. Avoid putting
4860 committed RPCs into the replay list at all to avoid this issue.
4863 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
4865 Description: System hang on startup
4866 Details : This bug allowed the liblustre (e.g. catamount) client to
4867 return to the app before handling all startup RPCs. This
4868 could leave the node unresponsive to lustre network traffic
4869 and manifested as a server ptllnd timeout.
4871 Severity : enhancement
4873 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
4874 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
4875 boolean to control whether to print peer debug info when a
4876 client's RPC times out.
4879 Frequency : only for kernels with patches from Lustre below 1.4.3
4881 Description: Remove old rdonly API
4882 Details : Remove old rdonly API which unused from at least lustre 1.4.3
4885 Frequency : only for devices with external journals
4887 Description: Set external device read-only also
4888 Details : During a commanded failover stop, we set the disk device
4889 read-only while the server shuts down. We now also set any
4890 external journal device read-only at the same time.
4893 Frequency : when upgrading from 1.4 while trying to change parameters
4895 Description: The wrong (new) MDC name was used when setting parameters for
4896 upgraded MDT's. Also allows changing of OSC (and MDC)
4897 parameters if --writeconf is specified at tunefs upgrade time.
4900 Frequency : when setting specific ost indicies
4902 Description: QOS code breaks on skipped indicies
4903 Details : Add checks for missing OST indicies in the QOS code, so OSTs
4904 created with --index need not be sequential.
4906 Severity : enhancement
4908 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
4909 Details : The uninit_groups feature works in conjunction with the kernel
4910 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
4911 up the pass1 processing of e2fsck. This is a read-only feature
4912 in ldiskfs2 only, so older kernels and current ldiskfs cannot
4913 mount filesystems that have had this feature enabled.
4915 Severity : enhancement
4917 Description: Improve multi-block allocation algorithm to avoid fragmentation
4918 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
4919 allocation locality and avoid filesystem fragmentation.
4921 ------------------------------------------------------------------------------
4923 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
4925 * Support for kernels:
4926 2.4.21-47.0.1.EL (RHEL 3)
4927 2.6.5-7.283 (SLES 9)
4928 2.6.9-42.0.10.EL (RHEL 4)
4929 2.6.12.6 vanilla (kernel.org)
4930 2.6.16.27-0.9 (SLES 10)
4931 * Recommended e2fsprogs version: 1.39.cfs5
4933 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
4937 Frequency : occasional, depends on client load and configuration
4938 Bugzilla : 12181, 12203
4939 Description: data loss for recently-modified files
4941 Details : In some cases it is possible that recently written or created
4942 files may not be written to disk in a timely manner (this should
4943 normally be within 30s unless client IO load is very high).
4944 The problem appears as zero-length files or files that are a
4945 multiple of 1MB in size after a client crash or client eviction
4946 that are missing data at the end of the file.
4948 This problem is more likely to be hit on clients where files are
4949 repeatedly created and unlinked in the same directory, clients
4950 have a large amount of RAM, have many CPUs, the filesystem has
4951 many OSTs, the clients are rebooted frequently, and/or the files
4952 are not accessed by other nodes after being written.
4954 The presence of the problem can be detected by looking at
4955 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
4956 smaller than the second (nr_unused) then dirty files will not
4957 be flushed automatically to disk. "sync; sleep 10" should be
4958 run several times on the node before unmounting it to update
4959 Lustre (this is also safe to run on nodes without this problem).
4961 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
4962 that can cause this same problem, so customers using that kernel
4963 also need to update the kernel in addition to Lustre. In order
4964 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
4966 It is normal that files written just before a client crash (less
4967 than 30s) may not yet have been flushed to disk, even for local
4971 Frequency : frequent on thin XT3 nodes
4973 Description: UUID collision on thin XT3 Linux nodes
4974 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
4975 randomly, since we relied on an insufficiently-seeded PRNG.
4980 Description: OSS hangs after "All ost request buffers busy"
4981 Details : A deadlock between quota and journal operations caused OSS
4982 hangs after printing "All ost request buffers busy."
4985 Frequency : always on liblustre builds
4987 Description: Cleanup compiler warnings on liblustre
4990 Frequency : always on liblustre builds on XT3
4992 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
4996 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
4997 > 32000 subdirectories
4998 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
4999 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
5000 subdirectory count crosses 32000. This will aid e2fsck to
5001 correctly handle more than 32000 subdirectories.
5004 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
5006 Description: System hang on startup
5007 Details : This bug allowed the liblustre (e.g. catamount) client to
5008 return to the app before handling all startup RPCs. This
5009 could leave the node unresponsive to lustre network traffic
5010 and manifested as a server ptllnd timeout.
5012 Severity : enhancement
5014 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
5015 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
5016 boolean to control whether to print peer debug info when a
5017 client's RPC times out.
5022 Description: make O_SYNC working on 2.6 kernels
5023 Details : 2.6 kernels use different method for mark pages for write,
5024 so need add a code to lustre for O_SYNC work.
5029 Description: Failure to close file and release space on NFS
5030 Details : Put inode details into lock acquired in ll_intent_file_open.
5031 Use mdc_intent_lock in ll_intent_open to properly
5032 detect all kind of errors unhandled by mdc_enqueue.
5037 Description: proc file read during shutdown sometimes raced obd removal,
5039 Details : Add lock to prevent obd access after proc file removal.
5042 Frequency : Only for files larger than 4GB on 32-bit clients.
5044 Description: improperly doing page alignment of locks
5045 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
5046 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
5051 Description: RPCs being resent when they shouldn't be
5052 Details : Some RPCs that should not be resent are being resent. This
5053 can cause inconsistencies in the RPC state machine. Do not
5054 resend such requests.
5057 Frequency : rare, only with NFS export
5059 Description: Crash on NFS re-export node
5060 Details : under very unusual load conditions an assertion is hit in
5061 ll_intent_file_open()
5064 Frequency : only if OST filesystem is corrupted
5066 Description: client incorrectly hits assertion in ptlrpc_replay_req()
5067 Details : for a short time RPCs with bulk IO are in the replay list,
5068 but replay of bulk IOs is unimplemented. If the OST filesystem
5069 is corrupted due to disk cache incoherency and then replay is
5070 started it is possible to trip an assertion. Avoid putting
5071 committed RPCs into the replay list at all to avoid this issue.
5076 Description: large O_DIRECT requests fail under memory pressure/fragmentation
5077 Details : Large single O_DIRECT read and write calls can fail to allocate
5078 a sufficiently large buffer to process the request. In case of
5079 allocation failure the allocation is retried with a smaller
5080 buffer and broken into smaller requests.
5082 Severity : enhancement
5084 Description: Add -o localflock option to simulate old noflock behaviour.
5085 Details : This will achieve local-only flock/fcntl locks coherentness.
5090 Description: versioning check is incomplete
5091 Details : Checking the version difference of client vs. server, report
5092 error if the gap is too big.
5097 Description: add support PG_writeback bit
5098 Details : add support for PG_writeback bit for Lustre, for more carefull
5099 work with page cache in 2.6 kernel. This also fix some deadlocks
5100 and remove hack for work O_SYNC with 2.6 kernel.
5102 Severity : enhancement
5104 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
5105 Details : The uninit_groups feature works in conjunction with the kernel
5106 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
5107 up the pass1 processing of e2fsck. This is a read-only feature
5108 in ldiskfs2 only, so older kernels and current ldiskfs cannot
5109 mount filesystems that have had this feature enabled.
5111 Severity : enhancement
5113 Description: Improve multi-block allocation algorithm to avoid fragmentation
5114 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
5115 allocation locality and avoid filesystem fragmentation.
5117 ------------------------------------------------------------------------------
5119 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
5121 * Support for kernels:
5122 2.6.9-42.0.3.EL (RHEL 4)
5123 2.6.5-7.276 (SLES 9)
5124 2.4.21-47.0.1.EL (RHEL 3)
5125 2.6.12.6 vanilla (kernel.org)
5126 2.6.16.21-0.8 (SLES10)
5127 * Recommended e2fsprogs version: 1.39.cfs2-0
5129 * The backwards-compatible /proc/sys/portals symlink has been removed
5130 in this release. Before upgrading, please ensure that you change
5131 any configuration scripts or /etc/sysctl.conf files that access
5132 /proc/sys/portals/* or sysctl portals.* to use the corresponding
5133 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
5134 in advance of the upgrade on any system running Lustre 1.4.6 or
5135 newer, since /proc/sys/lnet was added in that version.
5136 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
5140 Frequency : only when quota is used
5142 Description: avoid scanning export list for quota master
5143 Details : Change the algorithms to avoid scanning export list in order
5144 to improve the efficiency.
5147 Frequency : MDS failover only, very rarely
5149 Description: "went back in time" messages on mds failover
5150 Details : The greatest transno may be lost when the current operation
5151 finishes with an error (transno==0) and the client's last_rcvd
5152 record is over-written. Save the greatest transno in the
5153 mds_last_transno for this case.
5156 Frequency : always for specific kernels and striping counts
5158 Description: client may get "Matching packet too big" without ACL support
5159 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
5160 when trying to access files in certain configurations. The
5161 clients should in fact be denied when mounting because they do
5162 not understand ACLs.
5165 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
5167 Description: many clients connecting with IO in progress causes connect timeouts
5168 Details : Avoid synchronous journal commits to avoid delays caused by many
5169 clients connecting/disconnecting when bulk IO is in progress.
5170 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
5171 OST_IO_PORTAL to avoid delays behind potentially many pending
5175 Frequency : occasionally with multiple writers to a single file
5177 Description: shared writes to file may result in wrong size reported by stat()
5178 Details : Allow growing of kms when extent lock is cancelled
5181 Frequency : always with random mmap IO to multi-striped file
5183 Description: mmap write might be lost if we are writing to a 'hole' in stripe
5184 Details : Only if the hole is at the end of OST object so that kms is too
5185 small. Fix is to increase kms accordingly in ll_nopage.
5188 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
5190 Description: writes to a missing object would leak memory on the OST
5191 Details : If there is an inconsistency between the MDS and OST filesystems,
5192 such that the MDS references an object that doesn't exist, writes
5193 to that object will leak memory due to incorrect cleanup in the
5194 error handling path, eventually running out of memory on the OST.
5199 Description: Creating too long symlink causes lustre errors
5200 Details : Check symlink and name lengths before sending requests to MDS.
5203 Frequency : only if flock is enabled (not on by default)
5205 Description: posix locks not released on fd closure on 2.6.9+
5206 Details : We failed to add posix locks to list of inode locks on 2.6.9+
5207 kernels, this caused such locks not to be released on fd close and
5208 then assertions on fs unmount about still used locks.
5211 Frequency : MDS failover only, very rarely
5213 Description: clients may get ASSERTION(granted_lock != NULL)
5214 Details : When request was taking a long time, and a client was resending
5215 a getattr by name lock request. The were multiple lock requests
5216 with the same client lock handle and
5217 mds_getattr_name->fixup_handle_for_resent_request found one of the
5218 lock handles but later failed with ASSERTION(granted_lock != NULL).
5223 Description: handle->h_buffer_credits > 0, assertion failure
5224 Details : h_buffer_credits is zero after truncate, causing assertion
5225 failure. This patch extends the transaction or creates a new
5229 Frequency : NFS re-export or patchless client
5230 Bugzilla : 11179, 10796
5231 Description: Crash on NFS re-export node (__d_move)
5232 Details : We do not want to hash the dentry if we don't have a lock.
5233 But if this dentry is later used in d_move, we'd hit uninitialised
5234 list head d_hash, so we just do this to init d_hash field but
5235 leave dentry unhashed.
5238 Frequency : NFS re-export or patchless client
5240 Description: NFS exports has problem with symbolic link
5241 Details : lustre client didn't properly install dentry when re-exported
5242 to NFS or running patchless client.
5245 Frequency : NFS re-export or patchless client
5247 Description: Various nfs/patchless fixes.
5248 Details : fixes reuse disconected alias for lookup process - this fixes
5249 warning "find_exported_dentry: npd != pd",
5250 fix permission error with open files at nfs.
5251 fix apply umask when do revalidate.
5254 Frequency : occasional
5256 Description: Crash on NFS re-export node
5257 Details : calling clear_page() on the wrong pointer triggered oops in
5258 generic_mapping_read().
5261 Frequency : rarely, using O_DIRECT IO
5263 Description: unaligned directio crashes client with LASSERT
5264 Details : check for unaligned buffers before trying any requests.
5267 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
5269 Description: stale data returned from RAID cache
5270 Details : If only a small amount of IO is done to the RAID device before
5271 reading it again it is possible to get stale data from the RAID
5272 cache instead of reading it from disk.
5275 Frequency : always for sles10 kernel
5277 Description: sles10 support
5278 Details : ll_follow_link: compile fixes and using of nd_set_link
5279 under newer kernels.
5282 Frequency : depends on arch, kernel and compiler version, always on sles10
5285 Description: recursive or deep enough symlinks cause stack overflow
5286 Details : getting rid of large stack-allocated variable in
5290 Frequency : depends on hardware
5292 Description: lustre write performance loss in the SLES10 kernel
5293 Details : the performance loss is caused by using of write barriers in the
5294 ext3 code. The SLES10 kernel turns barrier support on by
5295 default. The fix is to undo that change for ldiskfs.
5297 ------------------------------------------------------------------------------
5299 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
5301 * Support for kernels:
5302 2.6.9-42.0.3EL (RHEL 4)
5303 2.6.5-7.276 (SLES 9)
5304 2.4.21-47.0.1.EL (RHEL 3)
5305 2.6.12.6 vanilla (kernel.org)
5309 Frequency : quota enabled and large files being deleted
5311 Description: releasing more than 4GB of quota at once hangs OST
5312 Details : If a user deletes more than 4GB of files on a single OST it
5313 will cause the OST to spin in an infinite loop. Release
5314 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
5319 Description: statfs data retrieved from /proc may be stale or zero
5320 Details : When reading per-device statfs data from /proc, in the
5321 {kbytes,files}_{total,free,avail} files, it may appear
5322 as zero or be out of date.
5325 Frequency : systems with MD RAID1 external journal devices
5327 Description: lconf's call to blkid is confused by RAID1 journal devices
5328 Details : Use the "blkid -l" flag to locate the MD RAID device instead
5329 of returning all block devices that match the journal UUID.
5332 Frequency : always, for aggregate stripe size over 4GB
5334 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
5335 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
5336 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
5339 Frequency : always if "lfs find" used on a local file/directory
5341 Description: "lfs find" segfaults if used on a local file/directory
5342 Details : The case where a directory component was not specified wasn't
5343 handled correctly. Handle this properly.
5346 Frequency : always on ppc64
5348 Description: the write to an ext3 filesystem mounted with mballoc got stuck
5349 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
5350 perform endianness conversion.
5353 Frequency : rarely (truncate to non-zero file size after write under load)
5354 Bugzilla : 10730, 10687
5355 Description: Files padded with zeros to next 4K multiple
5356 Details : With filesystems mounted using the "extents" option (2.6 kernels)
5357 it is possible that files that are truncated to a non-zero size
5358 immediately after being written are filled with zero bytes beyond
5359 the truncated size. No file data is lost.
5361 Severity : enhancement
5363 Description: Allow recovery/failover for liblustre clients.
5364 Details : liblustre clients were unaware of failover configurations until
5367 Severity : enhancement
5369 Description: user file locks should fail when not mounting with flock option
5370 Details : Set up an error-returning stub in ll_file_operations.lock field
5371 to prevent incorrect behaviour when client is mounted without
5372 flock option. Also, set up properly f_op->flock field for
5376 Frequency : always on ia64
5378 Description: "lfs df" loops on printing out MDS statfs information
5379 Details : The obd_ioctl_data was not initialized and in some systems
5380 this caused a failure during the ioctl that did not return
5381 an error. Initialize the struct and return an error on failure.
5384 Frequency : SLES 9 only
5386 Description: Error of copying files with lustre special EAs as root
5387 Details : Client side always return success for setxattr call for lustre
5388 special xattr (currently only "trusted.lov").
5391 Frequency : rarely on clusters with both ia64+i386 clients
5393 Description: ia64+i686 clients doing shared IO on the same file may LBUG
5394 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
5395 clients are doing concurrent writes to the same file it is
5396 possible that the ia64 clients may LASSERT because the OST
5397 extent locks are not PAGE_SIZE aligned. Ensure that grown
5398 locks are always aligned on the request boundary.
5401 Frequency : specific use, occasional
5403 Description: Overwriting in use executable truncates on-disk binary image
5404 Details : If one node attempts to overwrite an executable in use by
5405 another node, we now correctly return ETXTBSY instead of
5406 truncating the file.
5408 Severity : enhancement
5410 Description: Async OSC create to avoid the blocking unnecessarily.
5411 Details : If a OST has no remain object, system will block on the creating
5412 when need to create a new object on this OST. Now, ways use
5413 pre-created objects when available, instead of blocking on an
5414 empty osc while others are not empty. If we must block, we block
5415 for the shortest possible period of time.
5420 Description: chmod on Lustre root is propagated to other clients
5421 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
5422 invalid by the follow_mount time.
5425 Frequency : liblustre clients only
5427 Description: Race in 'instant cancel' lock handling could lead to such locks
5428 never to be granted in case of SMP MDS
5429 Details : Do not destroy not yet granted but cbpending locks in
5433 Frequency : replay/resend of open
5435 Description: non null lock assetion failure in mds_intent_policy
5436 Details : Trying to replay/resend lockless open requests resulted in
5437 mds_open() returning 0 with no lock. Now it sets a flag if
5438 a lock is going to be returned.
5440 Severity : enhancement
5442 Description: Checksum enhancements
5443 Details : New checksum enhancements allow for resending RPCs that failed
5446 Severity : enhancement
5448 Description: Tunables on number of dirty pages in cacche
5449 Details : Allow to set limit on number of dirty pages cached.
5454 Description: client crash on unmount - lock still has references
5455 Details : In some error handling cases it was possible to leak a lock
5456 reference on a client while accessing a file. This was not
5457 harmful to the client during operation, but would cause the
5458 client to crash when the filesystem is unmounted.
5461 Frequency : specific case, rare
5463 Description: ETXTBSY on mds though file not in use by client
5464 Details : ETXTBSY is no longer incorrectly returned when attempting to
5465 chmod or chown a directory that the user previously tried to
5466 execute or a currently-executing binary.
5469 Frequency : extremely rare except on liblustre-based clients
5471 Description: Lustre space not freed when files are deleted
5472 Details : Clean up open-unlinked files after client eviction. Previously
5473 the unlink was skipped and the files remained as orphans.
5478 Description: OST failure "would be an LBUG" in waiting_locks_callback()
5479 Details : In some cases it was possible to send a blocking callback to
5480 a client doing a glimpse, even though that client didn't get
5481 a lock granted. When the glimpse lock is cancelled on the OST
5482 the freed lock is left on the waiting list and corrupted the list.
5485 Frequency : all core dumps
5487 Description: Broke core dumps to lustre
5488 Details : Negative dentry may be unhashed if parent does not have UPDATE
5489 lock, but some callers, e.g. do_coredump, expect dentry to be
5490 hashed after successful create, hash it in ll_create_it.
5492 ------------------------------------------------------------------------------
5494 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
5496 * Support for kernels:
5497 2.6.9-42.0.2.EL (RHEL 4)
5498 2.6.5-7.276 (SLES 9)
5499 2.4.21-40.EL (RHEL 3)
5500 2.6.12.6 vanilla (kernel.org)
5504 Frequency : always on RHEL 3
5506 Description: Number of open files grows over time
5507 Details : The number of open files grows over time, whether or not
5508 Lustre is started. This was due to a filp leak introduced
5509 by one of our kernel patches.
5511 ------------------------------------------------------------------------------
5513 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
5515 * Support for kernels:
5516 2.6.9-42.EL (RHEL 4)
5517 2.6.5-7.267 (SLES 9)
5518 2.4.21-40.EL (RHEL 3)
5519 2.6.12.6 vanilla (kernel.org)
5524 Bugzilla : 5719, 9635, 9792, 9684
5525 Description: OST (or MDS) trips assertions in (re)connection under heavy load
5526 Details : If a server is under heavy load and cannot reply to new
5527 connection requests before the client resends the (re)connect,
5528 the connection handling code can behave badly if two service
5529 threads are concurrently handing separate (re)connections from
5530 the same client. Add better locking to the connection handling
5531 code, and ensure that only a single connection will be processed
5532 for a given client UUID, even if the lock is dropped.
5534 Severity : enhancement
5536 Description: add TCP zero-copy support to kernel
5537 Details : Add support to the kernel TCP stack to allow zero-copy bulk
5538 sends if the hardware supports scatter-gather and checksumming.
5539 This allows socklnd to do client-write and server-read more
5540 efficiently and reduce CPU utilization from skbuf copying.
5543 Frequency : only if NFS exporting from client
5545 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
5546 Details : If setting attributes on a file created under NFS that had
5547 never been opened it would be possible to oops the client
5548 if the file had no objects.
5552 Bugzilla : 9326, 10402, 10897
5553 Description: client crash in ptlrpcd_wake() thread when sending async RPC
5554 Details : It is possible that ptlrpcd_wake() dereferences a freed async
5555 RPC. In rare cases the ptlrpcd thread alread processed the RPC
5556 before ptlrpcd_wake() was called and the request was freed.
5559 Frequency : always for liblustre
5561 Description: liblustre client does MDS+OSTs setattr RPC for each write
5562 Details : When doing a write from a liblustre client, the client
5563 incorrectly issued an RPC to the MDS and each OST the file was
5564 striped over in order to update the timestamps. When writing
5565 with small chunks and many clients this could overwhelm the MDS
5566 with RPCs. In all cases it would slow down the write because
5567 these RPCs are unnecessary.
5569 Severity : enhancement
5571 Description: allow number of MDS service threads to be changed at module load
5572 Details : It is now possible to change the number of MDS service threads
5573 running. Adding "options mds mds_num_threads={N}" to the MDS's
5574 /etc/modprobe.conf will set the number of threads for the next
5575 time Lustre is restarted (assuming the "mds" module is also
5576 reloaded at that time). The default number of threads will
5577 stay the same, 32 for most systems.
5582 Description: OST crash if filesystem is unformatted or corrupt
5583 Details : If an OST is started on a device that has never been formatted
5584 or if the filesystem is corrupt and cannot even mount then the
5585 error handling cleanup routines would dereference a NULL pointer.
5590 Description: NULL pointer deref in llap_from_page.
5591 Details : get_cache_page_nowait can return a page with NULL (or otherwise
5592 incorrect) mapping if the page was truncated/reclaimed while it was
5593 searched for. Check for this condition and skip such pages when
5594 doing readahead. Introduce extra check to llap_from_page() to
5595 verify page->mapping->host is non-NULL (so page is not anonymous).
5598 Frequency : Sometimes when using sys_sendfile
5600 Description: "page not covered by a lock" warnings from ll_readpage
5601 Details : sendfile called ll_readpage without right page locks present.
5602 Now we introduced ll_file_sendfile that does necessary locking
5603 around call to generic_file_sendfile() much like we do in
5607 Frequency : with certain MDS communication failures at client mount time
5609 Description: NULL pointer deref after failed client mount
5610 Details : a client connection request may delayed by the network layer
5611 and not be sent until after the PTLRPC layer has timed out the
5612 request. If the client fails the mount immediately it will try
5613 to clean up before the network times out the request. Add a
5614 reference from the request import to the obd device and delay
5615 the cleanup until the network drops the request.
5618 Frequency : occasionally during client (re)connect
5620 Description: assertion failure during client (re)connect
5621 Details : processing a client connection request may be delayed by the
5622 client or server longer than the client connect timeout. This
5623 causes the client to resend the connection request. If the
5624 original connection request is replied in this interval, the
5625 client may trip an assertion failure in ptlrpc_connect_interpret()
5626 which thought it would be the only running connect process.
5629 Frequency : only with obd_echo servers and clients that are rebooted
5631 Description: kernel BUG accessing uninitialized data structure
5632 Details : When running an obd_echo server it did not start the ping_evictor
5633 thread, and when a client was evicted an uninitialized data
5634 structure was accessed. Start the ping_evictor in the RPC
5635 service startup instead of the OBD startup.
5637 Severity : enhancement
5638 Bugzilla : 10193 (patchless)
5639 Description: Remove dependency on various unexported kernel interfaces.
5640 Details : No longer need reparent_to_init, exit_mm, exit_files,
5641 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
5644 Frequency : rare (only users of deprecated and unsupported LDAP config)
5646 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
5647 Details : LDAP apparently contains 'lustreName' attributes instead of
5648 'name'. A simple remapping of the name is sufficient.
5651 Frequency : rare (only with non-default dump_on_timeout debug enabled)
5653 Description: waiting_locks_callback trips kernel BUG if client is evicted
5654 Details : Running with the dump_on_timeout debug flag turned on makes
5655 it possible that the waiting_locks_callback() can try to dump
5656 the Lustre kernel debug logs from an interrupt handler. Defer
5657 this log dumping to the expired_lock_main() thread.
5659 Severity : enhancement
5661 Description: Support NFS exporting on 2.6 kernels.
5662 Details : Implement non-rawops metadata methods for NFS server to use without
5663 changing NFS server code.
5666 Frequency : very rare (synthetic metadata workload only)
5668 Description: two racing renames might cause an MDS thread to deadlock
5669 Details : Running the "racer" program may cause one MDS thread to rename
5670 a file from being the source of a rename to being the target of
5671 a rename at exactly the same time that another thread is doing
5672 so, and the second thread has already enqueued these locks after
5673 doing a lookup of the target and is trying to relock them in
5674 order. Ensure that we don't try to re-lock the same resource.
5677 Frequency : only very large systems with liblustre clients
5679 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
5680 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
5681 all OSTs in parallel. This allows the eviction of stale liblustre
5682 clients to proceed much faster than if they were done in series,
5683 and also offers similar improvements for other set_info RPCs.
5688 Description: excessive CPU usage during initial read phase on client
5689 Details : During the initial read phase on a client, it would agressively
5690 retry readahead on the file, consuming too much CPU and impacting
5691 performance (since 1.4.5.8). Improve the readahead algorithm
5692 to avoid this, and also improve some other common cases (read
5693 of small files in particular, where "small" is files smaller than
5694 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
5699 Description: MDS crash when receiving packet with unknown intent.
5700 Details : Do not LBUG in unknown intent case, just return -EFAULT
5702 Severity : enhancement
5703 Bugzilla : 9293, 9385
5704 Description: MDS RPCs are serialised on client. This is unnecessary for some.
5705 Details : Do not serialize getattr (non-intent version) and statfs.
5708 Frequency : occasional, when OST network is overloaded/intermittent
5710 Description: client evicted by OST after bulk IO timeout
5711 Details : If a client sends a bulk IO request (read or write) the OST
5712 may evict the client if it is unresposive to its data GET/PUT
5713 request. This is incorrect if the network is overloaded (takes
5714 too long to transfer the RPC data) or dropped the OST GET/PUT
5715 request. There is no need to evict the client at all, since
5716 the pinger and/or lock callbacks will handle this, and the
5717 client can restart the bulk request.
5720 Frequency : Always when mmapping file with no objects
5722 Description: client crashes when mmapping file with no objects
5723 Details : Check that we actually have objects in a file before doing any
5724 operations on objects in ll_vm_open, ll_vm_close and
5730 Description: Request leak when working with deleted CWD
5731 Details : Introduce advanced request refcount tracking for requests
5732 referenced from lustre intent.
5734 Severity : Enhancement
5736 Description: Cache open file handles on client.
5737 Details : MDS now will return special lock along with openhandle, if
5738 requested and client is allowed to hold openhandle, even if unused,
5739 until such a lock is revoked. Helps NFS a lot, since NFS is opening
5740 closing files for every read/write openration.
5742 Severity : Enhancement
5744 Description: Cache open negative dentries on client when possible.
5745 Details : Guard negative dentries with UPDATE lock on parent dir, drop
5746 negative dentries on lock revocation.
5751 Description: Remounting a client read-only wasn't possible with a zconf mount
5752 Details : It wasn't possible to remount a client read-only with llmount.
5754 Severity : enhancement
5755 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
5756 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
5757 MPICH is included to add Lustre-specific ADIO interfaces.
5758 This is based closely on the UFS ADIO layer and only differs
5759 in file creation, in order to allow the OST striping to be set.
5760 This is user-contributed code and not supported by CFS.
5765 Description: extended inode attributes (immutable, append-only) work improperly
5766 when 2.4 and 2.6 kernels are used on client/server or vice versa
5767 Details : Introduce kernel-independent values for these flags.
5769 Severity : enhancement
5772 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
5773 Details : Many of the /proc/ tunables can only be tuned at a megabyte
5774 granularity. Now, Fractional MB granularity is be supported,
5775 this is very useful for low memory system.
5777 Severity : enhancement
5779 Description: Getattr by fid
5780 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
5781 locks, avoids extra getattr rpc requests to MDS, allows '/' to
5782 have locks and avoids getattr rpc requests for it on every stat.
5785 Frequency : Always, for filesystems larger than 2TB
5787 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
5788 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
5789 filesystems larger than 2TB to be mounted with mballoc enabled.
5792 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
5794 Description: filesystem corruption for non-standard kernels and very large OSTs
5795 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
5796 filesystems larger than 2TB is mounted then the kernel will
5797 silently corrupt the start of the filesystem. CONFIG_LBD is
5798 enabled for all CFS-supported kernels, but the possibility of
5799 this happening with a modified kernel config exists.
5801 Severity : enhancement
5803 Description: add client O_DIRECT support for 2.6 kernels
5804 Details : It is now possible to do O_DIRECT reads and writes to files
5805 in the Lustre client mountpoint on 2.6 kernel clients.
5807 Severity : enhancement
5809 Description: parallel glimpse, setattr, statfs, punch, destroy requests
5810 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
5811 parallel, not waiting for response from every OST before sending
5812 a rpc to the next OST.
5817 Description: setattr vs write race when updating file timestamps
5818 Details : Client processes that update a file timestamp into the past
5819 right after writing to the file (e.g. tar) it is possible that
5820 the updated file modification time can be reset to the current
5821 time due to a race between processing the setattr and write RPC.
5823 Severity : enhancement
5825 Description: Bring 'lfs find' closer in line with regular Linux find.
5826 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
5827 -print0 options and obtains all the needed info through the lustre
5830 Severity : enhancement
5832 Description: support up to 1024 configured devices on one node
5833 Details : change obd_dev array from statically allocated to dynamically
5834 allocated structs as they are first used to reduce memory usage
5839 Description: Flush dirty partially truncated pages during truncate
5840 Details : Immediatelly flush partially truncated pages in filter_setattr,
5841 this way we completely avoid having any pages in page cache on OST
5842 and can retire ugly workarounds during writes to flush such pages.
5847 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
5848 Details : protect inode from truncation within vfs_unlink() context
5849 just take a reference before calling vfs_unlink() and release it
5850 when parent's i_sem is free.
5853 Frequency : always, if extents are used on OSTs
5855 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
5856 Details : OSTs using the extents format would not zero the high 16 bits of
5857 the index physical block number. This is not a problem for any
5858 OST filesystems smaller than 16TB, and no kernels support ext3
5859 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
5860 new/modified files) and can be fixed for existing filesystems
5861 with e2fsprogs-1.39-cfs1.
5866 Description: import connection selection may be incorrect if timer wraps
5867 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
5868 import connections to be ignored if the 32-bit jiffies counter
5869 wraps. Use a 64-bit jiffies counter.
5872 Frequency : during server recovery
5874 Description: crash after server is denying duplicate export
5875 Details : If clients are resending connect requests to the server, the
5876 server refuses to allow a client to connect multiple times.
5877 Fixed a bug in the handling of this case.
5880 Frequency : very large clusters immediately after boot
5882 Description: LNET request buffers exhausted under heavy short-term load
5883 Details : If a large number of client requests are generated on a service
5884 that has previously never seen so many requests it is possible
5885 that the request buffer growth cannot keep up with the spike in
5886 demand. Instead of dropping incoming requests, they are held in
5887 the LND until the RPC service can accept more requests.
5890 Frequency : Sometimes during replay
5892 Description: Assertion failure in ll_local_open after replay.
5893 Details : If replay happened on an open request reply before we were able
5894 to set replay handler, reply will become not swabbed tripping the
5895 assertion in ll_local_open. Now we set the handler right after
5896 recognising of open request
5899 Frequency : very rare
5901 Description: kernel reports "badness in vsnprintf"
5902 Details : Reading from the "recovery_status" /proc file in small chunks
5903 may cause a negative length in lprocfs_obd_rd_recovery_status()
5904 call to vsnprintf() (which is otherwise harmless). Exit early
5905 if there is no more space in the output buffer.
5907 Severity : enhancement
5909 Description: clear OBD RPC statistics by writing to them
5910 Details : It is now possible to clear the OBD RPC statistics by writing
5911 to the "stats" file.
5916 Description: Client mtime is not the same on different clients after utimes
5917 Details : In some cases, the client was using the utimes() syscall on
5918 a file cached on another node. The clients now validate the
5919 ctime from the MDS + OSTs to determine which one is right.
5924 Description: Inability to activate failout mode
5925 Details : lconf script incorrectly assumed that in python string's numeric
5926 value is used in comparisons.
5929 Frequency : always with multiple stripes per file
5931 Description: Inefficient object allocation for mutli-stripe files
5932 Details : When selecting which OSTs to stripe files over, for files with
5933 a stripe count that divides evenly into the number of OSTs,
5934 the MDS is always picking the same starting OST for each file.
5935 Return the OST selection heuristic to the original design.
5940 Description: mount failures may take full timeout to return an error
5941 Details : Under some heavy load conditions it is possible that a
5942 failed mount can wait for the full obd_timeout interval,
5943 possibly several minutes, before reporting an error.
5944 Instead return an error as soon as the status is known.
5946 ------------------------------------------------------------------------------
5948 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
5950 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
5951 INTEROPERATE with older versions automatically. Please read the
5952 user documentation before upgrading any part of a live system.
5953 * WARNING: Lustre networking configuration changes are required with
5954 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
5957 * Support for kernels:
5958 2.6.9-22.0.2.EL (RHEL 4)
5959 2.6.5-7.244 (SLES 9)
5960 2.6.12.6 vanilla (kernel.org)
5963 Severity : enhancement
5964 Bugzilla : 7981/8208
5965 Description: Introduced Lustre Networking (LNET)
5966 Details : LNET is new networking infrastructure for Lustre, it includes
5967 a reorganized network configuration mode (see the user
5968 documentation for full details) as well as support for routing
5969 between different network fabrics. Lustre Networking Devices
5970 (LNDs) for the supported network fabrics have also been
5971 created for this new infrastructure.
5973 Severity : enhancement
5974 Description: Introduced Access control lists
5975 Details : clients can set ACLs on files and directories in order to have
5976 more fine-grained permissions than the standard Unix UGO+RWX.
5977 The MDS must be started with the "-o acl" mount option.
5979 Severity : enhancement
5980 Description: Introduced filesystem quotas
5981 Details : Administrators may now establish per-user quotas on the
5984 Severity : enhancement
5986 Description: Configuration change for the XT3
5987 The PTLLND is now used to run Lustre over Portals on the XT3
5988 The configure option(s) --with-cray-portals are no longer used.
5989 Rather --with-portals=<path-to-portals-includes> is used to
5990 enable building on the XT3. In addition to enable XT3 specific
5991 features the option --enable-cray-xt3 must be used.
5996 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
5997 Details : A race between allocating a new llcd and re-getting the llcd_lock
5998 allowed another thread to grab newly-allocated llcd.
6000 Severity : enhancement
6002 Description: 2.6 OST async journal commit and locking fix to improve performance
6003 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
6004 2.6 kernels are now async as they already were in 2.4 kernels so
6005 that they can commit concurrently with the network bulk transfer.
6006 For block-allocated files the filter allocation semaphore is held
6007 to avoid filesystem fragmentation during allocation. BKL lock
6008 removed for 2.6 xattr operations where it is no longer needed.
6013 Description: lconf incorrectly determined whether two IP networks could talk
6014 Details : In some more complicated routing and multiple-network
6015 configurations, lconf will avoid trying to make a network
6016 connection to a disjoint part of the IP space. It was doing the
6017 math incorrectly for one set of cases.
6022 Description: Fix for potential infinite loop processing records in an llog.
6023 Details : If an llog record is corrupted/zeroed, it is possible to loop
6024 forever in llog_process(). Validate the llog record length
6025 and skip the remainder of the block on error.
6028 Frequency : occasional (liblustre only)
6030 Description: liblustre could not open files whose last component is a symlink
6031 Details : sysio_path_walk() would incorrectly pass the open intent to
6032 intermediate path components.
6035 Frequency : rare (liblustre only with non-standard tuning)
6036 Bugzilla : 7201 (7350)
6037 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
6038 Details : llu_lookup_finish_locks() tries to set lock data on a lock
6039 after it has been released, only do this for referenced locks
6041 Severity : enhancement
6043 Description: specifying an (invalid) directory default stripe_size of -1
6044 would reset the directory default striping
6045 Details : stripe_size -1 was used internally to signal directory stripe
6046 removal, now use "all default" to signal dir stripe removal
6047 as a directory striping of "all default" is not useful
6050 Frequency : common for large clusters running liblustre clients
6052 Description: doing an ls when liblustre clients are running is slow
6053 Details : sending a glimpse AST to a liblustre client waits for every AST
6054 to time out, as liblustre clients will not respond. Since they
6055 cannot cache data we refresh the OST lock LVB from disk instead.
6057 Severity : enhancement
6059 Description: doing an ls at the same time as file IO can be slow
6060 Details : enqueue and other "small" requests can be blocked behind many
6061 large IO requests. Create a new OST IO portal for non-IO
6062 requests so they can be processed faster.
6065 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
6067 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
6068 created file with mode 000
6069 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
6070 allow an ftruncate() as a non-root user to a file with mode 000.
6071 Lustre now respects this flag to disable mode checks when
6072 truncating a file owned by the user
6075 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
6078 Description: Revoking locks from clients that went dead or catatonic might take
6080 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
6081 cancellation of such locks instant on servers without waiting for
6082 any reply from clients. Clients drops these locks when cancel
6083 notification from server is received without replying.
6086 Frequency : liblustre-only, when liblustre client dies or becomes busy
6088 Description: Doing ls on Linux clients can take a long time with active
6090 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
6091 granting such locks to it in the first place if possible. Locks
6092 are taken by proxy on the OST during the read or write and
6093 dropped immediately afterward. Add connect flags handling, do
6094 not grant locks to liblustre clients for glimpse ASTs.
6096 Severity : enhancement
6098 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
6099 Details : Existing read-ahead algorithm is tuned for the case of streamlined
6100 sequential reads and behaves badly with applications doing random
6101 reads. Improve it by reading ahead at least read region, and
6102 avoiding excessive large RPC for small reads.
6104 Severity : enhancement
6106 Description: Creating more than 1000 files for a single job may cause a load
6107 imbalance on the OSTs if there are also a large number of OSTs.
6108 Details : qos_prep_create() uses an OST index reseed value that is an
6109 even multiple of the number of available OSTs so that if the
6110 reseed happens in the middle of the object allocation it will
6111 still utilize the OSTs as uniformly as possible.
6116 Description: OST or MDS may oops in ping_evictor_main()
6117 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
6118 but doesn't restart at beginning of obd_exports_timed list
6121 Severity : enhancement
6123 Description: improve by-nid export eviction on the MDS and OST
6124 Details : allow multiple exports with the same NID to be evicted at one
6125 time without re-searching the exports list.
6128 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
6130 Description: MDS may oops in groups_free()
6131 Details : in rare race conditions a newly allocated group_info struct is
6132 freed again, and this can be NULL. The 2.4 compatibility code
6133 for groups_free() checked for a NULL pointer, but 2.6 did not.
6136 Frequency : common for liblustre clients doing little filesystem IO
6137 Bugzilla : 9352, 7313
6138 Description: server may evict liblustre clients accessing contended locks
6139 Details : if a client is granted a lock or receives a completion AST
6140 with a blocking AST already set it would not reply to the AST
6141 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
6144 Frequency : lfs setstripe, only systems with more than 160 OSTs
6146 Description: unable to set striping with a starting offset beyond OST 160
6147 Details : llapi_create_file() incorrectly limited the starting stripe
6148 index to the maximum single-file stripe count.
6151 Frequency : LDAP users only
6153 Description: lconf did not handle in-kernel recovery with LDAP properly
6154 Details : lconf/LustreDB get_refs() is searching the wrong namespace
6156 Severity : enhancement
6158 Description: bind OST threads to NUMA nodes to improve performance
6159 Details : all OST threads are uniformly bound to CPUs on a single NUMA
6160 node and do their allocations there to localize memory access
6162 Severity : enhancement
6164 Description: llmount can determine client NID directly from Myrinet (GM)
6165 Details : the client NID code from gmnalnid was moved directly into
6166 llmount, removing the need to use this or specifying the
6167 client NID explicitly when mounting GM clients with zeroconf
6170 Frequency : if client is started with down MDS
6172 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
6173 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
6174 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
6175 ptlrpc_import_delayed_req() because we didn't check if the
6176 request was interrupted, and we also didn't break out of the
6177 event loop if there was no timeout
6182 Description: data loss during non-page-aligned writes to a single file from
6183 both multiple nodes and multiple threads on one node at same time
6184 Details : updates to KMS and lsm weren't protected by common lock. Resulting
6185 inconsistency led to false short-reads, that were cached and later
6186 used by ->prepare_write() to fill in partially written page,
6187 leading to data loss.
6190 Frequency : always, if lconf --abort_recovery used
6192 Description: lconf --abort_recovery fails with 'Operation not supported'
6193 Details : lconf was attempting to abort recovery on the MDT device and not
6196 Severity : enhancement
6198 Description: remove cleanup logs
6199 Details : replace lconf-generated cleanup logs with lustre internal
6200 cleanup routines. Eliminates the need for client-cleanup and
6203 Severity : enhancement
6205 Description: add support for EAs (user and system) on lustre filesystems
6206 Details : it is now possible to store extended attributes in the Lustre
6207 client filesystem, and with the user_xattr mount option it
6208 is possible to allow users to store EAs on their files also
6210 Severity : enhancement
6212 Description: Add possibility (config option) to show minimal available OST free
6214 Details : When compiled with --enable-mindf configure option, statfs(2)
6215 (and so, df) will return least minimal free space available from
6216 all OSTs as amount of free space on FS, instead of summary of
6217 free spaces of all OSTs.
6219 Severity : enhancement
6221 Description: do not expand extent locks acquired on OST-side
6222 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
6223 by server: they are not cached anyway.
6226 Frequency : when mmap is used/binaries executed from Lustre
6228 Description: Unmmap pages before throwing them away from read cache.
6229 Details : llap_shrink cache now attempts to unmap pages before discarding
6230 them (if unmapping failed - do not discard). SLES9 kernel has
6231 extra checks that trigger if this unmapping is not done first.
6236 Description: lconf didn't resolve symlinks before checking to see whether a
6237 given mountpoint was already in use
6240 Frequency : when migrating failover services
6241 Bugzilla : 6395, 9514
6242 Description: When migrating a subset of services from a node (e.g. failback
6243 from a failover service node) the remaining services would
6244 time out and evict clients.
6245 Details : lconf --force (implied by --failover) sets the global obd_timeout
6246 to 5 seconds in order to quickly disconnect, but this caused
6247 other RPCs to time out too quickly. Do not change the global
6248 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
6250 Severity : enhancement
6251 Frequency : if MDS is started with down OST
6252 Bugzilla : 9439,5706
6253 Description: Allow startup/shutdown of an MDS without depending on the
6254 availability of the OSTs.
6255 Details : Asynchronously call mds_lov_synchronize during MDS startup.
6256 Add appropriate locking and lov-osc refcounts for safe
6257 cleaning. Add osc abort_inflight calls in case the
6258 synchronize never started.
6261 Frequency : occasional (Cray XT3 only)
6263 Description: root not authorized to access files in CRAY_PORTALS environment
6264 Details : The client process capabilities were not honoured on the MDS in
6265 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
6266 been accessed by an authorized user then root was able to access
6267 the file on the local client also. The root user capabilities
6268 are now allowed on the MDS, as this environment has secure UID.
6271 Frequency : occasional
6273 Description: ldiskfs "too long searching" message happens too often
6274 Details : A debugging message (otherwise harmless) prints too often on
6275 the OST console. This has been reduced to only happen when
6276 there are fragmentation problems on the filesystem.
6281 Description: Division by zero in statfs when all OSCs are inactive
6282 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
6283 lov_statfs divides by value returned by lov_get_stripecnt().
6287 Bugzilla : 9489, 3273
6288 Description: First write from each client to each OST was only 4kB in size,
6289 to initialize client writeback cache, which caused sub-optimal
6290 RPCs and poor layout on disk for the first writen file.
6291 Details : Clients now request an initial cache grant at (re)connect time
6292 and so that they can start streaming writes to the cache right
6293 away and always do full-sized RPCs if there is enough data.
6294 If the OST is rebooted the client also re-establishes its grant
6295 so that client cached writes will be honoured under the grant.
6300 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
6301 Details : Now I/O RPCs go to different portal number and (presumably) fast
6302 lock requests (and glimses) and other RPCs get their own service
6303 threads pool that should be able to service those RPCs
6306 Severity : enhancement
6308 Description: Ability to exchange lustre version between client and servers and
6309 issue warnings at client side if client is too old. Also for
6310 liblustre clients there is ability to refuse connection of too old
6312 Details : New 'version' field is added to connect data structure that is
6313 filled with version info. That info is later checked by server and
6317 Frequency : rare, liblustre only.
6318 Bugzilla : 9296, 9581
6319 Description: Two simultaneous writes from liblustre at offset within same page
6320 might proceed at the same time overwriting eachother with stale
6322 Details : I/O lock withing llu_file_prwv was released too early, before data
6323 actually was hitting the wire. Extended lock-holding time until
6324 server acknowledges receiving data.
6327 Frequency : extremely rare. Never observed in practice.
6329 Description: avoid generating lustre_handle cookie of 0.
6330 Details : class_handle_hash() generates handle cookies by incrementing
6331 global counter, and can hit 0 occasionaly (this is unlikely, but
6332 not impossible, because initial value of cookie counter is
6333 selected randonly). Value of 0 is used as a sentinel meaning
6334 "unassigned handle" --- avoid it. Also coalesce two critical
6335 sections in this function into one.
6337 Severity : enhancement
6339 Description: allow liblustre clients to delegate truncate locking to OST
6340 Details : To avoid overhead of locking, liblustre client instructs OST to
6341 take extent lock in ost_punch() on client's behalf. New connection
6342 flag is added to handle backward compatibility.
6344 Severity : enhancement
6345 Bugzilla : 4928, 7341, 9758
6346 Description: allow number of OST service threads to be specified
6347 Details : a module parameter allows the number of OST service threads
6348 to be specified via "options ost ost_num_threads={N}" in the
6349 OSS's /etc/modules.conf or /etc/modprobe.conf.
6353 Bugzilla : 6146, 9635, 9895
6354 Description: servers crash with bad pointer in target_handle_connect()
6355 Details : In rare cases when a client is reconnecting it was possible that
6356 the connection request was the last reference for that export.
6357 We would temporarily drop the export reference and get a new
6358 one, but this may have been the last reference and the export
6359 was just destroyed. Get new reference before dropping old one.
6361 Severity : enhancement
6362 Frequency : if client is started with failover MDS
6364 Description: Allow multiple MDS hostnames in the mount command
6365 Details : Try to read the configuration from all specified MDS
6366 hostnames during a client mount in case the "primary"
6369 Severity : enhancement
6371 Description: Stop sending data to evicted clients as soon as possible.
6372 Details : Check if the client we are about to send or are sending data to
6373 was evicted already. (Check is done every second of waiting,
6374 for which l_wait_event interface was extended to allow checking
6375 of exit condition at specified intervals).
6378 Frequency : rare, normally only when NFS exporting is done from client
6380 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
6382 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
6385 Frequency : always, if srand() is called before liblustre initialization
6387 Description: Liblustre uses system PRNG disturbing its usage by user application
6388 Details : Introduce internal to lustre fast and high-quality PRNG for
6389 lustre usage and make liblustre and some other places in generic
6390 lustre code to use it.
6392 Severity : enhancement
6393 Bugzilla : 9477, 9557, 9870
6394 Description: Verify that the MDS configuration logs are updated when xml is
6395 Details : Check if the .xml configuration logs are newer than the config
6396 logs stored on the MDS and report an error if this is the case.
6397 Request --write-conf, or allow starting with --old_conf.
6399 Severity : enhancement
6401 Description: Handle symlinks in the path when checking if Lustre is mounted.
6402 Details : Resolve intermediate symlinks when checking if a client has
6403 mounted a filesystem to avoid duplicate client mounts.
6408 Description: lconf can hit an error exception but still return success.
6409 Details : The lconf command catches the Command error exception at the top
6410 level script context and will exit with the associated exit
6411 status, but doesn't ensure that this exit status is non-zero.
6416 Description: failure of ptlrpc thread startup can cause oops
6417 Details : Starting a ptlrpc service thread can fail if there are a large
6418 number of threads or the server memory is very fragmented.
6419 Handle this without oopsing.
6422 Frequency : always, only if liblustre and non-default acceptor port was used
6424 Description: liblustre cannot connect to servers with non-default acceptor port
6425 Details : tcpnal_set_default_params() was not called and was therefore
6426 ignoring the environment varaible TCPNAL_PORT, as well as other
6427 TCPNAL_ environment variables
6432 Description: two objects could be created on the same OST for a single file
6433 Details : If an OST is down, in some cases it was possible to create two
6434 objects on a single OST for a single file. No problems other
6435 than potential performance impact and spurious error messages.
6439 Bugzilla : 5681, 9562
6440 Description: Client may oops in ll_unhash_aliases
6441 Details : Client dcache may become inconsistent in race condition.
6442 In some cases "getcwd" can fail if the current directory is
6448 Description: Inode refcounting problems in NFS export code
6449 Details : link_raw functions used to call d_instantiate without obtaining
6450 extra inode reference first.
6454 Bugzilla : 9942, 9903
6455 Description: Referencing freed requests leading to crash, memleaks with NFS.
6456 Details : We used to require that call to ll_revalidate_it was always
6457 followed by ll_lookup_it. Also with revalidate_special() it is
6458 possible to call ll_revalidate_it() twice for the same dentry
6459 even if first occurence returned success. This fix changes semantic
6460 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
6461 reference on a request referred from the intent.
6462 ll_intent_release() then releases such a request.
6465 Frequency : rare, normally benchmark loads only
6467 Description: unlinked inodes were kept in memory on the client
6468 Details : If a client is repeatedly creating and unlinking files it
6469 can accumulate a lot of stale inodes in the inode slab cache.
6470 If there is no other client load running this can cause the
6471 client node to run out of memory. Instead flush old inodes
6472 from client cache that have the same inode number as a new inode.
6475 Frequency : SLES9 2.6.5 kernel and long filenames only
6476 Bugzilla : 9969, 10379
6477 Description: utime reports stale NFS file handle
6478 Details : SLES9 uses out-of-dentry names in some cases, which confused
6479 the lustre dentry revalidation. Change it to always use the
6483 Frequency : rare, unless heavy write-truncate concurrency is continuous
6484 Bugzilla : 4180, 6984, 7171, 9963, 9331
6485 Description: OST becomes very slow and/or deadlocked during object unlink
6486 Details : filter_destroy() was holding onto the parent directory lock
6487 while truncating+unlinking objects. For very large objects this
6488 may block other threads for a long time and slow overall OST
6489 responsiveness. It may also be possible to get a lock ordering
6490 deadlock in this case, or run out of journal credits because of
6491 the combined truncate+unlink. Solution is to do object truncate
6492 first in one transaction without parent lock, and then do the
6493 final unlink in a new transaction with the parent lock. This
6494 reduces the lock hold time dramatically.
6497 Frequency : rare, 2.4 kernels only
6499 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
6500 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
6501 before unmounting their filesystem in order to avoid deadlock.
6502 The kernel_locked() function in 2.4 kernels only checks whether
6503 the kernel lock is held, not whether it is this process that is
6504 holding it as 2.6 kernels do.
6509 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
6510 Details : The client ptlrpc code may be trying to reconnect to a down
6511 server before a previous connection attempt has timed out.
6512 Increase the reconnect interval to be longer than the connection
6513 timeout interval to avoid sending duplicate connections to
6517 Frequency : echo_client brw_test command
6519 Description: fix echo_client to work with OST preallocated code
6520 Details : OST preallocation code (5137) didn't take echo_client IO path
6521 into account: echo_client calls filter methods outside of any
6522 OST thread and, hence, there is no per-thread preallocated
6523 pages and buffers to use. Solution: hijack pga pages for IO. As
6524 a byproduct, this avoids unnecessary data copying.
6528 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
6529 Description: Client can oops in mdc_commit_close() after open replay
6530 Details : It was possible for the MDS to return an open request with no
6531 transaction number in mds_finish_transno() if the client was
6532 evicted, but without actually returning an error. Clients
6533 would later try to replay that open and may trip an assertion
6534 Simplify the client close codepath, and always return an error
6535 from the MDS in case the open is not successful.
6538 Frequency : rare, 2.6 OSTs only
6540 Description: OST may deadlock under high load on fragmented files
6541 Details : If there was a heavy load and highly-fragmented OST filesystems
6542 it was possible to have all the OST threads deadlock waiting on
6543 allocation of biovecs, because the biovecs were not released
6544 until the entire RPC IO was completed. Instead, release biovecs
6545 as soon as they are complete to ensure forward IO progress.
6547 Severity : enhancement
6549 Description: Support for specifying external journal device at mount
6550 Details : If an OST or MDS device is formatted with an external journal
6551 device, this device major/minor is stored in the ext3 superblock
6552 and may not be valid for failover. Allow detecting and
6553 specifying the external journal at mount time.
6558 Description: Mounting an MDS with pending unlinked files may cause oops
6559 Details : target_finish_recovery() calls mds_postrecov() which returned
6560 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
6561 considers this an error and immediately begins cleaning up the
6562 lov, just after starting the mds_lov process
6564 Severity : enhancement
6566 Description: Implement 'lfs df' to report actual free space on per-OST basis
6567 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
6568 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
6569 usage of objects; '-h' to report in human readable format.
6571 ------------------------------------------------------------------------------
6573 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
6580 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
6581 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
6582 at startup during orphan unlinking. Instead, skip buddy update
6583 before setup, as it will be regenerated after recovery is complete.
6588 Description: If an OST is inactive, its locks might reference stale inodes.
6589 Details : lov_change_cbdata() must iterate over all namespaces, even if
6590 they are inactive to clear inode references from the lock.
6592 Severity : enhancement
6593 Frequency : occasional, if non-standard max_dirty_mb used
6595 Description: Client will block write RPCs if not enough grant
6596 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
6597 then the client will block writes while waiting for another RPC
6598 to complete instead of consuming its dirty limit. With change
6599 we get improved performance when max_dirty_mb is small.
6601 Severity : enhancement
6602 Bugzilla : 3389, 6253
6603 Description: Add support for supplementary groups on the MDS.
6604 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
6605 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
6606 lookups for user supplementary groups into a cache.
6610 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
6611 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
6612 for liblustre O_EXCL is enforced by MDS.
6616 Description: readdir never returns NULL in liblustre.
6617 Details : Corrected llu_iop_getdirentries logic, to return offset of next
6618 dentry in struct dirent.
6622 Frequency : liblustre only, depends on application IO pattern
6623 Description: liblustre clients evicted if not contacting servers
6624 Details : Don't put liblustre clients into the ping_evictor list, so
6625 they will not be evicted by the pinger ever.
6627 Severity : enhancement
6629 Description: Add ability to evict clients by NID from MDS.
6630 Details : By echoing "nid:$NID" string into
6631 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
6632 $NID would be instantly evicted from this MDS and from all active
6633 OSTs connected to it.
6637 Description: Do not query file size twice, somewhat slowing stat(2) calls.
6638 Details : lookup_it_finish() used to query file size from OSTs that was not
6643 Description: service threads change working directory to that of init
6644 Details : Starting lustre service threads may pin the working directory
6645 of the parent thread, making that filesystem busy. Threads
6646 now change to the working directory of init to avoid this.
6650 Frequency : during shutdown only
6651 Description: shutdown with a failed MDS or OST can cause unmount to hang
6652 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
6657 Frequency : 2.6 only
6658 Description: chmod/chown may include an extra supplementary group
6659 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
6660 supplementary group and if none is specified this is used.
6663 Bugzilla : 5479 (6816)
6665 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
6666 Details : If lookup is in progress on a file that is unlinked we might try
6667 to revalidate the inode and fail in revalidate after lookup is
6668 complete and ll_file_open() enqueues the open again but
6669 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
6672 Frequency : always, if lconf --abort_recovery used
6674 Description: lconf --abort_recovery fails with 'Operation not supported'
6675 Details : lconf was attempting to abort recovery on the MDT device and not
6678 ------------------------------------------------------------------------------
6680 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
6685 Frequency : rare (only unsupported configurations with a node running as an
6687 Bugzilla : 6514, 5137
6688 Description: Mounting a Lustre file system on a node running as an OST could
6690 Details : OSTs now preallocates memory needed to write out data at
6691 startup, instead of when needed, to avoid having to
6692 allocate memory in possibly low memory situations.
6693 Specifically, if the file system is mounted on on OST,
6694 memory pressure could force it to try to write out data,
6695 which it needed to allocate memory to do. Due to the low
6696 memory, it would be unable to do so and the node would
6697 become unresponsive.
6699 Severity : enhancement
6701 Description: Addition of lconf --service command line option
6702 Details : lconf now accepts a '--service <arg>' option, which is
6703 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
6705 Severity : enhancement
6707 Description: Failover mode is now the default for OSTs.
6708 Details : By default, OSTs will now run in failover mode. To return to
6709 the old behaviour, add '--failout' to the lmc line for OSTs.
6711 Severity : enhancement
6713 Description: Health checks are now provided for MDS and OSTs
6714 Details : Additional detailed health check information on MSD and OSTs
6715 is now provided through the procfs health_check value.
6718 Frequency : occasional, depends on IO load
6720 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
6721 after numerous create/delete cycles
6722 Details : The ext3 inode allocation policy would not allocate new inodes
6723 very well on the OSTs because there are no new directories
6724 being created. Instead we look for groups with free space if
6725 the parent directories are nearly full.
6730 Description: Network or server problems during mount may cause partially
6731 mounted clients instead of returning an error.
6732 Details : The config llog parsing code may overwrite the error return
6733 code during mount error handling, returning success instead
6739 Description: MDS can fail to allocate large reply buffers
6740 Details : After long uptimes the MDS can fail to allocate large reply
6741 buffers (e.g. zconf client mount config records) due to memory
6742 fragmentation or consumption by the buffer cache. Preallocate
6743 some large reply buffers so that these replies can be sent even
6744 under memory pressure.
6748 Frequency : rare (liblustre)
6749 Description: fsx running with liblustre complained that using truncate() to
6750 extend the file doesn't work. This patch corrects that issue.
6751 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
6752 ATTR_SIZE and lsm use in llu_setattr_raw.
6756 Frequency : rare, only 2.6 kernels
6757 Description: Unusual file access patterns on the MDS may result in inode
6758 data being lost in very rare circumstances.
6759 Details : Bad interaction between the ea-in-inode patch and the "no-read"
6760 code in the 2.6 kernel caused the inode and/or EA data not to
6761 be read from disk, causing single-file corruption.
6765 Frequency : rare, only 2.6 filesystems using extents
6766 Description: Heavy concurrent write and delete load may cause data corruption.
6767 Details : It was possible under high-load situations to have an extent
6768 metadata block in the block device cache from a just-unlinked
6769 file overwrite a newly-allocated data block. We now unmap any
6770 metadata buffers that alias just-allocated data blocks.
6774 Frequency : filesystems with default stripe_count larger than 77
6775 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
6776 Details : lconf specifies an inode size of 4096 bytes when the default
6777 stripe_count is larger than 77. This conflicts with the default
6778 inode density of 1 per 4096 bytes. Allocate smaller inodes in
6779 this case to avoid pinning too much memory for large EAs.
6781 ------------------------------------------------------------------------------
6783 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
6788 Frequency : rare (extremely heavy IO load with hundreds of clients)
6790 Description: Client is evicted, gets IO error writing to file
6791 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
6792 caused clients to be evicted for AST timeouts. The fixes in
6793 bug 5192 mean we no longer need to have such short AST timeouts
6794 so ldlm_timeout has been increased.
6797 Frequency : occasional during --force or --failover shutdown under load
6798 Bugzilla : 5949, 4834
6799 Description: Server oops/LBUG if stopped with --force or --failover under load
6800 Details : a collection of import/export refcount and cleanup ordering
6801 issues fixed for safer force cleanup
6804 Frequency : only filesystems larger than 120 OSTs
6805 Bugzilla : 5990, 6223
6806 Description: lfs getstripe would oops on a very large filesystem
6807 Details : lov_getconfig used kfree on vmalloc'd memory
6810 Frequency : only filesystems exporting via NFS to Solaris 10 clients
6811 Bugzilla : 6242, 6243
6812 Description: reading from files that had been truncated to a non-zero size
6813 but never opened returned no data
6814 Details : ll_file_read() reads zeros from no-object files to EOF
6819 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
6820 Details : There's little harm in aborting MDS/OSS recovery and letting it
6821 try again, so I removed the LASSERT and return an error instead.
6823 Severity : enhancement
6825 Description: New debugging infrastructure for tracking down data corruption
6826 Details : The I/O checksum code was replaced to: (a) control it at runtime,
6827 (b) cover more of the client-side code path, and (c) try to narrow
6828 down where problems occurred
6832 Bugzilla : 3819, 4364, 4397, 6313
6833 Description: Racing close and eviction MDS could cause assertion in mds_close
6834 Details : It was possible to get multiple mfd references during close and
6835 client eviction, leading to one thread referencing a freed mfd.
6837 Severity: : enhancement
6838 Bugzilla : 3262, 6359
6839 Description: Attempts to reconnect to servers are now more aggressive.
6840 Details : This builds on the enhanced upcall-less recovery that was added
6841 in 1.4.2. When trying to reconnect to servers, clients will
6842 now try each server in the failover group every 10 seconds. By
6843 default, clients would previously try one server every 25 seconds.
6848 Description: After recovery, certain operations trigger a failed
6849 assertion on a client.
6850 Details : Failing over an mds, using lconf -d --failover, while a
6851 client was doing a readdir() call would cause the client to
6852 LBUG after recovery completed and the readdir() was resent.
6854 Severity : enhancement
6856 Description: Default groups are now added by lconf
6857 Details : You can now run lconf --group <servicename> without having to
6858 manually add groups with lmc.
6861 Frequency : occasional
6863 Description: Nodes with an elan id of 0 trigger a failed assertion
6866 Frequency : always when accessing e.g. tty/console device nodes
6868 Description: tty and some other devices nodes cannot be used on lustre
6869 Details : file's private_data field is used by device data and lustre
6870 values in there got lost. New field was added to struct file to
6871 store fs-specific private data.
6874 Frequency : when exporting Lustre via NFS
6876 Description: NFSD failed occasionally when looking up a path component
6877 Details : NFSD is looking up ".." which was broken in ext3 directories
6878 that had grown large enough to become hashed.
6881 Frequency : Clusters with multiple interfaces not on the same subnet
6883 Description: Nodes will repeatedly try to reconnect to an interface which it
6884 cannot reach and report an error to the log.
6885 Details : Extra peer list entries will be created by lconf with some peers
6886 unreachable. lconf now validates the peer before adding it.
6889 Frequency : Only if a default stripe is set on the filesystem root.
6891 Description: Setting a default stripe on the filesystem root prevented the
6892 filesystem from being remounted.
6893 Details : The client was sending extra request flags in the root getattr
6894 request and did not allocate a reply buffer for the dir EA.
6897 Frequency : occasional, higher if lots of files are accessed by one client
6898 Bugzilla : 6159, 6097
6899 Description: Client trips assertion regarding lsm mismatch/magic
6900 Details : While revalidating inodes the VFS looks up inodes with ifind()
6901 and in rare cases can find an inode that is being freed.
6902 The ll_test_inode() code will free the lsm during ifind()
6903 when it finds an existing inode and then the VFS later attaches
6904 this free lsm to a new inode.
6908 Bugzilla : 6422, 7030
6909 Description: MDS deadlock between mkdir and client eviction
6910 Details : Creating a new file via mkdir or mknod (starting a transaction
6911 and getting the ns lock) can deadlock with client eviction
6912 (gets ns lock and trying to finish a synchronous transaction).
6915 Frequency : occasional
6916 Description: While starting a server, the fsfilt_ext3 module could not be
6918 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
6919 kernels. Previously, lconf would still use the ext3 name
6920 when trying to load modules. Now, it will correctly use
6921 ext3 on 2.4 and ldiskfs on 2.6.
6923 Severity : enhancement
6924 Description: The default stripe count has been changed to 1
6925 Details : The interpretation of the default stripe count (0, to lfs
6926 or lmc) has been changed to mean striping across a single
6927 OST, rather than all available. For general usage we have
6928 found a stripe count of 1 or 2 works best.
6930 Severity : enhancement
6931 Description: Add support for compiling against Cray portals.
6932 Details : Conditional compiling for some areas that are different
6936 Frequency : occasional
6937 Bugzilla : 6409, 6834
6938 Description: Creating files with an explicit stripe count may lead to
6939 a failed assertion on the MDS
6940 Details : If some OSTs are full or unavailable, creating files may
6941 trigger a failed assertion on the MDS. Now, Lustre will
6942 try to use other servers or return an error to the
6946 Frequency : occasional
6948 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
6949 caused lock timeout during readahead (since 1.4.2).
6950 Details : Processes doing readahead might match a lock that hasn't been
6951 granted yet if there are overlapping and conflicting lock
6952 requests. The readahead process waits on ungranted lock
6953 (original lock is CBPENDING), while OST waits for that process
6954 to cancel CBPENDING read lock and eventually evicts client.
6956 Severity : enhancement
6958 Description: Initial enabling of flock support for clients
6959 Details : Implements fcntl advisory locking and file status functions.
6960 This feature is provided as an optional mount flag (default
6961 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
6962 locking are implemented yet, and those that are are not guaranteed
6963 to be completely correct in production environments.
6964 mount -t lustre -o [flock|noflock] ...
6967 Frequency : occasional
6969 Description: OSTs running 2.4 kernels but with extents enabled might trip an
6970 assertion in the ext3 JBD (journaling) layer.
6971 Details : The b_committed_data struct is protected by the big kernel lock
6972 in 2.4 kernels, serializing journal_commit_transaction() and
6973 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
6974 access to this struct. In 2.6 kernels there is finer grained
6975 locking to improve SMP performance of the JBD layer.
6979 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
6981 -----------------------------------------------------------------------------
6983 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
6985 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
6986 versions of Lustre. Please update all servers and clients to
6987 version 1.4.2 or later at the same time. You must also run
6988 "lconf --write-conf {config}.xml" on the MDS while it is stopped
6989 to update the configuration logs.
6991 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
6992 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
6993 - swab llog records of type '0' so we get proper header size/idx (5861)
6994 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
6995 - fix rename of one directory over another leaking an inode (5953)
6996 - avoid SetPageDirty on 2.6 (5981)
6997 - don't re-add just-being-destroyed locks to the waiting list (5653)
6998 - when creating new directories, inherit the parent's custom
6999 striping settings if present parent (3048)
7000 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
7001 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
7002 - don't hold client locks on temporary worklist from l_lru (5666)
7003 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
7004 - automatically evict dead clients (5921)
7005 - Update file size properly in create+truncate+fstat case (6196)
7006 - Do not unhash mountpoint dentries, do not allow removal of
7008 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
7009 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
7010 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
7011 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
7012 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
7013 - Correctly handle reads of files with no objects (6243)
7014 - lctl recover will also mark a device active if deactivate used (5933)
7016 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
7017 - allow --write-conf on an MDS with different nettype than client (5619)
7018 - don't write config llogs to MDS for mounts not from that MDS (5617)
7019 - lconf should create multiple TCP connections from a client (5201)
7020 - init scripts are now turned off by default; run chkconfig --on
7021 lustre and chkconfig --on lustrefs to use them
7022 - upcalls are no longer needed for clients to recover to failover
7024 - add --abort-recovery option to lconf to abort recovery on device
7026 - add support for an arbitrary number of OSTs (3026)
7027 - Quota support protocol changes.
7028 - forward compatibility changes to wire structs (6007)
7029 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
7030 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
7031 - improved "lustre status" script
7032 - initialize blocksize for non-regular files (6062)
7033 - added --disable-server and --disable-client configure options (5782)
7034 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
7035 - Vanilla 2.4.29 support
7036 - increase maximum number of obd devices to 520 (6242)
7037 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
7038 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
7039 SLES 9 2.6.5 kernels
7040 - limit stripes per file to 160 (the maximum EA size) (6093)
7042 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
7045 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
7046 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
7047 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
7048 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
7049 - small ext3 extents cleanups and fixes (5733)
7050 - improved mballoc code, several small races and bugs fixed (5733, 5638)
7051 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
7052 - avoid needless client->OST connect, fix handle mismatch (5317)
7053 - fix DLM error path that led to out-of-sync client, long delays (5779)
7054 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
7055 - fix several locking issues related to i_size (5492,5624,5654,5672)
7056 - don't move pending lock onto export if it is already evicted (5683)
7057 - fix kernel oops when creating .foo in unlinked directory (5548)
7058 - fix deadlock in obdfilter statistics vs. object create (5811)
7059 - use time_{before,after} to avoid timer jiffies wrap (5882)
7060 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
7061 - Do not leak request if server was not able to process it (5154)
7062 - If mds_open unable to find parent dir, make that negative lookup(5154)
7063 - don't create new directories with extent-mapping (5909, 5936)
7065 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
7066 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
7067 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
7068 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
7069 - support for RHEL4 kernel on i686 (5773)
7070 - provide error messages when incompatible logs are encountered (5898)
7072 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
7073 * version 1.4.0.10 (1.4.1 release candidate 1)
7075 - don't keep a lock reference when lock is not granted (4238)
7076 - unsafe list practices (rarely) led to infinite eviction loop (4908)
7077 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
7078 - drop import inflight refcount on signal_completed_replay error (5255)
7079 - unlock page after async write error during send (3677)
7080 - handle missing objects in filter_preprw_read properly (5265)
7081 - no transno return for symlink open, don't save no-trasno open (3440)
7082 - don't try to complete elan receive that already failed (4012)
7083 - free RPC server reply state on error (5406)
7084 - clean up thread from ptlrpc_start_thread() on error (5160)
7085 - readahead could read extra page into cache that wasn't ejected (5388)
7086 - prevent races in class_attach/setup/cleanup/detach (5260)
7087 - don't dereference de->d_inode after l_dput of de (5458)
7088 - use "int" for stripe value returned from lock_to_stripe (5544)
7089 - mballoc allocation and error-checking fixes in 2.6 (5504)
7090 - block device patches to fix I/O request sizes in 2.6 (5482)
7091 - look up hostnames for IB nals (5602)
7092 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
7093 - don't start multiple acceptors for the same port (5277)
7094 - fix incorrect LASSERT in mds_getattr_name (5635)
7095 - export a proc file for general "ping" checking (5628)
7096 - fix "lfs check" to not block when the MDS is down (5628)
7098 - service request history (4965)
7099 - put {ll,lov,osc}_async_page structs in a single slab (4699)
7100 - create an "evict_client" /proc entry on OSTs, like the MDS has
7101 - fix mount usage message, return errors per mount(8) (5168)
7102 - change grep [] to grep "[]" in tests so they work in more UMLs
7103 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
7104 - remove ancient LOV_MAGIC_V0 EA support (5047)
7105 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
7106 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
7107 - allow readahead allocations to fail when low on memory (5383)
7108 - mmap locking landed again, after considerable improvement (2828)
7109 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
7111 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
7114 - send OST transaction number in read/write reply to free req (4966)
7115 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
7116 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
7117 - the watchdog thread now runs as interruptible (5246)
7118 - flock/lockf fixes (but it's still disabled, pending 5135)
7119 - don't use EXT3 constants in llite code (5094)
7120 - memory shortage at startup could cause assertion (5176)
7122 - reorganization of lov code
7123 - single portals codebase
7125 - add extents/mballoc support (5025)
7126 - direct I/O reads in the obdfilter (4048)
7127 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
7129 tbd Cluster File Systems, Inc. <info@clusterfs.com>
7132 - send OST transaction number in read/write reply to free req (4966)
7133 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
7134 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
7135 - the watchdog thread now runs as interruptible (5246)
7136 - handle missing objects in filter_preprw_read properly (5265)
7137 - unsafe list practices (rarely) led to infinite eviction loop (4908)
7138 - drop import inflight refcount on signal_completed_replay error (5255)
7139 - unlock page after async write error during send (3677)
7140 - return original error code on reconstructed replies (3761)
7141 - no transno return for symlink open, don't save no-trasno open (3440)
7143 - add pid to ldlm debugging output (4922)
7144 - bump the watchdog timeouts -- we can't handle 30sec yet
7145 - extra debugging for orphan dentry/inode bug (5259)
7147 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
7150 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
7151 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
7152 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
7153 - fix NULL dereference in /proc/sys/portals/routes (4827)
7154 - allow failed mdc_close() operations to be interrupted (4561)
7155 - stop precreate on OST before MDS would time out on it (4778)
7156 - don't send partial-page writes before EOF from client (4410)
7157 - discard client grant for sub-page writes on large-page clients (4520)
7158 - don't free dentries not owned by NFS code, check generation (4806)
7159 - fix lsm leak if mds_create_objects() fails (4801)
7160 - limit debug_daemon file size, always print CERROR messages (4789)
7161 - use transno after validating reply (3892)
7162 - process timed out requests if import state changes (3754)
7163 - update mtime on OST during writes, return in glimpse (4829)
7164 - add mkfsoptions to LDAP (4679)
7165 - use ->max_readahead method instead of zapping global ra (5039)
7166 - don't interrupt __l_wait_event() during strace
7168 - add software watchdogs to catch hung threads quickly (4941)
7169 - make lustrefs init script start after nfs is mounted
7170 - fix CWARN/ERROR duplication (4930)
7171 - return async write errors to application if possible (2248)
7172 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
7173 - print NAL number in %x format (4645)
7174 - update barely-supported suse-2.4.21-171 series (4842)
7175 - support for sles 9 %post scripts
7176 - support for building 2.6 kernel-source packages
7177 - support for sles km_* packages
7179 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
7182 - ignore -ENOENT errors in osc_destroy (3639)
7183 - notify osc create thread that OSC is being cleaned up (4600)
7184 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
7185 - reconstruct ost_handle() like mds_handle() (4657)
7186 - create a new thread to do import eviction to avoid deadlock (3969)
7187 - let lconf resolve symlinked-to devices (4629)
7188 - don't unlink "objects" from directory with default EA (4554)
7189 - hold socknal file ref over connect in case target is down (4394)
7190 - allow more than 32000 subdirectories in a single directory (3244)
7191 - fix blocks count for O_DIRECT writes (3751)
7192 - OST returns ENOSPC from object create when no space left (4539)
7193 - don't send truncate RPC if file size isn't changing (4410)
7194 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
7195 - bind to privileged port in socknal and tcpnal (3689)
7197 - rate limit CERROR/CWARN console message to avoid overload (4519)
7198 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
7199 - basic mmap support (3918)
7200 - kernel patch series update from b1_4 (4711)
7202 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
7205 - avoid crash during MDS cleanup with OST shut down (2775)
7206 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
7207 - don't use bad inodes on the MDS (3744)
7208 - dynamic object preallocation to improve recovery speed (4236)
7209 - don't hold spinlock over lock dumping or change debug flags (4401)
7210 - don't zero obd_dev when it is force cleaned (3651)
7211 - print grants to console if they go negative (4431)
7212 - "lctl deactivate" will stop automatic recovery attempts (3406)
7213 - look for existing locks in ldlm_handle_enqueue() (3764)
7214 - don't resolve lock handle twice in recovery avoiding race (4401)
7215 - revalidate should check working dir is a directory (4134)
7217 - don't always mark "slow" obdfilter messages as errors (4418)
7219 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
7222 - don't close LustreDB during write_conf until it is done (3860)
7223 - fix typo in lconf for_each_profile (3821)
7224 - allow dumping logs from multiple threads at one time (3820)
7225 - don't allow multiple threads in OSC recovery (3812)
7226 - fix debug_size parameters (3864)
7227 - fix mds_postrecov to initialize import for llog ctxt (3121)
7228 - replace config semaphore with spinlock (3306)
7229 - be sure to send a reply for a CANCEL rpc with bad export (3863)
7230 - don't allow enqueue to complete on a destroyed export (3822)
7231 - down write_lock before checking llog header bitmap (3825)
7232 - recover from lock replay timeout (3764)
7233 - up llog sem before sending rpc (3652)
7234 - reduce ns lock hold times when setting kms (3267)
7235 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
7236 - fix NULL deref and obd_dev leak on setup error (3312)
7237 - replace some LBUG about llog ops with error handling (3841)
7238 - don't match INVALID dentries from d_lookup and spin (3784)
7239 - hold dcache_lock while marking dentries INVALID and hashing (4255)
7240 - fix invalid assertion in ptlrpc_set_wait (3880)
7242 - add libwrap support for the TCP acceptor (3996)
7243 - add /proc/sys/portals/routes for non-root route listing (3994)
7244 - allow setting MDS UUID in .xml (2580)
7245 - print the stack of a process that LBUGs (4228)
7247 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
7250 - don't cleanup request in ll_file_open() on failed MDS open (3430)
7251 - make sure to unset replay flag from failed open requests (3440)
7252 - if default stripe count is 0, use OST count for inode size (3636)
7253 - update parent mtime/ctime on client for create/unlink (2611)
7254 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
7255 - free recovery state on server during a forced cleanup (3571)
7256 - unregister_reply for resent reqs (3063)
7257 - loop back devices mounting and status check on 2.6 (3563)
7258 - fix resource-creation race that can provoke i_size == 0 (3513)
7259 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
7260 - more debugging for page-accounting assertion (3746)
7261 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
7262 - avoid deadlock after precreation failure (3758)
7263 - fix race and lock order deadlock in orphan handling (3450, 3750)
7264 - add validity checks when grabbing inodes from l_ast_data (3599)
7266 - add /proc/.../recovery_status to obdfilter (3428)
7267 - lightweight CDEBUG infrastructure, debug daemon (3668)
7268 - change default OSC RPC parameters to be better on small clusters
7269 - turn off OST read cache for files smaller than 32MB
7270 - install man pages and include them in rpms (3100)
7271 - add new init script for (un)mounting lustre filesystems (2593)
7272 - run chkconfig in %post for init scripts (3701)
7273 - drop scimac NAL (unmaintained)
7275 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
7278 - clean kiobufs before and after use (3485)
7279 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
7280 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
7281 - change init script to fail more clearly if not run as root (1528)
7282 - allow clients to reconnect during replay (1742)
7283 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
7284 - don't do DNS lookups on NIDs too small for IP addresses (3442)
7285 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
7286 - fix cond_resched (3554)
7287 - only evict unfinished clients after recovery (3515)
7288 - allow bulk resend, prevent data loss (3570)
7289 - dynamic ptlrpc request buffer allocation (2102)
7290 - don't allow unlinking open directory if it isn't empty (2904)
7291 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
7292 - remove extraneous obd dereference causing LASSERT failure (3334)
7293 - don't use get_cycles() when creating temp. files on the mds (3156)
7294 - hold i_sem when setting i_size in ll_extent_lock() (3564)
7295 - handle EEXIST for set-stripe, set proper directory name (3336)
7297 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
7298 - fix ksocknal_fmb_callback() error messages (2918)
7300 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
7303 - don't copy lvb into (possibly NULL) reply on error (2983)
7304 - don't deref dentry after dput, don't free lvb on error (2922)
7305 - use the kms to determine writeback rpc length (2947)
7306 - increment oti_logcookies when osc is inactive (2948)
7307 - update client's i_blocks count via lvb messages (2543)
7308 - handle intent open/close of special files properly (1557)
7309 - mount MDS with errors=remount-ro, like obdfilter (2009)
7310 - initialize lock handle to avoid ASSERT on error cleanup (3057)
7311 - don't use cancelling-locks' kms values (2947)
7312 - use highest lock extent for kms, not last one (2925)
7313 - don't dereference ERR_PTR() dentry in error handling path (3107)
7314 - fix thread race in portals_debug_dumplog() (3122)
7315 - create lprocfs device entries at setup instead of at attach (1519)
7316 - common AST error handler, don't evict client on completion race (3145)
7317 - zero nameidata in detach_mnt in 2.6 (3118)
7318 - verify d_inode after revalidate_special is valid in 2.6 (3116)
7319 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
7320 - initialize RPC timeout timer earlier for 2.6 (3219)
7321 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
7322 - print nal/nid for unknown nid (3258)
7323 - additional checks for oscc recovery before doing precreate (3284)
7324 - fix ll_extent_lock() error return code for 64-bit systems (3043)
7325 - don't crash in mdc_close for bad permissions on open (3285)
7326 - zero i_rdev for non-device files (3147)
7327 - clear page->private before handing to FS, better assertion (3119)
7328 - tune the read pipeline (3236)
7329 - fix incorrect decref of invalidated dentry (2350)
7330 - provide read-ahead stats and refine rpc in flight stats (3328)
7331 - don't hold journal transaction open across create RPC (3313)
7332 - update atime on MDS at close time (3265)
7333 - close LDAP connection when recovering to avoid server load (3315)
7334 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
7335 - don't leak open file on MDS after open resend (3325)
7336 - serialize filter_precreate and filter_destroy_precreated (3329)
7337 - loop device shouldn't call sync_dev() for nul device (3092)
7338 - clear page cache after eviction (2766)
7339 - resynchronize MDS->OST in background (2824)
7340 - refuse to mount the same filesystem twice on same mountpoint (3394)
7341 - allow llmount to create routes for mounting behind routers (3320)
7342 - push lock cancellation to blocking thread for glimpse ASTs (3409)
7343 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
7344 - fix rare problem with rename on htree directories (3417)
7346 - allow default OST striping configuration per directory (1414)
7347 - fix compilation for qswnal for 2.6 kernels (3125)
7348 - increase maximum number of MDS request buffers for large systems
7349 - change liblustreapi to be useful for external progs like lfsck (3098)
7350 - increase local configuration timeout for slow disks (3353)
7351 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
7353 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
7356 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
7357 - don't overwrite extent policy data in reply if lock was blocked (2901)
7358 - drop filter export grants atomically with removal from device (2663)
7359 - del obd_self_export from work_list in class_disconnect_exports (2908)
7360 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
7361 - swab reply message in mdc_close, other PPC fixes (2464)
7362 - fix destroying of named logs (2325)
7363 - overwrite old logs when running lconf --write_conf (2264)
7364 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
7365 - fix race in target_handle_connect (2898)
7366 - mds_reint_create() should take same inode create lock (2926)
7367 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
7368 - don't close files for self_export to avoid uninitialized obd (2936)
7369 - allow MDS with the same name as client node (2939)
7370 - hold dentry reference for closed log files for unlink (2325)
7371 - reserve space for all logs during transactions (2059)
7372 - don't evict page beyond end of stripe extent (2925)
7373 - don't oops on a deleted current working directory (2399)
7374 - handle hard links to targets without a parent properly (2517)
7375 - don't dereference NULL lock when racing during eviction (2867)
7376 - don't grow lock extents when lots of conflicting locks (2919)
7378 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
7381 - account for cache space usage on clients to avoid data loss (974)
7382 - lfsck support in lustre kernel code (2349)
7383 - reduce journal credits needed for BRW writes (2370)
7384 - orphan handling to avoid losing space on client/server crashes
7385 - ptlrpcd can be blocked, stopping ALL progress (2477)
7386 - use lock value blocks to assist in proper KMS, faster stat (1021)
7387 - takes i_sem instead of DLM locks internally on obdfilter (2720)
7388 - recovery for initial connections (2355)
7389 - fixes for mds_cleanup_orphans (1934)
7390 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
7391 - block all file creations until orphan recovery completes (1901)
7392 - client remove rq_connection from request struct (2423)
7393 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
7394 - recovery timer race (2670)
7395 - mdc_close recovey bug (2532)
7396 - ptlrpc cleanup bug (2710)
7397 - mds timeout on local locks (2588)
7398 - namespace lock held during RPCs (2431)
7399 - handle interrupted sync write properly (2503)
7400 - don't try to handle a message that hasn't been replied to (2699)
7401 - client assert failure during cleanup after abort recovery (2701)
7402 - leak mdc device after failed mount (2712)
7403 - ptlrpc_check_set allows timedout requests to complete (2714)
7404 - wait for inflight reqs when ptlrpcd finishes (2710)
7405 - make sure unregistered services are removed from the srv_list
7406 - reset bulk XID's when resending them (caught by 1138 test)
7407 - unregister_bulk after timeout
7408 - fix lconf error (2694)
7409 - handle write after unfinished setstripe, stripe-only getstripe (2388)
7410 - readahead locks pages, leaves pending causing memory pressure (2673)
7411 - increase OST request buffers to 4096 on large machines (2729)
7412 - fix up permission of existing directories in simple_mkdir (2661)
7413 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
7414 - don't assign transno to errored transactions (2742)
7415 - don't delete objects on OST if given a bogus objid from MDS (2751)
7416 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
7417 - if rq_no_resend, then timeout request after recovery (2432)
7418 - fix MDS llog_logid record size, 64-bit array alignment (2733)
7419 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
7420 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
7421 - MDS recovery shouldn't delete objects that it has given out (2730)
7422 - if enqueue arrives after completion, don't clobber LVB (2819)
7423 - don't unlock pages twice when trigger_group_io returns error (2814)
7424 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
7425 - don't write pages to disk if there was an error (1450)
7426 - don't ping imports that have recovery disabled (2676)
7427 - take buffered bytes into account when balancing socknal conn (2817)
7428 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
7429 - reconnect unlink llog connection after MDS reconnects to OST (2816)
7430 - remove little-endian swabbing of llog records (1987)
7431 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
7432 - retry reposting request buffers if they fail (1191)
7433 - grow extent at grant time to avoid granting a revoked lock (2809)
7434 - lock revoke doesn't evict page if covered by a second lock (2765)
7435 - disable VM readahead to avoid reading outside lock extents (2805)
7437 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
7438 - updated kernel patches for hp-2.4.20 kernel (2681)
7440 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
7443 - fix truncated write corruption (2366)
7444 - fix for failed assertion in iopen_connect_dentry (1792,2517)
7446 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
7447 - ensure len(uuid) < 37 in lmc (1171)
7448 - fix ia64 OOPS in llog_test (2255)
7449 - zero end of page at obdfilter for partial page writes (2648)
7450 - don't leave stale dentries around after renames (bug 2428)
7451 - fix timeouts when evicting a client with a single lock held (2642)
7452 - set deadline for the initial HELLO message to drain (2634)
7453 - print out dotted-quad IP addresses in the socknal (2302)
7455 - additional debugging for MDS client eviction problem (2443)
7456 - fix mkfsoptions support for osts (2603, 2604)
7458 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
7461 - add series for the vanilla 2.6.0 kernel
7462 - add series for the vanilla 2.4.24 kernel
7463 - add series for a cray x86/64 UL kernel drop
7464 - fix xattr patches for the vanilla 2.4.19 series
7466 - generate true UUIDs in lmc (1171)
7467 - have portals stack dumping break in UML (2466)
7468 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
7469 - allocate with _NOFS in ldlm to avoid deadlock (1933)
7470 - wake callback waiting threads on client eviction (2460)
7471 - Add --ptldebug and --subsystem to lmc (1719)
7472 - update assertion to allow safe interrupt allocation
7473 - set rq_no_resend for cancel requests (2432)
7474 - recalculate ptlrpcd timeout after resend (2494)
7475 - call vfs_rmdir when removing pending directories (2368)
7476 - fix renaming a file to itself (2429)
7477 - lmc creates a default one-stripe lov (2454)
7478 - expand procfs space to handle large clusters (2326)
7479 - increase UML stack to avoid overflow
7480 - update lconf's list of debug and subsystem masks
7481 - fix lfs find --obd (2510)
7482 - /proc tunable for disabling filter read caching (2591)
7483 - stop rpm packages from altering slapd.conf (2301)
7484 - disable nagle in the socknal under 0conf (2578)
7485 - choose mds inode size based on stripe count (2572)
7486 - fix kernel-source rpm problems (2516)
7488 - add --disable-doc to avoid pdf generation (2421)
7489 - update documentation, tests, type-os, comments
7490 - avoid format warnings on ia64
7491 - remove the TOE NAL
7492 - tiny code cleanups by removing unused fields
7494 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
7497 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
7498 - remove the most blatant lies from BUILDING file (2371)
7499 - change default debug level to reasonable production setting
7500 - reduce client side cache size to reduce cache flush time
7501 - reduce max RPCs in flight to avoid unnecessary file fragmentation
7502 - make TCP zerocopy and pinger support enabled by default (2476)
7503 - sync writes completed after process exits caused crashes (2319)
7504 - maintain correct mount count on the MDS (2356)
7505 - backout 1557, because 2316 wasn't really fixed
7506 - better file I/O statistics gathering in /proc
7507 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
7508 - another kernel patch to fix zero-copy TCP function export
7509 - don't take duplicate lock when processing re-sent getattr (2420)
7510 - lctl uses obd_self_export instead of creating new conn (2353)
7511 - MDS/OST recovery case which requires object creation asserted (2425)
7512 - move lfs from /usr/sbin to /usr/bin in packages
7513 - fix race between mds_client_add and mds_client_free (2417)
7514 - use kmalloc instead of slabs in portals (2430)
7515 - don't create duplicate records when a failover MDS is present (2442)
7516 - remove unnecessary mount age check (2332)
7517 - don't remove directory inodes from locks prematurely (2451)
7518 - don't break if MDS service name is the same as hostname (2103)
7519 - fix races in client write RPC generation when cache full (2482)
7521 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
7524 - remove now-unused request->rq_obd (278)
7525 - if an allocation fails, print out how much memory we've used (1933)
7526 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
7527 - add the "configurable stack size" patch to most series files (1256)
7528 - ability to write large log records, for 100+ OST configs (2306)
7529 - fix NULL deref when filter_prep fails (2314)
7530 - fix operator precedence error in filter_sync
7531 - dynamic allocation of socknal TX descriptors (2315)
7532 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
7533 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
7534 - ensure that CWARN messages in Portals always get to the syslog
7535 - __init/__exit are not for prototype decls (ldlm_init/exit)
7536 - x86-64 compile warning fixes
7537 - fix gateway LMC keyword conflict (2318)
7538 - fix MDS lock inversions in getattr/reint paths (1844)
7539 - fix a rare lock re-ordering bug, which caused deadlock (2322)
7540 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
7541 - DLM race condition prevented some lock evictions (2328)
7542 - ENOMEM detection and retry on socknal sends (2230)
7543 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
7544 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
7545 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
7546 - better error messages when a client is rejected during recovery (1505)
7547 - avoid cancelling locks which were never granted, after failure (2330)
7548 - fix i_sem/journal inversion in mds_client_add (2333)
7549 - fix truncate/getattr lock cycle deadlock (2334)
7550 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
7551 - fix two rare exit paths which could leak an l_lock() ref (2321)
7552 - fencepost error in MDS/OST orphan recovery (2226)
7553 - make log record alignment 8 bytes (1988)
7554 - lstripe now fails when requested offset > ost_count (2237)
7555 - ensure that all kernel series have a complete list.h (1607)
7556 - fix crashes in special-file operations (2316)
7557 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
7559 - allow configurable automake binary, for testing new versions
7560 - small update to the lfs documentation
7562 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
7564 * fix negative export reference count in fsfilt_sync (2312)
7566 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
7567 * release candidate 0.9.1
7569 - orphans are moved into the PENDING directory for possible recovery
7570 - replayed opens now open by fid for orphan/rename safety (1042)
7571 - last close of an orphan inode generates a transno (683)
7572 - chdir() and mount() now pin the directory entry (1020)
7573 - avoid CERROR in normal ll_setattr_raw() error case (1500)
7574 - discard very old requests without processing them (1502)
7575 - remove some common, well-understood CERRORs (1505)
7576 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
7577 - clear "grant" flags in OST replies until OST grant code lands (1644)
7578 - fix read performance by not clobbering i_blksize on client (1598)
7579 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
7580 - make LRU size a /proc tunable, clears locks when reduced (707)
7581 - fix some lprocfs rot that prevented ptlbd from loading (1732)
7582 - server locks take references on exports now (1558)
7583 - build fixes for 2.4.20-rh trees (1663)
7584 - return an error from lov_create if all OSCs are inactive (1751)
7585 - fix import levels when a reconnect happens without a timeout (1597)
7586 - exit early from mds_open if we get a lookup error (1749)
7587 - partial page read at EOF wouldn't wait for disk before sending (1642)
7588 - avoid NULL deref in obdfilter when reading page past EOF (1592)
7589 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
7590 - fix LBUG in ll_it_open_error with rc = -2 (1861)
7591 - write/truncate lock inversion (1639)
7592 - Don't auto-load obdclass, portals modules during cleanup (1495)
7593 - fix timestamps from jumping to "now" (1763)
7594 - extra journal assertions (1648)
7595 - add an extra multiunlink test (1771)
7596 - fix read_record/write_record API (1776)
7597 - fix leak of offset_extent, possible incorrect i_size later (1772)
7598 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
7599 - Debugging for the kqswnal_get_idle_tx problems (1820)
7600 - Allow recovery to be attempted multiple times (1536)
7601 - Write out MDS last_rcvd file after it is first created (1600)
7602 - Fix tx_descriptor leak in failed transmit situations (1827)
7603 - ext3 journaling fixes for assertion failure after IO error (1871)
7604 - class_export_put() on freed export after completion AST error (1896)
7605 - Fix revalidate looping in VFS (1322)
7606 - Don't access a freed export during MDS_REINT timeout (1521)
7607 - Add open-unlink recovery support on the MDS (1673,1764)
7608 - Return an error if no MDS data was read from last_rcvd (1946)
7609 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
7610 - Don't setup a disk device that doesn't match exported UUID (317)
7611 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
7612 - avoid committing NULL handle in force close
7613 - local.sh is now a one-stripe LOV configuration
7614 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
7615 - fix ext3 htree duplicate directory entry corruption (1516)
7616 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
7617 - update to new LOV EA format (2097)
7618 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
7620 2003-06-15 Phil Schwan <phil@clusterfs.com>
7623 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
7624 - per-import/export recovery handling (958, 931, 959)
7625 - multiple last-rcvd slots, for serving multiple FSes (949)
7626 - connections are again shared between multiple imp/exports (963, 964)
7627 - "umount -f" would hang if any requests needed to be sent (393, 978)
7628 - avoid pinning large req buffer by copying for queued messages (989)
7629 - add "uuid" to "lctl device" command to help upcalls (991)
7630 - "open" RPCs with transnos would confuse recovery counters (1037)
7631 - do proper endian conversion of all wire messages (288, 340, 891)
7632 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
7633 - call waiting locks callback from LDLM recovery thread (1127, 1151)
7634 - fix ptlrpc_connection leak in target_handle_connect (1174)
7635 - fix import refcounting bug in OST and MDS cleanup (1134)
7636 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
7637 - fix very unlikely obd_types race condition (501)
7638 - remove osc_open hack for echo_client (1187)
7639 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
7640 - a failure in read_inode2 leads to deadlock (1139)
7641 - cancel ack-locks as soon as transaction is committed (1072)
7642 - fix major leaks and crashes in the bulk I/O path (937, 1057)
7643 - make sure to commitrw after any preprw to avoid deadlock (1162)
7644 - failing to execute a file in a lustre FS would lock inode (1203)
7645 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
7646 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
7647 - obd_uuid2tgt would walk off the end of the list (1255)
7648 - on IA64 the osc would give portals incorrect bulk size (1258)
7649 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
7650 - fix lock inversion caused by new llite matching code (1282)
7651 - limit the number of dirty pages on a client to 10MB (1286)
7652 - timed out locks were not being corrected cancelled (1289)
7653 - fix O_DIRECT above 4GB on IA-32 (1292)
7654 * major user-visible changes
7655 - fail out/fail over policy now controlled by the upcall (993)
7657 - add OBD_PING to check server availability and failure (954)
7658 - lustre messages are now sent in sending host order (288, 340, 891)
7659 - add eadatalen to MDS getattr reply (340)
7660 - OST read replies may contain second buffer, with per-page status (593)
7662 2003-03-11 Phil Schwan <phil@clusterfs.com>
7665 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
7666 - failed open()s could cause deadlock; fixed (867, 869)
7667 - stop cancelling OST locks when files are closed (481)
7668 - overlapping XID spaces caused network corruption (851, 853)
7669 - fix unsafe fsfilt counter arithmetic; change to atomic_t
7670 - setattr_raw added, to do single-RPC, server-side setattrs
7671 - lmc/lconf syntax change for OST UUIDs
7672 - fix crashy race condition between ptlrpc_free_req and osc_close
7673 - don't use request in mdc_enqueue if we hit a timeout (889)
7674 - don't set the inode i_size for regular files from the MDS (896)
7675 - handle out of order completion AST (842)
7676 - don't LBUG if a lock request times out after receiving AST (913)
7677 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
7678 - if a bad lock AST arrives, send an error instead of dropping entirely
7679 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
7680 - fix leak in bulk IO when only partially completed (899, 900, 926)
7681 - fix O_DIRECT for ia64 (55)
7682 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
7683 - C-z after timeout could hang a process forever; fixed (977)
7685 - client-side I/O cache (678, 924, 929, 941, 970)
7687 - READPAGE and SETATTRs which don't take server-side locks get
7690 2003-02-11 Phil Schwan <phil@clusterfs.com>
7693 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
7694 - Fix fsfilt_extN_readpage() to read a full page of directory
7695 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
7696 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
7697 - Fix imbalanced LOV object allocation and out-of-bound access (469)
7698 - Most intent operations were removed, in favour of a new RPC mode
7699 that does a single RPC to the server and bypasses most of the VFS
7700 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
7701 - Aggressively cancel local locks on DLM servers
7702 - mds_reint_unlink sends EA to the client if it's the last nlink.
7703 client uses that EA to unlink OST objects.
7704 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
7705 - recursive symlinks were fixed (439)
7706 - fixed NULL deref in DEBUG_REQ
7707 - filter_update_lastobjid no longer calls sync, which annoyed extN
7708 - fixed multi-client small-writes to a single file problem (445)
7709 - fixed mtime updates during file writes (607)
7710 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
7711 - fixed bug in obd_brw_read/write() (under guise of testing 367)
7712 - fixed Linux OST size reporting problem (444, 656)
7713 - OST now updates object mtime with writes or setattr (607, 619)
7714 - client verifies file size before zeroing page past EOF (445)
7715 - OST now writes last allocated objid to disk with allocation (108)
7716 - LOV on echo now works (409)
7718 - mds_reint_unlink sends a new buffer, with the EA included. this
7719 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
7720 set if a regular file was being unlinked, and it was the last link
7721 - use PtlGet from the target for bulk writes (315)
7722 - OST now updates object mtime with writes or setattr (607, 619)
7723 - LDLM now has a grant-time callback to revalidate locked items, if
7725 - Many MDS operations were reorganized to combat race conditions
7727 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
7728 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
7730 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
7733 - Fully reactivate OST imports after reconnection (512, others)
7734 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
7735 - More graceful error handling for truncating on dead OST (515)
7736 - Don't error out unless we're actually accessing dead stripes (474)
7737 - Fix garbage sizes when stripes are missing (410)
7738 - LRU counters were broken, causing constant lock purge (433, 432)
7739 - garbage on read from stripes with failed OSTs (441)
7740 - mark OSCs as active before reconnecting during recovery (438)
7741 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
7742 - lfind did not preserve OST order in output (443)
7743 - symlinks cause hung clients, incorrect data (439)
7744 - stop dereferencing request after dropping refcount (457)
7745 - don't LASSERT(spin_is_locked) on non-SMP (455)
7746 - fixes for many rename() bugs
7747 - fstat didn't correctly synchronize attributes (399)
7748 - server must handle lock cancellation during blocking AST prep (487)
7749 - bulk descriptors were free()d too soon (511)
7750 - fix paths in lconf, which would load incorrect modules (451, 507)
7751 - fix confusing lconf 'host not found' error message (386)
7752 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
7753 - fix race condition in mdc_blocking_ast() for inode access (526)
7754 - fix lov_unpackmd() unpacking wrong number of stripes (537)
7755 - fix lov_set_osc_active() marking wrong OSC inactive (440)
7756 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
7757 - fix multiple writes of stripe MD to MDS (358, maybe 519)
7758 - fix lstripe in several ways (kernel side) (527)
7759 - fix request leak in ldlm_cli_enqueue (262)
7760 - incorrect OSC was marked inactive after OST failure
7761 - call mds_fs_cleanup before unmounting filesystem (524)
7762 - fix races between taking ns_lock and ldlm_lock_change_resource
7763 - fix races updating LOV export open file list
7764 - fix lov_enqueue error path, avoid decref-ing bad lock handle
7765 - fix recovery NULL deref in ldlm_cli_cancel_unused
7766 - fix some DLM races by using new hash table for lock handles (419)
7767 - permit the client to specify desired inodes, at replay
7768 - duplicate requests when we queue them for replay reintegration
7769 - fix last_rcvd offset calculation
7770 - sync after each recovered transaction, so we always make progress
7771 - never, not always, ERESTART requests without transnos
7772 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
7774 - skip replay if the MDS says that the client is already connected
7775 - don't check for a recovery-enabled export to match lctl's UUID
7776 - don't INC_USE_COUNT for phantom exports
7777 - don't crash when cleaning up phantom exports (567)
7778 - don't double-finish or set replay data for errored mdc_open requests
7779 - abort requests when they time out, so we don't get old replies
7780 - send/receive replies for AST messages again
7781 - if the client says that it doesn't have the lock, cancel it on the
7783 - if we timeout during I/O, don't try to cancel an in-use lock; instead
7784 mark it as destroyed, it will all work out when decref is called
7785 - fix module use counts (22, 581)
7787 - ASTs now expect a reply (server cancels lock on error reply)
7789 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
7792 - fix many simultaneous client startup (392)
7793 - fix dentry->d_it clobbering
7794 - credentials weren't being shipped for readdir/getattr operations
7795 - remove invalid assertions triggered during some concurrent MD
7797 - proper Lustre versions added (336, 389)
7798 - fix memory leak for create error case (398)
7799 - fix LOV locking bug that would get cli/srv out of sync
7800 - fix echo client over LOV (409)
7801 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
7802 - fix double-O_EXCL intent crash (424)
7803 - avoid sending multiple lock CANCELs (352)
7805 - MDS can do multi-client recovery (modulo bugs in new code)
7807 - many updates, edits, cleanups
7809 2002-11-18 Phil Schwan <phil@clusterfs.com>
7812 - fix null d_it dereference (346)
7813 - fix full OST/dbench hang (333)
7814 - fix permission problem with file removal (286)
7815 - fix removal of OSCs from LOV when they fail
7816 - fix NULL deref during bulk timeout (214)
7817 - fix problems related to multiple filesystems on one MDS (241)
7818 - fixed serious subtle metadata locking bugs
7819 - free locks on clients when inodes are removed due to memory
7821 - fix inode pointer in lock data (285)
7822 - partial support for multiple MDS on a single host (241)
7823 - data locks weren't cancelled at clear_inode time (290, 311)
7824 - intent locks could lead to unbounded lock growth (205)
7825 - added a maximum lock count, an LRU list, and a flusher
7826 - fix multiple rename (365)
7827 - properly abstracted the echo client
7828 - OSC locked 1 byte too many; fixed
7829 - rewrote brw callback code:
7830 - fixed recovery bugs related to LOVs (306)
7831 - fixed too-many-pages-in-one-write crash (191)
7832 - fixed (again) crash in sync_io_timeout (214)
7833 - probably fixed callback-related race (385)
7835 - Add capability to MDS protocol
7836 - LDLM cancellations and callbacks on different portals
7838 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
7841 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
7842 - defer open object destruction to close time (601981, 138)
7843 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
7844 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
7845 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
7846 - don't call obd_setattr in ll_file_release for destroyed objects
7847 * protocol change to lustre_msg: move |version| and add |flags|
7848 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
7849 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
7850 * added replay of create, unlink, link and rename operations during
7851 MDS failover; recovery should be much more robust now
7852 * remove failed OSCs from LOVs (only lov_create uses this so far)
7853 * the lustre-HOWTO was brought (more) up to date (582544)
7855 2002-10-23 Phil Schwan <phil@clusterfs.com>
7858 - in-use dentries weren't being reused properly (617851)
7859 - prevent multiple LDLM setup (599178)
7860 - fix LOV size calculations for truncate (617853)
7861 - fix client handling of MDS intent errors (POSIX)
7862 - fix permission bug in lovstripe.c test (624321)
7863 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
7864 - truncate past end of file could corrupt data
7865 - proper cleanup after timeouts, crashes, etc (592524, 550815)
7866 - a race in recovery could return ETIMEDOUT to apps (623947)
7867 - building outside the source directory was fixed
7868 * the lustre-HOWTO was brought (more) up to date (582544)
7869 * major progress was made on recovery functionality
7871 2002-10-10 Phil Schwan <phil@clusterfs.com>
7874 - recovery deadlock fix
7875 - rm -rf causes LBUG fix (617817)
7876 - file open by multiple tasks fix (618962)
7877 - directory permissions bugs (602707 and 620007)
7878 - journal_stop fixed with locking (611313)
7879 - O_APPEND failures resolved (618273, perhaps 614459)
7880 - lconf PATH fix (619770)
7881 - IA64 build fix (621450)
7882 - RPC buffer sizes scale with amount of memory
7884 2002-10-01 Phil Schwan <phil@clusterfs.com>
7887 - locks would be cancelled without throwing away data pages,
7888 resulting in inconsistent data (605627)
7889 - inode attributes were not always being refreshed (605627, 612449)
7890 - lconf now continues to cleanup after lctl reports an error
7891 - MDS now enforces user permissions (602707)
7892 - lprocfs cleanup fixed, but not yet enabled (614157)
7893 - fixed infinite server hang, should a client not respond to an AST
7894 - avoid going into recovery if user calls readlink() with a buffer
7895 that's too small (613941)
7896 - AST RPCs no longer require replies (614867) -- this may be changed
7897 - don't crash server if client sends an IOV that's too big (611336)
7898 - fixed lock conversion deadlock (611892)
7899 - fixed the following of symlinks (614622)
7900 * recovery: the server can remove locks from a client that dies, other
7901 clients can make progress
7902 * more extN patch fixes
7903 * compile-time configurable ptlrpc buffer allocations
7905 - collaborative read cache document
7906 - Lustre Lite Performance CDR document-in-progress
7908 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
7911 - fix typo in patch-2.4.18
7913 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
7916 - clear ptlrpc request each time in handle_incoming_request()
7917 - unlink of files now destroys the object on the OST
7919 2002-09-19 Peter Braam <braam@clusterfs.com>
7921 * add hard link support
7922 * change obdfile creation method
7923 * kernel patch changed
7925 2002-09-19 Peter Braam <braam@clusterfs.com>
7928 - stack overflow bug in extN fixed
7930 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
7932 * documentation updates
7933 - add man pages for config tools
7934 - update tests/README to describe testing with new config tools
7935 - finish metadata API descriptions
7936 * bug fixes and cleanups
7937 - statfs workaround for 16TB limit
7938 - LOV stripe allocation improved, can stripe on subset of OSTs
7939 - LOV file size/IO offset was wrong for files > 4GB in size
7940 - object EA data was being dropped, caused files to be unreadable
7941 - memory overflow with non-LOV OST caused memory corruption
7942 - fixed regression tests to work with new config tools, obdfilter
7943 - fixed bug when directory size became larger than 1 block
7944 - fixed bug (for single client case) when PWD was deleted
7945 - invalidate local directory pages when doing intent-based ops
7946 - avoid LDLM oops when lock callback contained bad data
7948 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
7950 * documentation updates
7951 * bug fixes and cleanups
7952 - configuration tools
7955 - 64-bit compile warnings
7956 - 64-bit internal statfs data
7958 * test_brw on persistent OST devices
7960 * lprocfs (disabled)
7962 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
7964 * documentation updates
7965 * bug fixes and cleanups
7966 * configuration tools
7968 2002-08-30 Peter J. Braam <braam@clusterfs.com>
7971 * many small fixes to 0_5_4
7972 * io/network handling
7973 * thinkos in MDS operations
7975 2002-08-24 Peter J. Braam <braam@clusterfs.com>
7978 * crucial basic fixes to 0.5.3
7979 * IOR, Iozone work over Elan
7982 2002-08-07 Phil Schwan <phil@clusterfs.com>
7983 * version 0_5_3, our first alpha
7984 * we use the new Portals iovs
7985 * documentation updates
7986 * bug fixes and cleanups
7987 * small changes in the DLM wire protocol
7989 2002-07-25 Peter J. Braam <braam@clusterfs.com>
7990 * version 0_5_1 with some initial stability,
7991 * locking on MD and file I/O.
7992 * documentation updates
7993 * several bug fixes since 0.5.0
7994 * small changes in wire protocol
7996 2002-07-18 Phil Schwan <phil@clusterfs.com>
7998 * delivered as Lustre Light Alpha
7999 * fixed a crash after handling invalid MDS requests
8000 * fixed directory pages for architectures with non-4k pages sizes
8002 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
8003 * release version v0_4_4
8004 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
8006 * Fixes a number of interruption problems with OST operations.
8007 * Update documentation for portals header changes
8008 * Move all wire protocol structs/defines to lustre_idl.h
8009 * Fixes symlink length bug.
8010 * Add tcpdump to repository.
8012 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
8013 * release version v0_4_3
8014 * Fixes statfs for inodes on extN.
8015 * Fixes bug in runtests which would delete /etc/hosts.
8016 * Use 64-bit object IDs wherever possible (not into VFS though)
8017 Remove ost_get_info, which is unused by lustre, and out of date.
8019 2002-07-03 Peter Braam <braam@clusterfs.com>
8020 * release version v0_4_2 Fixes a lookup error (type not passed)
8021 * move forward to head of Portals
8022 * move forward to latest Lustre kernel
8024 2002-06-25 Peter Braam <braam@clusterfs.com>
8025 * release version v0_4_1. Hopefully stable on single node use.