1 tbd Sun Microsystems, Inc.
3 * Server support for kernels:
4 2.6.18-164.6.1.el5 (RHEL 5)
5 2.6.18-164.6.1.0.1.el5 (OEL 5)
6 * Client support for unpatched kernels:
7 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
8 2.6.18-164.6.1.el5 (RHEL 5),
9 2.6.18-164.6.1.0.1.el5 (OEL 5)
10 2.6.16.60-0.42.8 (SLES 10),
12 * Recommended e2fsprogs version: 1.41.6.sun1
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
18 Description: New acc-sm script and various test-framework enhancments.
22 Description: Previously landed patch was reverted.
26 Description: Update SLES10 kernel to 2.6.10.60-0.42.8 (patchless)
30 Description: Add ldiskfs maxdirsize mount option.
34 Description: Fix mkfs.lustre support for 16TB LUNs.
38 Description: Add cascading_rw to lustre mpi tests.
42 Description: Store SOM data in EA.
43 Details : Also several SOM related cleanups.
47 Description: Add refcount to md_open_data
51 Description: Use refcount on lov_request_set to prevent memory corruption.
55 Description: Statahead: use parent dir's i_mutex for synchronization.
59 Description: loadgen improvements.
63 Description: Enable peer health detection in ksocklnd and ko2iblnd.
67 Description: Control DCACHE_LUSTRE_INVALID flag with MDS_INODE_LOOKCUP lock
71 Description: Fix race on obd_devcie:md_stats
75 Description: Fix SLES 11 client builds.
76 Details : Resolved uninitalized pointer gcc warnings.
80 Description: Allow non-root access for "lfs check".
81 Details : Added a check in obd_class_ioctl() for OBD_IOC_PING_TARGET.
85 Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
87 Severity : enhancement
89 Description: Update kernel to RHEL5.4 2.6.18-164.6.1.el5 and
90 OEL5 2.6.18-164.6.1.0.1.el5(Both in-kernel OFED enabled).
94 Description: Sanity test_27p fails.
95 Details : OBD_FAIL_OST_ENOSPC has been moved to filter_statfs().
96 Sanity tests 27 were cleaned up and improved.
100 Description: Unify req format on client/servers
101 Details : Use new req_capsule API [almost] everywhere instead of old PTLRPC
102 buffers and swabbers approach..
107 Description: drop join file stuff.
109 Severity : enhancement
111 Description: Add 2.6.27 and SLES11 (patchless client) support
115 Description: Fix in ptlrpc_expire_one_request() to print the signed time
118 Severity : enhancement
120 Description: Build kernels (RHEL5, OEL5 and SLES10/11) using the vendor's own
123 Severity : enhancement
125 Description: Remove set_info(KEY_UNLINKED) from MDS/OSC
127 Severity : enhancement
129 Description: correctly handle big reply message.
130 Details : send LNet event if reply is bigger then buffer and adjust this buffer
135 Description: Drop unnecessary __GFP_NOMEMALLOC flag from filter_get_page()
137 Severity : enhancement
139 Description: Add support for OEL5.
143 Description: Conf-sanity.sh 50g test - deactivated OST should not cause a panic.
147 Description: client could not reconnect to OST because of an active request.
148 Details : The client now retries to reconnect to the same server, if a
149 connect request failed with EBUSY or -EAGAIN. Test case 35b
150 conf-sanity.sh has been created.
154 Description: Don't allow make backward step on assignin osc next id.
155 Details : race between allocation next id and ll_sync thread can be cause
156 of set wrong osc next id and can be kill valid ost objects.
160 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
164 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
168 Description: Error handling in osc_statfs_interpret() has been improved.
169 Details : Check in osc_statfs_interpret() for EBADR.
173 Description: Do not send statfs() requests to OSTs disabled by administrator.
174 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
176 Severity : enhancement
178 Description: Added sanity-quota test to for lfs(1) to work with numeric uid/gid.
180 Severity : enhancement
182 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
184 Severity : enhancement
186 Description: Enhancement for lfs(1) command to use numeric uid/gid.
190 Description: Improved cleanup procedure for conf-sanity test_46a.
194 Description: Do not start bulk transfer, if client tried to reconnect.
195 Details : Do not start bulk transfer, if client tried to reconnect,
196 but failed due to active rpcs. Check for exp_abort_active_req
197 has been implemented.
201 Description: Prevent inconsistences between linux and lustre mount structures.
202 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
203 Make the sleep interruptible.
205 Severity : enhancement
207 Description: provide server to client comms path
208 Details : server to client communications path via new LDLM_SET_INFO rpc
212 Description: truncate starts GFP_FS allocation under transaction causing deadlock
213 Details : ldiskfs_truncate calls grab_cache_page which may start page
214 allocation under an open transaction. This may lead to
215 calling prune_icache with consequent lustre reentrance.
219 Description: send correctly lsm on open replay
220 Details : MDS is trust to LSM size on replay open, but client can set wrong size
225 Description: enable client interface failover
226 Details : When a child reconnects from another NID, properly update export
227 nid hash position and ldlm reverse import.
229 Severity : enhancement
231 Description: Communicate OST degraded/readonly state via statfs to MDS
232 Details : Flags in the statfs returned from OSTs indicate whether the
233 OST is in a degraded RAID state, or if the filesystem has
234 turned read-only after a filesystem error is detected.
239 Description: don't panic if EPROTO was hit when reading symlink
240 Details : correctly handling request reference in error cases.
242 Severity : enhancement
244 Description: Add LustreNetLink, a kernel-userspace communcation path.
246 Severity : enhancement
248 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
253 Description: access to llog context before init.
254 Details : move handling CATALOGS file at osc layer and forbid access to llog
258 Frequency : rare, only when client is writing to OST on local node
260 Description: Avoid deadlock for local client writes
261 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
262 memory freeing context. This allows OST threads to set the
263 PF_MEMALLOC flag on task structures in order to allocate memory
264 from reserved pools and complete IO.
265 Use GFP_HIGHUSER for OST allocations for non-local client writes,
266 so that the OST threads generate memory pressure and allow
267 inactive pages to be reclaimed.
269 Severity : enhancement
271 Description: Update kernel to RHEL5.3 2.6.18-128.1.14.el5.
276 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
277 Details : move ldlm namespace creation in setup phase to avoid grab
278 _lprocfs_lock with cli_sem held.
282 Description: Temporarily disable grant shrink.
283 Details : Disable the feature for debugging.
286 Frequency : only when formatting test filesystems
288 Description: Unable to run several mkfs.lustre on loop devices at the same time
289 Details : mkfs.lustre returns error 256 on the concurrent loop devices
290 formatting. The solution is to proper handle the error.
292 Severity : enhancement
294 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
296 Severity : enhancement
298 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
303 Description: resolve race between obd_disconnect and class_disconnect_exports
304 Details : if obd_disconnect will be called to already disconnected export he
305 forget release one reference and osc module can't unloaded.
307 Severity : enhancement
309 Description: Allow tuning service thread via /proc
310 Details : For each service a new
311 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
312 created that can be used to set min/max thread counts, and get the
313 current number of running threads.
318 Descriptoin: don't return error if have partially created objects for file.
319 Details : lov_update_create_set uses set->set_success as index for created
320 objects, so if some requests will be failed, they will have hole
321 at end of array and qos_shrink_lsm for allocate correct lsm.
323 Severity : enhancement
325 Description: Update OFED support to 1.4.1
330 Description: Reduce small size read RPC
331 Details : Set read-ahead limit for every file and only do read-ahead when
332 available read-ahead pages are bigger than 1M to avoid small size
335 Severity : enhancement
337 Description: move AT tunable parameters for more consistent usage
338 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
340 Severity : enhancement
342 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
343 Details : allow skip disconnected ost for send statfs request and hide error
349 Description: fix racy locking of mballoc block bitmaps causing BUG
350 Details : The locking of the mballoc buddy bitmap and the in-memory
351 block bitmap was using two different spin locks in some
352 cases. This made it possible to incorrectly access the
353 mballoc bitmap while another process was modifying it,
354 causing a sanity assertion to fail. While no on-disk corruption
355 was reported, there was some risk of this happening.
358 Frequency : rare, on llog test 6
360 Descriptoin: don't allow connect to already connected import
361 Details : allowing connect to already connected import is hide connecting problem.
366 Descriptoin: allow kill process which wait statahead result
367 Details : for some reasons 'ls' can stick in waiting result from statahead,
368 in this case need way for kill this process.
370 Severity : enhancement
372 Description: Add state history info file, enhance import info file
373 Details : Track import connection state changes in a new osc/mdc proc file;
374 add overview-type data to the osc/mdc import proc file.
376 Severity : enhancement
378 Description: MDS create should not wait for statfs RPC while holding DLM lock.
380 Severity : enhancement
382 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
387 Descriptoin: OST grant too much space to client even there are not enough space.
388 Details : Client will shrink its grant cache to OST if there are no write
389 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
390 this grant cache if there are already not enough avaible space
391 (left_space < total_clients * 32M).
394 Frequency : start MDS on uncleanly shutdowned MDS device
396 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
397 Details : stay in waiting mds<>ost recovery finished produce random bugs
398 due race between two ll_sync thread for one lov target. send
399 ACTIVATE event only if connect realy finished and import have
403 Frequency : rare, connect and disconnect target at same time
405 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
406 Details : don't call obd_disconnect under lov_lock. this long time
407 operation and can block ptlrpcd which answer to connect request.
412 Descriptoin: don't lose wakeup for imp_recovery_waitq
413 Details : recover_import_no_retry or invalidate_import and import_close can
414 both sleep on imp_recovery_waitq, but we was send only one wakeup
418 Frequency : always with long access acl
420 Descriptoin: mds can't pack reply with long acl.
421 Details : mds don't control size of acl but they limited by reint/getattr
424 Severity : enhancement
426 Description: Update to SLES10 kernel-2.6.16.60-0.33.
428 Severity : enhancement
430 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
433 Frequency : start MDS on uncleanly shutdowned MDS device
435 Descriptoin: aborting recovery hang on MDS
436 Details : don't throttle destroy RPCs for the MDT.
439 Frequency : on remount
441 Description: external journal device not working after the remount
442 Details : clear dev_rdonly flag for external journal devices in
448 Description: shutdown vs evict race
449 Details : client_disconnect_export vs connect request race.
450 if client will evicted at this time - we start invalidate
451 thread without referece to import and import can be freed
455 Frequency : rare, need acl's on inode.
457 Description: client can't handle ost additional correctly
458 Details : if ost was added after client connected to mds client can have
459 hit lnet_try_match_md ... to big messages to wide striped files.
460 in this case need teach client to handle config events about add
461 lov target and update client max ea size at that event.
463 Severity : enhancement
465 Description: Changelogs
466 Details : Changelogs are a lightweight mechanism to track filesystem
467 metadata and namespace changes. The changelog is recorded
468 permanently on the MDTs, and is periodically "consumed" / purged
469 when records are no longer needed.
471 Severity : enhancement
473 Description: compact fld format with extents
474 Details : Store range of seq rather than every seq in FLD. Seq
475 controller update FLD rather than clients. In Case of CMD, mdt0
476 has FLD, all other metadata server act as non persistent proxy
477 for FLD queries and cache fld entries in fld cache.
482 Description: don't skip ost target if they assigned to file
483 Details : Drop slow OSCs if we can, but not for requested start idx.
484 This means "if OSC is slow and it is not the requested
485 start OST, then it can be skipped, otherwise skip it only
486 if it is inactive/recovering/out-of-space.
490 Description: more cleanup in mds_lov
491 Details : not send LOV EA under replay, we can't know about they size at this
492 time. Don't allow client connect to mds before any ost connected,
493 for avoid problems with LOV EA size and returning EIO to client.
495 Severity : enhancement
497 Description: Interoperability at server side (Disk interoperability)
499 Severity : enhancement
501 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
503 Severity : enhancement
505 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
507 Severity : enhancement
509 Description: New client IO stack (CLIO).
511 Severity : enhancement
513 Description: Commit on sharing. Eliminate inter-client dependencies between
514 uncommitted transactions by doing transaction commits.
515 Thereby clients may recovery independently.
518 Frequency : Create a symlink file with a very long name
520 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
521 Details : If there is no extra space in the request for early cancels,
522 ldlm_req_handles_avail() returns 0 instead of a negative value.
524 Severity : enhancement
526 Description: Add /proc entry for import status
527 Details : The mdc, osc, and mgc import directories now have
528 an import directory that contains useful import data for debugging
531 Severity : enhancement
533 Description: Re-disable certain /proc logging
534 Details : Enable and disable client's offset_stats, extents_stats and
535 extents_stats_per_process stats logging on the fly.
538 Frequency : Only on FC kernels 2.6.22+
540 Description: oops in statahead
541 Details : Do not drop reference count for the dentry from VFS when lookup,
542 VFS will do that by itself.
544 Severity : enhancement
546 Description: Generic /proc file permissions
547 Details : Set /Proc file permissions in a more generic way to enable non-
548 root users operate on some /proc files.
552 Description: Hitting mdc_commit_close() ASSERTION
553 Details : Properly handle request reference release in
554 ll_release_openhandle().
558 Description: quota recovery deadlock during mds failover
559 Details : This patch includes att18982, att18236, att18237 in bz14840.
561 1. fix osts hang when mds does failover with quotaon
562 2. prevent watchdog storm when osts threads wait for the
567 Frequency : only patchless client
568 Description: add workaround for race between add/remove dentry from hash
570 Severity : enhancement
572 Description: Allow OST glimpses to return PW locks
576 Description: LBUG when llog conf file is full
577 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
581 Description: Prevent import from entering FULL state when server in recovery
585 Description: service mount cannot take device name with ":"
586 Details : Only when device name contains ":/" will mount treat it as
592 Description: replace ptlrpcd with the statahead thread to interpret the async
593 statahead RPC callback
597 Frequency : on recovery
598 Description: I/O failures after umount during fail back
599 Details : if client reconnected to restarted server we need join to recovery
600 instead of find server handler is changed and process self
601 eviction with cancel all locks.
603 Severity : enhancement
605 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
607 Severity : enhancement
609 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
611 Severity : enhancement
613 Description: Upcall on Lustre log has been dumped
614 Details : Allow for a user mode script to be called once a Lustre log has
615 been dumped. It passes the filename of the dumped log to the
616 script, the location of the script can be specified via
617 /proc/sys/lnet/debug_log_upcall.
622 Description: avoid idr_remove called for id which is not allocated.
623 Details : Move assigment s_dev for clustered nfs to end of initialization,
624 to avoid problem with error handling.
629 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
630 Details : When connection is reused this not moved from CONN_UNUSED_HASH
631 into CONN_USED_HASH and this prodice warning when put connection
632 again in unused hash.
635 Severity : enhancement
637 Description: File striping can now be set to use an arbitrary pool of OSTs.
639 Severity : enhancement
641 Description: Export bytes_read/bytes_write count on OSC/OST.
645 Description: Early reply size mismatch, MGC loses connection
646 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
647 the connect flags are properly negotiated.
652 Description: quotas are not honored with O_DIRECT
653 Details : all writes with the flag O_DIRECT will use grants which leads to
654 this problem. Now using OBD_BRW_SYNC to guard this.
658 Description: add quota statistics
659 Details : 1. sort out quota proc entries and proc code.
660 2. add quota statistics
662 Severity : enhancement
664 Description: enable quota support for HEAD.
668 Description: Properly propagate oinfo flags from lov to osc for statfs
669 Details : restore missing copy oi_flags to lov requests.
671 Severity : enhancement
673 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
674 : l_getgroups(8), lst(8), routerstat(8)
675 Details : included man pages for llobdstat(8), llstat(8),
676 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
678 Severity : enhancement
680 Description: configure's --enable-quota should check the
681 : kernel .config for CONFIG_QUOTA
682 Details : configure is terminated if --enable-quota is passed but
683 : no quota support is in kernel
687 Description: Remove portals compatibility
688 Details : Remove portals compatibility, not interoperable with releases
693 Description: Resolve device initialization race
694 Details : Prevent proc handler from accessing devices added to the
695 obd_devs array but yet be intialized.
697 Severity : enhancement
699 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
701 Severity : enhancement
703 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
708 Description: Using wrong pointer in osc_brw_prep_request
709 Details : Access to array[-1] can produce panic if kernel compiled with
710 CONFIG_PAGE_ALLOC enabled
714 Description: Client runs out of low memory
715 Details : Consider only lowmem when counting initial number of llap pages
719 Description: *optional* service tags registration
720 Details : if the "service tags" package is installed on a Lustre node
721 When the filesystem is mounted, a local-node service tag will
722 be created. See http://inventory.sun.com/ for more information
723 about the Service Tags asset management system.
727 Description: Kernel BUG tries to release flock
728 Details : Lustre does not destroy flock lock before last reference goes
729 away. So always drop flock locks when client is evicted and
730 perform unlock regardless of successfulness of speaking to MDS.
734 Description: add refcount protection for osc callbacks, avoid panic on shutdown
738 Description: sanity test 65a fails if stripecount of -1 is set
739 Details : handle -1 striping on filesystem in ll_dirstripe_verify
744 Description: ASSERTION(CheckWriteback(page,cmd)) failed
745 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
748 Severity : enhancement
750 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
754 Description: do not process already freed flock
755 Details : flock can possibly be freed by another thread before it reaches
756 to ldlm_flock_completion_ast.
760 Description: LBUG during stress test
761 Details : Need properly lock accesses the flock deadlock detection list.
765 Description: oops in page fault handler
766 Details : kernel page fault handler can return two special 'pages' in error
767 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
771 Description: timeout with invalidate import.
772 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
773 should be handled by ptlrpcd. This produce long age waiting and
774 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
776 Severity : enhancement
778 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
782 Description: enable MGS and MDT services start separately
783 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
784 a co-located MGS without starting the MGS, which is a complement
785 to 'nosvc' mount option.
789 Frequency : after recovery
790 Description: precreate to many object's after del orphan.
791 Details : del orphan st in oscc last_id == next_id and this triger growing
792 count of precreated objects. Set flag LOW to skip increase count
793 of precreated objects.
797 Frequency : rare, on clear nid stats
798 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
799 Details : when clean nid stats sometimes try destroy live entry,
800 and this produce panic in free.
804 Description: Stack overflow during MDS log replay
805 ease stack pressure by using a thread dealing llog_process.
809 Description: wait until IO finished before start new when do lock cancel.
810 Details : VM protocol want old IO finished before start new, in this case
811 need wait until PG_writeback is cleared until check dirty flag and
812 call writepages in lock cancel callback.
814 Severity : enhancement
816 Description: using special macro for print time and cleanup in includes.
820 Description: mds_mfd_close() ASSERTION(rc == 0)
821 Details : In mds_mfd_close(), we need protect inode's writecount change
822 within its orphan write semaphore to prevent possible races.
826 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
830 Frequency : rare, on shutdown ost
831 Description: don't hit live lock with umount ost.
832 Details : shrink_dcache_parent can be in long loop with destroy dentries,
833 use shrink_dcache_sb instead.
837 Description: don't panic with use echo client
838 Details : echo client pass NULL as client nid pointer and this produce null
843 Description: fix build on ppc32
844 Details : compile code with -m64 flag produce wrong object file for ppc32.
848 Description: add message levels for liblustreapi
852 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
853 Details : Move the 'good_osts' check before the 'total_bavail' check. This
854 will result in an -EAGAIN and in the exit call path we call
855 alloc_rr() which will with increasing aggressiveness attempt to
856 aquire precreated objects on the minimum number of required OSCs.
860 Description: Use old size assignment to avoid deadlock
861 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
862 the scheduling while holding a spinlock. We do not need locking
863 for size in ll_update_inode() because size is only updated from
864 the MDS for directories or files without objects, so there is no
865 other place to do the update, and concurrent access to such inodes
866 are protected by the inode lock.
870 Description: resolve "_IOWR redefined" build error on SLES10
874 Description: dump the memory debugging after all modules are unloaded to
875 suppress false negative in conf_sanity test 39
877 Severity : enhancement
879 Description: build kernel-ib packages for OFED 1.3 in our release cycle
884 Description: fix SLES kernel versioning
885 Details : the kernel version for our SLES 10 kernel did not include a "-"
886 before the "smp" at the end. while this was not a problem in
887 general, it did mean that software trying to use the kernel
888 version to try to detect a vendor specific kernel would fail.
889 this was most evident by the OFED build scripts.
893 Description: Don't update lov_desc members until making sure they are valid
894 Details : When updating lov_desc members via proc fs, need fix their
895 validities before doing the real update.
899 Description: don't put request into delay list while invalidate in flight.
900 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
901 import in flight. this produce timeout for invalidate and sometimes
902 can cause stale data.
906 Frequency : on ppc only
907 Description: not convert ost objects for directory because it's not exist.
908 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
912 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
914 Severity : enhancement
916 Description: Update to support 2.6.22.14 vanilla kernel.
920 Frequency : rare, on recovery
921 Description: read procfs can produce deadlock in some situation
922 Details : Holding lprocfs lock which send rpc can produce block for destroy
923 obd objects and this also block reconnect with -EALREADY. This
924 isn't fix all lprocfs bugs - but make it rare.
926 Severity : enhancement
928 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
931 Frequency : frequent on X2 node
933 Description: mdc_set_open_replay_data LBUG
934 Details : Set replay data for requests that are eligible for replay.
938 Description: lustre_mgs: operation 101 on unconnected MGS
939 Details : When MGC is disconnected from MGS long enough, MGS will evict the
940 MGC, and late on MGC cannot successfully connect to MGS and a lot
941 of the error messages complaining that MGS is not connected.
945 Frequency : on network error
946 Description: panic with double free request if network error
947 Details : mdc_finish_enqueue is finish request if any network error ocuring,
948 but it's true only for synchronus enqueue, for async enqueue
949 (via ptlrpcd) this incorrect and ptlrpcd want finish request
952 Severity : enhancement
954 Description: client-side metadata stat-ahead during readdir(directory readahead)
955 Details : perform client-side metadata stat-ahead when the client detects
956 readdir and sequential stat of dir entries therein
959 Frequency : on start mds
961 Description: Implement get_info(last_id) in obdfilter.
964 Frequency : occasional
966 Description: Correctly check stale fid, not start epoch if ost not support SOM
967 Details : open with flag O_CREATE need set old fid in op_fid3 because
968 op_fid2 was overwritten with new generated fid, but mds can answer
969 with one of these two fids and both is not stale. Setattr
970 incorrectly started an epoch and assume will be called
971 done_writing, but without SOM done_writing ever being called.
974 Frequency : rare, depends on device drivers and load
976 Description: MDS or OSS nodes crash due to stack overflow
977 Details : Code changes in 1.8.0 increased the stack usage of some functions.
978 In some cases, in conjunction with device drivers that use a lot
979 of stack the MDS (or possibly OSS) service threads could overflow
980 the stack. One change which was identified to consume additional
981 stack has been reworked to avoid the extra stack usage.
984 Frequency : occasional
986 Description: Do not fail import if osc_interpret_create gets -EAGAIN
987 Details : If osc_interpret_create got -EAGAIN it immediately exits and
988 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
989 oscc_has_objects and see OSC has no objests and call
990 oscc_internal_create to resend create request.
992 Severity : enhancement
994 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
996 Severity : enhancement
998 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
1001 Frequency : very rare
1003 Description: Fix replay if there is an un-replied request and open
1004 Details : In some cases, older replay request will revert the
1005 mcd->mcd_last_xid on MDS which is used to record the client's
1006 latest sent request.
1008 Severity : enhancement
1010 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
1012 Severity : enhancement
1014 Description: Add rhel5 support to HEAD.
1016 Serverity : enhancement
1018 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
1023 Description: Don't allow skipping OSTs if index has been specified.
1024 Details : Don't allow skipping OSTs if index has been specified, make
1025 locking in internal create lots better.
1029 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
1030 Details : access to bitfield in structure is always rounded to long
1031 and this produce problem with not atomic change any bit.
1035 Description: Lustre make rpms failed.
1036 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
1037 builds Lustre rpms from tarball.
1040 Frequency : rare on shutdown ost
1042 Description: If llog cancel was not send before clean_exports phase, this can
1043 produce deadlock in llog code.
1044 Details : If llog thread has last reference to obd and call class_import_put
1045 this produce deadlock because llog_cleanup_commit_master wait when
1046 last llog_commit_thread exited, but this never success because was
1047 called from llog_commit_thread.
1051 Description: allow userland application know is lost one of stripes.
1052 Details : fill lvb_blocks with error code on ost and return it to
1053 application if error flag found.
1057 Description: NULL lov_tgts causing MDS oops
1058 Details : more safe checks for NULL lov_tgts for avoid oops.
1060 Severity : enhancement
1062 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
1066 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
1067 Details : If an OST is down the MDS will hang indefinitely in
1068 obd_statfs_rqset() waiting for the statfs data. While for
1069 MDS QOS usage of statfs, it should not stuck in waiting.
1071 Severity : enhancement
1073 Description: remote_acl support
1074 Details : Support ACL-based permission check for remote user.
1075 Support setfacl/getfacl for remote user with the utils
1076 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
1077 the system "{s,g}etfacl" utils.
1079 Severity : enhancement
1081 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
1083 Severity : enhancement
1085 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
1089 Description: make mgs_setparam() handle fsname containing dash
1090 Details : fsname containing a dash does not work with lctl conf_param
1092 Severity : enhancement
1094 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
1096 Severity : enhancement
1098 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
1102 Frequency : rare, at shutdown
1103 Description: access already free / zero obd_namespace.
1104 Details : if client_disconnect_export was called without force flag set,
1105 and exist connect request in flight, this can produce access to
1106 NULL pointer (or already free pointer) when connect_interpret
1107 store ocd flags in obd_namespace.
1111 Frequency : only at startup
1112 Description: not alloc memory with spinlock held.
1113 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
1114 if any spinlock held.
1116 Severity : enhancement
1118 Description: make lustre randomly fail allocating memory
1119 Details : Make lustre randomly failed allocating memory for testing purpose.
1121 Severity : enhancement
1123 Description: lost problems with lov objid file
1124 Details : Fixes some scability and access to not inited memory problems
1125 in work with lov objdid file.
1130 Description: lfs find does not continue on file error
1131 Details : Continue other files processing when a file/dir is absent.
1135 Description: Inconsistent usage of lustre_pack_reply()
1136 Details : Standardize the usage of lustre_pack_reply() such that it
1137 always generate a CERROR on failure.
1140 Frequency : occasional
1142 Description: MDS hang or stay in waiting lock
1143 Details : If client receive lock with CBPENDING flag ldlm need send lock
1144 cancel as separate rpc, to avoid situation when cancel request
1145 can't processed due all i/o threads stay in wait lock.
1149 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
1153 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
1157 Description: Update to SLES9 kernel-2.6.5-7.287.3.
1161 Description: Update to RHEL4 latest kernel.
1163 Severity : enhancement
1165 Description: Build SLES10 patchless client fails
1166 Details : The configure was broken by run ./configure with
1167 --with-linux-obj=.... argument for patchless client. When the
1168 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
1169 can't search header adequately. Use absolute path such as
1170 -I($LINUX)/include instead.
1174 Description: interrupt oig_wait produce painc on resend.
1175 Details : brw_redo_request can be used for resend requests from ptlrpcd and
1176 private set, and this produce situation when rq_ptlrpcd_data not
1177 copyed to new allocated request and triggered LBUG on assert
1178 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
1179 ptlrpcd set if request is changed and can be safety changed to use
1184 Description: LASSERT_{REQ,REP}SWAB macros are buggy
1185 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
1186 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
1187 these macros and replace them with their difinitions instead.
1191 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
1192 Details : Update which_patch & target file for SLES10 latest kernel.
1195 Description: Remove client patches from SLES 10 kernel.
1196 Details : This causes SLES 10 clients to behave as patchless clients
1197 even on a Lustre-patched (server) kernel.
1199 Severity : enhancement
1201 Description: self-adjustable client's lru lists
1202 Details : use adaptive algorithm for managing client cached locks lru
1203 lists according to current server load, other client's work
1204 pattern, memory activities, etc. Both, server and client
1205 side namespaces provide number of proc tunables for controlling
1208 Severity : enhancement
1210 Description: light-weight GSS support
1211 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
1212 and reduce performance overhead.
1214 Severity : enhancement
1216 Description: Linux keyring support
1217 Details : Support using service of Linux keyring for Lustre GSS internal
1218 context refresh/cache mechanism.
1222 Description: Fix errors in lfs documentation
1223 Details : Fixes man pages
1227 Description: don't use GFP_* in generic Lustre code.
1228 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
1232 Description: obdclass is limited by single OBD_ALLOC(idarray)
1233 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
1237 Description: warnings with build patchless client with vanila 2.6.19 and up
1238 Details : change old ctl_table style and replace ctl_table/ctl_table_header
1239 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1243 Frequency : Only for SLES
1244 Description: sanity_quota fail test_1
1245 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1246 file, which makes TSTID[2] inunique.
1250 Frequency : Only for SLES9
1251 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1252 Details : Update target/ChangeLog/which_patch .
1256 Frequency : Only for test_10 in sanity-quota.sh
1257 Description: a bug in quota test code
1258 Details : checking a test flag in a wrong place
1262 Frequency : Only for test_13 in sanity-quota.sh
1263 Description: a bug in quota test script
1264 Details : shouldn't check overfull allocated quota
1268 Description: jbd statistics
1269 Details : Port older jbd statistics patch for sles10
1271 Severity : enhancement
1273 Description: Add printing inode info into message about error in writepage.
1278 Description: reply_lock_interpret crash due to race with it and lock cancel.
1279 Details : Do not replay locks that are being cancelled. Do not reference
1280 locks by their address during replay, just by their handle.
1284 Frequency : When flocks are used.
1285 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1286 Details : Flock locks might destroy just granted lock if it could be merged
1287 with another existing flock, this is done in completion handler,
1288 so teach ldlm_cli_enquque_fini that this is a valid case for
1294 Description: Oops in read and write path when failing to allocate lock.
1295 Details : Check if lock allocation failed and return error back.
1299 Description: lstripe command fails for valid OST index
1300 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1301 instead of lov->desc.ld_active_tgt_count.
1305 Description: sanity.sh failed test 103
1306 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1311 Description: df doesn't work properly if diskfs blocksize != 4K
1312 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1316 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1317 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1319 Severity : enhancement
1321 Description: Adaptive timeouts
1322 Details : RPC timeouts adapt to changing server load and network
1323 conditions to reduce resend attempts and improve recovery time.
1327 Description: llapi_file_create() does not allow some changes
1328 Details : add llapi_file_open() that allows specifying the mode and
1329 open flags, and also returns an open file handle.
1333 Description: merge and cleanup kernel patches.
1334 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1338 Description: Add journal checksum support.(Kernel part)
1339 Details : The journal checksum feature adds two new flags i.e
1340 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1341 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1342 indicates that the commit block contains the checksum for
1343 the blocks described by the descriptor blocks. Now commit
1344 record can be sent to disk without waiting for descriptor
1345 blocks to be written to disk. This behavior is controlled
1346 using JBD2_FEATURE_ASYNC_COMMIT flag.
1350 Description: OSS needs mutliple precreate threads
1351 Details : Add ability to start more than one create thread per OSS.
1355 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1356 Details : execute lfs setstripe on client
1360 Description: mds_obd_create error creating tmp object
1361 Details : When the user sets quota on root, llog will be affected and can't
1362 create files and write files.
1365 Frequency : Always on ia64 patchless client, and possibly others.
1367 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1368 Details : This allows the patchless client to be loaded on architectures
1369 without this export.
1373 Description: RedHat Update kernel for RHEL5
1374 Details : Add the kernel config file for RHEL5.
1378 Description: RedHat Update kernel for RHEL5
1379 Details : Modify the kernel config file more closer RHEL5.
1383 Description: Build failure against Centos5 (RHEL5)
1384 Details : Define PAGE_SIZE when it isn't present.
1388 Description: test_7 on sanity-quota.sh can't be used on separate machines
1389 Details : add facet to handle it
1393 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1394 Details : with NFS, the anon dentry's parent was set to itself in
1395 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1396 corresponding dentry other than use rec->ur_name.
1398 Severity : enhancement
1400 Description: lfs setstripe enhancement
1401 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1405 Description: enable data checksumming by default
1406 Details : enable checksum by default, allow --disable-checksum
1407 configure option and "-o nochecksum" mount option. Checksums
1408 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1412 Description: lustre support for RHEL5
1413 Details : Add support for RHEL5.
1417 Description: Client eviction due to failover config
1418 Details : after a connection loss, the lustre client should attempt to
1419 reconnect to the last active server first before trying the
1420 other potential connections.
1424 Description: when mds and osts use different quota unit(32bit and 64bit),
1425 quota will be released repeatly.
1426 Details : void sending multiple quota reqs to mds, which will keep the
1427 status between the reqs.
1431 Description: osts not allocated evenly to files
1432 Details : change the condition to increase offset_idx
1435 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1437 Description: Data corruption for OSTs that are formatted larger than 2TB
1439 Details : When generating the bio request for lustre file writes the
1440 sector number would overflow a temporary variable before being
1441 used for the IO. The data reads correctly from Lustre (which
1442 will overflow in a similar manner) but other file data or
1443 filesystem metadata may be corrupted in some cases.
1447 Description: Tune the kernel for good SCSI performance.
1448 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1449 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1454 Description: rewrite ext2-derived code in obdclass/uuid.c
1455 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1456 from scratch preserving functionality.
1460 Description: "lfs find -obd UUID" prints directories
1461 Details : "lfs find -obd UUID" will return all directory names instead
1462 of just file names. It is incorrect because the directories
1463 do not reside on the OSTs.
1467 Description: use i_size_read and i_size_write in 2.6 port
1468 Details : replace inode->i_size access with i_size_read/write()
1471 Frequency : when removing large files
1473 Description: scheduling issue during removal of large Lustre files
1474 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1475 It causes scheduling issues when removing large files (17TB in the
1479 Frequency : only with liblustre clients on XT3
1481 Description: evictions taking too long
1482 Details : allow llrd to evict clients directly on OSTs
1485 Frequency : only on ppc
1487 Description: /proc/fs/lustre/devices broken on ppc
1488 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1489 We should make sure the type of 'index' is loff_t and then cast
1490 explicitly as needed below. Do not assign an explicitly cast
1494 Frequency : only for rhel5
1496 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1497 Details : Modify the target file & which_kernel.
1499 Severity : enhancement
1501 Description: omit set fsid for export NFS
1502 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1503 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1507 Frequency : Always, for kernels after 2.6.16
1508 Description: Fix warning idr_remove called for id=.. which is not allocated.
1509 Details : Last kernels save old s_dev before kill super and not allow
1510 to restore from callback - restore it before call kill_anon_super.
1514 Description: buffer overruns could theoretically occur
1515 Details : llapi_semantic_traverse() modifies the "path" argument by
1516 appending values to the end of the origin string, and a
1517 overrun may occur. Adding buffer overrun check in liblustreapi.
1521 Description: Fix error on 'ls .' at the top of the Lustre mount.
1522 Details : Don't revalidate dentry if it is a root dentry.
1526 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1527 Details : Modify vm-tunables-rhel4.patch.
1531 Description: Kernel config for 2.6.18-vanilla.
1532 Details : Modify targets/2.6-vanilla.target.in.
1533 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1534 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1535 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1536 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1540 Description: improve handling recoverable errors
1541 Details : if request processig with error which can be recoverable on server
1542 request should be resend, otherwise page released from cache and
1547 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1548 Details : Modify target file & which_patch.
1549 A flaw was found in the IA32 system call emulation provided
1550 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1551 value could be stored in the %RAX register, which could trigger an
1552 out-of-bounds system call table access. An untrusted local user
1553 could exploit this flaw to run code in the kernel
1554 (ie a root privilege escalation). (CVE-2007-4573).
1558 Description: O_DIRECT bypasses client statistics.
1559 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1560 read_ahead_stats were updated. copied the stats section from
1561 osc_send_oap_rpc() into async_internal().
1565 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1569 Description: change order of libsysio includes
1570 Details : '#include sysio.h' should always come before '#include xtio.h'
1574 Description: handle "serious error: objid * already exists" more gracefully
1575 Details : If LAST_ID value on disk is smaller than the objects existing in
1576 the O/0/d* directories, it indicates disk corruption and causes an
1577 LBUG(). If the object is 0-length, then we should use the existing
1578 object. This will help to avoid a full fsck in most cases.
1580 Severity : enhancement
1582 Description: adapt the lustre_config script to support the upgrade case
1583 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1584 server targets to 1.6 in parallel.
1588 Description: To avoid grant space > avaible space when the disk is almost
1589 full. Without this patch you might see the error "grant XXXX >
1590 available" or some LBUG about grant, when the disk is almost
1592 Details : In filter_check_grant, for non_grant cache write, we should
1593 check the left space by if (*left > ungranted + bytes), instead
1594 of (*left > ungranted), because only we are sure the left space
1595 is enough for another "bytes", then the ungrant space should be
1596 increase. In client, we should update cl_avail_grant only there
1597 is OBD_MD_FLGRANT in the reply.
1601 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1602 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1603 described in CVE-2007-4573. This problem could allow untrusted
1604 local users to gain root access.
1607 Frequency : when using O_DIRECT and quotas
1609 Description: Incorrect file ownership on O_DIRECT output files
1610 Details : block usage reported by 'lfs quota' does not take into account
1611 files that have been written with O_DIRECT.
1616 Description: touch file failed when fs is not full
1617 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1618 otherwise we can get ENOSP while fs is not full.
1622 Description: parallel lock callbacks
1623 Details : Instead of sending blocking and completion callbacks as separated
1624 requests, adding them to a set and sending in parallel.
1627 Frequency : only for Cray XT3
1628 Bugzilla : 12829/13455
1629 Description: Changing primary group doesn't change the group lustre assigns to
1631 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1632 overridden with the primary group provided by the group upcall,
1633 whereas the supplied fsgid can be trusted if it is in the list of
1634 supplementary groups returned by the group upcall.
1636 Severity : enhancement
1638 Description: Allow masking D_WARNING, D_ERROR messages from console
1639 Details : Console messages can now be disabled via lnet.printk.
1643 Description: User code with malformed file open parameter crashes client node
1644 Details : Before packing join_file req, all the related reference should be
1645 checked carefully in case some malformed flags cause fake
1646 join_file req on client.
1650 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1651 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1652 mode is equal to the granted mode. However, it should also check
1653 that LDLM_FL_CP_REQD is not set.
1657 Description: Heavy nfs access might result in deadlocks
1658 Details : After ELC code landed, it is now improper to enqueue any mds
1659 locks under och_sem, because enqueue might want to decide to
1660 cancel open locks for same inode we are holding och_sem for.
1664 Description: Client eviction while running blogbench
1665 Details : A lot of unlink operations with concurrent I/O can lead to a
1666 deadlock causing evictions. To address the problem, the number of
1667 oustanding OST_DESTROY requests is now throttled to
1668 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1669 callbacks are processed in priority.
1673 Description: enable ACLs on MDS by default
1674 Details : ACLs must be enabled on MDS by default.
1677 Frequency : PPC/PPC64 only
1679 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1680 Details : fix duplicated definitions between asm-ppc64/types.h and
1681 lustre_types.h on PPC.
1684 Frequency : PPC/PPC64 only
1686 Description: asm-ppc/segment.h does not exist
1687 Details : fix compile issue on PPC.
1691 Description: better handle error messages in extents code
1694 Frequency : RHEL4 only
1696 Description: mkfs is very slow on IA64/RHEL4
1697 Details : A performance regression has been discovered in the MPT Fusion
1698 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1699 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1700 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1702 Severity : enhancement
1704 Description: SNMP support enhancement
1705 Details : Adding total number of sampled request for an MDS node in snmp
1708 Severity : enhancement
1710 Description: Optimize ldlm waiting list processing for PR extent locks
1711 Details : When processing waiting list for read extent lock and meeting read
1712 lock that is same or wider to it that is not contended, skip
1713 processing rest of the list and immediatelly return current
1714 status of conflictness, since we are guaranteed there are no
1715 conflicting locks in the rest of the list.
1719 Description: Time out and refuse to reconnect
1720 Details : When the failover node is the primary node, it is possible
1721 to have two identical connections in imp_conn_list. We must
1722 compare not conn's pointers but NIDs, otherwise we can defeat
1723 connection throttling.
1727 Description: port llog fixes from b1_6 into HEAD
1728 Details : Port llog reference couting and some llog cleanups from b1_6
1729 (bug 10800) into HEAD, for protect from panic and access to already
1730 free llog structures.
1734 Description: Detect stride IO mode in read-ahead
1735 Details : When a client does stride read, read-ahead should detect that and
1736 read-ahead pages according to the detected stride pattern.
1740 Description: data checksumming impacts single node performance
1741 Details : add support for several checksum algorithm. Currently, only CRC32
1742 and Adler-32 are supported. The checksum type can be changed on
1743 the fly via /proc/fs/lustre/osc/*/checksum_type.
1747 Description: use adler32 for page checksums
1748 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1753 Description: build for x2 fails
1754 Details : fix compile issue on Cray systems.
1758 Description: Properly match for duplicate locks
1759 Details : Due to different lock order from skiplists code, we need to
1760 traverse entire list for now
1763 Frequency : only on PPC/SLES10
1765 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1766 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1767 be defined. Add a hack in mkfs_lustre.c to work around this compile
1772 Description: LASSERT on MDS when client holding flock lock dies
1773 Details : ldlm pool logic depends on number of granted locks equal to
1774 number of released locks which is not true for flock locks, so
1775 just exclude such locks from consideration.
1779 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1780 Details : Use fsfilt_sync() for both the whole filesystem sync and
1781 individual file sync to eliminate dangerous inode locking
1782 with I_LOCK that can lead to a deadlock.
1786 Description: performance in 1.6.3
1787 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1788 sure that 1MB requests are not fragmented by the block layer.
1790 Severity : enhancement
1792 Description: organize the server-side client stats on per-nid basis
1793 Details : Change the structure of stats under obdfilter and mds to
1801 The "uuid"s file would list the uuids of _active_ exports.
1802 And the clear entry is to clear all stats and stale nids.
1804 Severity : enhancement
1806 Description: eliminate client locks in face of contention
1807 Details : file contention detection and lockless i/o implementation
1808 for contended files.
1812 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1815 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1817 Description: MDS may lose file striping (and hence file data) in some cases
1818 Details : If there are additional extended attributes stored on the MDS,
1819 in particular ACLs, SELinux, or user attributes (if user_xattr
1820 is specified for the client mount options) then there is a risk
1821 of attribute loss. Additionally, the Lustre file striping
1822 needs to be larger than default (e.g. striped over all OSTs),
1823 and an additional attribute must be stored initially in the
1824 inode and then increase in size enough to be moved to the
1825 external attribute block (e.g. ACL growing in size) for file
1830 Description: skiplist implementation simplification
1831 Details : skiplists are used to group compatible locks on granted list
1832 that was implemented as tracking first and last lock of each
1833 lock group the patch changes that to using doubly linked lists
1837 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1838 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1839 because LASSERT on both the data supplied by a client, and the
1840 data on disk is dangerous and incorrect.
1842 Severity : enhancement
1844 Description: Slow truncate/writes to huge files at high offsets.
1845 Details : Directly associate cached pages to lock that protect those pages,
1846 this allows us to quickly find what pages to write and remove
1847 once lock callback is received.
1851 Description: more ldlm soft lockups
1852 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1853 starve other threads from the resource lock for a long time in
1854 case of long waiting queue, so change the debug level from
1855 D_OTHER to the less frequently used D_INFO.
1857 Severity : enhancement
1859 Description: add -gid, -group, -uid, -user options to lfs find
1863 Description: Hung threads in invalidate_inode_pages2_range
1864 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1865 once one is completed. As a result, some RPCs are stuck in the
1866 queue and are never sent.
1870 Description: filter threads hungs on waiting journal commit
1871 Details : Cleanup filter group llog code, then only filter group llog will
1872 be only created in the MDS/OST syncing process.
1876 Description: Procfs and llog threads access destoryed import sometimes.
1877 Details : Sync the import destoryed process with procfs and llog threads by
1878 the import refcount and semaphore.
1880 Severity : enhancement
1882 Description: openlock cache of b1_6 port to HEAD
1887 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1888 Details : If insertion of an extent fails, then discard the inode
1889 preallocation and free data blocks else it can lead to duplicate
1894 Description: don't always update ctime in ext3_xattr_set_handle()
1895 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1896 In some cases the ctime should not be updated, for example for
1897 2.0->1.8 compatibility it is necessary to delete an xattr and it
1898 should not update the ctime.
1902 Bugzilla : 15713/16362
1903 Description: Assertion in iopen_connect_dentry in 1.6.3
1904 Details : looking up an inode via iopen with the wrong generation number can
1905 populate the dcache with a disconneced dentry while the inode
1906 number is in the process of being reallocated. This causes an
1907 assertion failure in iopen since the inode's dentry list contains
1908 both a connected and disconnected dentry.
1912 Description: assertion failure in ldlm_handle2lock()
1913 Details : fix a race between class_handle_unhash() and class_handle2object()
1914 introduced in lustre 1.6.5 by bug 13622.
1919 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1920 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1925 Description: oops in sd_iostats_seq_show()
1926 Details : unloading/reloading the scsi low level driver triggers a kernel
1927 bug when trying to access the sd iostat file.
1932 Description: Kernel panics during QLogic driver reload
1933 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1934 patch, causing memory corruption.
1939 Description: journal_dev option does not work in b1_6
1940 Details : pass mount option during pre-mount.
1942 Severity : enhancement
1944 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1945 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1946 extent information of a file. It can be used to map logical blocks
1947 in a file to physical blocks in the block device.
1951 Description: LDLM soft lockups - improvement
1952 Details : It is be possible to send the lock handle along with each read
1953 or write request because the client is already doing a lock match
1954 itself so there isn't any reason the OST should have to re-do that
1958 Frequency : only on Cray X2
1960 Description: X2 build failures
1961 Details : fix build failures on Cray X2.
1965 Description: xid & resent requests
1966 Details : Initialize RPC XID from clock at startup (randomly if clock is
1969 Severity : enhancement
1971 Description: Add lustre_start utility to start or stop multiple Lustre servers
1976 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1977 Details : In case of memory pressure, list_del() can be called twice on
1978 req->rq_history_list, causing a kernel oops.
1982 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1983 Details : fix stack overflow in the distributed lock manager by defering
1984 export eviction after a failed AST to the elt thread instead of
1985 handling it in the dlm interpret routine.
1989 Description: Convert some comments to new format.
1990 Details : Update documenting comments to match doxygen conventions.
1994 Description: Grammar fixes.
1995 Details : A couple of trivial spelling fixes.
1999 Description: OSD_COUNTERS-mandatory
2000 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
2001 to simplify the code.
2005 Description: simplify cmm_device freeing logic.
2006 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
2010 Description: Add lockdep support to dt_object_operations locking interface.
2011 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
2012 parameter indicating lock ordering. Update mdd code to use new
2017 Description: Introduce failloc constants for lockless IO tests.
2018 Details : Add two new failloc constants to test lockless IO. Only one of
2019 them in implemented---another is checked in yet to be landed
2024 Description: Add lockdep support for inode mutex.
2025 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
2026 in the situations where Lustre has to lock more than one inode
2031 Description: Add optional invariants checking support.
2032 Details : Add new LINVRNT() macro, optional on new --enable-invariants
2033 configure switch. This macro is to be used for consistency and
2034 sanity checks that are too expensive to be left in `production'
2039 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
2040 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
2045 Description: Add lockdep support for ldlm_lock and ldlm_resource.
2046 Details : Use spin_lock_nested() in (the only) situation where more than
2047 one ldlm_lock is locked simultaneously. Also, fix possible
2048 dead-lock in ldlm_lock_change_resource() by enforcing particular
2053 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
2054 Details : Instead of specifying each ldlm_lock call-back through separate
2055 parameter, wrap them into struct ldlm_callback_suite.
2059 Description: Kill join_lru obd method and its callers.
2060 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
2061 in memory---a requirement implemented through ->o_join_lru() obd
2062 method in HEAD. Remove this method and its users.
2066 Description: Add asynchronous ldlm ENQUEUE completion handler.
2067 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
2068 case that stock ldlm_completion_ast() cannot handle as it waits
2069 until lock is granted. Introduce new ldlm_completion_ast_async()
2070 for this. Also comment ldlm_completion_ast().
2074 Description: ldlm_error <-> errno conversion.
2075 Details : Add functions to map (rather arbitrary) between LDLM error codes
2076 and standard errno values. CLIO needs this to prevent LDLM specific
2077 constants from escaping ldlm and osc.
2081 Description: Kill unused ldlm_handle2lock_ns() function.
2082 Details : Kill unused ldlm_handle2lock_ns() function.
2086 Description: Add lu_ref support to ldlm_lock
2087 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
2088 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
2089 and ldlm_resource. LDLM interface has to be changed, because code that
2090 releases a reference on a lock, has to "know" what reference this is.
2091 In the most frequent case
2093 lock = ldlm_handle2lock(handle);
2095 LDLM_LOCK_PUT(lock);
2097 no changes are required. When any other reference (received _not_ from
2098 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
2099 instead of LDLM_LOCK_PUT().
2101 Arguably, changes are pervasive, and interface requires some discipline
2102 for proper use. On the other hand, it was very instrumental in finding
2103 a few leaked lock references.
2107 Description: Add ldlm_lock_addref_try().
2108 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
2109 attempts to addref a lock that might be being canceled
2114 Description: Add ldlm_weigh_callback().
2115 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
2116 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
2117 hard-coded `number of pages' logic.
2121 Description: Add lockdep annotations to llog code.
2122 Details : Use appropriately tagged _nested() locking calls in the places
2123 where llog takes more than one ->lgh_lock lock.
2127 Description: Add loi_kms_set().
2128 Details : Wrap kms updates into a helper function.
2132 Description: Constify instances of struct lsm_operations.
2133 Details : Constify instances of struct lsm_operations.
2137 Description: lu_conf support.
2138 Details : On a server, a file system object is uniquely identified
2139 by a fid, which is sufficient to locate and load all object
2140 state (inode). On a client, on the other hand, more data are
2141 necessary instantiate an object. Change lu_object_find() and
2142 friends to take additional `lu_conf' argument describing object.
2143 Typically this includes layout information.
2147 Description: lu_context fixes.
2148 Details : Introduce new lu_context functions that are needed on the client
2149 side, where some system threads (ptlrpcd) are shared by multiple
2150 modules, and so cannot be stopped during module shutdown.
2154 Description: Add start and stop methods to lu_device_type_operations.
2155 Details : Introduce two new methods in lu_device_type_operations, that are
2156 invoked when first instance of a given type is created and last one
2157 is destroyed respectively. This is need by CLIO.
2161 Description: Add lu_ref support to struct lu_device.
2162 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
2163 track leaked references.
2167 Description: Introduce lu_kmem_descr.
2168 Details : lu_kmem_descr and its companion interface allow to create
2169 and destroy a number of kmem caches at once.
2173 Description: Fix lu_object finalization race.
2174 Details : Fix a race between lu_object_find() finding an object and its
2175 concurrent finalization. This race is (most likely) not possible
2176 on the server, but might happen on the client.
2180 Description: Introduce lu_ref interface.
2181 Details : lu_ref is a debugging module allowing to track references to
2182 a given object. It is quite cpu expensive, and has to be
2183 explicitly enabled with --enable-lu_ref. See usage description
2188 Description: Factor lu_site procfs stats into a separate function.
2189 Details : Separate lu_site stats printing code into a separate function
2190 that can be reused on a client.
2194 Description: Constify instances of struct {lu,dt,md}_device_operations.
2195 Details : Constify instances of struct {lu,dt,md}_device_operations.
2199 Description: Introduce struct md_site and move meta-data specific parts of
2200 struct lu_site here.
2201 Details : Move md-specific fields out of struct lu_site into special struct
2202 md_site, so that lu_site can be used on a client.
2206 Description: Kill mdd_lov_destroy().
2207 Details : Remove unused mdd code.
2211 Description: Add st_block checking to multistat.c.
2212 Details : Add st_block checking to multistat.c.
2216 Description: Add lu_ref support to struct obd_device.
2217 Details : Add lu_ref tracking to obd_device.
2221 Description: Kill obd_set_fail_loc().
2222 Details : Remove unused code.
2226 Description: Add special type for ptlrpc_request interpret functions.
2227 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
2228 there. Actual usage will be in CLIO.
2232 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
2233 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
2234 doesn't work with lockdep.
2238 Description: Add rwv.c test program.
2239 Details : New testing program exercising readv(2) and writev(2) (Qian).
2243 Description: Add sendfile.c test program.
2244 Details : New testing program exercising sendfile(2) (Jay).
2248 Description: Ratelimit a message that can be very frequent.
2249 Details : Ratelimit a memory allocation failure message that can
2254 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2255 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2259 Description: Liblustre build fixes.
2260 Details : Liblustre build fixes.
2264 Description: libcfs: add cfs_{need,cond}_resched() interface.
2265 Details : libcfs: add cfs_{need,cond}_resched() definition and
2266 implementations for Linux, NT, and liblustre.
2268 Severity : enhancement
2270 Description: More exported tunables for mballoc
2271 Details : Add support for tunable preallocation window and new tunables for
2272 large/small requests
2276 Description: Detect corruption of block bitmap and checking for preallocations
2277 Details : Checks validity of on-disk block bitmap. Also it does better
2278 checking of number of applied preallocations. When corruption is
2279 found, it turns filesystem readonly to prevent further corruptions.
2283 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2284 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2285 window to 0. If it does hit cache miss, and read-pattern is still
2286 stride-io mode, does not reset the stride window, but also does not
2287 increase the stride window length in this case.
2291 Frequency : only for big-endian servers
2292 Description: Check if system is big-endian while mounting fs with extents feature
2293 Details : Mounting a filesystem with extents feature will fail on big-endian
2294 systems since ext3-based ldiskfs is not supported on big-endian
2295 systems. This can be over-riden with "bigendian_extents" mount option.
2297 Severity : enhancement
2299 Description: The root squash functionality
2300 Details : A security feature, which is to prevent users from being able
2301 to mount lustre on their desktop, run as root, and delete
2302 all of the files in the filesystem. The goal is accomplished by
2303 remapping user id (UID) and group id (GID) of the root user to
2304 a UID and GID specified by the system administartor via Lustre
2305 configuration management server (MGS). The functionality also
2306 allows to specify sets of clients for which the remapping does
2311 Description: Excessive recovery window
2312 Details : With AT enabled, the recovery window can be excessively long (6000+
2313 seconds). To address this problem, we no longer use
2314 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2315 timeout no longer depends on the service time, it is set to
2316 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2317 time via pb_service_time.
2321 Description: Watchdog triggered on MDS failover
2322 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2323 the OSTs know that the MDS "UUID" can be reused for the same export
2324 from a different NID, so we do not need to wait for the export to be
2328 Frequency : rare, only if using MMP with Linux RAID
2330 Description: MMP doesn't work with Linux RAID
2331 Details : While using HA for Lustre servers with Linux RAID, it is possible
2332 that MMP will not detect multiple mounts. To make this work we
2333 need to unplug the device queue in RAID when the MMP block is being
2334 written. Also while reading the MMP block, we should read it from
2335 disk and not the cached one.
2337 Severity : enhancement
2339 Description: open file using fid
2340 Details : A file can be opened using just its fid, like
2341 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2344 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2347 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2348 Details : add proper locking for creation of jbd_4k slab cache
2352 Description: MMP check in ext3_remount() fails without displaying any error
2353 Details : When multiple mount protection fails during remount, proper error
2356 Severity : enhancement
2358 Description: Allow stripe size to be up to 4G-64k
2359 Details : Fix math logic to allow large stripe sizes.
2363 Description: add check for >8TB ldiskfs filesystems
2364 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2365 Don't allow >8TB ldiskfs filesystems to be mounted without
2366 force_over_8tb mount option
2371 Descriptoin: OSS complains "too many blocks" and crashes
2372 Details : In filter_commitrw_write():
2373 1. lquota_chkquota();
2374 2. obdo_from_inode();
2375 3. lquota_pending_commit();
2376 "2." has changed uid and gid in this case, then caused
2377 lquota_chkquota() and lquota_pending_commit() to use
2378 different uid and gid. That is the root of the bug.
2382 Descriptoin: getstripe doesn't report stripe count = -1 sanely
2383 Details : Detect and print out the 'stripe count = -1' case properly.
2387 Description: Out or order replies might be lost on replay
2388 Details : In ptlrpc_retain_replayable_request if we cannot find retained
2389 request with tid smaller then one currently being added, add it
2390 to the start, not end of the list.
2392 --------------------------------------------------------------------------------
2394 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2396 * Support for kernels:
2397 2.6.5-7.283 (SLES 9),
2398 2.6.9-55.EL (RHEL 4),
2399 2.6.16.46-0.14 (SLES 10),
2400 2.6.18.8 vanilla (kernel.org)
2401 * Client support for unpatched kernels:
2402 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2403 2.6.16 - 2.6.22 vanilla (kernel.org)
2404 * Due to recently discovered recovery problems, we do not recommend
2405 using patchless RHEL 4 clients with this or any earlier release.
2406 * Recommended e2fsprogs version: 1.39.cfs8
2407 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2408 * Starting with this release, the ldiskfs backing filesystem required
2409 by Lustre is now in its own package, lustre-ldiskfs. This package
2410 should be installed. It is versioned separately from Lustre and
2411 may be released separately in future.
2415 Description: block reactivating mgc import until all deactivates complete
2416 Details : Fix race when failing back MDT/MGS to itself (testing)
2419 Frequency : at statup only
2421 Description: mds_lov_synchronize race leads to various problems
2422 Details : simultaneous MDT->OST connections at startup can cause the
2423 sync to abort, leaving the OSC in a bad state.
2425 Severity : enhancement
2427 Description: add optional extra BUILD_VERSION info
2428 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2429 to override the lustre version.
2432 Frequency : 2.6.18 servers only
2434 Description: ll_kern_mount() doesn't release the module reference
2435 Details : The ldiskfs module reference count never drops down to 0
2436 because ll_kern_mount() doesn't release the module reference.
2441 Description: server LBUG when using old ost_num_threads parameter
2442 Details : Accept the old ost_num_threads parameter but warn that it
2443 is deprecated, and fix an off-by-one error that caused an LBUG.
2448 Description: Transient SCSI error results in persistent IO issue
2449 Details : iobuf->dr_error is not reinitialized to 0 between two
2453 Frequency : sometimes when underlying device returns I/O errors
2455 Description: OSTs not going read-only during write failures
2456 Details : OSTs are not remounted read-only when the journal commit threads
2457 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2458 instead of the ext3 wrappers.
2462 Description: poor connect scaling with increasing client count
2463 Details : Don't run filter_grant_sanity_check for more than 100 exports
2464 to improve scaling for large numbers of clients.
2467 Frequency : SLES10 only
2469 Description: sanity-quota.sh quotacheck failed: rc = -22
2470 Details : Quotas cannot be enabled on SLES10.
2473 Frequency : liblustre clients only
2475 Description: getdirentries does not give error when run on compute nodes
2476 Details : getdirentries does not fail when the size specified as an argument
2477 is too small to contain at least one entry
2479 Severity : enhancement
2481 Description: Add LNET router traceability for debug purposes
2482 Details : If a checksum failure occurs with a router as part of the
2483 IO path, the NID of the last router that forwarded the bulk data
2484 is printed so it can be identified.
2489 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2490 Details : Due to a race condition, liblustre clients were occasionally
2491 evicted incorrectly.
2493 Severity : enhancement
2495 Description: lfs setstripe use optional parameters instead of postional
2498 Severity : enhancement
2500 Description: Nanosecond timestamp support for ldiskfs
2501 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2502 resolution timestamps. There is not yet support for this at
2503 the Lustre filesystem level.
2506 Frequency : during server recovery
2508 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2509 Details : request with rq_no_resend flag not awake l_wait_event if they get
2513 Frequency : nfs export on patchless client
2515 Description: connectathon hang when test nfs export over patchless client
2516 Details : Disconnected dentry cannot be found with lookup, so we do not need
2517 to unhash it or make it invalid
2520 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2521 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2522 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2527 Description: open req refcounting wrong on reconnect
2528 Details : If reconnect happened between getting open reply from server and
2529 call to mdc_set_replay_data in ll_file_open, we will schedule
2530 replay for unreferenced request that we are about to free.
2531 Subsequent close will crash in variety of ways.
2532 Check that request is still eligible for replay in
2533 mdc_set_replay_data().
2538 Description: disable writes to filesystem when reading health_check file
2539 Details : the default for reading the health_check proc file has changed
2540 to NOT do a journal transaction and write to disk, because this
2541 can cause reads of the /proc file to hang and block HA state
2542 checking on a healthy but otherwise heavily loaded system. It
2543 is possible to return to the previous behaviour during configure
2544 with --enable-health-write.
2546 Severity : enhancement
2548 Description: 64-bit inode version
2549 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2550 to the inode. This will be required for version-based recovery.
2555 Description: MDS fails to start if a duplicate client export is detected
2556 Details : in some rare cases it was possible for a client to connect to
2557 an MDS multiple times. Upon recovery the MDS would detect this
2558 and fail during startup. Handle this more gracefully.
2560 Severity : enhancement
2562 Description: Add -o localflock option to simulate old noflock
2564 Details : This will achieve local-only flock/fcntl locks
2570 Description: log_commit_thread vs filter_destroy race leads to crash
2571 Details : Take import reference before releasing llog record semaphore
2576 Description: Wrong request locking in request set processing
2577 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2578 imp_delayed_list, in this place should be used imp_lock.
2581 Frequency : when reconnection
2583 Description: Grant Leak when osc reconnect to OST
2584 Details : When osc reconnect ost, OST(filter) should check whether it
2585 should grant more space to client by comparing fed_grant and
2586 cl_avail_grant, and return the granted space to client instead
2587 of "new granted" space, because client will call osc_init_grant
2588 to update the client grant space info.
2591 Frequency : when client reconnect to OST
2593 Description: Grant Leak when osc do resend and replay bulk write
2594 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2595 bulk write request, because the grant info while be sync between
2596 OSC and OST when reconnect, and we should ignore the grant info
2597 these of resend/replay write req.
2602 Description: Grant space more than avaiable left space sometimes.
2603 Details : When then OST is about to be full, if two bulk writing from
2604 different clients came to OST. Accord the avaliable space of the
2605 OST, the first req should be permitted, and the second one
2606 should be denied by ENOSPC. But if the seconde arrived before
2607 the first one is commited. The OST might wrongly permit second
2608 writing, which will cause grant space > avaiable space.
2611 Frequency : when client is evicted
2613 Description: Grant might be wrongly erased when osc is evicted by OST
2614 Details : when the import is evicted by server, it will fork another
2615 thread ptlrpc_invalidate_import_thread to invalidate the
2616 import, where the grant will be set to 0. While the original
2617 thread will update the grant it got when connecting. So if
2618 the former happened latter, the grant will be wrongly errased
2619 because of this race.
2624 Description: Checking Stale with correct fid
2625 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2626 to check whether it is stale, because sometimes, we want the
2627 enqueue happened anyway, and op_data.fid2 will not be initialized.
2629 Severity : enhancement
2631 Description: update patchless client
2632 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2635 Frequency : only with 2.4 kernel
2637 Description: random memory corruption
2638 Details : size of struct ll_inode_info is to big for union inode.u and this
2639 can be cause of random memory corruption.
2644 Description: Memory leak in recovery
2645 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2646 It should also check obd_fail before fsfilt_start, otherwise if
2647 fsfilt_start return -EROFS,(failover mds during mds recovery).
2648 then the req will return with repmsg->transno = 0 and rc = EROFS.
2649 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2650 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2651 be freed no matter whether fsfilt_commit success or not.
2654 Frequency : only with huge count clients
2656 Description: Prevents from taking the superblock lock in llap_from_page for
2658 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2659 for llap_from_page call in ll_removepage prevents from taking the
2660 superblock lock for a soon died page.
2665 Description: Not check open intent error before release open handle
2666 Details : in some rare cases, the open intent error is not checked before
2667 release open handle, which may cause
2668 ASSERTION(open_req->rq_transno != 0), because it tries to release
2669 the failed open handle.
2674 Description: Set cat log bitmap only after create log success.
2675 Details : in some rare cases, the cat log bitmap is set too early. and it
2676 should be set only after create log success.
2680 Description: Accessing a block bevice can re-enable I/O when Lustre is
2681 tearing down a device.
2682 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2687 Description: service threads may hog cpus when there are a lot of requests
2688 Details : Insert cond_resched to give other threads a chance to use some CPU
2693 Description: the cat log was not initialized in recovery
2694 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2695 unlink log on mds will not be initialized until mds post
2696 recovery. And also in mds post recovery, the unlink log will
2697 initialization will be done asynchronausly, so there will be race
2698 between add unlink log and unlink log initialization.
2702 Description: brw_stats were being printed incorrectly
2703 Details : brw_stats were being printed as log2 but all of them were not
2704 recorded as log2. Also remove some code duplication arising from
2705 filter_tally_{read,write}.
2709 Frequency : rare, only in recovery.
2710 Description: ASSERTION(req->rq_type != LI_POISON) failed
2711 Details : imp_lock should be held while iterating over imp_sending_list for
2712 prevent destroy request after get timeout in ptlrpc_queue_wait.
2716 Description: replay-single.sh test 52 fails
2717 Details : A lock's skiplist need to be cleanup when it being unlinked
2718 from its resource list.
2722 Description: Short directio read returns full requested size rather than
2724 Details : Direct I/O operations should return actual amount of bytes
2725 transferred rather than requested size.
2727 Severity : enhancement
2729 Description: metadata RPC reduction (e.g. for rm performance)
2730 Details : decrease the amount of synchronous RPC between clients and servers
2731 by canceling conflicing lock before the operation on the client
2732 and packing thier handles into the main operation RPC to server.
2734 Severity : enhancement
2736 Description: Async OSC create to avoid the blocking unnecessarily.
2737 Details : If a OST has no remain object, system will block on the creating
2738 when need to create a new object on this OST. Now, ways use
2739 pre-created objects when available, instead of blocking on an
2740 empty osc while others are not empty. If we must block, we block
2741 for the shortest possible period of time.
2745 Description: improve handling recoverable errors
2746 Details : if request processig with error which can be recoverable on server
2747 request should be resend, otherwise page released from cache and
2750 Severity : enhancement
2752 Description: refine locking for avoid write wrong info into lov_objid file
2753 Details : fix possible races with add new target and write/update data in
2756 --------------------------------------------------------------------------------
2758 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2763 Frequency : on some architectures
2765 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2766 Details : Uninitialized flags sometimes cause configuration commands to
2770 Frequency : patchless clients only
2772 Description: missing __iget() symbol export
2773 Details : The __iget() symbol export is missing. To avoid the need for
2774 this on patchless clients the deathrow inode reaper is turned
2775 off, and we depend on the VM to clean up old inodes. This
2776 dependency was during via the fix for bug 12181.
2778 --------------------------------------------------------------------------------
2780 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2782 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2783 INTEROPERATE with older versions automatically. In many cases a
2784 special upgrade step is needed. Please read the
2785 user documentation before upgrading any part of a live system.
2786 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2787 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2788 * WARNING: Lustre configuration and startup changes are required with
2789 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2793 Severity : enhancement
2795 Description: MountConf
2796 Details : Lustre configuration is now managed via mkfs and mount
2797 commands instead of lmc and lconf. New obd types (MGS, MGC)
2798 are added for dynamic configuration management. See
2799 https://mail.clusterfs.com/wikis/lustre/MountConf for
2802 Severity : enhancement
2804 Description: dynamic OST addition
2805 Details : OSTs can now be added to a live filesystem
2807 Severity : enhancement
2809 Description: startup order invariance
2810 Details : MDTs and OSTs can be started in any order. Clients only
2811 require the MDT to complete startup.
2813 Severity : enhancement
2815 Description: parallel, asynchronous orphan cleanup
2816 Details : orphan cleanup is now performed in separate threads for each
2817 OST, allowing parallel non-blocking operation.
2819 Severity : enhancement
2821 Description: optimized stripe assignment
2822 Details : stripe assignments are now made based on ost space available,
2823 ost previous usage, and OSS previous usage, in order to try
2824 to optimize storage space and networking resources.
2826 Severity : enhancement
2828 Description: Permanently set tunables
2829 Details : All writable /proc/fs/lustre tunables can now be permanently
2830 set on a per-server basis, at mkfs time or on a live system.
2832 Severity : enhancement
2834 Description: Lustre message v2
2835 Details : Add lustre message format v2.
2837 Severity : enhancement
2839 Description: client OST exclusion list
2840 Details : Clients can be started with a list of OSTs that should be
2841 declared "inactive" for known non-responsive OSTs.
2845 Description: SPEC SFS validation failure on NFS v2 over lustre.
2846 Details : Changes the blocksize for regular files to be 2x RPC size,
2847 and not depend on stripe size.
2849 Severity : enhancement
2851 Description: Multiple MD RPCs in flight.
2852 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2853 To avoid overly-overloading MDS, introduce a limit on number of
2854 MDS RPCs in flight for a single client and add /proc controls
2855 to adjust this limit.
2857 Severity : enhancement
2859 Description: client read/write statistics
2860 Details : Add client read/write call usage stats for performance
2861 analysis of user processes.
2862 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2863 file access. extents_stats shows chunk size distribution.
2864 extents_stats_per_process show chunk size distribution per
2867 Severity : enhancement
2869 Description: mds statistics
2870 Details : Add detailed mds operations statistics in
2871 /proc/fs/lustre/mds/*/stats.
2875 Description: Failure of copying files with lustre special EAs.
2876 Details : Client side always return success for setxattr call for lustre
2877 special xattr (currently only "trusted.lov").
2882 Description: Refcount LNET uuids
2883 Details : The global LNET uuid list grew linearly with every startup;
2884 refcount repeated list entries instead of always adding to
2887 Severity : enhancement
2889 Description: Dynamic service threads
2890 Details : Within a small range, start extra service threads
2891 automatically when the request queue builds up.
2894 Frequency : mixed-endian client/server environments
2896 Description: mixed-endian crashes
2897 Details : The new msg_v2 system had some failures in mixed-endian
2900 Severity : enhancement
2902 Description: Easy OST removal
2903 Details : OSTs can be permanently deactivated with e.g. 'lctl
2904 conf_param lustre-OST0001.osc.active=0'
2906 Severity : enhancement
2908 Description: MGS proc entries
2909 Details : Added basic proc entries for the MGS showing what filesystems
2912 Severity : enhancement
2914 Description: provide MGS failover
2915 Details : Added config lock reacquisition after MGS server failover.
2917 Severity : enhancement
2919 Description: add Linux 2.4 support
2920 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2924 Description: a large application tries to do I/O to the same resource and dies
2925 in the middle of it.
2926 Details : Check the req->rq_arrival time after the call to
2927 ost_brw_lock_get(), but before we do anything about
2928 processing it & sending the BULK transfer request. This
2929 should help move old stale pending locks off the queue as
2930 quickly as obd_timeout.
2933 Frequency : when an incorrect nid is specified during startup
2935 Description: ptlrpc connect to non-existant node causes kernel crash
2936 Details : LNET can't be re-entered from an event callback, which
2937 happened when we expire a message after the export has been
2938 cleaned up. Instead, hand the zombie cleanup off to another
2941 Severity : enhancement
2943 Description: plain/inodebits lock performance improvement
2944 Details : Grouping plain/inodebits in granted list by their request modes
2945 and bits policy, thus improving the performance of search through
2949 Frequency : only if OST filesystem is corrupted
2951 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2952 Details : for a short time RPCs with bulk IO are in the replay list,
2953 but replay of bulk IOs is unimplemented. If the OST filesystem
2954 is corrupted due to disk cache incoherency and then replay is
2955 started it is possible to trip an assertion. Avoid putting
2956 committed RPCs into the replay list at all to avoid this issue.
2959 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2961 Description: System hang on startup
2962 Details : This bug allowed the liblustre (e.g. catamount) client to
2963 return to the app before handling all startup RPCs. This
2964 could leave the node unresponsive to lustre network traffic
2965 and manifested as a server ptllnd timeout.
2967 Severity : enhancement
2969 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2970 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2971 boolean to control whether to print peer debug info when a
2972 client's RPC times out.
2975 Frequency : only for kernels with patches from Lustre below 1.4.3
2977 Description: Remove old rdonly API
2978 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2981 Frequency : only for devices with external journals
2983 Description: Set external device read-only also
2984 Details : During a commanded failover stop, we set the disk device
2985 read-only while the server shuts down. We now also set any
2986 external journal device read-only at the same time.
2989 Frequency : when upgrading from 1.4 while trying to change parameters
2991 Description: The wrong (new) MDC name was used when setting parameters for
2992 upgraded MDT's. Also allows changing of OSC (and MDC)
2993 parameters if --writeconf is specified at tunefs upgrade time.
2996 Frequency : when setting specific ost indicies
2998 Description: QOS code breaks on skipped indicies
2999 Details : Add checks for missing OST indicies in the QOS code, so OSTs
3000 created with --index need not be sequential.
3002 Severity : enhancement
3004 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3005 Details : The uninit_groups feature works in conjunction with the kernel
3006 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3007 up the pass1 processing of e2fsck. This is a read-only feature
3008 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3009 mount filesystems that have had this feature enabled.
3011 Severity : enhancement
3013 Description: Improve multi-block allocation algorithm to avoid fragmentation
3014 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3015 allocation locality and avoid filesystem fragmentation.
3017 ------------------------------------------------------------------------------
3019 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
3021 * Support for kernels:
3022 2.4.21-47.0.1.EL (RHEL 3)
3023 2.6.5-7.283 (SLES 9)
3024 2.6.9-42.0.10.EL (RHEL 4)
3025 2.6.12.6 vanilla (kernel.org)
3026 2.6.16.27-0.9 (SLES 10)
3027 * Recommended e2fsprogs version: 1.39.cfs5
3029 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3033 Frequency : occasional, depends on client load and configuration
3034 Bugzilla : 12181, 12203
3035 Description: data loss for recently-modified files
3037 Details : In some cases it is possible that recently written or created
3038 files may not be written to disk in a timely manner (this should
3039 normally be within 30s unless client IO load is very high).
3040 The problem appears as zero-length files or files that are a
3041 multiple of 1MB in size after a client crash or client eviction
3042 that are missing data at the end of the file.
3044 This problem is more likely to be hit on clients where files are
3045 repeatedly created and unlinked in the same directory, clients
3046 have a large amount of RAM, have many CPUs, the filesystem has
3047 many OSTs, the clients are rebooted frequently, and/or the files
3048 are not accessed by other nodes after being written.
3050 The presence of the problem can be detected by looking at
3051 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
3052 smaller than the second (nr_unused) then dirty files will not
3053 be flushed automatically to disk. "sync; sleep 10" should be
3054 run several times on the node before unmounting it to update
3055 Lustre (this is also safe to run on nodes without this problem).
3057 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
3058 that can cause this same problem, so customers using that kernel
3059 also need to update the kernel in addition to Lustre. In order
3060 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
3062 It is normal that files written just before a client crash (less
3063 than 30s) may not yet have been flushed to disk, even for local
3067 Frequency : frequent on thin XT3 nodes
3069 Description: UUID collision on thin XT3 Linux nodes
3070 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
3071 randomly, since we relied on an insufficiently-seeded PRNG.
3076 Description: OSS hangs after "All ost request buffers busy"
3077 Details : A deadlock between quota and journal operations caused OSS
3078 hangs after printing "All ost request buffers busy."
3081 Frequency : always on liblustre builds
3083 Description: Cleanup compiler warnings on liblustre
3086 Frequency : always on liblustre builds on XT3
3088 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
3092 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
3093 > 32000 subdirectories
3094 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
3095 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
3096 subdirectory count crosses 32000. This will aid e2fsck to
3097 correctly handle more than 32000 subdirectories.
3100 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3102 Description: System hang on startup
3103 Details : This bug allowed the liblustre (e.g. catamount) client to
3104 return to the app before handling all startup RPCs. This
3105 could leave the node unresponsive to lustre network traffic
3106 and manifested as a server ptllnd timeout.
3108 Severity : enhancement
3110 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3111 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
3112 boolean to control whether to print peer debug info when a
3113 client's RPC times out.
3118 Description: make O_SYNC working on 2.6 kernels
3119 Details : 2.6 kernels use different method for mark pages for write,
3120 so need add a code to lustre for O_SYNC work.
3125 Description: Failure to close file and release space on NFS
3126 Details : Put inode details into lock acquired in ll_intent_file_open.
3127 Use mdc_intent_lock in ll_intent_open to properly
3128 detect all kind of errors unhandled by mdc_enqueue.
3133 Description: proc file read during shutdown sometimes raced obd removal,
3135 Details : Add lock to prevent obd access after proc file removal.
3138 Frequency : Only for files larger than 4GB on 32-bit clients.
3140 Description: improperly doing page alignment of locks
3141 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
3142 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
3147 Description: RPCs being resent when they shouldn't be
3148 Details : Some RPCs that should not be resent are being resent. This
3149 can cause inconsistencies in the RPC state machine. Do not
3150 resend such requests.
3153 Frequency : rare, only with NFS export
3155 Description: Crash on NFS re-export node
3156 Details : under very unusual load conditions an assertion is hit in
3157 ll_intent_file_open()
3160 Frequency : only if OST filesystem is corrupted
3162 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3163 Details : for a short time RPCs with bulk IO are in the replay list,
3164 but replay of bulk IOs is unimplemented. If the OST filesystem
3165 is corrupted due to disk cache incoherency and then replay is
3166 started it is possible to trip an assertion. Avoid putting
3167 committed RPCs into the replay list at all to avoid this issue.
3172 Description: large O_DIRECT requests fail under memory pressure/fragmentation
3173 Details : Large single O_DIRECT read and write calls can fail to allocate
3174 a sufficiently large buffer to process the request. In case of
3175 allocation failure the allocation is retried with a smaller
3176 buffer and broken into smaller requests.
3178 Severity : enhancement
3180 Description: Add -o localflock option to simulate old noflock behaviour.
3181 Details : This will achieve local-only flock/fcntl locks coherentness.
3186 Description: versioning check is incomplete
3187 Details : Checking the version difference of client vs. server, report
3188 error if the gap is too big.
3193 Description: add support PG_writeback bit
3194 Details : add support for PG_writeback bit for Lustre, for more carefull
3195 work with page cache in 2.6 kernel. This also fix some deadlocks
3196 and remove hack for work O_SYNC with 2.6 kernel.
3198 Severity : enhancement
3200 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3201 Details : The uninit_groups feature works in conjunction with the kernel
3202 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3203 up the pass1 processing of e2fsck. This is a read-only feature
3204 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3205 mount filesystems that have had this feature enabled.
3207 Severity : enhancement
3209 Description: Improve multi-block allocation algorithm to avoid fragmentation
3210 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3211 allocation locality and avoid filesystem fragmentation.
3213 ------------------------------------------------------------------------------
3215 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
3217 * Support for kernels:
3218 2.6.9-42.0.3.EL (RHEL 4)
3219 2.6.5-7.276 (SLES 9)
3220 2.4.21-47.0.1.EL (RHEL 3)
3221 2.6.12.6 vanilla (kernel.org)
3222 2.6.16.21-0.8 (SLES10)
3223 * Recommended e2fsprogs version: 1.39.cfs2-0
3225 * The backwards-compatible /proc/sys/portals symlink has been removed
3226 in this release. Before upgrading, please ensure that you change
3227 any configuration scripts or /etc/sysctl.conf files that access
3228 /proc/sys/portals/* or sysctl portals.* to use the corresponding
3229 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
3230 in advance of the upgrade on any system running Lustre 1.4.6 or
3231 newer, since /proc/sys/lnet was added in that version.
3232 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3236 Frequency : only when quota is used
3238 Description: avoid scanning export list for quota master
3239 Details : Change the algorithms to avoid scanning export list in order
3240 to improve the efficiency.
3243 Frequency : MDS failover only, very rarely
3245 Description: "went back in time" messages on mds failover
3246 Details : The greatest transno may be lost when the current operation
3247 finishes with an error (transno==0) and the client's last_rcvd
3248 record is over-written. Save the greatest transno in the
3249 mds_last_transno for this case.
3252 Frequency : always for specific kernels and striping counts
3254 Description: client may get "Matching packet too big" without ACL support
3255 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
3256 when trying to access files in certain configurations. The
3257 clients should in fact be denied when mounting because they do
3258 not understand ACLs.
3261 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3263 Description: many clients connecting with IO in progress causes connect timeouts
3264 Details : Avoid synchronous journal commits to avoid delays caused by many
3265 clients connecting/disconnecting when bulk IO is in progress.
3266 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3267 OST_IO_PORTAL to avoid delays behind potentially many pending
3271 Frequency : occasionally with multiple writers to a single file
3273 Description: shared writes to file may result in wrong size reported by stat()
3274 Details : Allow growing of kms when extent lock is cancelled
3277 Frequency : always with random mmap IO to multi-striped file
3279 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3280 Details : Only if the hole is at the end of OST object so that kms is too
3281 small. Fix is to increase kms accordingly in ll_nopage.
3284 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3286 Description: writes to a missing object would leak memory on the OST
3287 Details : If there is an inconsistency between the MDS and OST filesystems,
3288 such that the MDS references an object that doesn't exist, writes
3289 to that object will leak memory due to incorrect cleanup in the
3290 error handling path, eventually running out of memory on the OST.
3295 Description: Creating too long symlink causes lustre errors
3296 Details : Check symlink and name lengths before sending requests to MDS.
3299 Frequency : only if flock is enabled (not on by default)
3301 Description: posix locks not released on fd closure on 2.6.9+
3302 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3303 kernels, this caused such locks not to be released on fd close and
3304 then assertions on fs unmount about still used locks.
3307 Frequency : MDS failover only, very rarely
3309 Description: clients may get ASSERTION(granted_lock != NULL)
3310 Details : When request was taking a long time, and a client was resending
3311 a getattr by name lock request. The were multiple lock requests
3312 with the same client lock handle and
3313 mds_getattr_name->fixup_handle_for_resent_request found one of the
3314 lock handles but later failed with ASSERTION(granted_lock != NULL).
3319 Description: handle->h_buffer_credits > 0, assertion failure
3320 Details : h_buffer_credits is zero after truncate, causing assertion
3321 failure. This patch extends the transaction or creates a new
3325 Frequency : NFS re-export or patchless client
3326 Bugzilla : 11179, 10796
3327 Description: Crash on NFS re-export node (__d_move)
3328 Details : We do not want to hash the dentry if we don't have a lock.
3329 But if this dentry is later used in d_move, we'd hit uninitialised
3330 list head d_hash, so we just do this to init d_hash field but
3331 leave dentry unhashed.
3334 Frequency : NFS re-export or patchless client
3336 Description: NFS exports has problem with symbolic link
3337 Details : lustre client didn't properly install dentry when re-exported
3338 to NFS or running patchless client.
3341 Frequency : NFS re-export or patchless client
3343 Description: Various nfs/patchless fixes.
3344 Details : fixes reuse disconected alias for lookup process - this fixes
3345 warning "find_exported_dentry: npd != pd",
3346 fix permission error with open files at nfs.
3347 fix apply umask when do revalidate.
3350 Frequency : occasional
3352 Description: Crash on NFS re-export node
3353 Details : calling clear_page() on the wrong pointer triggered oops in
3354 generic_mapping_read().
3357 Frequency : rarely, using O_DIRECT IO
3359 Description: unaligned directio crashes client with LASSERT
3360 Details : check for unaligned buffers before trying any requests.
3363 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3365 Description: stale data returned from RAID cache
3366 Details : If only a small amount of IO is done to the RAID device before
3367 reading it again it is possible to get stale data from the RAID
3368 cache instead of reading it from disk.
3371 Frequency : always for sles10 kernel
3373 Description: sles10 support
3374 Details : ll_follow_link: compile fixes and using of nd_set_link
3375 under newer kernels.
3378 Frequency : depends on arch, kernel and compiler version, always on sles10
3381 Description: recursive or deep enough symlinks cause stack overflow
3382 Details : getting rid of large stack-allocated variable in
3386 Frequency : depends on hardware
3388 Description: lustre write performance loss in the SLES10 kernel
3389 Details : the performance loss is caused by using of write barriers in the
3390 ext3 code. The SLES10 kernel turns barrier support on by
3391 default. The fix is to undo that change for ldiskfs.
3393 ------------------------------------------------------------------------------
3395 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3397 * Support for kernels:
3398 2.6.9-42.0.3EL (RHEL 4)
3399 2.6.5-7.276 (SLES 9)
3400 2.4.21-47.0.1.EL (RHEL 3)
3401 2.6.12.6 vanilla (kernel.org)
3405 Frequency : quota enabled and large files being deleted
3407 Description: releasing more than 4GB of quota at once hangs OST
3408 Details : If a user deletes more than 4GB of files on a single OST it
3409 will cause the OST to spin in an infinite loop. Release
3410 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3415 Description: statfs data retrieved from /proc may be stale or zero
3416 Details : When reading per-device statfs data from /proc, in the
3417 {kbytes,files}_{total,free,avail} files, it may appear
3418 as zero or be out of date.
3421 Frequency : systems with MD RAID1 external journal devices
3423 Description: lconf's call to blkid is confused by RAID1 journal devices
3424 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3425 of returning all block devices that match the journal UUID.
3428 Frequency : always, for aggregate stripe size over 4GB
3430 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3431 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3432 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3435 Frequency : always if "lfs find" used on a local file/directory
3437 Description: "lfs find" segfaults if used on a local file/directory
3438 Details : The case where a directory component was not specified wasn't
3439 handled correctly. Handle this properly.
3442 Frequency : always on ppc64
3444 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3445 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3446 perform endianness conversion.
3449 Frequency : rarely (truncate to non-zero file size after write under load)
3450 Bugzilla : 10730, 10687
3451 Description: Files padded with zeros to next 4K multiple
3452 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3453 it is possible that files that are truncated to a non-zero size
3454 immediately after being written are filled with zero bytes beyond
3455 the truncated size. No file data is lost.
3457 Severity : enhancement
3459 Description: Allow recovery/failover for liblustre clients.
3460 Details : liblustre clients were unaware of failover configurations until
3463 Severity : enhancement
3465 Description: user file locks should fail when not mounting with flock option
3466 Details : Set up an error-returning stub in ll_file_operations.lock field
3467 to prevent incorrect behaviour when client is mounted without
3468 flock option. Also, set up properly f_op->flock field for
3472 Frequency : always on ia64
3474 Description: "lfs df" loops on printing out MDS statfs information
3475 Details : The obd_ioctl_data was not initialized and in some systems
3476 this caused a failure during the ioctl that did not return
3477 an error. Initialize the struct and return an error on failure.
3480 Frequency : SLES 9 only
3482 Description: Error of copying files with lustre special EAs as root
3483 Details : Client side always return success for setxattr call for lustre
3484 special xattr (currently only "trusted.lov").
3487 Frequency : rarely on clusters with both ia64+i386 clients
3489 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3490 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3491 clients are doing concurrent writes to the same file it is
3492 possible that the ia64 clients may LASSERT because the OST
3493 extent locks are not PAGE_SIZE aligned. Ensure that grown
3494 locks are always aligned on the request boundary.
3497 Frequency : specific use, occasional
3499 Description: Overwriting in use executable truncates on-disk binary image
3500 Details : If one node attempts to overwrite an executable in use by
3501 another node, we now correctly return ETXTBSY instead of
3502 truncating the file.
3504 Severity : enhancement
3506 Description: Async OSC create to avoid the blocking unnecessarily.
3507 Details : If a OST has no remain object, system will block on the creating
3508 when need to create a new object on this OST. Now, ways use
3509 pre-created objects when available, instead of blocking on an
3510 empty osc while others are not empty. If we must block, we block
3511 for the shortest possible period of time.
3516 Description: chmod on Lustre root is propagated to other clients
3517 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3518 invalid by the follow_mount time.
3521 Frequency : liblustre clients only
3523 Description: Race in 'instant cancel' lock handling could lead to such locks
3524 never to be granted in case of SMP MDS
3525 Details : Do not destroy not yet granted but cbpending locks in
3529 Frequency : replay/resend of open
3531 Description: non null lock assetion failure in mds_intent_policy
3532 Details : Trying to replay/resend lockless open requests resulted in
3533 mds_open() returning 0 with no lock. Now it sets a flag if
3534 a lock is going to be returned.
3536 Severity : enhancement
3538 Description: Checksum enhancements
3539 Details : New checksum enhancements allow for resending RPCs that failed
3542 Severity : enhancement
3544 Description: Tunables on number of dirty pages in cacche
3545 Details : Allow to set limit on number of dirty pages cached.
3550 Description: client crash on unmount - lock still has references
3551 Details : In some error handling cases it was possible to leak a lock
3552 reference on a client while accessing a file. This was not
3553 harmful to the client during operation, but would cause the
3554 client to crash when the filesystem is unmounted.
3557 Frequency : specific case, rare
3559 Description: ETXTBSY on mds though file not in use by client
3560 Details : ETXTBSY is no longer incorrectly returned when attempting to
3561 chmod or chown a directory that the user previously tried to
3562 execute or a currently-executing binary.
3565 Frequency : extremely rare except on liblustre-based clients
3567 Description: Lustre space not freed when files are deleted
3568 Details : Clean up open-unlinked files after client eviction. Previously
3569 the unlink was skipped and the files remained as orphans.
3574 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3575 Details : In some cases it was possible to send a blocking callback to
3576 a client doing a glimpse, even though that client didn't get
3577 a lock granted. When the glimpse lock is cancelled on the OST
3578 the freed lock is left on the waiting list and corrupted the list.
3581 Frequency : all core dumps
3583 Description: Broke core dumps to lustre
3584 Details : Negative dentry may be unhashed if parent does not have UPDATE
3585 lock, but some callers, e.g. do_coredump, expect dentry to be
3586 hashed after successful create, hash it in ll_create_it.
3588 ------------------------------------------------------------------------------
3590 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3592 * Support for kernels:
3593 2.6.9-42.0.2.EL (RHEL 4)
3594 2.6.5-7.276 (SLES 9)
3595 2.4.21-40.EL (RHEL 3)
3596 2.6.12.6 vanilla (kernel.org)
3600 Frequency : always on RHEL 3
3602 Description: Number of open files grows over time
3603 Details : The number of open files grows over time, whether or not
3604 Lustre is started. This was due to a filp leak introduced
3605 by one of our kernel patches.
3607 ------------------------------------------------------------------------------
3609 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3611 * Support for kernels:
3612 2.6.9-42.EL (RHEL 4)
3613 2.6.5-7.276 (SLES 9)
3614 2.4.21-40.EL (RHEL 3)
3615 2.6.12.6 vanilla (kernel.org)
3620 Bugzilla : 5719, 9635, 9792, 9684
3621 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3622 Details : If a server is under heavy load and cannot reply to new
3623 connection requests before the client resends the (re)connect,
3624 the connection handling code can behave badly if two service
3625 threads are concurrently handing separate (re)connections from
3626 the same client. Add better locking to the connection handling
3627 code, and ensure that only a single connection will be processed
3628 for a given client UUID, even if the lock is dropped.
3630 Severity : enhancement
3632 Description: add TCP zero-copy support to kernel
3633 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3634 sends if the hardware supports scatter-gather and checksumming.
3635 This allows socklnd to do client-write and server-read more
3636 efficiently and reduce CPU utilization from skbuf copying.
3639 Frequency : only if NFS exporting from client
3641 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3642 Details : If setting attributes on a file created under NFS that had
3643 never been opened it would be possible to oops the client
3644 if the file had no objects.
3647 Frequency : always for liblustre
3649 Description: liblustre client does MDS+OSTs setattr RPC for each write
3650 Details : When doing a write from a liblustre client, the client
3651 incorrectly issued an RPC to the MDS and each OST the file was
3652 striped over in order to update the timestamps. When writing
3653 with small chunks and many clients this could overwhelm the MDS
3654 with RPCs. In all cases it would slow down the write because
3655 these RPCs are unnecessary.
3657 Severity : enhancement
3659 Description: allow number of MDS service threads to be changed at module load
3660 Details : It is now possible to change the number of MDS service threads
3661 running. Adding "options mds mds_num_threads={N}" to the MDS's
3662 /etc/modprobe.conf will set the number of threads for the next
3663 time Lustre is restarted (assuming the "mds" module is also
3664 reloaded at that time). The default number of threads will
3665 stay the same, 32 for most systems.
3670 Description: OST crash if filesystem is unformatted or corrupt
3671 Details : If an OST is started on a device that has never been formatted
3672 or if the filesystem is corrupt and cannot even mount then the
3673 error handling cleanup routines would dereference a NULL pointer.
3678 Description: NULL pointer deref in llap_from_page.
3679 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3680 incorrect) mapping if the page was truncated/reclaimed while it was
3681 searched for. Check for this condition and skip such pages when
3682 doing readahead. Introduce extra check to llap_from_page() to
3683 verify page->mapping->host is non-NULL (so page is not anonymous).
3686 Frequency : Sometimes when using sys_sendfile
3688 Description: "page not covered by a lock" warnings from ll_readpage
3689 Details : sendfile called ll_readpage without right page locks present.
3690 Now we introduced ll_file_sendfile that does necessary locking
3691 around call to generic_file_sendfile() much like we do in
3695 Frequency : with certain MDS communication failures at client mount time
3697 Description: NULL pointer deref after failed client mount
3698 Details : a client connection request may delayed by the network layer
3699 and not be sent until after the PTLRPC layer has timed out the
3700 request. If the client fails the mount immediately it will try
3701 to clean up before the network times out the request. Add a
3702 reference from the request import to the obd device and delay
3703 the cleanup until the network drops the request.
3706 Frequency : occasionally during client (re)connect
3708 Description: assertion failure during client (re)connect
3709 Details : processing a client connection request may be delayed by the
3710 client or server longer than the client connect timeout. This
3711 causes the client to resend the connection request. If the
3712 original connection request is replied in this interval, the
3713 client may trip an assertion failure in ptlrpc_connect_interpret()
3714 which thought it would be the only running connect process.
3717 Frequency : only with obd_echo servers and clients that are rebooted
3719 Description: kernel BUG accessing uninitialized data structure
3720 Details : When running an obd_echo server it did not start the ping_evictor
3721 thread, and when a client was evicted an uninitialized data
3722 structure was accessed. Start the ping_evictor in the RPC
3723 service startup instead of the OBD startup.
3725 Severity : enhancement
3726 Bugzilla : 10193 (patchless)
3727 Description: Remove dependency on various unexported kernel interfaces.
3728 Details : No longer need reparent_to_init, exit_mm, exit_files,
3729 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3732 Frequency : rare (only users of deprecated and unsupported LDAP config)
3734 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3735 Details : LDAP apparently contains 'lustreName' attributes instead of
3736 'name'. A simple remapping of the name is sufficient.
3739 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3741 Description: waiting_locks_callback trips kernel BUG if client is evicted
3742 Details : Running with the dump_on_timeout debug flag turned on makes
3743 it possible that the waiting_locks_callback() can try to dump
3744 the Lustre kernel debug logs from an interrupt handler. Defer
3745 this log dumping to the expired_lock_main() thread.
3747 Severity : enhancement
3749 Description: Support NFS exporting on 2.6 kernels.
3750 Details : Implement non-rawops metadata methods for NFS server to use without
3751 changing NFS server code.
3754 Frequency : very rare (synthetic metadata workload only)
3756 Description: two racing renames might cause an MDS thread to deadlock
3757 Details : Running the "racer" program may cause one MDS thread to rename
3758 a file from being the source of a rename to being the target of
3759 a rename at exactly the same time that another thread is doing
3760 so, and the second thread has already enqueued these locks after
3761 doing a lookup of the target and is trying to relock them in
3762 order. Ensure that we don't try to re-lock the same resource.
3765 Frequency : only very large systems with liblustre clients
3767 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3768 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3769 all OSTs in parallel. This allows the eviction of stale liblustre
3770 clients to proceed much faster than if they were done in series,
3771 and also offers similar improvements for other set_info RPCs.
3776 Description: excessive CPU usage during initial read phase on client
3777 Details : During the initial read phase on a client, it would agressively
3778 retry readahead on the file, consuming too much CPU and impacting
3779 performance (since 1.4.5.8). Improve the readahead algorithm
3780 to avoid this, and also improve some other common cases (read
3781 of small files in particular, where "small" is files smaller than
3782 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3787 Description: MDS crash when receiving packet with unknown intent.
3788 Details : Do not LBUG in unknown intent case, just return -EFAULT
3790 Severity : enhancement
3791 Bugzilla : 9293, 9385
3792 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3793 Details : Do not serialize getattr (non-intent version) and statfs.
3796 Frequency : occasional, when OST network is overloaded/intermittent
3798 Description: client evicted by OST after bulk IO timeout
3799 Details : If a client sends a bulk IO request (read or write) the OST
3800 may evict the client if it is unresposive to its data GET/PUT
3801 request. This is incorrect if the network is overloaded (takes
3802 too long to transfer the RPC data) or dropped the OST GET/PUT
3803 request. There is no need to evict the client at all, since
3804 the pinger and/or lock callbacks will handle this, and the
3805 client can restart the bulk request.
3808 Frequency : Always when mmapping file with no objects
3810 Description: client crashes when mmapping file with no objects
3811 Details : Check that we actually have objects in a file before doing any
3812 operations on objects in ll_vm_open, ll_vm_close and
3818 Description: Request leak when working with deleted CWD
3819 Details : Introduce advanced request refcount tracking for requests
3820 referenced from lustre intent.
3822 Severity : Enhancement
3824 Description: Cache open file handles on client.
3825 Details : MDS now will return special lock along with openhandle, if
3826 requested and client is allowed to hold openhandle, even if unused,
3827 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3828 closing files for every read/write openration.
3830 Severity : Enhancement
3832 Description: Cache open negative dentries on client when possible.
3833 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3834 negative dentries on lock revocation.
3839 Description: Remounting a client read-only wasn't possible with a zconf mount
3840 Details : It wasn't possible to remount a client read-only with llmount.
3842 Severity : enhancement
3843 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3844 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3845 MPICH is included to add Lustre-specific ADIO interfaces.
3846 This is based closely on the UFS ADIO layer and only differs
3847 in file creation, in order to allow the OST striping to be set.
3848 This is user-contributed code and not supported by CFS.
3853 Description: extended inode attributes (immutable, append-only) work improperly
3854 when 2.4 and 2.6 kernels are used on client/server or vice versa
3855 Details : Introduce kernel-independent values for these flags.
3857 Severity : enhancement
3860 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3861 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3862 granularity. Now, Fractional MB granularity is be supported,
3863 this is very useful for low memory system.
3865 Severity : enhancement
3867 Description: Getattr by fid
3868 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3869 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3870 have locks and avoids getattr rpc requests for it on every stat.
3873 Frequency : Always, for filesystems larger than 2TB
3875 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3876 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3877 filesystems larger than 2TB to be mounted with mballoc enabled.
3880 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3882 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3883 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3884 filesystems larger than 2TB is mounted then the kernel will
3885 silently corrupt the start of the filesystem. CONFIG_LBD is
3886 enabled for all CFS-supported kernels, but the possibility of
3887 this happening with a modified kernel config exists.
3889 Severity : enhancement
3891 Description: add client O_DIRECT support for 2.6 kernels
3892 Details : It is now possible to do O_DIRECT reads and writes to files
3893 in the Lustre client mountpoint on 2.6 kernel clients.
3895 Severity : enhancement
3897 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3898 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3899 parallel, not waiting for response from every OST before sending
3900 a rpc to the next OST.
3905 Description: setattr vs write race when updating file timestamps
3906 Details : Client processes that update a file timestamp into the past
3907 right after writing to the file (e.g. tar) it is possible that
3908 the updated file modification time can be reset to the current
3909 time due to a race between processing the setattr and write RPC.
3911 Severity : enhancement
3913 Description: Bring 'lfs find' closer in line with regular Linux find.
3914 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3915 -print0 options and obtains all the needed info through the lustre
3918 Severity : enhancement
3920 Description: support up to 1024 configured devices on one node
3921 Details : change obd_dev array from statically allocated to dynamically
3922 allocated structs as they are first used to reduce memory usage
3927 Description: Flush dirty partially truncated pages during truncate
3928 Details : Immediatelly flush partially truncated pages in filter_setattr,
3929 this way we completely avoid having any pages in page cache on OST
3930 and can retire ugly workarounds during writes to flush such pages.
3935 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3936 Details : protect inode from truncation within vfs_unlink() context
3937 just take a reference before calling vfs_unlink() and release it
3938 when parent's i_sem is free.
3943 Description: last_id value checked outside lock on OST caused LASSERT failure
3944 Details : If there were multiple MDS->OST object precreate requests in
3945 flight, it was possible that the OST's last object id was checked
3946 outside a lock and incorrectly tripped an assertion. Move checks
3947 inside locks, and discard old precreate requests.
3950 Frequency : always, if extents are used on OSTs
3952 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3953 Details : OSTs using the extents format would not zero the high 16 bits of
3954 the index physical block number. This is not a problem for any
3955 OST filesystems smaller than 16TB, and no kernels support ext3
3956 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3957 new/modified files) and can be fixed for existing filesystems
3958 with e2fsprogs-1.39-cfs1.
3963 Description: import connection selection may be incorrect if timer wraps
3964 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3965 import connections to be ignored if the 32-bit jiffies counter
3966 wraps. Use a 64-bit jiffies counter.
3969 Frequency : very large clusters immediately after boot
3971 Description: LNET request buffers exhausted under heavy short-term load
3972 Details : If a large number of client requests are generated on a service
3973 that has previously never seen so many requests it is possible
3974 that the request buffer growth cannot keep up with the spike in
3975 demand. Instead of dropping incoming requests, they are held in
3976 the LND until the RPC service can accept more requests.
3979 Frequency : Sometimes during replay
3981 Description: Assertion failure in ll_local_open after replay.
3982 Details : If replay happened on an open request reply before we were able
3983 to set replay handler, reply will become not swabbed tripping the
3984 assertion in ll_local_open. Now we set the handler right after
3985 recognising of open request
3988 Frequency : very rare
3990 Description: kernel reports "badness in vsnprintf"
3991 Details : Reading from the "recovery_status" /proc file in small chunks
3992 may cause a negative length in lprocfs_obd_rd_recovery_status()
3993 call to vsnprintf() (which is otherwise harmless). Exit early
3994 if there is no more space in the output buffer.
3996 Severity : enhancement
3998 Description: clear OBD RPC statistics by writing to them
3999 Details : It is now possible to clear the OBD RPC statistics by writing
4000 to the "stats" file.
4005 Description: Client mtime is not the same on different clients after utimes
4006 Details : In some cases, the client was using the utimes() syscall on
4007 a file cached on another node. The clients now validate the
4008 ctime from the MDS + OSTs to determine which one is right.
4013 Description: Inability to activate failout mode
4014 Details : lconf script incorrectly assumed that in pythong string's numeric
4015 value is used in comparisons.
4018 Frequency : always with multiple stripes per file
4020 Description: Inefficient object allocation for mutli-stripe files
4021 Details : When selecting which OSTs to stripe files over, for files with
4022 a stripe count that divides evenly into the number of OSTs,
4023 the MDS is always picking the same starting OST for each file.
4024 Return the OST selection heuristic to the original design.
4029 Description: mount failures may take full timeout to return an error
4030 Details : Under some heavy load conditions it is possible that a
4031 failed mount can wait for the full obd_timeout interval,
4032 possibly several minutes, before reporting an error.
4033 Instead return an error as soon as the status is known.
4035 Frequency : quota enabled and large files being deleted
4037 Description: releasing more than 4GB of quota at once hangs OST
4038 Details : If a user deletes more than 4GB of files on a single OST it
4039 will cause the OST to spin in an infinite loop. Release
4040 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
4045 Description: statfs data retrieved from /proc may be stale or zero
4046 Details : When reading per-device statfs data from /proc, in the
4047 {kbytes,files}_{total,free,avail} files, it may appear
4048 as zero or be out of date.
4051 Frequency : systems with MD RAID1 external journal devices
4053 Description: lconf's call to blkid is confused by RAID1 journal devices
4054 Details : Use the "blkid -l" flag to locate the MD RAID device instead
4055 of returning all block devices that match the journal UUID.
4058 Frequency : always, for aggregate stripe size over 4GB
4060 Description: assertion fails when trying to use 4GB stripe size
4061 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
4062 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
4065 Frequency : always on ppc64
4067 Description: the first write on an ext3 filesystem with mballoc got stuck
4068 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
4069 perform endianness conversion.
4071 ------------------------------------------------------------------------------
4073 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
4075 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
4076 INTEROPERATE with older versions automatically. Please read the
4077 user documentation before upgrading any part of a live system.
4078 * WARNING: Lustre networking configuration changes are required with
4079 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
4082 * Support for kernels:
4083 2.6.9-22.0.2.EL (RHEL 4)
4084 2.6.5-7.244 (SLES 9)
4085 2.6.12.6 vanilla (kernel.org)
4088 Severity : enhancement
4089 Bugzilla : 7981/8208
4090 Description: Introduced Lustre Networking (LNET)
4091 Details : LNET is new networking infrastructure for Lustre, it includes
4092 a reorganized network configuration mode (see the user
4093 documentation for full details) as well as support for routing
4094 between different network fabrics. Lustre Networking Devices
4095 (LNDs) for the supported network fabrics have also been
4096 created for this new infrastructure.
4098 Severity : enhancement
4099 Description: Introduced Access control lists
4100 Details : clients can set ACLs on files and directories in order to have
4101 more fine-grained permissions than the standard Unix UGO+RWX.
4102 The MDS must be started with the "-o acl" mount option.
4104 Severity : enhancement
4105 Description: Introduced filesystem quotas
4106 Details : Administrators may now establish per-user quotas on the
4109 Severity : enhancement
4111 Description: Configuration change for the XT3
4112 The PTLLND is now used to run Lustre over Portals on the XT3
4113 The configure option(s) --with-cray-portals are no longer used.
4114 Rather --with-portals=<path-to-portals-includes> is used to
4115 enable building on the XT3. In addition to enable XT3 specific
4116 features the option --enable-cray-xt3 must be used.
4121 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
4122 Details : A race between allocating a new llcd and re-getting the llcd_lock
4123 allowed another thread to grab newly-allocated llcd.
4125 Severity : enhancement
4127 Description: 2.6 OST async journal commit and locking fix to improve performance
4128 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
4129 2.6 kernels are now async as they already were in 2.4 kernels so
4130 that they can commit concurrently with the network bulk transfer.
4131 For block-allocated files the filter allocation semaphore is held
4132 to avoid filesystem fragmentation during allocation. BKL lock
4133 removed for 2.6 xattr operations where it is no longer needed.
4138 Description: lconf incorrectly determined whether two IP networks could talk
4139 Details : In some more complicated routing and multiple-network
4140 configurations, lconf will avoid trying to make a network
4141 connection to a disjoint part of the IP space. It was doing the
4142 math incorrectly for one set of cases.
4147 Description: Fix for potential infinite loop processing records in an llog.
4148 Details : If an llog record is corrupted/zeroed, it is possible to loop
4149 forever in llog_process(). Validate the llog record length
4150 and skip the remainder of the block on error.
4153 Frequency : occasional (liblustre only)
4155 Description: liblustre could not open files whose last component is a symlink
4156 Details : sysio_path_walk() would incorrectly pass the open intent to
4157 intermediate path components.
4160 Frequency : rare (liblustre only with non-standard tuning)
4161 Bugzilla : 7201 (7350)
4162 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
4163 Details : llu_lookup_finish_locks() tries to set lock data on a lock
4164 after it has been released, only do this for referenced locks
4166 Severity : enhancement
4168 Description: specifying an (invalid) directory default stripe_size of -1
4169 would reset the directory default striping
4170 Details : stripe_size -1 was used internally to signal directory stripe
4171 removal, now use "all default" to signal dir stripe removal
4172 as a directory striping of "all default" is not useful
4175 Frequency : common for large clusters running liblustre clients
4177 Description: doing an ls when liblustre clients are running is slow
4178 Details : sending a glimpse AST to a liblustre client waits for every AST
4179 to time out, as liblustre clients will not respond. Since they
4180 cannot cache data we refresh the OST lock LVB from disk instead.
4182 Severity : enhancement
4184 Description: doing an ls at the same time as file IO can be slow
4185 Details : enqueue and other "small" requests can be blocked behind many
4186 large IO requests. Create a new OST IO portal for non-IO
4187 requests so they can be processed faster.
4190 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
4192 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
4193 created file with mode 000
4194 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
4195 allow an ftruncate() as a non-root user to a file with mode 000.
4196 Lustre now respects this flag to disable mode checks when
4197 truncating a file owned by the user
4200 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
4203 Description: Revoking locks from clients that went dead or catatonic might take
4205 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
4206 cancellation of such locks instant on servers without waiting for
4207 any reply from clients. Clients drops these locks when cancel
4208 notification from server is received without replying.
4211 Frequency : liblustre-only, when liblustre client dies or becomes busy
4213 Description: Doing ls on Linux clients can take a long time with active
4215 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
4216 granting such locks to it in the first place if possible. Locks
4217 are taken by proxy on the OST during the read or write and
4218 dropped immediately afterward. Add connect flags handling, do
4219 not grant locks to liblustre clients for glimpse ASTs.
4221 Severity : enhancement
4223 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
4224 Details : Existing read-ahead algorithm is tuned for the case of streamlined
4225 sequential reads and behaves badly with applications doing random
4226 reads. Improve it by reading ahead at least read region, and
4227 avoiding excessive large RPC for small reads.
4229 Severity : enhancement
4231 Description: Creating more than 1000 files for a single job may cause a load
4232 imbalance on the OSTs if there are also a large number of OSTs.
4233 Details : qos_prep_create() uses an OST index reseed value that is an
4234 even multiple of the number of available OSTs so that if the
4235 reseed happens in the middle of the object allocation it will
4236 still utilize the OSTs as uniformly as possible.
4241 Description: OST or MDS may oops in ping_evictor_main()
4242 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
4243 but doesn't restart at beginning of obd_exports_timed list
4246 Severity : enhancement
4248 Description: improve by-nid export eviction on the MDS and OST
4249 Details : allow multiple exports with the same NID to be evicted at one
4250 time without re-searching the exports list.
4253 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
4255 Description: MDS may oops in groups_free()
4256 Details : in rare race conditions a newly allocated group_info struct is
4257 freed again, and this can be NULL. The 2.4 compatibility code
4258 for groups_free() checked for a NULL pointer, but 2.6 did not.
4261 Frequency : common for liblustre clients doing little filesystem IO
4262 Bugzilla : 9352, 7313
4263 Description: server may evict liblustre clients accessing contended locks
4264 Details : if a client is granted a lock or receives a completion AST
4265 with a blocking AST already set it would not reply to the AST
4266 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4269 Frequency : lfs setstripe, only systems with more than 160 OSTs
4271 Description: unable to set striping with a starting offset beyond OST 160
4272 Details : llapi_create_file() incorrectly limited the starting stripe
4273 index to the maximum single-file stripe count.
4276 Frequency : LDAP users only
4278 Description: lconf did not handle in-kernel recovery with LDAP properly
4279 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4281 Severity : enhancement
4283 Description: bind OST threads to NUMA nodes to improve performance
4284 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4285 node and do their allocations there to localize memory access
4287 Severity : enhancement
4289 Description: llmount can determine client NID directly from Myrinet (GM)
4290 Details : the client NID code from gmnalnid was moved directly into
4291 llmount, removing the need to use this or specifying the
4292 client NID explicitly when mounting GM clients with zeroconf
4295 Frequency : if client is started with down MDS
4297 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4298 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4299 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4300 ptlrpc_import_delayed_req() because we didn't check if the
4301 request was interrupted, and we also didn't break out of the
4302 event loop if there was no timeout
4307 Description: data loss during non-page-aligned writes to a single file from
4308 both multiple nodes and multiple threads on one node at same time
4309 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4310 inconsistency led to false short-reads, that were cached and later
4311 used by ->prepare_write() to fill in partially written page,
4312 leading to data loss.
4315 Frequency : always, if lconf --abort_recovery used
4317 Description: lconf --abort_recovery fails with 'Operation not supported'
4318 Details : lconf was attempting to abort recovery on the MDT device and not
4321 Severity : enhancement
4323 Description: remove cleanup logs
4324 Details : replace lconf-generated cleanup logs with lustre internal
4325 cleanup routines. Eliminates the need for client-cleanup and
4328 Severity : enhancement
4330 Description: add support for EAs (user and system) on lustre filesystems
4331 Details : it is now possible to store extended attributes in the Lustre
4332 client filesystem, and with the user_xattr mount option it
4333 is possible to allow users to store EAs on their files also
4335 Severity : enhancement
4337 Description: Add possibility (config option) to show minimal available OST free
4339 Details : When compiled with --enable-mindf configure option, statfs(2)
4340 (and so, df) will return least minimal free space available from
4341 all OSTs as amount of free space on FS, instead of summary of
4342 free spaces of all OSTs.
4344 Severity : enhancement
4346 Description: do not expand extent locks acquired on OST-side
4347 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4348 by server: they are not cached anyway.
4351 Frequency : when mmap is used/binaries executed from Lustre
4353 Description: Unmmap pages before throwing them away from read cache.
4354 Details : llap_shrink cache now attempts to unmap pages before discarding
4355 them (if unmapping failed - do not discard). SLES9 kernel has
4356 extra checks that trigger if this unmapping is not done first.
4361 Description: lconf didn't resolve symlinks before checking to see whether a
4362 given mountpoint was already in use
4365 Frequency : when migrating failover services
4366 Bugzilla : 6395, 9514
4367 Description: When migrating a subset of services from a node (e.g. failback
4368 from a failover service node) the remaining services would
4369 time out and evict clients.
4370 Details : lconf --force (implied by --failover) sets the global obd_timeout
4371 to 5 seconds in order to quickly disconnect, but this caused
4372 other RPCs to time out too quickly. Do not change the global
4373 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4375 Severity : enhancement
4376 Frequency : if MDS is started with down OST
4377 Bugzilla : 9439,5706
4378 Description: Allow startup/shutdown of an MDS without depending on the
4379 availability of the OSTs.
4380 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4381 Add appropriate locking and lov-osc refcounts for safe
4382 cleaning. Add osc abort_inflight calls in case the
4383 synchronize never started.
4386 Frequency : occasional (Cray XT3 only)
4388 Description: root not authorized to access files in CRAY_PORTALS environment
4389 Details : The client process capabilities were not honoured on the MDS in
4390 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4391 been accessed by an authorized user then root was able to access
4392 the file on the local client also. The root user capabilities
4393 are now allowed on the MDS, as this environment has secure UID.
4396 Frequency : occasional
4398 Description: ldiskfs "too long searching" message happens too often
4399 Details : A debugging message (otherwise harmless) prints too often on
4400 the OST console. This has been reduced to only happen when
4401 there are fragmentation problems on the filesystem.
4406 Description: Division by zero in statfs when all OSCs are inactive
4407 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4408 lov_statfs divides by value returned by lov_get_stripecnt().
4412 Bugzilla : 9489, 3273
4413 Description: First write from each client to each OST was only 4kB in size,
4414 to initialize client writeback cache, which caused sub-optimal
4415 RPCs and poor layout on disk for the first writen file.
4416 Details : Clients now request an initial cache grant at (re)connect time
4417 and so that they can start streaming writes to the cache right
4418 away and always do full-sized RPCs if there is enough data.
4419 If the OST is rebooted the client also re-establishes its grant
4420 so that client cached writes will be honoured under the grant.
4425 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4426 Details : Now I/O RPCs go to different portal number and (presumably) fast
4427 lock requests (and glimses) and other RPCs get their own service
4428 threads pool that should be able to service those RPCs
4431 Severity : enhancement
4433 Description: Ability to exchange lustre version between client and servers and
4434 issue warnings at client side if client is too old. Also for
4435 liblustre clients there is ability to refuse connection of too old
4437 Details : New 'version' field is added to connect data structure that is
4438 filled with version info. That info is later checked by server and
4442 Frequency : rare, liblustre only.
4443 Bugzilla : 9296, 9581
4444 Description: Two simultaneous writes from liblustre at offset within same page
4445 might proceed at the same time overwriting eachother with stale
4447 Details : I/O lock withing llu_file_prwv was released too early, before data
4448 actually was hitting the wire. Extended lock-holding time until
4449 server acknowledges receiving data.
4452 Frequency : extremely rare. Never observed in practice.
4454 Description: avoid generating lustre_handle cookie of 0.
4455 Details : class_handle_hash() generates handle cookies by incrementing
4456 global counter, and can hit 0 occasionaly (this is unlikely, but
4457 not impossible, because initial value of cookie counter is
4458 selected randonly). Value of 0 is used as a sentinel meaning
4459 "unassigned handle" --- avoid it. Also coalesce two critical
4460 sections in this function into one.
4462 Severity : enhancement
4464 Description: allow liblustre clients to delegate truncate locking to OST
4465 Details : To avoid overhead of locking, liblustre client instructs OST to
4466 take extent lock in ost_punch() on client's behalf. New connection
4467 flag is added to handle backward compatibility.
4469 Severity : enhancement
4470 Bugzilla : 4928, 7341, 9758
4471 Description: allow number of OST service threads to be specified
4472 Details : a module parameter allows the number of OST service threads
4473 to be specified via "options ost ost_num_threads={N}" in the
4474 OSS's /etc/modules.conf or /etc/modprobe.conf.
4478 Bugzilla : 6146, 9635, 9895
4479 Description: servers crash with bad pointer in target_handle_connect()
4480 Details : In rare cases when a client is reconnecting it was possible that
4481 the connection request was the last reference for that export.
4482 We would temporarily drop the export reference and get a new
4483 one, but this may have been the last reference and the export
4484 was just destroyed. Get new reference before dropping old one.
4486 Severity : enhancement
4487 Frequency : if client is started with failover MDS
4489 Description: Allow multiple MDS hostnames in the mount command
4490 Details : Try to read the configuration from all specified MDS
4491 hostnames during a client mount in case the "primary"
4494 Severity : enhancement
4496 Description: Stop sending data to evicted clients as soon as possible.
4497 Details : Check if the client we are about to send or are sending data to
4498 was evicted already. (Check is done every second of waiting,
4499 for which l_wait_event interface was extended to allow checking
4500 of exit condition at specified intervals).
4503 Frequency : rare, normally only when NFS exporting is done from client
4505 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4507 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4510 Frequency : always, if srand() is called before liblustre initialization
4512 Description: Liblustre uses system PRNG disturbing its usage by user application
4513 Details : Introduce internal to lustre fast and high-quality PRNG for
4514 lustre usage and make liblustre and some other places in generic
4515 lustre code to use it.
4517 Severity : enhancement
4518 Bugzilla : 9477, 9557, 9870
4519 Description: Verify that the MDS configuration logs are updated when xml is
4520 Details : Check if the .xml configuration logs are newer than the config
4521 logs stored on the MDS and report an error if this is the case.
4522 Request --write-conf, or allow starting with --old_conf.
4524 Severity : enhancement
4526 Description: Handle symlinks in the path when checking if Lustre is mounted.
4527 Details : Resolve intermediate symlinks when checking if a client has
4528 mounted a filesystem to avoid duplicate client mounts.
4533 Description: lconf can hit an error exception but still return success.
4534 Details : The lconf command catches the Command error exception at the top
4535 level script context and will exit with the associated exit
4536 status, but doesn't ensure that this exit status is non-zero.
4541 Description: failure of ptlrpc thread startup can cause oops
4542 Details : Starting a ptlrpc service thread can fail if there are a large
4543 number of threads or the server memory is very fragmented.
4544 Handle this without oopsing.
4547 Frequency : always, only if liblustre and non-default acceptor port was used
4549 Description: liblustre cannot connect to servers with non-default acceptor port
4550 Details : tcpnal_set_default_params() was not called and was therefore
4551 ignoring the environment varaible TCPNAL_PORT, as well as other
4552 TCPNAL_ environment variables
4557 Description: two objects could be created on the same OST for a single file
4558 Details : If an OST is down, in some cases it was possible to create two
4559 objects on a single OST for a single file. No problems other
4560 than potential performance impact and spurious error messages.
4564 Bugzilla : 5681, 9562
4565 Description: Client may oops in ll_unhash_aliases
4566 Details : Client dcache may become inconsistent in race condition.
4567 In some cases "getcwd" can fail if the current directory is
4573 Description: Inode refcounting problems in NFS export code
4574 Details : link_raw functions used to call d_instantiate without obtaining
4575 extra inode reference first.
4579 Bugzilla : 9942, 9903
4580 Description: Referencing freed requests leading to crash, memleaks with NFS.
4581 Details : We used to require that call to ll_revalidate_it was always
4582 followed by ll_lookup_it. Also with revalidate_special() it is
4583 possible to call ll_revalidate_it() twice for the same dentry
4584 even if first occurence returned success. This fix changes semantic
4585 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4586 reference on a request referred from the intent.
4587 ll_intent_release() then releases such a request.
4590 Frequency : rare, normally benchmark loads only
4592 Description: unlinked inodes were kept in memory on the client
4593 Details : If a client is repeatedly creating and unlinking files it
4594 can accumulate a lot of stale inodes in the inode slab cache.
4595 If there is no other client load running this can cause the
4596 client node to run out of memory. Instead flush old inodes
4597 from client cache that have the same inode number as a new inode.
4600 Frequency : SLES9 2.6.5 kernel and long filenames only
4601 Bugzilla : 9969, 10379
4602 Description: utime reports stale NFS file handle
4603 Details : SLES9 uses out-of-dentry names in some cases, which confused
4604 the lustre dentry revalidation. Change it to always use the
4608 Frequency : rare, unless heavy write-truncate concurrency is continuous
4609 Bugzilla : 4180, 6984, 7171, 9963, 9331
4610 Description: OST becomes very slow and/or deadlocked during object unlink
4611 Details : filter_destroy() was holding onto the parent directory lock
4612 while truncating+unlinking objects. For very large objects this
4613 may block other threads for a long time and slow overall OST
4614 responsiveness. It may also be possible to get a lock ordering
4615 deadlock in this case, or run out of journal credits because of
4616 the combined truncate+unlink. Solution is to do object truncate
4617 first in one transaction without parent lock, and then do the
4618 final unlink in a new transaction with the parent lock. This
4619 reduces the lock hold time dramatically.
4622 Frequency : rare, 2.4 kernels only
4624 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4625 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4626 before unmounting their filesystem in order to avoid deadlock.
4627 The kernel_locked() function in 2.4 kernels only checks whether
4628 the kernel lock is held, not whether it is this process that is
4629 holding it as 2.6 kernels do.
4634 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4635 Details : The client ptlrpc code may be trying to reconnect to a down
4636 server before a previous connection attempt has timed out.
4637 Increase the reconnect interval to be longer than the connection
4638 timeout interval to avoid sending duplicate connections to
4642 Frequency : echo_client brw_test command
4644 Description: fix echo_client to work with OST preallocated code
4645 Details : OST preallocation code (5137) didn't take echo_client IO path
4646 into account: echo_client calls filter methods outside of any
4647 OST thread and, hence, there is no per-thread preallocated
4648 pages and buffers to use. Solution: hijack pga pages for IO. As
4649 a byproduct, this avoids unnecessary data copying.
4653 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4654 Description: Client can oops in mdc_commit_close() after open replay
4655 Details : It was possible for the MDS to return an open request with no
4656 transaction number in mds_finish_transno() if the client was
4657 evicted, but without actually returning an error. Clients
4658 would later try to replay that open and may trip an assertion
4659 Simplify the client close codepath, and always return an error
4660 from the MDS in case the open is not successful.
4663 Frequency : rare, 2.6 OSTs only
4665 Description: OST may deadlock under high load on fragmented files
4666 Details : If there was a heavy load and highly-fragmented OST filesystems
4667 it was possible to have all the OST threads deadlock waiting on
4668 allocation of biovecs, because the biovecs were not released
4669 until the entire RPC IO was completed. Instead, release biovecs
4670 as soon as they are complete to ensure forward IO progress.
4672 Severity : enhancement
4674 Description: Support for specifying external journal device at mount
4675 Details : If an OST or MDS device is formatted with an external journal
4676 device, this device major/minor is stored in the ext3 superblock
4677 and may not be valid for failover. Allow detecting and
4678 specifying the external journal at mount time.
4683 Description: Mounting an MDS with pending unlinked files may cause oops
4684 Details : target_finish_recovery() calls mds_postrecov() which returned
4685 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4686 considers this an error and immediately begins cleaning up the
4687 lov, just after starting the mds_lov process
4689 Severity : enhancement
4691 Description: Implement 'lfs df' to report actual free space on per-OST basis
4692 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4693 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4694 usage of objects; '-h' to report in human readable format.
4696 ------------------------------------------------------------------------------
4698 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4705 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4706 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4707 at startup during orphan unlinking. Instead, skip buddy update
4708 before setup, as it will be regenerated after recovery is complete.
4713 Description: If an OST is inactive, its locks might reference stale inodes.
4714 Details : lov_change_cbdata() must iterate over all namespaces, even if
4715 they are inactive to clear inode references from the lock.
4717 Severity : enhancement
4718 Frequency : occasional, if non-standard max_dirty_mb used
4720 Description: Client will block write RPCs if not enough grant
4721 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4722 then the client will block writes while waiting for another RPC
4723 to complete instead of consuming its dirty limit. With change
4724 we get improved performance when max_dirty_mb is small.
4726 Severity : enhancement
4727 Bugzilla : 3389, 6253
4728 Description: Add support for supplementary groups on the MDS.
4729 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4730 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4731 lookups for user supplementary groups into a cache.
4735 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4736 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4737 for liblustre O_EXCL is enforced by MDS.
4741 Description: readdir never returns NULL in liblustre.
4742 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4743 dentry in struct dirent.
4747 Frequency : liblustre only, depends on application IO pattern
4748 Description: liblustre clients evicted if not contacting servers
4749 Details : Don't put liblustre clients into the ping_evictor list, so
4750 they will not be evicted by the pinger ever.
4752 Severity : enhancement
4754 Description: Add ability to evict clients by NID from MDS.
4755 Details : By echoing "nid:$NID" string into
4756 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4757 $NID would be instantly evicted from this MDS and from all active
4758 OSTs connected to it.
4762 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4763 Details : lookup_it_finish() used to query file size from OSTs that was not
4768 Description: service threads change working directory to that of init
4769 Details : Starting lustre service threads may pin the working directory
4770 of the parent thread, making that filesystem busy. Threads
4771 now change to the working directory of init to avoid this.
4775 Frequency : during shutdown only
4776 Description: shutdown with a failed MDS or OST can cause unmount to hang
4777 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4782 Frequency : 2.6 only
4783 Description: chmod/chown may include an extra supplementary group
4784 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4785 supplementary group and if none is specified this is used.
4788 Bugzilla : 5479 (6816)
4790 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4791 Details : If lookup is in progress on a file that is unlinked we might try
4792 to revalidate the inode and fail in revalidate after lookup is
4793 complete and ll_file_open() enqueues the open again but
4794 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4797 Frequency : always, if lconf --abort_recovery used
4799 Description: lconf --abort_recovery fails with 'Operation not supported'
4800 Details : lconf was attempting to abort recovery on the MDT device and not
4803 ------------------------------------------------------------------------------
4805 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4810 Frequency : rare (only unsupported configurations with a node running as an
4812 Bugzilla : 6514, 5137
4813 Description: Mounting a Lustre file system on a node running as an OST could
4815 Details : OSTs now preallocates memory needed to write out data at
4816 startup, instead of when needed, to avoid having to
4817 allocate memory in possibly low memory situations.
4818 Specifically, if the file system is mounted on on OST,
4819 memory pressure could force it to try to write out data,
4820 which it needed to allocate memory to do. Due to the low
4821 memory, it would be unable to do so and the node would
4822 become unresponsive.
4824 Severity : enhancement
4826 Description: Addition of lconf --service command line option
4827 Details : lconf now accepts a '--service <arg>' option, which is
4828 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4830 Severity : enhancement
4832 Description: Failover mode is now the default for OSTs.
4833 Details : By default, OSTs will now run in failover mode. To return to
4834 the old behaviour, add '--failout' to the lmc line for OSTs.
4836 Severity : enhancement
4838 Description: Health checks are now provided for MDS and OSTs
4839 Details : Additional detailed health check information on MSD and OSTs
4840 is now provided through the procfs health_check value.
4843 Frequency : occasional, depends on IO load
4845 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4846 after numerous create/delete cycles
4847 Details : The ext3 inode allocation policy would not allocate new inodes
4848 very well on the OSTs because there are no new directories
4849 being created. Instead we look for groups with free space if
4850 the parent directories are nearly full.
4855 Description: Network or server problems during mount may cause partially
4856 mounted clients instead of returning an error.
4857 Details : The config llog parsing code may overwrite the error return
4858 code during mount error handling, returning success instead
4864 Description: MDS can fail to allocate large reply buffers
4865 Details : After long uptimes the MDS can fail to allocate large reply
4866 buffers (e.g. zconf client mount config records) due to memory
4867 fragmentation or consumption by the buffer cache. Preallocate
4868 some large reply buffers so that these replies can be sent even
4869 under memory pressure.
4873 Frequency : rare (liblustre)
4874 Description: fsx running with liblustre complained that using truncate() to
4875 extend the file doesn't work. This patch corrects that issue.
4876 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4877 ATTR_SIZE and lsm use in llu_setattr_raw.
4881 Frequency : rare, only 2.6 kernels
4882 Description: Unusual file access patterns on the MDS may result in inode
4883 data being lost in very rare circumstances.
4884 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4885 code in the 2.6 kernel caused the inode and/or EA data not to
4886 be read from disk, causing single-file corruption.
4890 Frequency : rare, only 2.6 filesystems using extents
4891 Description: Heavy concurrent write and delete load may cause data corruption.
4892 Details : It was possible under high-load situations to have an extent
4893 metadata block in the block device cache from a just-unlinked
4894 file overwrite a newly-allocated data block. We now unmap any
4895 metadata buffers that alias just-allocated data blocks.
4899 Frequency : filesystems with default stripe_count larger than 77
4900 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4901 Details : lconf specifies an inode size of 4096 bytes when the default
4902 stripe_count is larger than 77. This conflicts with the default
4903 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4904 this case to avoid pinning too much memory for large EAs.
4906 ------------------------------------------------------------------------------
4908 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4913 Frequency : rare (extremely heavy IO load with hundreds of clients)
4915 Description: Client is evicted, gets IO error writing to file
4916 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4917 caused clients to be evicted for AST timeouts. The fixes in
4918 bug 5192 mean we no longer need to have such short AST timeouts
4919 so ldlm_timeout has been increased.
4922 Frequency : occasional during --force or --failover shutdown under load
4923 Bugzilla : 5949, 4834
4924 Description: Server oops/LBUG if stopped with --force or --failover under load
4925 Details : a collection of import/export refcount and cleanup ordering
4926 issues fixed for safer force cleanup
4929 Frequency : only filesystems larger than 120 OSTs
4930 Bugzilla : 5990, 6223
4931 Description: lfs getstripe would oops on a very large filesystem
4932 Details : lov_getconfig used kfree on vmalloc'd memory
4935 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4936 Bugzilla : 6242, 6243
4937 Description: reading from files that had been truncated to a non-zero size
4938 but never opened returned no data
4939 Details : ll_file_read() reads zeros from no-object files to EOF
4944 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4945 Details : There's little harm in aborting MDS/OSS recovery and letting it
4946 try again, so I removed the LASSERT and return an error instead.
4948 Severity : enhancement
4950 Description: New debugging infrastructure for tracking down data corruption
4951 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4952 (b) cover more of the client-side code path, and (c) try to narrow
4953 down where problems occurred
4957 Bugzilla : 3819, 4364, 4397, 6313
4958 Description: Racing close and eviction MDS could cause assertion in mds_close
4959 Details : It was possible to get multiple mfd references during close and
4960 client eviction, leading to one thread referencing a freed mfd.
4962 Severity: : enhancement
4963 Bugzilla : 3262, 6359
4964 Description: Attempts to reconnect to servers are now more aggressive.
4965 Details : This builds on the enhanced upcall-less recovery that was added
4966 in 1.4.2. When trying to reconnect to servers, clients will
4967 now try each server in the failover group every 10 seconds. By
4968 default, clients would previously try one server every 25 seconds.
4973 Description: After recovery, certain operations trigger a failed
4974 assertion on a client.
4975 Details : Failing over an mds, using lconf -d --failover, while a
4976 client was doing a readdir() call would cause the client to
4977 LBUG after recovery completed and the readdir() was resent.
4979 Severity : enhancement
4981 Description: Default groups are now added by lconf
4982 Details : You can now run lconf --group <servicename> without having to
4983 manually add groups with lmc.
4986 Frequency : occasional
4988 Description: Nodes with an elan id of 0 trigger a failed assertion
4991 Frequency : always when accessing e.g. tty/console device nodes
4993 Description: tty and some other devices nodes cannot be used on lustre
4994 Details : file's private_data field is used by device data and lustre
4995 values in there got lost. New field was added to struct file to
4996 store fs-specific private data.
4999 Frequency : when exporting Lustre via NFS
5001 Description: NFSD failed occasionally when looking up a path component
5002 Details : NFSD is looking up ".." which was broken in ext3 directories
5003 that had grown large enough to become hashed.
5006 Frequency : Clusters with multiple interfaces not on the same subnet
5008 Description: Nodes will repeatedly try to reconnect to an interface which it
5009 cannot reach and report an error to the log.
5010 Details : Extra peer list entries will be created by lconf with some peers
5011 unreachable. lconf now validates the peer before adding it.
5014 Frequency : Only if a default stripe is set on the filesystem root.
5016 Description: Setting a default stripe on the filesystem root prevented the
5017 filesystem from being remounted.
5018 Details : The client was sending extra request flags in the root getattr
5019 request and did not allocate a reply buffer for the dir EA.
5022 Frequency : occasional, higher if lots of files are accessed by one client
5023 Bugzilla : 6159, 6097
5024 Description: Client trips assertion regarding lsm mismatch/magic
5025 Details : While revalidating inodes the VFS looks up inodes with ifind()
5026 and in rare cases can find an inode that is being freed.
5027 The ll_test_inode() code will free the lsm during ifind()
5028 when it finds an existing inode and then the VFS later attaches
5029 this free lsm to a new inode.
5033 Bugzilla : 6422, 7030
5034 Description: MDS deadlock between mkdir and client eviction
5035 Details : Creating a new file via mkdir or mknod (starting a transaction
5036 and getting the ns lock) can deadlock with client eviction
5037 (gets ns lock and trying to finish a synchronous transaction).
5040 Frequency : occasional
5041 Description: While starting a server, the fsfilt_ext3 module could not be
5043 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
5044 kernels. Previously, lconf would still use the ext3 name
5045 when trying to load modules. Now, it will correctly use
5046 ext3 on 2.4 and ldiskfs on 2.6.
5048 Severity : enhancement
5049 Description: The default stripe count has been changed to 1
5050 Details : The interpretation of the default stripe count (0, to lfs
5051 or lmc) has been changed to mean striping across a single
5052 OST, rather than all available. For general usage we have
5053 found a stripe count of 1 or 2 works best.
5055 Severity : enhancement
5056 Description: Add support for compiling against Cray portals.
5057 Details : Conditional compiling for some areas that are different
5061 Frequency : occasional
5062 Bugzilla : 6409, 6834
5063 Description: Creating files with an explicit stripe count may lead to
5064 a failed assertion on the MDS
5065 Details : If some OSTs are full or unavailable, creating files may
5066 trigger a failed assertion on the MDS. Now, Lustre will
5067 try to use other servers or return an error to the
5071 Frequency : occasional
5073 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
5074 caused lock timeout during readahead (since 1.4.2).
5075 Details : Processes doing readahead might match a lock that hasn't been
5076 granted yet if there are overlapping and conflicting lock
5077 requests. The readahead process waits on ungranted lock
5078 (original lock is CBPENDING), while OST waits for that process
5079 to cancel CBPENDING read lock and eventually evicts client.
5081 Severity : enhancement
5083 Description: Initial enabling of flock support for clients
5084 Details : Implements fcntl advisory locking and file status functions.
5085 This feature is provided as an optional mount flag (default
5086 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
5087 locking are implemented yet, and those that are are not guaranteed
5088 to be completely correct in production environments.
5089 mount -t lustre -o [flock|noflock] ...
5092 Frequency : occasional
5094 Description: OSTs running 2.4 kernels but with extents enabled might trip an
5095 assertion in the ext3 JBD (journaling) layer.
5096 Details : The b_committed_data struct is protected by the big kernel lock
5097 in 2.4 kernels, serializing journal_commit_transaction() and
5098 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
5099 access to this struct. In 2.6 kernels there is finer grained
5100 locking to improve SMP performance of the JBD layer.
5104 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
5109 Description: merge and cleanup kernel patches.
5112 -----------------------------------------------------------------------------
5114 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
5116 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
5117 versions of Lustre. Please update all servers and clients to
5118 version 1.4.2 or later at the same time. You must also run
5119 "lconf --write-conf {config}.xml" on the MDS while it is stopped
5120 to update the configuration logs.
5122 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
5123 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
5124 - swab llog records of type '0' so we get proper header size/idx (5861)
5125 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
5126 - fix rename of one directory over another leaking an inode (5953)
5127 - avoid SetPageDirty on 2.6 (5981)
5128 - don't re-add just-being-destroyed locks to the waiting list (5653)
5129 - when creating new directories, inherit the parent's custom
5130 striping settings if present parent (3048)
5131 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
5132 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
5133 - don't hold client locks on temporary worklist from l_lru (5666)
5134 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
5135 - automatically evict dead clients (5921)
5136 - Update file size properly in create+truncate+fstat case (6196)
5137 - Do not unhash mountpoint dentries, do not allow removal of
5139 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
5140 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
5141 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
5142 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
5143 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
5144 - Correctly handle reads of files with no objects (6243)
5145 - lctl recover will also mark a device active if deactivate used (5933)
5147 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
5148 - allow --write-conf on an MDS with different nettype than client (5619)
5149 - don't write config llogs to MDS for mounts not from that MDS (5617)
5150 - lconf should create multiple TCP connections from a client (5201)
5151 - init scripts are now turned off by default; run chkconfig --on
5152 lustre and chkconfig --on lustrefs to use them
5153 - upcalls are no longer needed for clients to recover to failover
5155 - add --abort-recovery option to lconf to abort recovery on device
5157 - add support for an arbitrary number of OSTs (3026)
5158 - Quota support protocol changes.
5159 - forward compatibility changes to wire structs (6007)
5160 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
5161 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
5162 - improved "lustre status" script
5163 - initialize blocksize for non-regular files (6062)
5164 - added --disable-server and --disable-client configure options (5782)
5165 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
5166 - Vanilla 2.4.29 support
5167 - increase maximum number of obd devices to 520 (6242)
5168 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
5169 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
5170 SLES 9 2.6.5 kernels
5171 - limit stripes per file to 160 (the maximum EA size) (6093)
5173 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5176 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
5177 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
5178 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
5179 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
5180 - small ext3 extents cleanups and fixes (5733)
5181 - improved mballoc code, several small races and bugs fixed (5733, 5638)
5182 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
5183 - avoid needless client->OST connect, fix handle mismatch (5317)
5184 - fix DLM error path that led to out-of-sync client, long delays (5779)
5185 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
5186 - fix several locking issues related to i_size (5492,5624,5654,5672)
5187 - don't move pending lock onto export if it is already evicted (5683)
5188 - fix kernel oops when creating .foo in unlinked directory (5548)
5189 - fix deadlock in obdfilter statistics vs. object create (5811)
5190 - use time_{before,after} to avoid timer jiffies wrap (5882)
5191 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
5192 - Do not leak request if server was not able to process it (5154)
5193 - If mds_open unable to find parent dir, make that negative lookup(5154)
5194 - don't create new directories with extent-mapping (5909, 5936)
5196 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
5197 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
5198 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
5199 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
5200 - support for RHEL4 kernel on i686 (5773)
5201 - provide error messages when incompatible logs are encountered (5898)
5203 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
5204 * version 1.4.0.10 (1.4.1 release candidate 1)
5206 - don't keep a lock reference when lock is not granted (4238)
5207 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5208 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
5209 - drop import inflight refcount on signal_completed_replay error (5255)
5210 - unlock page after async write error during send (3677)
5211 - handle missing objects in filter_preprw_read properly (5265)
5212 - no transno return for symlink open, don't save no-trasno open (3440)
5213 - don't try to complete elan receive that already failed (4012)
5214 - free RPC server reply state on error (5406)
5215 - clean up thread from ptlrpc_start_thread() on error (5160)
5216 - readahead could read extra page into cache that wasn't ejected (5388)
5217 - prevent races in class_attach/setup/cleanup/detach (5260)
5218 - don't dereference de->d_inode after l_dput of de (5458)
5219 - use "int" for stripe value returned from lock_to_stripe (5544)
5220 - mballoc allocation and error-checking fixes in 2.6 (5504)
5221 - block device patches to fix I/O request sizes in 2.6 (5482)
5222 - look up hostnames for IB nals (5602)
5223 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
5224 - don't start multiple acceptors for the same port (5277)
5225 - fix incorrect LASSERT in mds_getattr_name (5635)
5226 - export a proc file for general "ping" checking (5628)
5227 - fix "lfs check" to not block when the MDS is down (5628)
5229 - service request history (4965)
5230 - put {ll,lov,osc}_async_page structs in a single slab (4699)
5231 - create an "evict_client" /proc entry on OSTs, like the MDS has
5232 - fix mount usage message, return errors per mount(8) (5168)
5233 - change grep [] to grep "[]" in tests so they work in more UMLs
5234 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
5235 - remove ancient LOV_MAGIC_V0 EA support (5047)
5236 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
5237 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
5238 - allow readahead allocations to fail when low on memory (5383)
5239 - mmap locking landed again, after considerable improvement (2828)
5240 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
5242 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
5245 - send OST transaction number in read/write reply to free req (4966)
5246 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5247 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5248 - the watchdog thread now runs as interruptible (5246)
5249 - flock/lockf fixes (but it's still disabled, pending 5135)
5250 - don't use EXT3 constants in llite code (5094)
5251 - memory shortage at startup could cause assertion (5176)
5253 - reorganization of lov code
5254 - single portals codebase
5256 - add extents/mballoc support (5025)
5257 - direct I/O reads in the obdfilter (4048)
5258 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5260 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5263 - send OST transaction number in read/write reply to free req (4966)
5264 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5265 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5266 - the watchdog thread now runs as interruptible (5246)
5267 - handle missing objects in filter_preprw_read properly (5265)
5268 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5269 - drop import inflight refcount on signal_completed_replay error (5255)
5270 - unlock page after async write error during send (3677)
5271 - return original error code on reconstructed replies (3761)
5272 - no transno return for symlink open, don't save no-trasno open (3440)
5274 - add pid to ldlm debugging output (4922)
5275 - bump the watchdog timeouts -- we can't handle 30sec yet
5276 - extra debugging for orphan dentry/inode bug (5259)
5278 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5281 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5282 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5283 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5284 - fix NULL dereference in /proc/sys/portals/routes (4827)
5285 - allow failed mdc_close() operations to be interrupted (4561)
5286 - stop precreate on OST before MDS would time out on it (4778)
5287 - don't send partial-page writes before EOF from client (4410)
5288 - discard client grant for sub-page writes on large-page clients (4520)
5289 - don't free dentries not owned by NFS code, check generation (4806)
5290 - fix lsm leak if mds_create_objects() fails (4801)
5291 - limit debug_daemon file size, always print CERROR messages (4789)
5292 - use transno after validating reply (3892)
5293 - process timed out requests if import state changes (3754)
5294 - update mtime on OST during writes, return in glimpse (4829)
5295 - add mkfsoptions to LDAP (4679)
5296 - use ->max_readahead method instead of zapping global ra (5039)
5297 - don't interrupt __l_wait_event() during strace
5299 - add software watchdogs to catch hung threads quickly (4941)
5300 - make lustrefs init script start after nfs is mounted
5301 - fix CWARN/ERROR duplication (4930)
5302 - return async write errors to application if possible (2248)
5303 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5304 - print NAL number in %x format (4645)
5305 - update barely-supported suse-2.4.21-171 series (4842)
5306 - support for sles 9 %post scripts
5307 - support for building 2.6 kernel-source packages
5308 - support for sles km_* packages
5310 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5313 - ignore -ENOENT errors in osc_destroy (3639)
5314 - notify osc create thread that OSC is being cleaned up (4600)
5315 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5316 - reconstruct ost_handle() like mds_handle() (4657)
5317 - create a new thread to do import eviction to avoid deadlock (3969)
5318 - let lconf resolve symlinked-to devices (4629)
5319 - don't unlink "objects" from directory with default EA (4554)
5320 - hold socknal file ref over connect in case target is down (4394)
5321 - allow more than 32000 subdirectories in a single directory (3244)
5322 - fix blocks count for O_DIRECT writes (3751)
5323 - OST returns ENOSPC from object create when no space left (4539)
5324 - don't send truncate RPC if file size isn't changing (4410)
5325 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5326 - bind to privileged port in socknal and tcpnal (3689)
5328 - rate limit CERROR/CWARN console message to avoid overload (4519)
5329 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5330 - basic mmap support (3918)
5331 - kernel patch series update from b1_4 (4711)
5333 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5336 - avoid crash during MDS cleanup with OST shut down (2775)
5337 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5338 - don't use bad inodes on the MDS (3744)
5339 - dynamic object preallocation to improve recovery speed (4236)
5340 - don't hold spinlock over lock dumping or change debug flags (4401)
5341 - don't zero obd_dev when it is force cleaned (3651)
5342 - print grants to console if they go negative (4431)
5343 - "lctl deactivate" will stop automatic recovery attempts (3406)
5344 - look for existing locks in ldlm_handle_enqueue() (3764)
5345 - don't resolve lock handle twice in recovery avoiding race (4401)
5346 - revalidate should check working dir is a directory (4134)
5348 - don't always mark "slow" obdfilter messages as errors (4418)
5350 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5353 - don't close LustreDB during write_conf until it is done (3860)
5354 - fix typo in lconf for_each_profile (3821)
5355 - allow dumping logs from multiple threads at one time (3820)
5356 - don't allow multiple threads in OSC recovery (3812)
5357 - fix debug_size parameters (3864)
5358 - fix mds_postrecov to initialize import for llog ctxt (3121)
5359 - replace config semaphore with spinlock (3306)
5360 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5361 - don't allow enqueue to complete on a destroyed export (3822)
5362 - down write_lock before checking llog header bitmap (3825)
5363 - recover from lock replay timeout (3764)
5364 - up llog sem before sending rpc (3652)
5365 - reduce ns lock hold times when setting kms (3267)
5366 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5367 - fix NULL deref and obd_dev leak on setup error (3312)
5368 - replace some LBUG about llog ops with error handling (3841)
5369 - don't match INVALID dentries from d_lookup and spin (3784)
5370 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5371 - fix invalid assertion in ptlrpc_set_wait (3880)
5373 - add libwrap support for the TCP acceptor (3996)
5374 - add /proc/sys/portals/routes for non-root route listing (3994)
5375 - allow setting MDS UUID in .xml (2580)
5376 - print the stack of a process that LBUGs (4228)
5378 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5381 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5382 - make sure to unset replay flag from failed open requests (3440)
5383 - if default stripe count is 0, use OST count for inode size (3636)
5384 - update parent mtime/ctime on client for create/unlink (2611)
5385 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5386 - free recovery state on server during a forced cleanup (3571)
5387 - unregister_reply for resent reqs (3063)
5388 - loop back devices mounting and status check on 2.6 (3563)
5389 - fix resource-creation race that can provoke i_size == 0 (3513)
5390 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5391 - more debugging for page-accounting assertion (3746)
5392 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5393 - avoid deadlock after precreation failure (3758)
5394 - fix race and lock order deadlock in orphan handling (3450, 3750)
5395 - add validity checks when grabbing inodes from l_ast_data (3599)
5397 - add /proc/.../recovery_status to obdfilter (3428)
5398 - lightweight CDEBUG infrastructure, debug daemon (3668)
5399 - change default OSC RPC parameters to be better on small clusters
5400 - turn off OST read cache for files smaller than 32MB
5401 - install man pages and include them in rpms (3100)
5402 - add new init script for (un)mounting lustre filesystems (2593)
5403 - run chkconfig in %post for init scripts (3701)
5404 - drop scimac NAL (unmaintained)
5406 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5409 - clean kiobufs before and after use (3485)
5410 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5411 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5412 - change init script to fail more clearly if not run as root (1528)
5413 - allow clients to reconnect during replay (1742)
5414 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5415 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5416 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5417 - fix cond_resched (3554)
5418 - only evict unfinished clients after recovery (3515)
5419 - allow bulk resend, prevent data loss (3570)
5420 - dynamic ptlrpc request buffer allocation (2102)
5421 - don't allow unlinking open directory if it isn't empty (2904)
5422 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5423 - remove extraneous obd dereference causing LASSERT failure (3334)
5424 - don't use get_cycles() when creating temp. files on the mds (3156)
5425 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5426 - handle EEXIST for set-stripe, set proper directory name (3336)
5428 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5429 - fix ksocknal_fmb_callback() error messages (2918)
5431 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5434 - don't copy lvb into (possibly NULL) reply on error (2983)
5435 - don't deref dentry after dput, don't free lvb on error (2922)
5436 - use the kms to determine writeback rpc length (2947)
5437 - increment oti_logcookies when osc is inactive (2948)
5438 - update client's i_blocks count via lvb messages (2543)
5439 - handle intent open/close of special files properly (1557)
5440 - mount MDS with errors=remount-ro, like obdfilter (2009)
5441 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5442 - don't use cancelling-locks' kms values (2947)
5443 - use highest lock extent for kms, not last one (2925)
5444 - don't dereference ERR_PTR() dentry in error handling path (3107)
5445 - fix thread race in portals_debug_dumplog() (3122)
5446 - create lprocfs device entries at setup instead of at attach (1519)
5447 - common AST error handler, don't evict client on completion race (3145)
5448 - zero nameidata in detach_mnt in 2.6 (3118)
5449 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5450 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5451 - initialize RPC timeout timer earlier for 2.6 (3219)
5452 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5453 - print nal/nid for unknown nid (3258)
5454 - additional checks for oscc recovery before doing precreate (3284)
5455 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5456 - don't crash in mdc_close for bad permissions on open (3285)
5457 - zero i_rdev for non-device files (3147)
5458 - clear page->private before handing to FS, better assertion (3119)
5459 - tune the read pipeline (3236)
5460 - fix incorrect decref of invalidated dentry (2350)
5461 - provide read-ahead stats and refine rpc in flight stats (3328)
5462 - don't hold journal transaction open across create RPC (3313)
5463 - update atime on MDS at close time (3265)
5464 - close LDAP connection when recovering to avoid server load (3315)
5465 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5466 - don't leak open file on MDS after open resend (3325)
5467 - serialize filter_precreate and filter_destroy_precreated (3329)
5468 - loop device shouldn't call sync_dev() for nul device (3092)
5469 - clear page cache after eviction (2766)
5470 - resynchronize MDS->OST in background (2824)
5471 - refuse to mount the same filesystem twice on same mountpoint (3394)
5472 - allow llmount to create routes for mounting behind routers (3320)
5473 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5474 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5475 - fix rare problem with rename on htree directories (3417)
5477 - allow default OST striping configuration per directory (1414)
5478 - fix compilation for qswnal for 2.6 kernels (3125)
5479 - increase maximum number of MDS request buffers for large systems
5480 - change liblustreapi to be useful for external progs like lfsck (3098)
5481 - increase local configuration timeout for slow disks (3353)
5482 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5484 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5487 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5488 - don't overwrite extent policy data in reply if lock was blocked (2901)
5489 - drop filter export grants atomically with removal from device (2663)
5490 - del obd_self_export from work_list in class_disconnect_exports (2908)
5491 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5492 - swab reply message in mdc_close, other PPC fixes (2464)
5493 - fix destroying of named logs (2325)
5494 - overwrite old logs when running lconf --write_conf (2264)
5495 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5496 - fix race in target_handle_connect (2898)
5497 - mds_reint_create() should take same inode create lock (2926)
5498 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5499 - don't close files for self_export to avoid uninitialized obd (2936)
5500 - allow MDS with the same name as client node (2939)
5501 - hold dentry reference for closed log files for unlink (2325)
5502 - reserve space for all logs during transactions (2059)
5503 - don't evict page beyond end of stripe extent (2925)
5504 - don't oops on a deleted current working directory (2399)
5505 - handle hard links to targets without a parent properly (2517)
5506 - don't dereference NULL lock when racing during eviction (2867)
5507 - don't grow lock extents when lots of conflicting locks (2919)
5509 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5512 - account for cache space usage on clients to avoid data loss (974)
5513 - lfsck support in lustre kernel code (2349)
5514 - reduce journal credits needed for BRW writes (2370)
5515 - orphan handling to avoid losing space on client/server crashes
5516 - ptlrpcd can be blocked, stopping ALL progress (2477)
5517 - use lock value blocks to assist in proper KMS, faster stat (1021)
5518 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5519 - recovery for initial connections (2355)
5520 - fixes for mds_cleanup_orphans (1934)
5521 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5522 - block all file creations until orphan recovery completes (1901)
5523 - client remove rq_connection from request struct (2423)
5524 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5525 - recovery timer race (2670)
5526 - mdc_close recovey bug (2532)
5527 - ptlrpc cleanup bug (2710)
5528 - mds timeout on local locks (2588)
5529 - namespace lock held during RPCs (2431)
5530 - handle interrupted sync write properly (2503)
5531 - don't try to handle a message that hasn't been replied to (2699)
5532 - client assert failure during cleanup after abort recovery (2701)
5533 - leak mdc device after failed mount (2712)
5534 - ptlrpc_check_set allows timedout requests to complete (2714)
5535 - wait for inflight reqs when ptlrpcd finishes (2710)
5536 - make sure unregistered services are removed from the srv_list
5537 - reset bulk XID's when resending them (caught by 1138 test)
5538 - unregister_bulk after timeout
5539 - fix lconf error (2694)
5540 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5541 - readahead locks pages, leaves pending causing memory pressure (2673)
5542 - increase OST request buffers to 4096 on large machines (2729)
5543 - fix up permission of existing directories in simple_mkdir (2661)
5544 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5545 - don't assign transno to errored transactions (2742)
5546 - don't delete objects on OST if given a bogus objid from MDS (2751)
5547 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5548 - if rq_no_resend, then timeout request after recovery (2432)
5549 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5550 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5551 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5552 - MDS recovery shouldn't delete objects that it has given out (2730)
5553 - if enqueue arrives after completion, don't clobber LVB (2819)
5554 - don't unlock pages twice when trigger_group_io returns error (2814)
5555 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5556 - don't write pages to disk if there was an error (1450)
5557 - don't ping imports that have recovery disabled (2676)
5558 - take buffered bytes into account when balancing socknal conn (2817)
5559 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5560 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5561 - remove little-endian swabbing of llog records (1987)
5562 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5563 - retry reposting request buffers if they fail (1191)
5564 - grow extent at grant time to avoid granting a revoked lock (2809)
5565 - lock revoke doesn't evict page if covered by a second lock (2765)
5566 - disable VM readahead to avoid reading outside lock extents (2805)
5568 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5569 - updated kernel patches for hp-2.4.20 kernel (2681)
5571 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5574 - fix truncated write corruption (2366)
5575 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5577 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5578 - ensure len(uuid) < 37 in lmc (1171)
5579 - fix ia64 OOPS in llog_test (2255)
5580 - zero end of page at obdfilter for partial page writes (2648)
5581 - don't leave stale dentries around after renames (bug 2428)
5582 - fix timeouts when evicting a client with a single lock held (2642)
5583 - set deadline for the initial HELLO message to drain (2634)
5584 - print out dotted-quad IP addresses in the socknal (2302)
5586 - additional debugging for MDS client eviction problem (2443)
5587 - fix mkfsoptions support for osts (2603, 2604)
5589 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5592 - add series for the vanilla 2.6.0 kernel
5593 - add series for the vanilla 2.4.24 kernel
5594 - add series for a cray x86/64 UL kernel drop
5595 - fix xattr patches for the vanilla 2.4.19 series
5597 - generate true UUIDs in lmc (1171)
5598 - have portals stack dumping break in UML (2466)
5599 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5600 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5601 - wake callback waiting threads on client eviction (2460)
5602 - Add --ptldebug and --subsystem to lmc (1719)
5603 - update assertion to allow safe interrupt allocation
5604 - set rq_no_resend for cancel requests (2432)
5605 - recalculate ptlrpcd timeout after resend (2494)
5606 - call vfs_rmdir when removing pending directories (2368)
5607 - fix renaming a file to itself (2429)
5608 - lmc creates a default one-stripe lov (2454)
5609 - expand procfs space to handle large clusters (2326)
5610 - increase UML stack to avoid overflow
5611 - update lconf's list of debug and subsystem masks
5612 - fix lfs find --obd (2510)
5613 - /proc tunable for disabling filter read caching (2591)
5614 - stop rpm packages from altering slapd.conf (2301)
5615 - disable nagle in the socknal under 0conf (2578)
5616 - choose mds inode size based on stripe count (2572)
5617 - fix kernel-source rpm problems (2516)
5619 - add --disable-doc to avoid pdf generation (2421)
5620 - update documentation, tests, type-os, comments
5621 - avoid format warnings on ia64
5622 - remove the TOE NAL
5623 - tiny code cleanups by removing unused fields
5625 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5628 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5629 - remove the most blatant lies from BUILDING file (2371)
5630 - change default debug level to reasonable production setting
5631 - reduce client side cache size to reduce cache flush time
5632 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5633 - make TCP zerocopy and pinger support enabled by default (2476)
5634 - sync writes completed after process exits caused crashes (2319)
5635 - maintain correct mount count on the MDS (2356)
5636 - backout 1557, because 2316 wasn't really fixed
5637 - better file I/O statistics gathering in /proc
5638 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5639 - another kernel patch to fix zero-copy TCP function export
5640 - don't take duplicate lock when processing re-sent getattr (2420)
5641 - lctl uses obd_self_export instead of creating new conn (2353)
5642 - MDS/OST recovery case which requires object creation asserted (2425)
5643 - move lfs from /usr/sbin to /usr/bin in packages
5644 - fix race between mds_client_add and mds_client_free (2417)
5645 - use kmalloc instead of slabs in portals (2430)
5646 - don't create duplicate records when a failover MDS is present (2442)
5647 - remove unnecessary mount age check (2332)
5648 - don't remove directory inodes from locks prematurely (2451)
5649 - don't break if MDS service name is the same as hostname (2103)
5650 - fix races in client write RPC generation when cache full (2482)
5652 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5655 - remove now-unused request->rq_obd (278)
5656 - if an allocation fails, print out how much memory we've used (1933)
5657 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5658 - add the "configurable stack size" patch to most series files (1256)
5659 - ability to write large log records, for 100+ OST configs (2306)
5660 - fix NULL deref when filter_prep fails (2314)
5661 - fix operator precedence error in filter_sync
5662 - dynamic allocation of socknal TX descriptors (2315)
5663 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5664 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5665 - ensure that CWARN messages in Portals always get to the syslog
5666 - __init/__exit are not for prototype decls (ldlm_init/exit)
5667 - x86-64 compile warning fixes
5668 - fix gateway LMC keyword conflict (2318)
5669 - fix MDS lock inversions in getattr/reint paths (1844)
5670 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5671 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5672 - DLM race condition prevented some lock evictions (2328)
5673 - ENOMEM detection and retry on socknal sends (2230)
5674 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5675 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5676 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5677 - better error messages when a client is rejected during recovery (1505)
5678 - avoid cancelling locks which were never granted, after failure (2330)
5679 - fix i_sem/journal inversion in mds_client_add (2333)
5680 - fix truncate/getattr lock cycle deadlock (2334)
5681 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5682 - fix two rare exit paths which could leak an l_lock() ref (2321)
5683 - fencepost error in MDS/OST orphan recovery (2226)
5684 - make log record alignment 8 bytes (1988)
5685 - lstripe now fails when requested offset > ost_count (2237)
5686 - ensure that all kernel series have a complete list.h (1607)
5687 - fix crashes in special-file operations (2316)
5688 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5690 - allow configurable automake binary, for testing new versions
5691 - small update to the lfs documentation
5693 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5695 * fix negative export reference count in fsfilt_sync (2312)
5697 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5698 * release candidate 0.9.1
5700 - orphans are moved into the PENDING directory for possible recovery
5701 - replayed opens now open by fid for orphan/rename safety (1042)
5702 - last close of an orphan inode generates a transno (683)
5703 - chdir() and mount() now pin the directory entry (1020)
5704 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5705 - discard very old requests without processing them (1502)
5706 - remove some common, well-understood CERRORs (1505)
5707 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5708 - clear "grant" flags in OST replies until OST grant code lands (1644)
5709 - fix read performance by not clobbering i_blksize on client (1598)
5710 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5711 - make LRU size a /proc tunable, clears locks when reduced (707)
5712 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5713 - server locks take references on exports now (1558)
5714 - build fixes for 2.4.20-rh trees (1663)
5715 - return an error from lov_create if all OSCs are inactive (1751)
5716 - fix import levels when a reconnect happens without a timeout (1597)
5717 - exit early from mds_open if we get a lookup error (1749)
5718 - partial page read at EOF wouldn't wait for disk before sending (1642)
5719 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5720 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5721 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5722 - write/truncate lock inversion (1639)
5723 - Don't auto-load obdclass, portals modules during cleanup (1495)
5724 - fix timestamps from jumping to "now" (1763)
5725 - extra journal assertions (1648)
5726 - add an extra multiunlink test (1771)
5727 - fix read_record/write_record API (1776)
5728 - fix leak of offset_extent, possible incorrect i_size later (1772)
5729 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5730 - Debugging for the kqswnal_get_idle_tx problems (1820)
5731 - Allow recovery to be attempted multiple times (1536)
5732 - Write out MDS last_rcvd file after it is first created (1600)
5733 - Fix tx_descriptor leak in failed transmit situations (1827)
5734 - ext3 journaling fixes for assertion failure after IO error (1871)
5735 - class_export_put() on freed export after completion AST error (1896)
5736 - Fix revalidate looping in VFS (1322)
5737 - Don't access a freed export during MDS_REINT timeout (1521)
5738 - Add open-unlink recovery support on the MDS (1673,1764)
5739 - Return an error if no MDS data was read from last_rcvd (1946)
5740 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5741 - Don't setup a disk device that doesn't match exported UUID (317)
5742 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5743 - avoid committing NULL handle in force close
5744 - local.sh is now a one-stripe LOV configuration
5745 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5746 - fix ext3 htree duplicate directory entry corruption (1516)
5747 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5748 - update to new LOV EA format (2097)
5749 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5751 2003-06-15 Phil Schwan <phil@clusterfs.com>
5754 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5755 - per-import/export recovery handling (958, 931, 959)
5756 - multiple last-rcvd slots, for serving multiple FSes (949)
5757 - connections are again shared between multiple imp/exports (963, 964)
5758 - "umount -f" would hang if any requests needed to be sent (393, 978)
5759 - avoid pinning large req buffer by copying for queued messages (989)
5760 - add "uuid" to "lctl device" command to help upcalls (991)
5761 - "open" RPCs with transnos would confuse recovery counters (1037)
5762 - do proper endian conversion of all wire messages (288, 340, 891)
5763 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5764 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5765 - fix ptlrpc_connection leak in target_handle_connect (1174)
5766 - fix import refcounting bug in OST and MDS cleanup (1134)
5767 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5768 - fix very unlikely obd_types race condition (501)
5769 - remove osc_open hack for echo_client (1187)
5770 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5771 - a failure in read_inode2 leads to deadlock (1139)
5772 - cancel ack-locks as soon as transaction is committed (1072)
5773 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5774 - make sure to commitrw after any preprw to avoid deadlock (1162)
5775 - failing to execute a file in a lustre FS would lock inode (1203)
5776 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5777 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5778 - obd_uuid2tgt would walk off the end of the list (1255)
5779 - on IA64 the osc would give portals incorrect bulk size (1258)
5780 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5781 - fix lock inversion caused by new llite matching code (1282)
5782 - limit the number of dirty pages on a client to 10MB (1286)
5783 - timed out locks were not being corrected cancelled (1289)
5784 - fix O_DIRECT above 4GB on IA-32 (1292)
5785 * major user-visible changes
5786 - fail out/fail over policy now controlled by the upcall (993)
5788 - add OBD_PING to check server availability and failure (954)
5789 - lustre messages are now sent in sending host order (288, 340, 891)
5790 - add eadatalen to MDS getattr reply (340)
5791 - OST read replies may contain second buffer, with per-page status (593)
5793 2003-03-11 Phil Schwan <phil@clusterfs.com>
5796 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5797 - failed open()s could cause deadlock; fixed (867, 869)
5798 - stop cancelling OST locks when files are closed (481)
5799 - overlapping XID spaces caused network corruption (851, 853)
5800 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5801 - setattr_raw added, to do single-RPC, server-side setattrs
5802 - lmc/lconf syntax change for OST UUIDs
5803 - fix crashy race condition between ptlrpc_free_req and osc_close
5804 - don't use request in mdc_enqueue if we hit a timeout (889)
5805 - don't set the inode i_size for regular files from the MDS (896)
5806 - handle out of order completion AST (842)
5807 - don't LBUG if a lock request times out after receiving AST (913)
5808 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5809 - if a bad lock AST arrives, send an error instead of dropping entirely
5810 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5811 - fix leak in bulk IO when only partially completed (899, 900, 926)
5812 - fix O_DIRECT for ia64 (55)
5813 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5814 - C-z after timeout could hang a process forever; fixed (977)
5816 - client-side I/O cache (678, 924, 929, 941, 970)
5818 - READPAGE and SETATTRs which don't take server-side locks get
5821 2003-02-11 Phil Schwan <phil@clusterfs.com>
5824 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5825 - Fix fsfilt_extN_readpage() to read a full page of directory
5826 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5827 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5828 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5829 - Most intent operations were removed, in favour of a new RPC mode
5830 that does a single RPC to the server and bypasses most of the VFS
5831 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5832 - Aggressively cancel local locks on DLM servers
5833 - mds_reint_unlink sends EA to the client if it's the last nlink.
5834 client uses that EA to unlink OST objects.
5835 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5836 - recursive symlinks were fixed (439)
5837 - fixed NULL deref in DEBUG_REQ
5838 - filter_update_lastobjid no longer calls sync, which annoyed extN
5839 - fixed multi-client small-writes to a single file problem (445)
5840 - fixed mtime updates during file writes (607)
5841 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5842 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5843 - fixed Linux OST size reporting problem (444, 656)
5844 - OST now updates object mtime with writes or setattr (607, 619)
5845 - client verifies file size before zeroing page past EOF (445)
5846 - OST now writes last allocated objid to disk with allocation (108)
5847 - LOV on echo now works (409)
5849 - mds_reint_unlink sends a new buffer, with the EA included. this
5850 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5851 set if a regular file was being unlinked, and it was the last link
5852 - use PtlGet from the target for bulk writes (315)
5853 - OST now updates object mtime with writes or setattr (607, 619)
5854 - LDLM now has a grant-time callback to revalidate locked items, if
5856 - Many MDS operations were reorganized to combat race conditions
5858 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5859 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5861 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5864 - Fully reactivate OST imports after reconnection (512, others)
5865 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5866 - More graceful error handling for truncating on dead OST (515)
5867 - Don't error out unless we're actually accessing dead stripes (474)
5868 - Fix garbage sizes when stripes are missing (410)
5869 - LRU counters were broken, causing constant lock purge (433, 432)
5870 - garbage on read from stripes with failed OSTs (441)
5871 - mark OSCs as active before reconnecting during recovery (438)
5872 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5873 - lfind did not preserve OST order in output (443)
5874 - symlinks cause hung clients, incorrect data (439)
5875 - stop dereferencing request after dropping refcount (457)
5876 - don't LASSERT(spin_is_locked) on non-SMP (455)
5877 - fixes for many rename() bugs
5878 - fstat didn't correctly synchronize attributes (399)
5879 - server must handle lock cancellation during blocking AST prep (487)
5880 - bulk descriptors were free()d too soon (511)
5881 - fix paths in lconf, which would load incorrect modules (451, 507)
5882 - fix confusing lconf 'host not found' error message (386)
5883 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5884 - fix race condition in mdc_blocking_ast() for inode access (526)
5885 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5886 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5887 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5888 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5889 - fix lstripe in several ways (kernel side) (527)
5890 - fix request leak in ldlm_cli_enqueue (262)
5891 - incorrect OSC was marked inactive after OST failure
5892 - call mds_fs_cleanup before unmounting filesystem (524)
5893 - fix races between taking ns_lock and ldlm_lock_change_resource
5894 - fix races updating LOV export open file list
5895 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5896 - fix recovery NULL deref in ldlm_cli_cancel_unused
5897 - fix some DLM races by using new hash table for lock handles (419)
5898 - permit the client to specify desired inodes, at replay
5899 - duplicate requests when we queue them for replay reintegration
5900 - fix last_rcvd offset calculation
5901 - sync after each recovered transaction, so we always make progress
5902 - never, not always, ERESTART requests without transnos
5903 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5905 - skip replay if the MDS says that the client is already connected
5906 - don't check for a recovery-enabled export to match lctl's UUID
5907 - don't INC_USE_COUNT for phantom exports
5908 - don't crash when cleaning up phantom exports (567)
5909 - don't double-finish or set replay data for errored mdc_open requests
5910 - abort requests when they time out, so we don't get old replies
5911 - send/receive replies for AST messages again
5912 - if the client says that it doesn't have the lock, cancel it on the
5914 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5915 mark it as destroyed, it will all work out when decref is called
5916 - fix module use counts (22, 581)
5918 - ASTs now expect a reply (server cancels lock on error reply)
5920 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5923 - fix many simultaneous client startup (392)
5924 - fix dentry->d_it clobbering
5925 - credentials weren't being shipped for readdir/getattr operations
5926 - remove invalid assertions triggered during some concurrent MD
5928 - proper Lustre versions added (336, 389)
5929 - fix memory leak for create error case (398)
5930 - fix LOV locking bug that would get cli/srv out of sync
5931 - fix echo client over LOV (409)
5932 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5933 - fix double-O_EXCL intent crash (424)
5934 - avoid sending multiple lock CANCELs (352)
5936 - MDS can do multi-client recovery (modulo bugs in new code)
5938 - many updates, edits, cleanups
5940 2002-11-18 Phil Schwan <phil@clusterfs.com>
5943 - fix null d_it dereference (346)
5944 - fix full OST/dbench hang (333)
5945 - fix permission problem with file removal (286)
5946 - fix removal of OSCs from LOV when they fail
5947 - fix NULL deref during bulk timeout (214)
5948 - fix problems related to multiple filesystems on one MDS (241)
5949 - fixed serious subtle metadata locking bugs
5950 - free locks on clients when inodes are removed due to memory
5952 - fix inode pointer in lock data (285)
5953 - partial support for multiple MDS on a single host (241)
5954 - data locks weren't cancelled at clear_inode time (290, 311)
5955 - intent locks could lead to unbounded lock growth (205)
5956 - added a maximum lock count, an LRU list, and a flusher
5957 - fix multiple rename (365)
5958 - properly abstracted the echo client
5959 - OSC locked 1 byte too many; fixed
5960 - rewrote brw callback code:
5961 - fixed recovery bugs related to LOVs (306)
5962 - fixed too-many-pages-in-one-write crash (191)
5963 - fixed (again) crash in sync_io_timeout (214)
5964 - probably fixed callback-related race (385)
5966 - Add capability to MDS protocol
5967 - LDLM cancellations and callbacks on different portals
5969 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5972 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5973 - defer open object destruction to close time (601981, 138)
5974 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5975 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5976 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5977 - don't call obd_setattr in ll_file_release for destroyed objects
5978 * protocol change to lustre_msg: move |version| and add |flags|
5979 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5980 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5981 * added replay of create, unlink, link and rename operations during
5982 MDS failover; recovery should be much more robust now
5983 * remove failed OSCs from LOVs (only lov_create uses this so far)
5984 * the lustre-HOWTO was brought (more) up to date (582544)
5986 2002-10-23 Phil Schwan <phil@clusterfs.com>
5989 - in-use dentries weren't being reused properly (617851)
5990 - prevent multiple LDLM setup (599178)
5991 - fix LOV size calculations for truncate (617853)
5992 - fix client handling of MDS intent errors (POSIX)
5993 - fix permission bug in lovstripe.c test (624321)
5994 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5995 - truncate past end of file could corrupt data
5996 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5997 - a race in recovery could return ETIMEDOUT to apps (623947)
5998 - building outside the source directory was fixed
5999 * the lustre-HOWTO was brought (more) up to date (582544)
6000 * major progress was made on recovery functionality
6002 2002-10-10 Phil Schwan <phil@clusterfs.com>
6005 - recovery deadlock fix
6006 - rm -rf causes LBUG fix (617817)
6007 - file open by multiple tasks fix (618962)
6008 - directory permissions bugs (602707 and 620007)
6009 - journal_stop fixed with locking (611313)
6010 - O_APPEND failures resolved (618273, perhaps 614459)
6011 - lconf PATH fix (619770)
6012 - IA64 build fix (621450)
6013 - RPC buffer sizes scale with amount of memory
6015 2002-10-01 Phil Schwan <phil@clusterfs.com>
6018 - locks would be cancelled without throwing away data pages,
6019 resulting in inconsistent data (605627)
6020 - inode attributes were not always being refreshed (605627, 612449)
6021 - lconf now continues to cleanup after lctl reports an error
6022 - MDS now enforces user permissions (602707)
6023 - lprocfs cleanup fixed, but not yet enabled (614157)
6024 - fixed infinite server hang, should a client not respond to an AST
6025 - avoid going into recovery if user calls readlink() with a buffer
6026 that's too small (613941)
6027 - AST RPCs no longer require replies (614867) -- this may be changed
6028 - don't crash server if client sends an IOV that's too big (611336)
6029 - fixed lock conversion deadlock (611892)
6030 - fixed the following of symlinks (614622)
6031 * recovery: the server can remove locks from a client that dies, other
6032 clients can make progress
6033 * more extN patch fixes
6034 * compile-time configurable ptlrpc buffer allocations
6036 - collaborative read cache document
6037 - Lustre Lite Performance CDR document-in-progress
6039 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6042 - fix typo in patch-2.4.18
6044 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6047 - clear ptlrpc request each time in handle_incoming_request()
6048 - unlink of files now destroys the object on the OST
6050 2002-09-19 Peter Braam <braam@clusterfs.com>
6052 * add hard link support
6053 * change obdfile creation method
6054 * kernel patch changed
6056 2002-09-19 Peter Braam <braam@clusterfs.com>
6059 - stack overflow bug in extN fixed
6061 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
6063 * documentation updates
6064 - add man pages for config tools
6065 - update tests/README to describe testing with new config tools
6066 - finish metadata API descriptions
6067 * bug fixes and cleanups
6068 - statfs workaround for 16TB limit
6069 - LOV stripe allocation improved, can stripe on subset of OSTs
6070 - LOV file size/IO offset was wrong for files > 4GB in size
6071 - object EA data was being dropped, caused files to be unreadable
6072 - memory overflow with non-LOV OST caused memory corruption
6073 - fixed regression tests to work with new config tools, obdfilter
6074 - fixed bug when directory size became larger than 1 block
6075 - fixed bug (for single client case) when PWD was deleted
6076 - invalidate local directory pages when doing intent-based ops
6077 - avoid LDLM oops when lock callback contained bad data
6079 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
6081 * documentation updates
6082 * bug fixes and cleanups
6083 - configuration tools
6086 - 64-bit compile warnings
6087 - 64-bit internal statfs data
6089 * test_brw on persistent OST devices
6091 * lprocfs (disabled)
6093 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
6095 * documentation updates
6096 * bug fixes and cleanups
6097 * configuration tools
6099 2002-08-30 Peter J. Braam <braam@clusterfs.com>
6102 * many small fixes to 0_5_4
6103 * io/network handling
6104 * thinkos in MDS operations
6106 2002-08-24 Peter J. Braam <braam@clusterfs.com>
6109 * crucial basic fixes to 0.5.3
6110 * IOR, Iozone work over Elan
6113 2002-08-07 Phil Schwan <phil@clusterfs.com>
6114 * version 0_5_3, our first alpha
6115 * we use the new Portals iovs
6116 * documentation updates
6117 * bug fixes and cleanups
6118 * small changes in the DLM wire protocol
6120 2002-07-25 Peter J. Braam <braam@clusterfs.com>
6121 * version 0_5_1 with some initial stability,
6122 * locking on MD and file I/O.
6123 * documentation updates
6124 * several bug fixes since 0.5.0
6125 * small changes in wire protocol
6127 2002-07-18 Phil Schwan <phil@clusterfs.com>
6129 * delivered as Lustre Light Alpha
6130 * fixed a crash after handling invalid MDS requests
6131 * fixed directory pages for architectures with non-4k pages sizes
6133 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
6134 * release version v0_4_4
6135 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
6137 * Fixes a number of interruption problems with OST operations.
6138 * Update documentation for portals header changes
6139 * Move all wire protocol structs/defines to lustre_idl.h
6140 * Fixes symlink length bug.
6141 * Add tcpdump to repository.
6143 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
6144 * release version v0_4_3
6145 * Fixes statfs for inodes on extN.
6146 * Fixes bug in runtests which would delete /etc/hosts.
6147 * Use 64-bit object IDs wherever possible (not into VFS though)
6148 Remove ost_get_info, which is unused by lustre, and out of date.
6150 2002-07-03 Peter Braam <braam@clusterfs.com>
6151 * release version v0_4_2 Fixes a lookup error (type not passed)
6152 * move forward to head of Portals
6153 * move forward to latest Lustre kernel
6155 2002-06-25 Peter Braam <braam@clusterfs.com>
6156 * release version v0_4_1. Hopefully stable on single node use.