1 tbd Sun Microsystems, Inc.
3 * Server support for kernels:
4 2.6.18-164.11.1.el5 (RHEL 5)
5 2.6.18-164.11.1.0.1.el5 (OEL 5)
6 * Client support for unpatched kernels:
7 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
8 2.6.18-164.11.1.el5 (RHEL 5),
9 2.6.18-164.11.1.0.1.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.
17 Description: decrease the usage of memory on clients.
18 Details : 1. On clients, recycle dentries and inodes unused.
19 2. Delete the code related to ll_deathrow(att 6215 in bug 1443). It
24 Description: llobdstat fix and enhancement
25 Details : add a counter to set a limit to how many samples will be returned
26 fix a wildcard in the path to limit to obdfilter stats only
30 Description: lfs check only allowed for root
31 Details : Code cleanups and added a regresion test of non-root use of
36 Description: MDS Oopses in ptlrpc_dispatch_difficult_reply()/__list_add()
37 Details : Fix hr->hr_index overload.
41 Description: Fix FIEMAP functionality.
45 Description: FC11 patchless client support added
49 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and
50 OEL5.4 kernel to 2.6.18-164.11.1.0.1.el5.
54 Description: New acc-sm script and various test-framework enhancments.
58 Description: Previously landed patch was reverted.
62 Description: Update SLES10 kernel to 2.6.16.60-0.42.8 (patchless)
66 Description: Add ldiskfs maxdirsize mount option.
70 Description: Fix mkfs.lustre support for 16TB LUNs.
74 Description: Print mmp_check_interval
75 Details : print mmp_check_interval and make it possible to abort
76 mount operation in case it takes too long.
80 Description: Add cascading_rw to lustre mpi tests.
84 Description: Store SOM data in EA.
85 Details : Also several SOM related cleanups.
89 Description: Add refcount to md_open_data
93 Description: Use refcount on lov_request_set to prevent memory corruption.
97 Description: Statahead: use parent dir's i_mutex for synchronization.
101 Description: loadgen improvements.
105 Description: Enable peer health detection in ksocklnd and ko2iblnd.
109 Description: Control DCACHE_LUSTRE_INVALID flag with MDS_INODE_LOOKCUP lock
113 Description: Fix race on obd_devcie:md_stats
117 Description: Fix SLES 11 client builds.
118 Details : Resolved uninitalized pointer gcc warnings.
122 Description: Allow non-root access for "lfs check".
123 Details : Added a check in obd_class_ioctl() for OBD_IOC_PING_TARGET.
127 Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
129 Severity : enhancement
131 Description: Update kernel to RHEL5.4 2.6.18-164.6.1.el5 and
132 OEL5 2.6.18-164.6.1.0.1.el5(Both in-kernel OFED enabled).
136 Description: Sanity test_27p fails.
137 Details : OBD_FAIL_OST_ENOSPC has been moved to filter_statfs().
138 Sanity tests 27 were cleaned up and improved.
142 Description: Unify req format on client/servers
143 Details : Use new req_capsule API [almost] everywhere instead of old PTLRPC
144 buffers and swabbers approach..
149 Description: drop join file stuff.
151 Severity : enhancement
153 Description: Add 2.6.27 and SLES11 (patchless client) support
157 Description: Fix in ptlrpc_expire_one_request() to print the signed time
160 Severity : enhancement
162 Description: Build kernels (RHEL5, OEL5 and SLES10/11) using the vendor's own
165 Severity : enhancement
167 Description: Remove set_info(KEY_UNLINKED) from MDS/OSC
169 Severity : enhancement
171 Description: correctly handle big reply message.
172 Details : send LNet event if reply is bigger then buffer and adjust this buffer
177 Description: Drop unnecessary __GFP_NOMEMALLOC flag from filter_get_page()
179 Severity : enhancement
181 Description: Add support for OEL5.
185 Description: Conf-sanity.sh 50g test - deactivated OST should not cause a panic.
189 Description: client could not reconnect to OST because of an active request.
190 Details : The client now retries to reconnect to the same server, if a
191 connect request failed with EBUSY or -EAGAIN. Test case 35b
192 conf-sanity.sh has been created.
196 Description: Don't allow make backward step on assignin osc next id.
197 Details : race between allocation next id and ll_sync thread can be cause
198 of set wrong osc next id and can be kill valid ost objects.
202 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
206 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
210 Description: Error handling in osc_statfs_interpret() has been improved.
211 Details : Check in osc_statfs_interpret() for EBADR.
215 Description: Do not send statfs() requests to OSTs disabled by administrator.
216 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
218 Severity : enhancement
220 Description: Added sanity-quota test to for lfs(1) to work with numeric uid/gid.
222 Severity : enhancement
224 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
226 Severity : enhancement
228 Description: Enhancement for lfs(1) command to use numeric uid/gid.
232 Description: Improved cleanup procedure for conf-sanity test_46a.
236 Description: Do not start bulk transfer, if client tried to reconnect.
237 Details : Do not start bulk transfer, if client tried to reconnect,
238 but failed due to active rpcs. Check for exp_abort_active_req
239 has been implemented.
243 Description: Prevent inconsistences between linux and lustre mount structures.
244 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
245 Make the sleep interruptible.
247 Severity : enhancement
249 Description: provide server to client comms path
250 Details : server to client communications path via new LDLM_SET_INFO rpc
254 Description: truncate starts GFP_FS allocation under transaction causing deadlock
255 Details : ldiskfs_truncate calls grab_cache_page which may start page
256 allocation under an open transaction. This may lead to
257 calling prune_icache with consequent lustre reentrance.
261 Description: send correctly lsm on open replay
262 Details : MDS is trust to LSM size on replay open, but client can set wrong size
267 Description: enable client interface failover
268 Details : When a child reconnects from another NID, properly update export
269 nid hash position and ldlm reverse import.
271 Severity : enhancement
273 Description: Communicate OST degraded/readonly state via statfs to MDS
274 Details : Flags in the statfs returned from OSTs indicate whether the
275 OST is in a degraded RAID state, or if the filesystem has
276 turned read-only after a filesystem error is detected.
281 Description: don't panic if EPROTO was hit when reading symlink
282 Details : correctly handling request reference in error cases.
284 Severity : enhancement
286 Description: Add LustreNetLink, a kernel-userspace communcation path.
288 Severity : enhancement
290 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
295 Description: access to llog context before init.
296 Details : move handling CATALOGS file at osc layer and forbid access to llog
300 Frequency : rare, only when client is writing to OST on local node
302 Description: Avoid deadlock for local client writes
303 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
304 memory freeing context. This allows OST threads to set the
305 PF_MEMALLOC flag on task structures in order to allocate memory
306 from reserved pools and complete IO.
307 Use GFP_HIGHUSER for OST allocations for non-local client writes,
308 so that the OST threads generate memory pressure and allow
309 inactive pages to be reclaimed.
311 Severity : enhancement
313 Description: Update kernel to RHEL5.3 2.6.18-128.1.14.el5.
318 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
319 Details : move ldlm namespace creation in setup phase to avoid grab
320 _lprocfs_lock with cli_sem held.
324 Description: Temporarily disable grant shrink.
325 Details : Disable the feature for debugging.
328 Frequency : only when formatting test filesystems
330 Description: Unable to run several mkfs.lustre on loop devices at the same time
331 Details : mkfs.lustre returns error 256 on the concurrent loop devices
332 formatting. The solution is to proper handle the error.
334 Severity : enhancement
336 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
338 Severity : enhancement
340 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
345 Description: resolve race between obd_disconnect and class_disconnect_exports
346 Details : if obd_disconnect will be called to already disconnected export he
347 forget release one reference and osc module can't unloaded.
349 Severity : enhancement
351 Description: Allow tuning service thread via /proc
352 Details : For each service a new
353 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
354 created that can be used to set min/max thread counts, and get the
355 current number of running threads.
360 Descriptoin: don't return error if have partially created objects for file.
361 Details : lov_update_create_set uses set->set_success as index for created
362 objects, so if some requests will be failed, they will have hole
363 at end of array and qos_shrink_lsm for allocate correct lsm.
365 Severity : enhancement
367 Description: Update OFED support to 1.4.1
372 Description: Reduce small size read RPC
373 Details : Set read-ahead limit for every file and only do read-ahead when
374 available read-ahead pages are bigger than 1M to avoid small size
377 Severity : enhancement
379 Description: move AT tunable parameters for more consistent usage
380 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
382 Severity : enhancement
384 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
385 Details : allow skip disconnected ost for send statfs request and hide error
391 Description: fix racy locking of mballoc block bitmaps causing BUG
392 Details : The locking of the mballoc buddy bitmap and the in-memory
393 block bitmap was using two different spin locks in some
394 cases. This made it possible to incorrectly access the
395 mballoc bitmap while another process was modifying it,
396 causing a sanity assertion to fail. While no on-disk corruption
397 was reported, there was some risk of this happening.
400 Frequency : rare, on llog test 6
402 Descriptoin: don't allow connect to already connected import
403 Details : allowing connect to already connected import is hide connecting problem.
408 Descriptoin: allow kill process which wait statahead result
409 Details : for some reasons 'ls' can stick in waiting result from statahead,
410 in this case need way for kill this process.
412 Severity : enhancement
414 Description: Add state history info file, enhance import info file
415 Details : Track import connection state changes in a new osc/mdc proc file;
416 add overview-type data to the osc/mdc import proc file.
418 Severity : enhancement
420 Description: MDS create should not wait for statfs RPC while holding DLM lock.
422 Severity : enhancement
424 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
429 Descriptoin: OST grant too much space to client even there are not enough space.
430 Details : Client will shrink its grant cache to OST if there are no write
431 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
432 this grant cache if there are already not enough avaible space
433 (left_space < total_clients * 32M).
436 Frequency : start MDS on uncleanly shutdowned MDS device
438 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
439 Details : stay in waiting mds<>ost recovery finished produce random bugs
440 due race between two ll_sync thread for one lov target. send
441 ACTIVATE event only if connect realy finished and import have
445 Frequency : rare, connect and disconnect target at same time
447 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
448 Details : don't call obd_disconnect under lov_lock. this long time
449 operation and can block ptlrpcd which answer to connect request.
454 Descriptoin: don't lose wakeup for imp_recovery_waitq
455 Details : recover_import_no_retry or invalidate_import and import_close can
456 both sleep on imp_recovery_waitq, but we was send only one wakeup
460 Frequency : always with long access acl
462 Descriptoin: mds can't pack reply with long acl.
463 Details : mds don't control size of acl but they limited by reint/getattr
466 Severity : enhancement
468 Description: Update to SLES10 kernel-2.6.16.60-0.33.
470 Severity : enhancement
472 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
475 Frequency : start MDS on uncleanly shutdowned MDS device
477 Descriptoin: aborting recovery hang on MDS
478 Details : don't throttle destroy RPCs for the MDT.
481 Frequency : on remount
483 Description: external journal device not working after the remount
484 Details : clear dev_rdonly flag for external journal devices in
490 Description: shutdown vs evict race
491 Details : client_disconnect_export vs connect request race.
492 if client will evicted at this time - we start invalidate
493 thread without referece to import and import can be freed
497 Frequency : rare, need acl's on inode.
499 Description: client can't handle ost additional correctly
500 Details : if ost was added after client connected to mds client can have
501 hit lnet_try_match_md ... to big messages to wide striped files.
502 in this case need teach client to handle config events about add
503 lov target and update client max ea size at that event.
505 Severity : enhancement
507 Description: Changelogs
508 Details : Changelogs are a lightweight mechanism to track filesystem
509 metadata and namespace changes. The changelog is recorded
510 permanently on the MDTs, and is periodically "consumed" / purged
511 when records are no longer needed.
513 Severity : enhancement
515 Description: compact fld format with extents
516 Details : Store range of seq rather than every seq in FLD. Seq
517 controller update FLD rather than clients. In Case of CMD, mdt0
518 has FLD, all other metadata server act as non persistent proxy
519 for FLD queries and cache fld entries in fld cache.
524 Description: don't skip ost target if they assigned to file
525 Details : Drop slow OSCs if we can, but not for requested start idx.
526 This means "if OSC is slow and it is not the requested
527 start OST, then it can be skipped, otherwise skip it only
528 if it is inactive/recovering/out-of-space.
532 Description: more cleanup in mds_lov
533 Details : not send LOV EA under replay, we can't know about they size at this
534 time. Don't allow client connect to mds before any ost connected,
535 for avoid problems with LOV EA size and returning EIO to client.
537 Severity : enhancement
539 Description: Interoperability at server side (Disk interoperability)
541 Severity : enhancement
543 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
545 Severity : enhancement
547 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
549 Severity : enhancement
551 Description: New client IO stack (CLIO).
553 Severity : enhancement
555 Description: Commit on sharing. Eliminate inter-client dependencies between
556 uncommitted transactions by doing transaction commits.
557 Thereby clients may recovery independently.
560 Frequency : Create a symlink file with a very long name
562 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
563 Details : If there is no extra space in the request for early cancels,
564 ldlm_req_handles_avail() returns 0 instead of a negative value.
566 Severity : enhancement
568 Description: Add /proc entry for import status
569 Details : The mdc, osc, and mgc import directories now have
570 an import directory that contains useful import data for debugging
573 Severity : enhancement
575 Description: Re-disable certain /proc logging
576 Details : Enable and disable client's offset_stats, extents_stats and
577 extents_stats_per_process stats logging on the fly.
580 Frequency : Only on FC kernels 2.6.22+
582 Description: oops in statahead
583 Details : Do not drop reference count for the dentry from VFS when lookup,
584 VFS will do that by itself.
586 Severity : enhancement
588 Description: Generic /proc file permissions
589 Details : Set /Proc file permissions in a more generic way to enable non-
590 root users operate on some /proc files.
594 Description: Hitting mdc_commit_close() ASSERTION
595 Details : Properly handle request reference release in
596 ll_release_openhandle().
600 Description: quota recovery deadlock during mds failover
601 Details : This patch includes att18982, att18236, att18237 in bz14840.
603 1. fix osts hang when mds does failover with quotaon
604 2. prevent watchdog storm when osts threads wait for the
609 Frequency : only patchless client
610 Description: add workaround for race between add/remove dentry from hash
612 Severity : enhancement
614 Description: Allow OST glimpses to return PW locks
618 Description: LBUG when llog conf file is full
619 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
623 Description: Prevent import from entering FULL state when server in recovery
627 Description: service mount cannot take device name with ":"
628 Details : Only when device name contains ":/" will mount treat it as
634 Description: replace ptlrpcd with the statahead thread to interpret the async
635 statahead RPC callback
639 Frequency : on recovery
640 Description: I/O failures after umount during fail back
641 Details : if client reconnected to restarted server we need join to recovery
642 instead of find server handler is changed and process self
643 eviction with cancel all locks.
645 Severity : enhancement
647 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
649 Severity : enhancement
651 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
653 Severity : enhancement
655 Description: Upcall on Lustre log has been dumped
656 Details : Allow for a user mode script to be called once a Lustre log has
657 been dumped. It passes the filename of the dumped log to the
658 script, the location of the script can be specified via
659 /proc/sys/lnet/debug_log_upcall.
664 Description: avoid idr_remove called for id which is not allocated.
665 Details : Move assigment s_dev for clustered nfs to end of initialization,
666 to avoid problem with error handling.
671 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
672 Details : When connection is reused this not moved from CONN_UNUSED_HASH
673 into CONN_USED_HASH and this prodice warning when put connection
674 again in unused hash.
677 Severity : enhancement
679 Description: File striping can now be set to use an arbitrary pool of OSTs.
681 Severity : enhancement
683 Description: Export bytes_read/bytes_write count on OSC/OST.
687 Description: Early reply size mismatch, MGC loses connection
688 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
689 the connect flags are properly negotiated.
694 Description: quotas are not honored with O_DIRECT
695 Details : all writes with the flag O_DIRECT will use grants which leads to
696 this problem. Now using OBD_BRW_SYNC to guard this.
700 Description: add quota statistics
701 Details : 1. sort out quota proc entries and proc code.
702 2. add quota statistics
704 Severity : enhancement
706 Description: enable quota support for HEAD.
710 Description: Properly propagate oinfo flags from lov to osc for statfs
711 Details : restore missing copy oi_flags to lov requests.
713 Severity : enhancement
715 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
716 : l_getgroups(8), lst(8), routerstat(8)
717 Details : included man pages for llobdstat(8), llstat(8),
718 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
720 Severity : enhancement
722 Description: configure's --enable-quota should check the
723 : kernel .config for CONFIG_QUOTA
724 Details : configure is terminated if --enable-quota is passed but
725 : no quota support is in kernel
729 Description: Remove portals compatibility
730 Details : Remove portals compatibility, not interoperable with releases
735 Description: Resolve device initialization race
736 Details : Prevent proc handler from accessing devices added to the
737 obd_devs array but yet be intialized.
739 Severity : enhancement
741 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
743 Severity : enhancement
745 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
750 Description: Using wrong pointer in osc_brw_prep_request
751 Details : Access to array[-1] can produce panic if kernel compiled with
752 CONFIG_PAGE_ALLOC enabled
756 Description: Client runs out of low memory
757 Details : Consider only lowmem when counting initial number of llap pages
761 Description: *optional* service tags registration
762 Details : if the "service tags" package is installed on a Lustre node
763 When the filesystem is mounted, a local-node service tag will
764 be created. See http://inventory.sun.com/ for more information
765 about the Service Tags asset management system.
769 Description: Kernel BUG tries to release flock
770 Details : Lustre does not destroy flock lock before last reference goes
771 away. So always drop flock locks when client is evicted and
772 perform unlock regardless of successfulness of speaking to MDS.
776 Description: add refcount protection for osc callbacks, avoid panic on shutdown
780 Description: sanity test 65a fails if stripecount of -1 is set
781 Details : handle -1 striping on filesystem in ll_dirstripe_verify
786 Description: ASSERTION(CheckWriteback(page,cmd)) failed
787 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
790 Severity : enhancement
792 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
796 Description: do not process already freed flock
797 Details : flock can possibly be freed by another thread before it reaches
798 to ldlm_flock_completion_ast.
802 Description: LBUG during stress test
803 Details : Need properly lock accesses the flock deadlock detection list.
807 Description: oops in page fault handler
808 Details : kernel page fault handler can return two special 'pages' in error
809 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
813 Description: timeout with invalidate import.
814 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
815 should be handled by ptlrpcd. This produce long age waiting and
816 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
818 Severity : enhancement
820 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
824 Description: enable MGS and MDT services start separately
825 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
826 a co-located MGS without starting the MGS, which is a complement
827 to 'nosvc' mount option.
831 Frequency : after recovery
832 Description: precreate to many object's after del orphan.
833 Details : del orphan st in oscc last_id == next_id and this triger growing
834 count of precreated objects. Set flag LOW to skip increase count
835 of precreated objects.
839 Frequency : rare, on clear nid stats
840 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
841 Details : when clean nid stats sometimes try destroy live entry,
842 and this produce panic in free.
846 Description: Stack overflow during MDS log replay
847 ease stack pressure by using a thread dealing llog_process.
851 Description: wait until IO finished before start new when do lock cancel.
852 Details : VM protocol want old IO finished before start new, in this case
853 need wait until PG_writeback is cleared until check dirty flag and
854 call writepages in lock cancel callback.
856 Severity : enhancement
858 Description: using special macro for print time and cleanup in includes.
862 Description: mds_mfd_close() ASSERTION(rc == 0)
863 Details : In mds_mfd_close(), we need protect inode's writecount change
864 within its orphan write semaphore to prevent possible races.
868 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
872 Frequency : rare, on shutdown ost
873 Description: don't hit live lock with umount ost.
874 Details : shrink_dcache_parent can be in long loop with destroy dentries,
875 use shrink_dcache_sb instead.
879 Description: don't panic with use echo client
880 Details : echo client pass NULL as client nid pointer and this produce null
885 Description: fix build on ppc32
886 Details : compile code with -m64 flag produce wrong object file for ppc32.
890 Description: add message levels for liblustreapi
894 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
895 Details : Move the 'good_osts' check before the 'total_bavail' check. This
896 will result in an -EAGAIN and in the exit call path we call
897 alloc_rr() which will with increasing aggressiveness attempt to
898 aquire precreated objects on the minimum number of required OSCs.
902 Description: Use old size assignment to avoid deadlock
903 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
904 the scheduling while holding a spinlock. We do not need locking
905 for size in ll_update_inode() because size is only updated from
906 the MDS for directories or files without objects, so there is no
907 other place to do the update, and concurrent access to such inodes
908 are protected by the inode lock.
912 Description: resolve "_IOWR redefined" build error on SLES10
916 Description: dump the memory debugging after all modules are unloaded to
917 suppress false negative in conf_sanity test 39
919 Severity : enhancement
921 Description: build kernel-ib packages for OFED 1.3 in our release cycle
926 Description: fix SLES kernel versioning
927 Details : the kernel version for our SLES 10 kernel did not include a "-"
928 before the "smp" at the end. while this was not a problem in
929 general, it did mean that software trying to use the kernel
930 version to try to detect a vendor specific kernel would fail.
931 this was most evident by the OFED build scripts.
935 Description: Don't update lov_desc members until making sure they are valid
936 Details : When updating lov_desc members via proc fs, need fix their
937 validities before doing the real update.
941 Description: don't put request into delay list while invalidate in flight.
942 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
943 import in flight. this produce timeout for invalidate and sometimes
944 can cause stale data.
948 Frequency : on ppc only
949 Description: not convert ost objects for directory because it's not exist.
950 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
954 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
956 Severity : enhancement
958 Description: Update to support 2.6.22.14 vanilla kernel.
962 Frequency : rare, on recovery
963 Description: read procfs can produce deadlock in some situation
964 Details : Holding lprocfs lock which send rpc can produce block for destroy
965 obd objects and this also block reconnect with -EALREADY. This
966 isn't fix all lprocfs bugs - but make it rare.
968 Severity : enhancement
970 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
973 Frequency : frequent on X2 node
975 Description: mdc_set_open_replay_data LBUG
976 Details : Set replay data for requests that are eligible for replay.
980 Description: lustre_mgs: operation 101 on unconnected MGS
981 Details : When MGC is disconnected from MGS long enough, MGS will evict the
982 MGC, and late on MGC cannot successfully connect to MGS and a lot
983 of the error messages complaining that MGS is not connected.
987 Frequency : on network error
988 Description: panic with double free request if network error
989 Details : mdc_finish_enqueue is finish request if any network error ocuring,
990 but it's true only for synchronus enqueue, for async enqueue
991 (via ptlrpcd) this incorrect and ptlrpcd want finish request
994 Severity : enhancement
996 Description: client-side metadata stat-ahead during readdir(directory readahead)
997 Details : perform client-side metadata stat-ahead when the client detects
998 readdir and sequential stat of dir entries therein
1001 Frequency : on start mds
1003 Description: Implement get_info(last_id) in obdfilter.
1006 Frequency : occasional
1008 Description: Correctly check stale fid, not start epoch if ost not support SOM
1009 Details : open with flag O_CREATE need set old fid in op_fid3 because
1010 op_fid2 was overwritten with new generated fid, but mds can answer
1011 with one of these two fids and both is not stale. Setattr
1012 incorrectly started an epoch and assume will be called
1013 done_writing, but without SOM done_writing ever being called.
1016 Frequency : rare, depends on device drivers and load
1018 Description: MDS or OSS nodes crash due to stack overflow
1019 Details : Code changes in 1.8.0 increased the stack usage of some functions.
1020 In some cases, in conjunction with device drivers that use a lot
1021 of stack the MDS (or possibly OSS) service threads could overflow
1022 the stack. One change which was identified to consume additional
1023 stack has been reworked to avoid the extra stack usage.
1026 Frequency : occasional
1028 Description: Do not fail import if osc_interpret_create gets -EAGAIN
1029 Details : If osc_interpret_create got -EAGAIN it immediately exits and
1030 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
1031 oscc_has_objects and see OSC has no objests and call
1032 oscc_internal_create to resend create request.
1034 Severity : enhancement
1036 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
1038 Severity : enhancement
1040 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
1043 Frequency : very rare
1045 Description: Fix replay if there is an un-replied request and open
1046 Details : In some cases, older replay request will revert the
1047 mcd->mcd_last_xid on MDS which is used to record the client's
1048 latest sent request.
1050 Severity : enhancement
1052 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
1054 Severity : enhancement
1056 Description: Add rhel5 support to HEAD.
1058 Serverity : enhancement
1060 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
1065 Description: Don't allow skipping OSTs if index has been specified.
1066 Details : Don't allow skipping OSTs if index has been specified, make
1067 locking in internal create lots better.
1071 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
1072 Details : access to bitfield in structure is always rounded to long
1073 and this produce problem with not atomic change any bit.
1077 Description: Lustre make rpms failed.
1078 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
1079 builds Lustre rpms from tarball.
1082 Frequency : rare on shutdown ost
1084 Description: If llog cancel was not send before clean_exports phase, this can
1085 produce deadlock in llog code.
1086 Details : If llog thread has last reference to obd and call class_import_put
1087 this produce deadlock because llog_cleanup_commit_master wait when
1088 last llog_commit_thread exited, but this never success because was
1089 called from llog_commit_thread.
1093 Description: allow userland application know is lost one of stripes.
1094 Details : fill lvb_blocks with error code on ost and return it to
1095 application if error flag found.
1099 Description: NULL lov_tgts causing MDS oops
1100 Details : more safe checks for NULL lov_tgts for avoid oops.
1102 Severity : enhancement
1104 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
1108 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
1109 Details : If an OST is down the MDS will hang indefinitely in
1110 obd_statfs_rqset() waiting for the statfs data. While for
1111 MDS QOS usage of statfs, it should not stuck in waiting.
1113 Severity : enhancement
1115 Description: remote_acl support
1116 Details : Support ACL-based permission check for remote user.
1117 Support setfacl/getfacl for remote user with the utils
1118 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
1119 the system "{s,g}etfacl" utils.
1121 Severity : enhancement
1123 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
1125 Severity : enhancement
1127 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
1131 Description: make mgs_setparam() handle fsname containing dash
1132 Details : fsname containing a dash does not work with lctl conf_param
1134 Severity : enhancement
1136 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
1138 Severity : enhancement
1140 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
1144 Frequency : rare, at shutdown
1145 Description: access already free / zero obd_namespace.
1146 Details : if client_disconnect_export was called without force flag set,
1147 and exist connect request in flight, this can produce access to
1148 NULL pointer (or already free pointer) when connect_interpret
1149 store ocd flags in obd_namespace.
1153 Frequency : only at startup
1154 Description: not alloc memory with spinlock held.
1155 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
1156 if any spinlock held.
1158 Severity : enhancement
1160 Description: make lustre randomly fail allocating memory
1161 Details : Make lustre randomly failed allocating memory for testing purpose.
1163 Severity : enhancement
1165 Description: lost problems with lov objid file
1166 Details : Fixes some scability and access to not inited memory problems
1167 in work with lov objdid file.
1172 Description: lfs find does not continue on file error
1173 Details : Continue other files processing when a file/dir is absent.
1177 Description: Inconsistent usage of lustre_pack_reply()
1178 Details : Standardize the usage of lustre_pack_reply() such that it
1179 always generate a CERROR on failure.
1182 Frequency : occasional
1184 Description: MDS hang or stay in waiting lock
1185 Details : If client receive lock with CBPENDING flag ldlm need send lock
1186 cancel as separate rpc, to avoid situation when cancel request
1187 can't processed due all i/o threads stay in wait lock.
1191 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
1195 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
1199 Description: Update to SLES9 kernel-2.6.5-7.287.3.
1203 Description: Update to RHEL4 latest kernel.
1205 Severity : enhancement
1207 Description: Build SLES10 patchless client fails
1208 Details : The configure was broken by run ./configure with
1209 --with-linux-obj=.... argument for patchless client. When the
1210 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
1211 can't search header adequately. Use absolute path such as
1212 -I($LINUX)/include instead.
1216 Description: interrupt oig_wait produce painc on resend.
1217 Details : brw_redo_request can be used for resend requests from ptlrpcd and
1218 private set, and this produce situation when rq_ptlrpcd_data not
1219 copyed to new allocated request and triggered LBUG on assert
1220 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
1221 ptlrpcd set if request is changed and can be safety changed to use
1226 Description: LASSERT_{REQ,REP}SWAB macros are buggy
1227 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
1228 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
1229 these macros and replace them with their difinitions instead.
1233 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
1234 Details : Update which_patch & target file for SLES10 latest kernel.
1237 Description: Remove client patches from SLES 10 kernel.
1238 Details : This causes SLES 10 clients to behave as patchless clients
1239 even on a Lustre-patched (server) kernel.
1241 Severity : enhancement
1243 Description: self-adjustable client's lru lists
1244 Details : use adaptive algorithm for managing client cached locks lru
1245 lists according to current server load, other client's work
1246 pattern, memory activities, etc. Both, server and client
1247 side namespaces provide number of proc tunables for controlling
1250 Severity : enhancement
1252 Description: light-weight GSS support
1253 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
1254 and reduce performance overhead.
1256 Severity : enhancement
1258 Description: Linux keyring support
1259 Details : Support using service of Linux keyring for Lustre GSS internal
1260 context refresh/cache mechanism.
1264 Description: Fix errors in lfs documentation
1265 Details : Fixes man pages
1269 Description: don't use GFP_* in generic Lustre code.
1270 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
1274 Description: obdclass is limited by single OBD_ALLOC(idarray)
1275 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
1279 Description: warnings with build patchless client with vanila 2.6.19 and up
1280 Details : change old ctl_table style and replace ctl_table/ctl_table_header
1281 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1285 Frequency : Only for SLES
1286 Description: sanity_quota fail test_1
1287 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1288 file, which makes TSTID[2] inunique.
1292 Frequency : Only for SLES9
1293 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1294 Details : Update target/ChangeLog/which_patch .
1298 Frequency : Only for test_10 in sanity-quota.sh
1299 Description: a bug in quota test code
1300 Details : checking a test flag in a wrong place
1304 Frequency : Only for test_13 in sanity-quota.sh
1305 Description: a bug in quota test script
1306 Details : shouldn't check overfull allocated quota
1310 Description: jbd statistics
1311 Details : Port older jbd statistics patch for sles10
1313 Severity : enhancement
1315 Description: Add printing inode info into message about error in writepage.
1320 Description: reply_lock_interpret crash due to race with it and lock cancel.
1321 Details : Do not replay locks that are being cancelled. Do not reference
1322 locks by their address during replay, just by their handle.
1326 Frequency : When flocks are used.
1327 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1328 Details : Flock locks might destroy just granted lock if it could be merged
1329 with another existing flock, this is done in completion handler,
1330 so teach ldlm_cli_enquque_fini that this is a valid case for
1336 Description: Oops in read and write path when failing to allocate lock.
1337 Details : Check if lock allocation failed and return error back.
1341 Description: lstripe command fails for valid OST index
1342 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1343 instead of lov->desc.ld_active_tgt_count.
1347 Description: sanity.sh failed test 103
1348 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1353 Description: df doesn't work properly if diskfs blocksize != 4K
1354 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1358 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1359 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1361 Severity : enhancement
1363 Description: Adaptive timeouts
1364 Details : RPC timeouts adapt to changing server load and network
1365 conditions to reduce resend attempts and improve recovery time.
1369 Description: llapi_file_create() does not allow some changes
1370 Details : add llapi_file_open() that allows specifying the mode and
1371 open flags, and also returns an open file handle.
1375 Description: merge and cleanup kernel patches.
1376 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1380 Description: Add journal checksum support.(Kernel part)
1381 Details : The journal checksum feature adds two new flags i.e
1382 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1383 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1384 indicates that the commit block contains the checksum for
1385 the blocks described by the descriptor blocks. Now commit
1386 record can be sent to disk without waiting for descriptor
1387 blocks to be written to disk. This behavior is controlled
1388 using JBD2_FEATURE_ASYNC_COMMIT flag.
1392 Description: OSS needs mutliple precreate threads
1393 Details : Add ability to start more than one create thread per OSS.
1397 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1398 Details : execute lfs setstripe on client
1402 Description: mds_obd_create error creating tmp object
1403 Details : When the user sets quota on root, llog will be affected and can't
1404 create files and write files.
1407 Frequency : Always on ia64 patchless client, and possibly others.
1409 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1410 Details : This allows the patchless client to be loaded on architectures
1411 without this export.
1415 Description: RedHat Update kernel for RHEL5
1416 Details : Add the kernel config file for RHEL5.
1420 Description: RedHat Update kernel for RHEL5
1421 Details : Modify the kernel config file more closer RHEL5.
1425 Description: Build failure against Centos5 (RHEL5)
1426 Details : Define PAGE_SIZE when it isn't present.
1430 Description: test_7 on sanity-quota.sh can't be used on separate machines
1431 Details : add facet to handle it
1435 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1436 Details : with NFS, the anon dentry's parent was set to itself in
1437 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1438 corresponding dentry other than use rec->ur_name.
1440 Severity : enhancement
1442 Description: lfs setstripe enhancement
1443 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1447 Description: enable data checksumming by default
1448 Details : enable checksum by default, allow --disable-checksum
1449 configure option and "-o nochecksum" mount option. Checksums
1450 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1454 Description: lustre support for RHEL5
1455 Details : Add support for RHEL5.
1459 Description: Client eviction due to failover config
1460 Details : after a connection loss, the lustre client should attempt to
1461 reconnect to the last active server first before trying the
1462 other potential connections.
1466 Description: when mds and osts use different quota unit(32bit and 64bit),
1467 quota will be released repeatly.
1468 Details : void sending multiple quota reqs to mds, which will keep the
1469 status between the reqs.
1473 Description: osts not allocated evenly to files
1474 Details : change the condition to increase offset_idx
1477 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1479 Description: Data corruption for OSTs that are formatted larger than 2TB
1481 Details : When generating the bio request for lustre file writes the
1482 sector number would overflow a temporary variable before being
1483 used for the IO. The data reads correctly from Lustre (which
1484 will overflow in a similar manner) but other file data or
1485 filesystem metadata may be corrupted in some cases.
1489 Description: Tune the kernel for good SCSI performance.
1490 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1491 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1496 Description: rewrite ext2-derived code in obdclass/uuid.c
1497 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1498 from scratch preserving functionality.
1502 Description: "lfs find -obd UUID" prints directories
1503 Details : "lfs find -obd UUID" will return all directory names instead
1504 of just file names. It is incorrect because the directories
1505 do not reside on the OSTs.
1509 Description: use i_size_read and i_size_write in 2.6 port
1510 Details : replace inode->i_size access with i_size_read/write()
1513 Frequency : when removing large files
1515 Description: scheduling issue during removal of large Lustre files
1516 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1517 It causes scheduling issues when removing large files (17TB in the
1521 Frequency : only with liblustre clients on XT3
1523 Description: evictions taking too long
1524 Details : allow llrd to evict clients directly on OSTs
1527 Frequency : only on ppc
1529 Description: /proc/fs/lustre/devices broken on ppc
1530 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1531 We should make sure the type of 'index' is loff_t and then cast
1532 explicitly as needed below. Do not assign an explicitly cast
1536 Frequency : only for rhel5
1538 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1539 Details : Modify the target file & which_kernel.
1541 Severity : enhancement
1543 Description: omit set fsid for export NFS
1544 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1545 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1549 Frequency : Always, for kernels after 2.6.16
1550 Description: Fix warning idr_remove called for id=.. which is not allocated.
1551 Details : Last kernels save old s_dev before kill super and not allow
1552 to restore from callback - restore it before call kill_anon_super.
1556 Description: buffer overruns could theoretically occur
1557 Details : llapi_semantic_traverse() modifies the "path" argument by
1558 appending values to the end of the origin string, and a
1559 overrun may occur. Adding buffer overrun check in liblustreapi.
1563 Description: Fix error on 'ls .' at the top of the Lustre mount.
1564 Details : Don't revalidate dentry if it is a root dentry.
1568 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1569 Details : Modify vm-tunables-rhel4.patch.
1573 Description: Kernel config for 2.6.18-vanilla.
1574 Details : Modify targets/2.6-vanilla.target.in.
1575 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1576 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1577 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1578 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1582 Description: improve handling recoverable errors
1583 Details : if request processig with error which can be recoverable on server
1584 request should be resend, otherwise page released from cache and
1589 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1590 Details : Modify target file & which_patch.
1591 A flaw was found in the IA32 system call emulation provided
1592 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1593 value could be stored in the %RAX register, which could trigger an
1594 out-of-bounds system call table access. An untrusted local user
1595 could exploit this flaw to run code in the kernel
1596 (ie a root privilege escalation). (CVE-2007-4573).
1600 Description: O_DIRECT bypasses client statistics.
1601 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1602 read_ahead_stats were updated. copied the stats section from
1603 osc_send_oap_rpc() into async_internal().
1607 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1611 Description: change order of libsysio includes
1612 Details : '#include sysio.h' should always come before '#include xtio.h'
1616 Description: handle "serious error: objid * already exists" more gracefully
1617 Details : If LAST_ID value on disk is smaller than the objects existing in
1618 the O/0/d* directories, it indicates disk corruption and causes an
1619 LBUG(). If the object is 0-length, then we should use the existing
1620 object. This will help to avoid a full fsck in most cases.
1622 Severity : enhancement
1624 Description: adapt the lustre_config script to support the upgrade case
1625 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1626 server targets to 1.6 in parallel.
1630 Description: To avoid grant space > avaible space when the disk is almost
1631 full. Without this patch you might see the error "grant XXXX >
1632 available" or some LBUG about grant, when the disk is almost
1634 Details : In filter_check_grant, for non_grant cache write, we should
1635 check the left space by if (*left > ungranted + bytes), instead
1636 of (*left > ungranted), because only we are sure the left space
1637 is enough for another "bytes", then the ungrant space should be
1638 increase. In client, we should update cl_avail_grant only there
1639 is OBD_MD_FLGRANT in the reply.
1643 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1644 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1645 described in CVE-2007-4573. This problem could allow untrusted
1646 local users to gain root access.
1649 Frequency : when using O_DIRECT and quotas
1651 Description: Incorrect file ownership on O_DIRECT output files
1652 Details : block usage reported by 'lfs quota' does not take into account
1653 files that have been written with O_DIRECT.
1658 Description: touch file failed when fs is not full
1659 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1660 otherwise we can get ENOSP while fs is not full.
1664 Description: parallel lock callbacks
1665 Details : Instead of sending blocking and completion callbacks as separated
1666 requests, adding them to a set and sending in parallel.
1669 Frequency : only for Cray XT3
1670 Bugzilla : 12829/13455
1671 Description: Changing primary group doesn't change the group lustre assigns to
1673 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1674 overridden with the primary group provided by the group upcall,
1675 whereas the supplied fsgid can be trusted if it is in the list of
1676 supplementary groups returned by the group upcall.
1678 Severity : enhancement
1680 Description: Allow masking D_WARNING, D_ERROR messages from console
1681 Details : Console messages can now be disabled via lnet.printk.
1685 Description: User code with malformed file open parameter crashes client node
1686 Details : Before packing join_file req, all the related reference should be
1687 checked carefully in case some malformed flags cause fake
1688 join_file req on client.
1692 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1693 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1694 mode is equal to the granted mode. However, it should also check
1695 that LDLM_FL_CP_REQD is not set.
1699 Description: Heavy nfs access might result in deadlocks
1700 Details : After ELC code landed, it is now improper to enqueue any mds
1701 locks under och_sem, because enqueue might want to decide to
1702 cancel open locks for same inode we are holding och_sem for.
1706 Description: Client eviction while running blogbench
1707 Details : A lot of unlink operations with concurrent I/O can lead to a
1708 deadlock causing evictions. To address the problem, the number of
1709 oustanding OST_DESTROY requests is now throttled to
1710 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1711 callbacks are processed in priority.
1715 Description: enable ACLs on MDS by default
1716 Details : ACLs must be enabled on MDS by default.
1719 Frequency : PPC/PPC64 only
1721 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1722 Details : fix duplicated definitions between asm-ppc64/types.h and
1723 lustre_types.h on PPC.
1726 Frequency : PPC/PPC64 only
1728 Description: asm-ppc/segment.h does not exist
1729 Details : fix compile issue on PPC.
1733 Description: better handle error messages in extents code
1736 Frequency : RHEL4 only
1738 Description: mkfs is very slow on IA64/RHEL4
1739 Details : A performance regression has been discovered in the MPT Fusion
1740 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1741 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1742 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1744 Severity : enhancement
1746 Description: SNMP support enhancement
1747 Details : Adding total number of sampled request for an MDS node in snmp
1750 Severity : enhancement
1752 Description: Optimize ldlm waiting list processing for PR extent locks
1753 Details : When processing waiting list for read extent lock and meeting read
1754 lock that is same or wider to it that is not contended, skip
1755 processing rest of the list and immediatelly return current
1756 status of conflictness, since we are guaranteed there are no
1757 conflicting locks in the rest of the list.
1761 Description: Time out and refuse to reconnect
1762 Details : When the failover node is the primary node, it is possible
1763 to have two identical connections in imp_conn_list. We must
1764 compare not conn's pointers but NIDs, otherwise we can defeat
1765 connection throttling.
1769 Description: port llog fixes from b1_6 into HEAD
1770 Details : Port llog reference couting and some llog cleanups from b1_6
1771 (bug 10800) into HEAD, for protect from panic and access to already
1772 free llog structures.
1776 Description: Detect stride IO mode in read-ahead
1777 Details : When a client does stride read, read-ahead should detect that and
1778 read-ahead pages according to the detected stride pattern.
1782 Description: data checksumming impacts single node performance
1783 Details : add support for several checksum algorithm. Currently, only CRC32
1784 and Adler-32 are supported. The checksum type can be changed on
1785 the fly via /proc/fs/lustre/osc/*/checksum_type.
1789 Description: use adler32 for page checksums
1790 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1795 Description: build for x2 fails
1796 Details : fix compile issue on Cray systems.
1800 Description: Properly match for duplicate locks
1801 Details : Due to different lock order from skiplists code, we need to
1802 traverse entire list for now
1805 Frequency : only on PPC/SLES10
1807 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1808 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1809 be defined. Add a hack in mkfs_lustre.c to work around this compile
1814 Description: LASSERT on MDS when client holding flock lock dies
1815 Details : ldlm pool logic depends on number of granted locks equal to
1816 number of released locks which is not true for flock locks, so
1817 just exclude such locks from consideration.
1821 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1822 Details : Use fsfilt_sync() for both the whole filesystem sync and
1823 individual file sync to eliminate dangerous inode locking
1824 with I_LOCK that can lead to a deadlock.
1828 Description: performance in 1.6.3
1829 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1830 sure that 1MB requests are not fragmented by the block layer.
1832 Severity : enhancement
1834 Description: organize the server-side client stats on per-nid basis
1835 Details : Change the structure of stats under obdfilter and mds to
1843 The "uuid"s file would list the uuids of _active_ exports.
1844 And the clear entry is to clear all stats and stale nids.
1846 Severity : enhancement
1848 Description: eliminate client locks in face of contention
1849 Details : file contention detection and lockless i/o implementation
1850 for contended files.
1854 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1857 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1859 Description: MDS may lose file striping (and hence file data) in some cases
1860 Details : If there are additional extended attributes stored on the MDS,
1861 in particular ACLs, SELinux, or user attributes (if user_xattr
1862 is specified for the client mount options) then there is a risk
1863 of attribute loss. Additionally, the Lustre file striping
1864 needs to be larger than default (e.g. striped over all OSTs),
1865 and an additional attribute must be stored initially in the
1866 inode and then increase in size enough to be moved to the
1867 external attribute block (e.g. ACL growing in size) for file
1872 Description: skiplist implementation simplification
1873 Details : skiplists are used to group compatible locks on granted list
1874 that was implemented as tracking first and last lock of each
1875 lock group the patch changes that to using doubly linked lists
1879 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1880 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1881 because LASSERT on both the data supplied by a client, and the
1882 data on disk is dangerous and incorrect.
1884 Severity : enhancement
1886 Description: Slow truncate/writes to huge files at high offsets.
1887 Details : Directly associate cached pages to lock that protect those pages,
1888 this allows us to quickly find what pages to write and remove
1889 once lock callback is received.
1893 Description: more ldlm soft lockups
1894 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1895 starve other threads from the resource lock for a long time in
1896 case of long waiting queue, so change the debug level from
1897 D_OTHER to the less frequently used D_INFO.
1899 Severity : enhancement
1901 Description: add -gid, -group, -uid, -user options to lfs find
1905 Description: Hung threads in invalidate_inode_pages2_range
1906 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1907 once one is completed. As a result, some RPCs are stuck in the
1908 queue and are never sent.
1912 Description: filter threads hungs on waiting journal commit
1913 Details : Cleanup filter group llog code, then only filter group llog will
1914 be only created in the MDS/OST syncing process.
1918 Description: Procfs and llog threads access destoryed import sometimes.
1919 Details : Sync the import destoryed process with procfs and llog threads by
1920 the import refcount and semaphore.
1922 Severity : enhancement
1924 Description: openlock cache of b1_6 port to HEAD
1929 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1930 Details : If insertion of an extent fails, then discard the inode
1931 preallocation and free data blocks else it can lead to duplicate
1936 Description: don't always update ctime in ext3_xattr_set_handle()
1937 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1938 In some cases the ctime should not be updated, for example for
1939 2.0->1.8 compatibility it is necessary to delete an xattr and it
1940 should not update the ctime.
1944 Bugzilla : 15713/16362
1945 Description: Assertion in iopen_connect_dentry in 1.6.3
1946 Details : looking up an inode via iopen with the wrong generation number can
1947 populate the dcache with a disconneced dentry while the inode
1948 number is in the process of being reallocated. This causes an
1949 assertion failure in iopen since the inode's dentry list contains
1950 both a connected and disconnected dentry.
1954 Description: assertion failure in ldlm_handle2lock()
1955 Details : fix a race between class_handle_unhash() and class_handle2object()
1956 introduced in lustre 1.6.5 by bug 13622.
1961 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1962 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1967 Description: oops in sd_iostats_seq_show()
1968 Details : unloading/reloading the scsi low level driver triggers a kernel
1969 bug when trying to access the sd iostat file.
1974 Description: Kernel panics during QLogic driver reload
1975 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1976 patch, causing memory corruption.
1981 Description: journal_dev option does not work in b1_6
1982 Details : pass mount option during pre-mount.
1984 Severity : enhancement
1986 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1987 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1988 extent information of a file. It can be used to map logical blocks
1989 in a file to physical blocks in the block device.
1993 Description: LDLM soft lockups - improvement
1994 Details : It is be possible to send the lock handle along with each read
1995 or write request because the client is already doing a lock match
1996 itself so there isn't any reason the OST should have to re-do that
2000 Frequency : only on Cray X2
2002 Description: X2 build failures
2003 Details : fix build failures on Cray X2.
2007 Description: xid & resent requests
2008 Details : Initialize RPC XID from clock at startup (randomly if clock is
2011 Severity : enhancement
2013 Description: Add lustre_start utility to start or stop multiple Lustre servers
2018 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
2019 Details : In case of memory pressure, list_del() can be called twice on
2020 req->rq_history_list, causing a kernel oops.
2024 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
2025 Details : fix stack overflow in the distributed lock manager by defering
2026 export eviction after a failed AST to the elt thread instead of
2027 handling it in the dlm interpret routine.
2031 Description: Convert some comments to new format.
2032 Details : Update documenting comments to match doxygen conventions.
2036 Description: Grammar fixes.
2037 Details : A couple of trivial spelling fixes.
2041 Description: OSD_COUNTERS-mandatory
2042 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
2043 to simplify the code.
2047 Description: simplify cmm_device freeing logic.
2048 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
2052 Description: Add lockdep support to dt_object_operations locking interface.
2053 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
2054 parameter indicating lock ordering. Update mdd code to use new
2059 Description: Introduce failloc constants for lockless IO tests.
2060 Details : Add two new failloc constants to test lockless IO. Only one of
2061 them in implemented---another is checked in yet to be landed
2066 Description: Add lockdep support for inode mutex.
2067 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
2068 in the situations where Lustre has to lock more than one inode
2073 Description: Add optional invariants checking support.
2074 Details : Add new LINVRNT() macro, optional on new --enable-invariants
2075 configure switch. This macro is to be used for consistency and
2076 sanity checks that are too expensive to be left in `production'
2081 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
2082 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
2087 Description: Add lockdep support for ldlm_lock and ldlm_resource.
2088 Details : Use spin_lock_nested() in (the only) situation where more than
2089 one ldlm_lock is locked simultaneously. Also, fix possible
2090 dead-lock in ldlm_lock_change_resource() by enforcing particular
2095 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
2096 Details : Instead of specifying each ldlm_lock call-back through separate
2097 parameter, wrap them into struct ldlm_callback_suite.
2101 Description: Kill join_lru obd method and its callers.
2102 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
2103 in memory---a requirement implemented through ->o_join_lru() obd
2104 method in HEAD. Remove this method and its users.
2108 Description: Add asynchronous ldlm ENQUEUE completion handler.
2109 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
2110 case that stock ldlm_completion_ast() cannot handle as it waits
2111 until lock is granted. Introduce new ldlm_completion_ast_async()
2112 for this. Also comment ldlm_completion_ast().
2116 Description: ldlm_error <-> errno conversion.
2117 Details : Add functions to map (rather arbitrary) between LDLM error codes
2118 and standard errno values. CLIO needs this to prevent LDLM specific
2119 constants from escaping ldlm and osc.
2123 Description: Kill unused ldlm_handle2lock_ns() function.
2124 Details : Kill unused ldlm_handle2lock_ns() function.
2128 Description: Add lu_ref support to ldlm_lock
2129 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
2130 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
2131 and ldlm_resource. LDLM interface has to be changed, because code that
2132 releases a reference on a lock, has to "know" what reference this is.
2133 In the most frequent case
2135 lock = ldlm_handle2lock(handle);
2137 LDLM_LOCK_PUT(lock);
2139 no changes are required. When any other reference (received _not_ from
2140 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
2141 instead of LDLM_LOCK_PUT().
2143 Arguably, changes are pervasive, and interface requires some discipline
2144 for proper use. On the other hand, it was very instrumental in finding
2145 a few leaked lock references.
2149 Description: Add ldlm_lock_addref_try().
2150 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
2151 attempts to addref a lock that might be being canceled
2156 Description: Add ldlm_weigh_callback().
2157 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
2158 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
2159 hard-coded `number of pages' logic.
2163 Description: Add lockdep annotations to llog code.
2164 Details : Use appropriately tagged _nested() locking calls in the places
2165 where llog takes more than one ->lgh_lock lock.
2169 Description: Add loi_kms_set().
2170 Details : Wrap kms updates into a helper function.
2174 Description: Constify instances of struct lsm_operations.
2175 Details : Constify instances of struct lsm_operations.
2179 Description: lu_conf support.
2180 Details : On a server, a file system object is uniquely identified
2181 by a fid, which is sufficient to locate and load all object
2182 state (inode). On a client, on the other hand, more data are
2183 necessary instantiate an object. Change lu_object_find() and
2184 friends to take additional `lu_conf' argument describing object.
2185 Typically this includes layout information.
2189 Description: lu_context fixes.
2190 Details : Introduce new lu_context functions that are needed on the client
2191 side, where some system threads (ptlrpcd) are shared by multiple
2192 modules, and so cannot be stopped during module shutdown.
2196 Description: Add start and stop methods to lu_device_type_operations.
2197 Details : Introduce two new methods in lu_device_type_operations, that are
2198 invoked when first instance of a given type is created and last one
2199 is destroyed respectively. This is need by CLIO.
2203 Description: Add lu_ref support to struct lu_device.
2204 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
2205 track leaked references.
2209 Description: Introduce lu_kmem_descr.
2210 Details : lu_kmem_descr and its companion interface allow to create
2211 and destroy a number of kmem caches at once.
2215 Description: Fix lu_object finalization race.
2216 Details : Fix a race between lu_object_find() finding an object and its
2217 concurrent finalization. This race is (most likely) not possible
2218 on the server, but might happen on the client.
2222 Description: Introduce lu_ref interface.
2223 Details : lu_ref is a debugging module allowing to track references to
2224 a given object. It is quite cpu expensive, and has to be
2225 explicitly enabled with --enable-lu_ref. See usage description
2230 Description: Factor lu_site procfs stats into a separate function.
2231 Details : Separate lu_site stats printing code into a separate function
2232 that can be reused on a client.
2236 Description: Constify instances of struct {lu,dt,md}_device_operations.
2237 Details : Constify instances of struct {lu,dt,md}_device_operations.
2241 Description: Introduce struct md_site and move meta-data specific parts of
2242 struct lu_site here.
2243 Details : Move md-specific fields out of struct lu_site into special struct
2244 md_site, so that lu_site can be used on a client.
2248 Description: Kill mdd_lov_destroy().
2249 Details : Remove unused mdd code.
2253 Description: Add st_block checking to multistat.c.
2254 Details : Add st_block checking to multistat.c.
2258 Description: Add lu_ref support to struct obd_device.
2259 Details : Add lu_ref tracking to obd_device.
2263 Description: Kill obd_set_fail_loc().
2264 Details : Remove unused code.
2268 Description: Add special type for ptlrpc_request interpret functions.
2269 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
2270 there. Actual usage will be in CLIO.
2274 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
2275 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
2276 doesn't work with lockdep.
2280 Description: Add rwv.c test program.
2281 Details : New testing program exercising readv(2) and writev(2) (Qian).
2285 Description: Add sendfile.c test program.
2286 Details : New testing program exercising sendfile(2) (Jay).
2290 Description: Ratelimit a message that can be very frequent.
2291 Details : Ratelimit a memory allocation failure message that can
2296 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2297 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2301 Description: Liblustre build fixes.
2302 Details : Liblustre build fixes.
2306 Description: libcfs: add cfs_{need,cond}_resched() interface.
2307 Details : libcfs: add cfs_{need,cond}_resched() definition and
2308 implementations for Linux, NT, and liblustre.
2310 Severity : enhancement
2312 Description: More exported tunables for mballoc
2313 Details : Add support for tunable preallocation window and new tunables for
2314 large/small requests
2318 Description: Detect corruption of block bitmap and checking for preallocations
2319 Details : Checks validity of on-disk block bitmap. Also it does better
2320 checking of number of applied preallocations. When corruption is
2321 found, it turns filesystem readonly to prevent further corruptions.
2325 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2326 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2327 window to 0. If it does hit cache miss, and read-pattern is still
2328 stride-io mode, does not reset the stride window, but also does not
2329 increase the stride window length in this case.
2333 Frequency : only for big-endian servers
2334 Description: Check if system is big-endian while mounting fs with extents feature
2335 Details : Mounting a filesystem with extents feature will fail on big-endian
2336 systems since ext3-based ldiskfs is not supported on big-endian
2337 systems. This can be over-riden with "bigendian_extents" mount option.
2339 Severity : enhancement
2341 Description: The root squash functionality
2342 Details : A security feature, which is to prevent users from being able
2343 to mount lustre on their desktop, run as root, and delete
2344 all of the files in the filesystem. The goal is accomplished by
2345 remapping user id (UID) and group id (GID) of the root user to
2346 a UID and GID specified by the system administartor via Lustre
2347 configuration management server (MGS). The functionality also
2348 allows to specify sets of clients for which the remapping does
2353 Description: Excessive recovery window
2354 Details : With AT enabled, the recovery window can be excessively long (6000+
2355 seconds). To address this problem, we no longer use
2356 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2357 timeout no longer depends on the service time, it is set to
2358 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2359 time via pb_service_time.
2363 Description: Watchdog triggered on MDS failover
2364 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2365 the OSTs know that the MDS "UUID" can be reused for the same export
2366 from a different NID, so we do not need to wait for the export to be
2370 Frequency : rare, only if using MMP with Linux RAID
2372 Description: MMP doesn't work with Linux RAID
2373 Details : While using HA for Lustre servers with Linux RAID, it is possible
2374 that MMP will not detect multiple mounts. To make this work we
2375 need to unplug the device queue in RAID when the MMP block is being
2376 written. Also while reading the MMP block, we should read it from
2377 disk and not the cached one.
2379 Severity : enhancement
2381 Description: open file using fid
2382 Details : A file can be opened using just its fid, like
2383 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2386 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2389 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2390 Details : add proper locking for creation of jbd_4k slab cache
2394 Description: MMP check in ext3_remount() fails without displaying any error
2395 Details : When multiple mount protection fails during remount, proper error
2398 Severity : enhancement
2400 Description: Allow stripe size to be up to 4G-64k
2401 Details : Fix math logic to allow large stripe sizes.
2405 Description: add check for >8TB ldiskfs filesystems
2406 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2407 Don't allow >8TB ldiskfs filesystems to be mounted without
2408 force_over_8tb mount option
2413 Descriptoin: OSS complains "too many blocks" and crashes
2414 Details : In filter_commitrw_write():
2415 1. lquota_chkquota();
2416 2. obdo_from_inode();
2417 3. lquota_pending_commit();
2418 "2." has changed uid and gid in this case, then caused
2419 lquota_chkquota() and lquota_pending_commit() to use
2420 different uid and gid. That is the root of the bug.
2424 Descriptoin: getstripe doesn't report stripe count = -1 sanely
2425 Details : Detect and print out the 'stripe count = -1' case properly.
2429 Description: Out or order replies might be lost on replay
2430 Details : In ptlrpc_retain_replayable_request if we cannot find retained
2431 request with tid smaller then one currently being added, add it
2432 to the start, not end of the list.
2434 --------------------------------------------------------------------------------
2436 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2438 * Support for kernels:
2439 2.6.5-7.283 (SLES 9),
2440 2.6.9-55.EL (RHEL 4),
2441 2.6.16.46-0.14 (SLES 10),
2442 2.6.18.8 vanilla (kernel.org)
2443 * Client support for unpatched kernels:
2444 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2445 2.6.16 - 2.6.22 vanilla (kernel.org)
2446 * Due to recently discovered recovery problems, we do not recommend
2447 using patchless RHEL 4 clients with this or any earlier release.
2448 * Recommended e2fsprogs version: 1.39.cfs8
2449 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2450 * Starting with this release, the ldiskfs backing filesystem required
2451 by Lustre is now in its own package, lustre-ldiskfs. This package
2452 should be installed. It is versioned separately from Lustre and
2453 may be released separately in future.
2457 Description: block reactivating mgc import until all deactivates complete
2458 Details : Fix race when failing back MDT/MGS to itself (testing)
2461 Frequency : at statup only
2463 Description: mds_lov_synchronize race leads to various problems
2464 Details : simultaneous MDT->OST connections at startup can cause the
2465 sync to abort, leaving the OSC in a bad state.
2467 Severity : enhancement
2469 Description: add optional extra BUILD_VERSION info
2470 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2471 to override the lustre version.
2474 Frequency : 2.6.18 servers only
2476 Description: ll_kern_mount() doesn't release the module reference
2477 Details : The ldiskfs module reference count never drops down to 0
2478 because ll_kern_mount() doesn't release the module reference.
2483 Description: server LBUG when using old ost_num_threads parameter
2484 Details : Accept the old ost_num_threads parameter but warn that it
2485 is deprecated, and fix an off-by-one error that caused an LBUG.
2490 Description: Transient SCSI error results in persistent IO issue
2491 Details : iobuf->dr_error is not reinitialized to 0 between two
2495 Frequency : sometimes when underlying device returns I/O errors
2497 Description: OSTs not going read-only during write failures
2498 Details : OSTs are not remounted read-only when the journal commit threads
2499 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2500 instead of the ext3 wrappers.
2504 Description: poor connect scaling with increasing client count
2505 Details : Don't run filter_grant_sanity_check for more than 100 exports
2506 to improve scaling for large numbers of clients.
2509 Frequency : SLES10 only
2511 Description: sanity-quota.sh quotacheck failed: rc = -22
2512 Details : Quotas cannot be enabled on SLES10.
2515 Frequency : liblustre clients only
2517 Description: getdirentries does not give error when run on compute nodes
2518 Details : getdirentries does not fail when the size specified as an argument
2519 is too small to contain at least one entry
2521 Severity : enhancement
2523 Description: Add LNET router traceability for debug purposes
2524 Details : If a checksum failure occurs with a router as part of the
2525 IO path, the NID of the last router that forwarded the bulk data
2526 is printed so it can be identified.
2531 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2532 Details : Due to a race condition, liblustre clients were occasionally
2533 evicted incorrectly.
2535 Severity : enhancement
2537 Description: lfs setstripe use optional parameters instead of postional
2540 Severity : enhancement
2542 Description: Nanosecond timestamp support for ldiskfs
2543 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2544 resolution timestamps. There is not yet support for this at
2545 the Lustre filesystem level.
2548 Frequency : during server recovery
2550 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2551 Details : request with rq_no_resend flag not awake l_wait_event if they get
2555 Frequency : nfs export on patchless client
2557 Description: connectathon hang when test nfs export over patchless client
2558 Details : Disconnected dentry cannot be found with lookup, so we do not need
2559 to unhash it or make it invalid
2562 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2563 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2564 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2569 Description: open req refcounting wrong on reconnect
2570 Details : If reconnect happened between getting open reply from server and
2571 call to mdc_set_replay_data in ll_file_open, we will schedule
2572 replay for unreferenced request that we are about to free.
2573 Subsequent close will crash in variety of ways.
2574 Check that request is still eligible for replay in
2575 mdc_set_replay_data().
2580 Description: disable writes to filesystem when reading health_check file
2581 Details : the default for reading the health_check proc file has changed
2582 to NOT do a journal transaction and write to disk, because this
2583 can cause reads of the /proc file to hang and block HA state
2584 checking on a healthy but otherwise heavily loaded system. It
2585 is possible to return to the previous behaviour during configure
2586 with --enable-health-write.
2588 Severity : enhancement
2590 Description: 64-bit inode version
2591 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2592 to the inode. This will be required for version-based recovery.
2597 Description: MDS fails to start if a duplicate client export is detected
2598 Details : in some rare cases it was possible for a client to connect to
2599 an MDS multiple times. Upon recovery the MDS would detect this
2600 and fail during startup. Handle this more gracefully.
2602 Severity : enhancement
2604 Description: Add -o localflock option to simulate old noflock
2606 Details : This will achieve local-only flock/fcntl locks
2612 Description: log_commit_thread vs filter_destroy race leads to crash
2613 Details : Take import reference before releasing llog record semaphore
2618 Description: Wrong request locking in request set processing
2619 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2620 imp_delayed_list, in this place should be used imp_lock.
2623 Frequency : when reconnection
2625 Description: Grant Leak when osc reconnect to OST
2626 Details : When osc reconnect ost, OST(filter) should check whether it
2627 should grant more space to client by comparing fed_grant and
2628 cl_avail_grant, and return the granted space to client instead
2629 of "new granted" space, because client will call osc_init_grant
2630 to update the client grant space info.
2633 Frequency : when client reconnect to OST
2635 Description: Grant Leak when osc do resend and replay bulk write
2636 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2637 bulk write request, because the grant info while be sync between
2638 OSC and OST when reconnect, and we should ignore the grant info
2639 these of resend/replay write req.
2644 Description: Grant space more than avaiable left space sometimes.
2645 Details : When then OST is about to be full, if two bulk writing from
2646 different clients came to OST. Accord the avaliable space of the
2647 OST, the first req should be permitted, and the second one
2648 should be denied by ENOSPC. But if the seconde arrived before
2649 the first one is commited. The OST might wrongly permit second
2650 writing, which will cause grant space > avaiable space.
2653 Frequency : when client is evicted
2655 Description: Grant might be wrongly erased when osc is evicted by OST
2656 Details : when the import is evicted by server, it will fork another
2657 thread ptlrpc_invalidate_import_thread to invalidate the
2658 import, where the grant will be set to 0. While the original
2659 thread will update the grant it got when connecting. So if
2660 the former happened latter, the grant will be wrongly errased
2661 because of this race.
2666 Description: Checking Stale with correct fid
2667 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2668 to check whether it is stale, because sometimes, we want the
2669 enqueue happened anyway, and op_data.fid2 will not be initialized.
2671 Severity : enhancement
2673 Description: update patchless client
2674 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2677 Frequency : only with 2.4 kernel
2679 Description: random memory corruption
2680 Details : size of struct ll_inode_info is to big for union inode.u and this
2681 can be cause of random memory corruption.
2686 Description: Memory leak in recovery
2687 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2688 It should also check obd_fail before fsfilt_start, otherwise if
2689 fsfilt_start return -EROFS,(failover mds during mds recovery).
2690 then the req will return with repmsg->transno = 0 and rc = EROFS.
2691 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2692 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2693 be freed no matter whether fsfilt_commit success or not.
2696 Frequency : only with huge count clients
2698 Description: Prevents from taking the superblock lock in llap_from_page for
2700 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2701 for llap_from_page call in ll_removepage prevents from taking the
2702 superblock lock for a soon died page.
2707 Description: Not check open intent error before release open handle
2708 Details : in some rare cases, the open intent error is not checked before
2709 release open handle, which may cause
2710 ASSERTION(open_req->rq_transno != 0), because it tries to release
2711 the failed open handle.
2716 Description: Set cat log bitmap only after create log success.
2717 Details : in some rare cases, the cat log bitmap is set too early. and it
2718 should be set only after create log success.
2722 Description: Accessing a block bevice can re-enable I/O when Lustre is
2723 tearing down a device.
2724 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2729 Description: service threads may hog cpus when there are a lot of requests
2730 Details : Insert cond_resched to give other threads a chance to use some CPU
2735 Description: the cat log was not initialized in recovery
2736 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2737 unlink log on mds will not be initialized until mds post
2738 recovery. And also in mds post recovery, the unlink log will
2739 initialization will be done asynchronausly, so there will be race
2740 between add unlink log and unlink log initialization.
2744 Description: brw_stats were being printed incorrectly
2745 Details : brw_stats were being printed as log2 but all of them were not
2746 recorded as log2. Also remove some code duplication arising from
2747 filter_tally_{read,write}.
2751 Frequency : rare, only in recovery.
2752 Description: ASSERTION(req->rq_type != LI_POISON) failed
2753 Details : imp_lock should be held while iterating over imp_sending_list for
2754 prevent destroy request after get timeout in ptlrpc_queue_wait.
2758 Description: replay-single.sh test 52 fails
2759 Details : A lock's skiplist need to be cleanup when it being unlinked
2760 from its resource list.
2764 Description: Short directio read returns full requested size rather than
2766 Details : Direct I/O operations should return actual amount of bytes
2767 transferred rather than requested size.
2769 Severity : enhancement
2771 Description: metadata RPC reduction (e.g. for rm performance)
2772 Details : decrease the amount of synchronous RPC between clients and servers
2773 by canceling conflicing lock before the operation on the client
2774 and packing thier handles into the main operation RPC to server.
2776 Severity : enhancement
2778 Description: Async OSC create to avoid the blocking unnecessarily.
2779 Details : If a OST has no remain object, system will block on the creating
2780 when need to create a new object on this OST. Now, ways use
2781 pre-created objects when available, instead of blocking on an
2782 empty osc while others are not empty. If we must block, we block
2783 for the shortest possible period of time.
2787 Description: improve handling recoverable errors
2788 Details : if request processig with error which can be recoverable on server
2789 request should be resend, otherwise page released from cache and
2792 Severity : enhancement
2794 Description: refine locking for avoid write wrong info into lov_objid file
2795 Details : fix possible races with add new target and write/update data in
2798 --------------------------------------------------------------------------------
2800 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2805 Frequency : on some architectures
2807 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2808 Details : Uninitialized flags sometimes cause configuration commands to
2812 Frequency : patchless clients only
2814 Description: missing __iget() symbol export
2815 Details : The __iget() symbol export is missing. To avoid the need for
2816 this on patchless clients the deathrow inode reaper is turned
2817 off, and we depend on the VM to clean up old inodes. This
2818 dependency was during via the fix for bug 12181.
2820 --------------------------------------------------------------------------------
2822 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2824 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2825 INTEROPERATE with older versions automatically. In many cases a
2826 special upgrade step is needed. Please read the
2827 user documentation before upgrading any part of a live system.
2828 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2829 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2830 * WARNING: Lustre configuration and startup changes are required with
2831 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2835 Severity : enhancement
2837 Description: MountConf
2838 Details : Lustre configuration is now managed via mkfs and mount
2839 commands instead of lmc and lconf. New obd types (MGS, MGC)
2840 are added for dynamic configuration management. See
2841 https://mail.clusterfs.com/wikis/lustre/MountConf for
2844 Severity : enhancement
2846 Description: dynamic OST addition
2847 Details : OSTs can now be added to a live filesystem
2849 Severity : enhancement
2851 Description: startup order invariance
2852 Details : MDTs and OSTs can be started in any order. Clients only
2853 require the MDT to complete startup.
2855 Severity : enhancement
2857 Description: parallel, asynchronous orphan cleanup
2858 Details : orphan cleanup is now performed in separate threads for each
2859 OST, allowing parallel non-blocking operation.
2861 Severity : enhancement
2863 Description: optimized stripe assignment
2864 Details : stripe assignments are now made based on ost space available,
2865 ost previous usage, and OSS previous usage, in order to try
2866 to optimize storage space and networking resources.
2868 Severity : enhancement
2870 Description: Permanently set tunables
2871 Details : All writable /proc/fs/lustre tunables can now be permanently
2872 set on a per-server basis, at mkfs time or on a live system.
2874 Severity : enhancement
2876 Description: Lustre message v2
2877 Details : Add lustre message format v2.
2879 Severity : enhancement
2881 Description: client OST exclusion list
2882 Details : Clients can be started with a list of OSTs that should be
2883 declared "inactive" for known non-responsive OSTs.
2887 Description: SPEC SFS validation failure on NFS v2 over lustre.
2888 Details : Changes the blocksize for regular files to be 2x RPC size,
2889 and not depend on stripe size.
2891 Severity : enhancement
2893 Description: Multiple MD RPCs in flight.
2894 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2895 To avoid overly-overloading MDS, introduce a limit on number of
2896 MDS RPCs in flight for a single client and add /proc controls
2897 to adjust this limit.
2899 Severity : enhancement
2901 Description: client read/write statistics
2902 Details : Add client read/write call usage stats for performance
2903 analysis of user processes.
2904 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2905 file access. extents_stats shows chunk size distribution.
2906 extents_stats_per_process show chunk size distribution per
2909 Severity : enhancement
2911 Description: mds statistics
2912 Details : Add detailed mds operations statistics in
2913 /proc/fs/lustre/mds/*/stats.
2917 Description: Failure of copying files with lustre special EAs.
2918 Details : Client side always return success for setxattr call for lustre
2919 special xattr (currently only "trusted.lov").
2924 Description: Refcount LNET uuids
2925 Details : The global LNET uuid list grew linearly with every startup;
2926 refcount repeated list entries instead of always adding to
2929 Severity : enhancement
2931 Description: Dynamic service threads
2932 Details : Within a small range, start extra service threads
2933 automatically when the request queue builds up.
2936 Frequency : mixed-endian client/server environments
2938 Description: mixed-endian crashes
2939 Details : The new msg_v2 system had some failures in mixed-endian
2942 Severity : enhancement
2944 Description: Easy OST removal
2945 Details : OSTs can be permanently deactivated with e.g. 'lctl
2946 conf_param lustre-OST0001.osc.active=0'
2948 Severity : enhancement
2950 Description: MGS proc entries
2951 Details : Added basic proc entries for the MGS showing what filesystems
2954 Severity : enhancement
2956 Description: provide MGS failover
2957 Details : Added config lock reacquisition after MGS server failover.
2959 Severity : enhancement
2961 Description: add Linux 2.4 support
2962 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2966 Description: a large application tries to do I/O to the same resource and dies
2967 in the middle of it.
2968 Details : Check the req->rq_arrival time after the call to
2969 ost_brw_lock_get(), but before we do anything about
2970 processing it & sending the BULK transfer request. This
2971 should help move old stale pending locks off the queue as
2972 quickly as obd_timeout.
2975 Frequency : when an incorrect nid is specified during startup
2977 Description: ptlrpc connect to non-existant node causes kernel crash
2978 Details : LNET can't be re-entered from an event callback, which
2979 happened when we expire a message after the export has been
2980 cleaned up. Instead, hand the zombie cleanup off to another
2983 Severity : enhancement
2985 Description: plain/inodebits lock performance improvement
2986 Details : Grouping plain/inodebits in granted list by their request modes
2987 and bits policy, thus improving the performance of search through
2991 Frequency : only if OST filesystem is corrupted
2993 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2994 Details : for a short time RPCs with bulk IO are in the replay list,
2995 but replay of bulk IOs is unimplemented. If the OST filesystem
2996 is corrupted due to disk cache incoherency and then replay is
2997 started it is possible to trip an assertion. Avoid putting
2998 committed RPCs into the replay list at all to avoid this issue.
3001 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3003 Description: System hang on startup
3004 Details : This bug allowed the liblustre (e.g. catamount) client to
3005 return to the app before handling all startup RPCs. This
3006 could leave the node unresponsive to lustre network traffic
3007 and manifested as a server ptllnd timeout.
3009 Severity : enhancement
3011 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3012 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
3013 boolean to control whether to print peer debug info when a
3014 client's RPC times out.
3017 Frequency : only for kernels with patches from Lustre below 1.4.3
3019 Description: Remove old rdonly API
3020 Details : Remove old rdonly API which unused from at least lustre 1.4.3
3023 Frequency : only for devices with external journals
3025 Description: Set external device read-only also
3026 Details : During a commanded failover stop, we set the disk device
3027 read-only while the server shuts down. We now also set any
3028 external journal device read-only at the same time.
3031 Frequency : when upgrading from 1.4 while trying to change parameters
3033 Description: The wrong (new) MDC name was used when setting parameters for
3034 upgraded MDT's. Also allows changing of OSC (and MDC)
3035 parameters if --writeconf is specified at tunefs upgrade time.
3038 Frequency : when setting specific ost indicies
3040 Description: QOS code breaks on skipped indicies
3041 Details : Add checks for missing OST indicies in the QOS code, so OSTs
3042 created with --index need not be sequential.
3044 Severity : enhancement
3046 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3047 Details : The uninit_groups feature works in conjunction with the kernel
3048 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3049 up the pass1 processing of e2fsck. This is a read-only feature
3050 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3051 mount filesystems that have had this feature enabled.
3053 Severity : enhancement
3055 Description: Improve multi-block allocation algorithm to avoid fragmentation
3056 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3057 allocation locality and avoid filesystem fragmentation.
3059 ------------------------------------------------------------------------------
3061 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
3063 * Support for kernels:
3064 2.4.21-47.0.1.EL (RHEL 3)
3065 2.6.5-7.283 (SLES 9)
3066 2.6.9-42.0.10.EL (RHEL 4)
3067 2.6.12.6 vanilla (kernel.org)
3068 2.6.16.27-0.9 (SLES 10)
3069 * Recommended e2fsprogs version: 1.39.cfs5
3071 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3075 Frequency : occasional, depends on client load and configuration
3076 Bugzilla : 12181, 12203
3077 Description: data loss for recently-modified files
3079 Details : In some cases it is possible that recently written or created
3080 files may not be written to disk in a timely manner (this should
3081 normally be within 30s unless client IO load is very high).
3082 The problem appears as zero-length files or files that are a
3083 multiple of 1MB in size after a client crash or client eviction
3084 that are missing data at the end of the file.
3086 This problem is more likely to be hit on clients where files are
3087 repeatedly created and unlinked in the same directory, clients
3088 have a large amount of RAM, have many CPUs, the filesystem has
3089 many OSTs, the clients are rebooted frequently, and/or the files
3090 are not accessed by other nodes after being written.
3092 The presence of the problem can be detected by looking at
3093 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
3094 smaller than the second (nr_unused) then dirty files will not
3095 be flushed automatically to disk. "sync; sleep 10" should be
3096 run several times on the node before unmounting it to update
3097 Lustre (this is also safe to run on nodes without this problem).
3099 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
3100 that can cause this same problem, so customers using that kernel
3101 also need to update the kernel in addition to Lustre. In order
3102 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
3104 It is normal that files written just before a client crash (less
3105 than 30s) may not yet have been flushed to disk, even for local
3109 Frequency : frequent on thin XT3 nodes
3111 Description: UUID collision on thin XT3 Linux nodes
3112 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
3113 randomly, since we relied on an insufficiently-seeded PRNG.
3118 Description: OSS hangs after "All ost request buffers busy"
3119 Details : A deadlock between quota and journal operations caused OSS
3120 hangs after printing "All ost request buffers busy."
3123 Frequency : always on liblustre builds
3125 Description: Cleanup compiler warnings on liblustre
3128 Frequency : always on liblustre builds on XT3
3130 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
3134 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
3135 > 32000 subdirectories
3136 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
3137 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
3138 subdirectory count crosses 32000. This will aid e2fsck to
3139 correctly handle more than 32000 subdirectories.
3142 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3144 Description: System hang on startup
3145 Details : This bug allowed the liblustre (e.g. catamount) client to
3146 return to the app before handling all startup RPCs. This
3147 could leave the node unresponsive to lustre network traffic
3148 and manifested as a server ptllnd timeout.
3150 Severity : enhancement
3152 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3153 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
3154 boolean to control whether to print peer debug info when a
3155 client's RPC times out.
3160 Description: make O_SYNC working on 2.6 kernels
3161 Details : 2.6 kernels use different method for mark pages for write,
3162 so need add a code to lustre for O_SYNC work.
3167 Description: Failure to close file and release space on NFS
3168 Details : Put inode details into lock acquired in ll_intent_file_open.
3169 Use mdc_intent_lock in ll_intent_open to properly
3170 detect all kind of errors unhandled by mdc_enqueue.
3175 Description: proc file read during shutdown sometimes raced obd removal,
3177 Details : Add lock to prevent obd access after proc file removal.
3180 Frequency : Only for files larger than 4GB on 32-bit clients.
3182 Description: improperly doing page alignment of locks
3183 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
3184 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
3189 Description: RPCs being resent when they shouldn't be
3190 Details : Some RPCs that should not be resent are being resent. This
3191 can cause inconsistencies in the RPC state machine. Do not
3192 resend such requests.
3195 Frequency : rare, only with NFS export
3197 Description: Crash on NFS re-export node
3198 Details : under very unusual load conditions an assertion is hit in
3199 ll_intent_file_open()
3202 Frequency : only if OST filesystem is corrupted
3204 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3205 Details : for a short time RPCs with bulk IO are in the replay list,
3206 but replay of bulk IOs is unimplemented. If the OST filesystem
3207 is corrupted due to disk cache incoherency and then replay is
3208 started it is possible to trip an assertion. Avoid putting
3209 committed RPCs into the replay list at all to avoid this issue.
3214 Description: large O_DIRECT requests fail under memory pressure/fragmentation
3215 Details : Large single O_DIRECT read and write calls can fail to allocate
3216 a sufficiently large buffer to process the request. In case of
3217 allocation failure the allocation is retried with a smaller
3218 buffer and broken into smaller requests.
3220 Severity : enhancement
3222 Description: Add -o localflock option to simulate old noflock behaviour.
3223 Details : This will achieve local-only flock/fcntl locks coherentness.
3228 Description: versioning check is incomplete
3229 Details : Checking the version difference of client vs. server, report
3230 error if the gap is too big.
3235 Description: add support PG_writeback bit
3236 Details : add support for PG_writeback bit for Lustre, for more carefull
3237 work with page cache in 2.6 kernel. This also fix some deadlocks
3238 and remove hack for work O_SYNC with 2.6 kernel.
3240 Severity : enhancement
3242 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3243 Details : The uninit_groups feature works in conjunction with the kernel
3244 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3245 up the pass1 processing of e2fsck. This is a read-only feature
3246 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3247 mount filesystems that have had this feature enabled.
3249 Severity : enhancement
3251 Description: Improve multi-block allocation algorithm to avoid fragmentation
3252 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3253 allocation locality and avoid filesystem fragmentation.
3255 ------------------------------------------------------------------------------
3257 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
3259 * Support for kernels:
3260 2.6.9-42.0.3.EL (RHEL 4)
3261 2.6.5-7.276 (SLES 9)
3262 2.4.21-47.0.1.EL (RHEL 3)
3263 2.6.12.6 vanilla (kernel.org)
3264 2.6.16.21-0.8 (SLES10)
3265 * Recommended e2fsprogs version: 1.39.cfs2-0
3267 * The backwards-compatible /proc/sys/portals symlink has been removed
3268 in this release. Before upgrading, please ensure that you change
3269 any configuration scripts or /etc/sysctl.conf files that access
3270 /proc/sys/portals/* or sysctl portals.* to use the corresponding
3271 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
3272 in advance of the upgrade on any system running Lustre 1.4.6 or
3273 newer, since /proc/sys/lnet was added in that version.
3274 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3278 Frequency : only when quota is used
3280 Description: avoid scanning export list for quota master
3281 Details : Change the algorithms to avoid scanning export list in order
3282 to improve the efficiency.
3285 Frequency : MDS failover only, very rarely
3287 Description: "went back in time" messages on mds failover
3288 Details : The greatest transno may be lost when the current operation
3289 finishes with an error (transno==0) and the client's last_rcvd
3290 record is over-written. Save the greatest transno in the
3291 mds_last_transno for this case.
3294 Frequency : always for specific kernels and striping counts
3296 Description: client may get "Matching packet too big" without ACL support
3297 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
3298 when trying to access files in certain configurations. The
3299 clients should in fact be denied when mounting because they do
3300 not understand ACLs.
3303 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3305 Description: many clients connecting with IO in progress causes connect timeouts
3306 Details : Avoid synchronous journal commits to avoid delays caused by many
3307 clients connecting/disconnecting when bulk IO is in progress.
3308 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3309 OST_IO_PORTAL to avoid delays behind potentially many pending
3313 Frequency : occasionally with multiple writers to a single file
3315 Description: shared writes to file may result in wrong size reported by stat()
3316 Details : Allow growing of kms when extent lock is cancelled
3319 Frequency : always with random mmap IO to multi-striped file
3321 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3322 Details : Only if the hole is at the end of OST object so that kms is too
3323 small. Fix is to increase kms accordingly in ll_nopage.
3326 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3328 Description: writes to a missing object would leak memory on the OST
3329 Details : If there is an inconsistency between the MDS and OST filesystems,
3330 such that the MDS references an object that doesn't exist, writes
3331 to that object will leak memory due to incorrect cleanup in the
3332 error handling path, eventually running out of memory on the OST.
3337 Description: Creating too long symlink causes lustre errors
3338 Details : Check symlink and name lengths before sending requests to MDS.
3341 Frequency : only if flock is enabled (not on by default)
3343 Description: posix locks not released on fd closure on 2.6.9+
3344 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3345 kernels, this caused such locks not to be released on fd close and
3346 then assertions on fs unmount about still used locks.
3349 Frequency : MDS failover only, very rarely
3351 Description: clients may get ASSERTION(granted_lock != NULL)
3352 Details : When request was taking a long time, and a client was resending
3353 a getattr by name lock request. The were multiple lock requests
3354 with the same client lock handle and
3355 mds_getattr_name->fixup_handle_for_resent_request found one of the
3356 lock handles but later failed with ASSERTION(granted_lock != NULL).
3361 Description: handle->h_buffer_credits > 0, assertion failure
3362 Details : h_buffer_credits is zero after truncate, causing assertion
3363 failure. This patch extends the transaction or creates a new
3367 Frequency : NFS re-export or patchless client
3368 Bugzilla : 11179, 10796
3369 Description: Crash on NFS re-export node (__d_move)
3370 Details : We do not want to hash the dentry if we don't have a lock.
3371 But if this dentry is later used in d_move, we'd hit uninitialised
3372 list head d_hash, so we just do this to init d_hash field but
3373 leave dentry unhashed.
3376 Frequency : NFS re-export or patchless client
3378 Description: NFS exports has problem with symbolic link
3379 Details : lustre client didn't properly install dentry when re-exported
3380 to NFS or running patchless client.
3383 Frequency : NFS re-export or patchless client
3385 Description: Various nfs/patchless fixes.
3386 Details : fixes reuse disconected alias for lookup process - this fixes
3387 warning "find_exported_dentry: npd != pd",
3388 fix permission error with open files at nfs.
3389 fix apply umask when do revalidate.
3392 Frequency : occasional
3394 Description: Crash on NFS re-export node
3395 Details : calling clear_page() on the wrong pointer triggered oops in
3396 generic_mapping_read().
3399 Frequency : rarely, using O_DIRECT IO
3401 Description: unaligned directio crashes client with LASSERT
3402 Details : check for unaligned buffers before trying any requests.
3405 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3407 Description: stale data returned from RAID cache
3408 Details : If only a small amount of IO is done to the RAID device before
3409 reading it again it is possible to get stale data from the RAID
3410 cache instead of reading it from disk.
3413 Frequency : always for sles10 kernel
3415 Description: sles10 support
3416 Details : ll_follow_link: compile fixes and using of nd_set_link
3417 under newer kernels.
3420 Frequency : depends on arch, kernel and compiler version, always on sles10
3423 Description: recursive or deep enough symlinks cause stack overflow
3424 Details : getting rid of large stack-allocated variable in
3428 Frequency : depends on hardware
3430 Description: lustre write performance loss in the SLES10 kernel
3431 Details : the performance loss is caused by using of write barriers in the
3432 ext3 code. The SLES10 kernel turns barrier support on by
3433 default. The fix is to undo that change for ldiskfs.
3435 ------------------------------------------------------------------------------
3437 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3439 * Support for kernels:
3440 2.6.9-42.0.3EL (RHEL 4)
3441 2.6.5-7.276 (SLES 9)
3442 2.4.21-47.0.1.EL (RHEL 3)
3443 2.6.12.6 vanilla (kernel.org)
3447 Frequency : quota enabled and large files being deleted
3449 Description: releasing more than 4GB of quota at once hangs OST
3450 Details : If a user deletes more than 4GB of files on a single OST it
3451 will cause the OST to spin in an infinite loop. Release
3452 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3457 Description: statfs data retrieved from /proc may be stale or zero
3458 Details : When reading per-device statfs data from /proc, in the
3459 {kbytes,files}_{total,free,avail} files, it may appear
3460 as zero or be out of date.
3463 Frequency : systems with MD RAID1 external journal devices
3465 Description: lconf's call to blkid is confused by RAID1 journal devices
3466 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3467 of returning all block devices that match the journal UUID.
3470 Frequency : always, for aggregate stripe size over 4GB
3472 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3473 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3474 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3477 Frequency : always if "lfs find" used on a local file/directory
3479 Description: "lfs find" segfaults if used on a local file/directory
3480 Details : The case where a directory component was not specified wasn't
3481 handled correctly. Handle this properly.
3484 Frequency : always on ppc64
3486 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3487 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3488 perform endianness conversion.
3491 Frequency : rarely (truncate to non-zero file size after write under load)
3492 Bugzilla : 10730, 10687
3493 Description: Files padded with zeros to next 4K multiple
3494 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3495 it is possible that files that are truncated to a non-zero size
3496 immediately after being written are filled with zero bytes beyond
3497 the truncated size. No file data is lost.
3499 Severity : enhancement
3501 Description: Allow recovery/failover for liblustre clients.
3502 Details : liblustre clients were unaware of failover configurations until
3505 Severity : enhancement
3507 Description: user file locks should fail when not mounting with flock option
3508 Details : Set up an error-returning stub in ll_file_operations.lock field
3509 to prevent incorrect behaviour when client is mounted without
3510 flock option. Also, set up properly f_op->flock field for
3514 Frequency : always on ia64
3516 Description: "lfs df" loops on printing out MDS statfs information
3517 Details : The obd_ioctl_data was not initialized and in some systems
3518 this caused a failure during the ioctl that did not return
3519 an error. Initialize the struct and return an error on failure.
3522 Frequency : SLES 9 only
3524 Description: Error of copying files with lustre special EAs as root
3525 Details : Client side always return success for setxattr call for lustre
3526 special xattr (currently only "trusted.lov").
3529 Frequency : rarely on clusters with both ia64+i386 clients
3531 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3532 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3533 clients are doing concurrent writes to the same file it is
3534 possible that the ia64 clients may LASSERT because the OST
3535 extent locks are not PAGE_SIZE aligned. Ensure that grown
3536 locks are always aligned on the request boundary.
3539 Frequency : specific use, occasional
3541 Description: Overwriting in use executable truncates on-disk binary image
3542 Details : If one node attempts to overwrite an executable in use by
3543 another node, we now correctly return ETXTBSY instead of
3544 truncating the file.
3546 Severity : enhancement
3548 Description: Async OSC create to avoid the blocking unnecessarily.
3549 Details : If a OST has no remain object, system will block on the creating
3550 when need to create a new object on this OST. Now, ways use
3551 pre-created objects when available, instead of blocking on an
3552 empty osc while others are not empty. If we must block, we block
3553 for the shortest possible period of time.
3558 Description: chmod on Lustre root is propagated to other clients
3559 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3560 invalid by the follow_mount time.
3563 Frequency : liblustre clients only
3565 Description: Race in 'instant cancel' lock handling could lead to such locks
3566 never to be granted in case of SMP MDS
3567 Details : Do not destroy not yet granted but cbpending locks in
3571 Frequency : replay/resend of open
3573 Description: non null lock assetion failure in mds_intent_policy
3574 Details : Trying to replay/resend lockless open requests resulted in
3575 mds_open() returning 0 with no lock. Now it sets a flag if
3576 a lock is going to be returned.
3578 Severity : enhancement
3580 Description: Checksum enhancements
3581 Details : New checksum enhancements allow for resending RPCs that failed
3584 Severity : enhancement
3586 Description: Tunables on number of dirty pages in cacche
3587 Details : Allow to set limit on number of dirty pages cached.
3592 Description: client crash on unmount - lock still has references
3593 Details : In some error handling cases it was possible to leak a lock
3594 reference on a client while accessing a file. This was not
3595 harmful to the client during operation, but would cause the
3596 client to crash when the filesystem is unmounted.
3599 Frequency : specific case, rare
3601 Description: ETXTBSY on mds though file not in use by client
3602 Details : ETXTBSY is no longer incorrectly returned when attempting to
3603 chmod or chown a directory that the user previously tried to
3604 execute or a currently-executing binary.
3607 Frequency : extremely rare except on liblustre-based clients
3609 Description: Lustre space not freed when files are deleted
3610 Details : Clean up open-unlinked files after client eviction. Previously
3611 the unlink was skipped and the files remained as orphans.
3616 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3617 Details : In some cases it was possible to send a blocking callback to
3618 a client doing a glimpse, even though that client didn't get
3619 a lock granted. When the glimpse lock is cancelled on the OST
3620 the freed lock is left on the waiting list and corrupted the list.
3623 Frequency : all core dumps
3625 Description: Broke core dumps to lustre
3626 Details : Negative dentry may be unhashed if parent does not have UPDATE
3627 lock, but some callers, e.g. do_coredump, expect dentry to be
3628 hashed after successful create, hash it in ll_create_it.
3630 ------------------------------------------------------------------------------
3632 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3634 * Support for kernels:
3635 2.6.9-42.0.2.EL (RHEL 4)
3636 2.6.5-7.276 (SLES 9)
3637 2.4.21-40.EL (RHEL 3)
3638 2.6.12.6 vanilla (kernel.org)
3642 Frequency : always on RHEL 3
3644 Description: Number of open files grows over time
3645 Details : The number of open files grows over time, whether or not
3646 Lustre is started. This was due to a filp leak introduced
3647 by one of our kernel patches.
3649 ------------------------------------------------------------------------------
3651 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3653 * Support for kernels:
3654 2.6.9-42.EL (RHEL 4)
3655 2.6.5-7.276 (SLES 9)
3656 2.4.21-40.EL (RHEL 3)
3657 2.6.12.6 vanilla (kernel.org)
3662 Bugzilla : 5719, 9635, 9792, 9684
3663 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3664 Details : If a server is under heavy load and cannot reply to new
3665 connection requests before the client resends the (re)connect,
3666 the connection handling code can behave badly if two service
3667 threads are concurrently handing separate (re)connections from
3668 the same client. Add better locking to the connection handling
3669 code, and ensure that only a single connection will be processed
3670 for a given client UUID, even if the lock is dropped.
3672 Severity : enhancement
3674 Description: add TCP zero-copy support to kernel
3675 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3676 sends if the hardware supports scatter-gather and checksumming.
3677 This allows socklnd to do client-write and server-read more
3678 efficiently and reduce CPU utilization from skbuf copying.
3681 Frequency : only if NFS exporting from client
3683 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3684 Details : If setting attributes on a file created under NFS that had
3685 never been opened it would be possible to oops the client
3686 if the file had no objects.
3689 Frequency : always for liblustre
3691 Description: liblustre client does MDS+OSTs setattr RPC for each write
3692 Details : When doing a write from a liblustre client, the client
3693 incorrectly issued an RPC to the MDS and each OST the file was
3694 striped over in order to update the timestamps. When writing
3695 with small chunks and many clients this could overwhelm the MDS
3696 with RPCs. In all cases it would slow down the write because
3697 these RPCs are unnecessary.
3699 Severity : enhancement
3701 Description: allow number of MDS service threads to be changed at module load
3702 Details : It is now possible to change the number of MDS service threads
3703 running. Adding "options mds mds_num_threads={N}" to the MDS's
3704 /etc/modprobe.conf will set the number of threads for the next
3705 time Lustre is restarted (assuming the "mds" module is also
3706 reloaded at that time). The default number of threads will
3707 stay the same, 32 for most systems.
3712 Description: OST crash if filesystem is unformatted or corrupt
3713 Details : If an OST is started on a device that has never been formatted
3714 or if the filesystem is corrupt and cannot even mount then the
3715 error handling cleanup routines would dereference a NULL pointer.
3720 Description: NULL pointer deref in llap_from_page.
3721 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3722 incorrect) mapping if the page was truncated/reclaimed while it was
3723 searched for. Check for this condition and skip such pages when
3724 doing readahead. Introduce extra check to llap_from_page() to
3725 verify page->mapping->host is non-NULL (so page is not anonymous).
3728 Frequency : Sometimes when using sys_sendfile
3730 Description: "page not covered by a lock" warnings from ll_readpage
3731 Details : sendfile called ll_readpage without right page locks present.
3732 Now we introduced ll_file_sendfile that does necessary locking
3733 around call to generic_file_sendfile() much like we do in
3737 Frequency : with certain MDS communication failures at client mount time
3739 Description: NULL pointer deref after failed client mount
3740 Details : a client connection request may delayed by the network layer
3741 and not be sent until after the PTLRPC layer has timed out the
3742 request. If the client fails the mount immediately it will try
3743 to clean up before the network times out the request. Add a
3744 reference from the request import to the obd device and delay
3745 the cleanup until the network drops the request.
3748 Frequency : occasionally during client (re)connect
3750 Description: assertion failure during client (re)connect
3751 Details : processing a client connection request may be delayed by the
3752 client or server longer than the client connect timeout. This
3753 causes the client to resend the connection request. If the
3754 original connection request is replied in this interval, the
3755 client may trip an assertion failure in ptlrpc_connect_interpret()
3756 which thought it would be the only running connect process.
3759 Frequency : only with obd_echo servers and clients that are rebooted
3761 Description: kernel BUG accessing uninitialized data structure
3762 Details : When running an obd_echo server it did not start the ping_evictor
3763 thread, and when a client was evicted an uninitialized data
3764 structure was accessed. Start the ping_evictor in the RPC
3765 service startup instead of the OBD startup.
3767 Severity : enhancement
3768 Bugzilla : 10193 (patchless)
3769 Description: Remove dependency on various unexported kernel interfaces.
3770 Details : No longer need reparent_to_init, exit_mm, exit_files,
3771 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3774 Frequency : rare (only users of deprecated and unsupported LDAP config)
3776 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3777 Details : LDAP apparently contains 'lustreName' attributes instead of
3778 'name'. A simple remapping of the name is sufficient.
3781 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3783 Description: waiting_locks_callback trips kernel BUG if client is evicted
3784 Details : Running with the dump_on_timeout debug flag turned on makes
3785 it possible that the waiting_locks_callback() can try to dump
3786 the Lustre kernel debug logs from an interrupt handler. Defer
3787 this log dumping to the expired_lock_main() thread.
3789 Severity : enhancement
3791 Description: Support NFS exporting on 2.6 kernels.
3792 Details : Implement non-rawops metadata methods for NFS server to use without
3793 changing NFS server code.
3796 Frequency : very rare (synthetic metadata workload only)
3798 Description: two racing renames might cause an MDS thread to deadlock
3799 Details : Running the "racer" program may cause one MDS thread to rename
3800 a file from being the source of a rename to being the target of
3801 a rename at exactly the same time that another thread is doing
3802 so, and the second thread has already enqueued these locks after
3803 doing a lookup of the target and is trying to relock them in
3804 order. Ensure that we don't try to re-lock the same resource.
3807 Frequency : only very large systems with liblustre clients
3809 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3810 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3811 all OSTs in parallel. This allows the eviction of stale liblustre
3812 clients to proceed much faster than if they were done in series,
3813 and also offers similar improvements for other set_info RPCs.
3818 Description: excessive CPU usage during initial read phase on client
3819 Details : During the initial read phase on a client, it would agressively
3820 retry readahead on the file, consuming too much CPU and impacting
3821 performance (since 1.4.5.8). Improve the readahead algorithm
3822 to avoid this, and also improve some other common cases (read
3823 of small files in particular, where "small" is files smaller than
3824 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3829 Description: MDS crash when receiving packet with unknown intent.
3830 Details : Do not LBUG in unknown intent case, just return -EFAULT
3832 Severity : enhancement
3833 Bugzilla : 9293, 9385
3834 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3835 Details : Do not serialize getattr (non-intent version) and statfs.
3838 Frequency : occasional, when OST network is overloaded/intermittent
3840 Description: client evicted by OST after bulk IO timeout
3841 Details : If a client sends a bulk IO request (read or write) the OST
3842 may evict the client if it is unresposive to its data GET/PUT
3843 request. This is incorrect if the network is overloaded (takes
3844 too long to transfer the RPC data) or dropped the OST GET/PUT
3845 request. There is no need to evict the client at all, since
3846 the pinger and/or lock callbacks will handle this, and the
3847 client can restart the bulk request.
3850 Frequency : Always when mmapping file with no objects
3852 Description: client crashes when mmapping file with no objects
3853 Details : Check that we actually have objects in a file before doing any
3854 operations on objects in ll_vm_open, ll_vm_close and
3860 Description: Request leak when working with deleted CWD
3861 Details : Introduce advanced request refcount tracking for requests
3862 referenced from lustre intent.
3864 Severity : Enhancement
3866 Description: Cache open file handles on client.
3867 Details : MDS now will return special lock along with openhandle, if
3868 requested and client is allowed to hold openhandle, even if unused,
3869 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3870 closing files for every read/write openration.
3872 Severity : Enhancement
3874 Description: Cache open negative dentries on client when possible.
3875 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3876 negative dentries on lock revocation.
3881 Description: Remounting a client read-only wasn't possible with a zconf mount
3882 Details : It wasn't possible to remount a client read-only with llmount.
3884 Severity : enhancement
3885 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3886 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3887 MPICH is included to add Lustre-specific ADIO interfaces.
3888 This is based closely on the UFS ADIO layer and only differs
3889 in file creation, in order to allow the OST striping to be set.
3890 This is user-contributed code and not supported by CFS.
3895 Description: extended inode attributes (immutable, append-only) work improperly
3896 when 2.4 and 2.6 kernels are used on client/server or vice versa
3897 Details : Introduce kernel-independent values for these flags.
3899 Severity : enhancement
3902 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3903 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3904 granularity. Now, Fractional MB granularity is be supported,
3905 this is very useful for low memory system.
3907 Severity : enhancement
3909 Description: Getattr by fid
3910 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3911 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3912 have locks and avoids getattr rpc requests for it on every stat.
3915 Frequency : Always, for filesystems larger than 2TB
3917 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3918 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3919 filesystems larger than 2TB to be mounted with mballoc enabled.
3922 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3924 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3925 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3926 filesystems larger than 2TB is mounted then the kernel will
3927 silently corrupt the start of the filesystem. CONFIG_LBD is
3928 enabled for all CFS-supported kernels, but the possibility of
3929 this happening with a modified kernel config exists.
3931 Severity : enhancement
3933 Description: add client O_DIRECT support for 2.6 kernels
3934 Details : It is now possible to do O_DIRECT reads and writes to files
3935 in the Lustre client mountpoint on 2.6 kernel clients.
3937 Severity : enhancement
3939 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3940 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3941 parallel, not waiting for response from every OST before sending
3942 a rpc to the next OST.
3947 Description: setattr vs write race when updating file timestamps
3948 Details : Client processes that update a file timestamp into the past
3949 right after writing to the file (e.g. tar) it is possible that
3950 the updated file modification time can be reset to the current
3951 time due to a race between processing the setattr and write RPC.
3953 Severity : enhancement
3955 Description: Bring 'lfs find' closer in line with regular Linux find.
3956 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3957 -print0 options and obtains all the needed info through the lustre
3960 Severity : enhancement
3962 Description: support up to 1024 configured devices on one node
3963 Details : change obd_dev array from statically allocated to dynamically
3964 allocated structs as they are first used to reduce memory usage
3969 Description: Flush dirty partially truncated pages during truncate
3970 Details : Immediatelly flush partially truncated pages in filter_setattr,
3971 this way we completely avoid having any pages in page cache on OST
3972 and can retire ugly workarounds during writes to flush such pages.
3977 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3978 Details : protect inode from truncation within vfs_unlink() context
3979 just take a reference before calling vfs_unlink() and release it
3980 when parent's i_sem is free.
3985 Description: last_id value checked outside lock on OST caused LASSERT failure
3986 Details : If there were multiple MDS->OST object precreate requests in
3987 flight, it was possible that the OST's last object id was checked
3988 outside a lock and incorrectly tripped an assertion. Move checks
3989 inside locks, and discard old precreate requests.
3992 Frequency : always, if extents are used on OSTs
3994 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3995 Details : OSTs using the extents format would not zero the high 16 bits of
3996 the index physical block number. This is not a problem for any
3997 OST filesystems smaller than 16TB, and no kernels support ext3
3998 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3999 new/modified files) and can be fixed for existing filesystems
4000 with e2fsprogs-1.39-cfs1.
4005 Description: import connection selection may be incorrect if timer wraps
4006 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
4007 import connections to be ignored if the 32-bit jiffies counter
4008 wraps. Use a 64-bit jiffies counter.
4011 Frequency : very large clusters immediately after boot
4013 Description: LNET request buffers exhausted under heavy short-term load
4014 Details : If a large number of client requests are generated on a service
4015 that has previously never seen so many requests it is possible
4016 that the request buffer growth cannot keep up with the spike in
4017 demand. Instead of dropping incoming requests, they are held in
4018 the LND until the RPC service can accept more requests.
4021 Frequency : Sometimes during replay
4023 Description: Assertion failure in ll_local_open after replay.
4024 Details : If replay happened on an open request reply before we were able
4025 to set replay handler, reply will become not swabbed tripping the
4026 assertion in ll_local_open. Now we set the handler right after
4027 recognising of open request
4030 Frequency : very rare
4032 Description: kernel reports "badness in vsnprintf"
4033 Details : Reading from the "recovery_status" /proc file in small chunks
4034 may cause a negative length in lprocfs_obd_rd_recovery_status()
4035 call to vsnprintf() (which is otherwise harmless). Exit early
4036 if there is no more space in the output buffer.
4038 Severity : enhancement
4040 Description: clear OBD RPC statistics by writing to them
4041 Details : It is now possible to clear the OBD RPC statistics by writing
4042 to the "stats" file.
4047 Description: Client mtime is not the same on different clients after utimes
4048 Details : In some cases, the client was using the utimes() syscall on
4049 a file cached on another node. The clients now validate the
4050 ctime from the MDS + OSTs to determine which one is right.
4055 Description: Inability to activate failout mode
4056 Details : lconf script incorrectly assumed that in pythong string's numeric
4057 value is used in comparisons.
4060 Frequency : always with multiple stripes per file
4062 Description: Inefficient object allocation for mutli-stripe files
4063 Details : When selecting which OSTs to stripe files over, for files with
4064 a stripe count that divides evenly into the number of OSTs,
4065 the MDS is always picking the same starting OST for each file.
4066 Return the OST selection heuristic to the original design.
4071 Description: mount failures may take full timeout to return an error
4072 Details : Under some heavy load conditions it is possible that a
4073 failed mount can wait for the full obd_timeout interval,
4074 possibly several minutes, before reporting an error.
4075 Instead return an error as soon as the status is known.
4077 Frequency : quota enabled and large files being deleted
4079 Description: releasing more than 4GB of quota at once hangs OST
4080 Details : If a user deletes more than 4GB of files on a single OST it
4081 will cause the OST to spin in an infinite loop. Release
4082 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
4087 Description: statfs data retrieved from /proc may be stale or zero
4088 Details : When reading per-device statfs data from /proc, in the
4089 {kbytes,files}_{total,free,avail} files, it may appear
4090 as zero or be out of date.
4093 Frequency : systems with MD RAID1 external journal devices
4095 Description: lconf's call to blkid is confused by RAID1 journal devices
4096 Details : Use the "blkid -l" flag to locate the MD RAID device instead
4097 of returning all block devices that match the journal UUID.
4100 Frequency : always, for aggregate stripe size over 4GB
4102 Description: assertion fails when trying to use 4GB stripe size
4103 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
4104 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
4107 Frequency : always on ppc64
4109 Description: the first write on an ext3 filesystem with mballoc got stuck
4110 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
4111 perform endianness conversion.
4113 ------------------------------------------------------------------------------
4115 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
4117 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
4118 INTEROPERATE with older versions automatically. Please read the
4119 user documentation before upgrading any part of a live system.
4120 * WARNING: Lustre networking configuration changes are required with
4121 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
4124 * Support for kernels:
4125 2.6.9-22.0.2.EL (RHEL 4)
4126 2.6.5-7.244 (SLES 9)
4127 2.6.12.6 vanilla (kernel.org)
4130 Severity : enhancement
4131 Bugzilla : 7981/8208
4132 Description: Introduced Lustre Networking (LNET)
4133 Details : LNET is new networking infrastructure for Lustre, it includes
4134 a reorganized network configuration mode (see the user
4135 documentation for full details) as well as support for routing
4136 between different network fabrics. Lustre Networking Devices
4137 (LNDs) for the supported network fabrics have also been
4138 created for this new infrastructure.
4140 Severity : enhancement
4141 Description: Introduced Access control lists
4142 Details : clients can set ACLs on files and directories in order to have
4143 more fine-grained permissions than the standard Unix UGO+RWX.
4144 The MDS must be started with the "-o acl" mount option.
4146 Severity : enhancement
4147 Description: Introduced filesystem quotas
4148 Details : Administrators may now establish per-user quotas on the
4151 Severity : enhancement
4153 Description: Configuration change for the XT3
4154 The PTLLND is now used to run Lustre over Portals on the XT3
4155 The configure option(s) --with-cray-portals are no longer used.
4156 Rather --with-portals=<path-to-portals-includes> is used to
4157 enable building on the XT3. In addition to enable XT3 specific
4158 features the option --enable-cray-xt3 must be used.
4163 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
4164 Details : A race between allocating a new llcd and re-getting the llcd_lock
4165 allowed another thread to grab newly-allocated llcd.
4167 Severity : enhancement
4169 Description: 2.6 OST async journal commit and locking fix to improve performance
4170 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
4171 2.6 kernels are now async as they already were in 2.4 kernels so
4172 that they can commit concurrently with the network bulk transfer.
4173 For block-allocated files the filter allocation semaphore is held
4174 to avoid filesystem fragmentation during allocation. BKL lock
4175 removed for 2.6 xattr operations where it is no longer needed.
4180 Description: lconf incorrectly determined whether two IP networks could talk
4181 Details : In some more complicated routing and multiple-network
4182 configurations, lconf will avoid trying to make a network
4183 connection to a disjoint part of the IP space. It was doing the
4184 math incorrectly for one set of cases.
4189 Description: Fix for potential infinite loop processing records in an llog.
4190 Details : If an llog record is corrupted/zeroed, it is possible to loop
4191 forever in llog_process(). Validate the llog record length
4192 and skip the remainder of the block on error.
4195 Frequency : occasional (liblustre only)
4197 Description: liblustre could not open files whose last component is a symlink
4198 Details : sysio_path_walk() would incorrectly pass the open intent to
4199 intermediate path components.
4202 Frequency : rare (liblustre only with non-standard tuning)
4203 Bugzilla : 7201 (7350)
4204 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
4205 Details : llu_lookup_finish_locks() tries to set lock data on a lock
4206 after it has been released, only do this for referenced locks
4208 Severity : enhancement
4210 Description: specifying an (invalid) directory default stripe_size of -1
4211 would reset the directory default striping
4212 Details : stripe_size -1 was used internally to signal directory stripe
4213 removal, now use "all default" to signal dir stripe removal
4214 as a directory striping of "all default" is not useful
4217 Frequency : common for large clusters running liblustre clients
4219 Description: doing an ls when liblustre clients are running is slow
4220 Details : sending a glimpse AST to a liblustre client waits for every AST
4221 to time out, as liblustre clients will not respond. Since they
4222 cannot cache data we refresh the OST lock LVB from disk instead.
4224 Severity : enhancement
4226 Description: doing an ls at the same time as file IO can be slow
4227 Details : enqueue and other "small" requests can be blocked behind many
4228 large IO requests. Create a new OST IO portal for non-IO
4229 requests so they can be processed faster.
4232 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
4234 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
4235 created file with mode 000
4236 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
4237 allow an ftruncate() as a non-root user to a file with mode 000.
4238 Lustre now respects this flag to disable mode checks when
4239 truncating a file owned by the user
4242 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
4245 Description: Revoking locks from clients that went dead or catatonic might take
4247 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
4248 cancellation of such locks instant on servers without waiting for
4249 any reply from clients. Clients drops these locks when cancel
4250 notification from server is received without replying.
4253 Frequency : liblustre-only, when liblustre client dies or becomes busy
4255 Description: Doing ls on Linux clients can take a long time with active
4257 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
4258 granting such locks to it in the first place if possible. Locks
4259 are taken by proxy on the OST during the read or write and
4260 dropped immediately afterward. Add connect flags handling, do
4261 not grant locks to liblustre clients for glimpse ASTs.
4263 Severity : enhancement
4265 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
4266 Details : Existing read-ahead algorithm is tuned for the case of streamlined
4267 sequential reads and behaves badly with applications doing random
4268 reads. Improve it by reading ahead at least read region, and
4269 avoiding excessive large RPC for small reads.
4271 Severity : enhancement
4273 Description: Creating more than 1000 files for a single job may cause a load
4274 imbalance on the OSTs if there are also a large number of OSTs.
4275 Details : qos_prep_create() uses an OST index reseed value that is an
4276 even multiple of the number of available OSTs so that if the
4277 reseed happens in the middle of the object allocation it will
4278 still utilize the OSTs as uniformly as possible.
4283 Description: OST or MDS may oops in ping_evictor_main()
4284 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
4285 but doesn't restart at beginning of obd_exports_timed list
4288 Severity : enhancement
4290 Description: improve by-nid export eviction on the MDS and OST
4291 Details : allow multiple exports with the same NID to be evicted at one
4292 time without re-searching the exports list.
4295 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
4297 Description: MDS may oops in groups_free()
4298 Details : in rare race conditions a newly allocated group_info struct is
4299 freed again, and this can be NULL. The 2.4 compatibility code
4300 for groups_free() checked for a NULL pointer, but 2.6 did not.
4303 Frequency : common for liblustre clients doing little filesystem IO
4304 Bugzilla : 9352, 7313
4305 Description: server may evict liblustre clients accessing contended locks
4306 Details : if a client is granted a lock or receives a completion AST
4307 with a blocking AST already set it would not reply to the AST
4308 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4311 Frequency : lfs setstripe, only systems with more than 160 OSTs
4313 Description: unable to set striping with a starting offset beyond OST 160
4314 Details : llapi_create_file() incorrectly limited the starting stripe
4315 index to the maximum single-file stripe count.
4318 Frequency : LDAP users only
4320 Description: lconf did not handle in-kernel recovery with LDAP properly
4321 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4323 Severity : enhancement
4325 Description: bind OST threads to NUMA nodes to improve performance
4326 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4327 node and do their allocations there to localize memory access
4329 Severity : enhancement
4331 Description: llmount can determine client NID directly from Myrinet (GM)
4332 Details : the client NID code from gmnalnid was moved directly into
4333 llmount, removing the need to use this or specifying the
4334 client NID explicitly when mounting GM clients with zeroconf
4337 Frequency : if client is started with down MDS
4339 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4340 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4341 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4342 ptlrpc_import_delayed_req() because we didn't check if the
4343 request was interrupted, and we also didn't break out of the
4344 event loop if there was no timeout
4349 Description: data loss during non-page-aligned writes to a single file from
4350 both multiple nodes and multiple threads on one node at same time
4351 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4352 inconsistency led to false short-reads, that were cached and later
4353 used by ->prepare_write() to fill in partially written page,
4354 leading to data loss.
4357 Frequency : always, if lconf --abort_recovery used
4359 Description: lconf --abort_recovery fails with 'Operation not supported'
4360 Details : lconf was attempting to abort recovery on the MDT device and not
4363 Severity : enhancement
4365 Description: remove cleanup logs
4366 Details : replace lconf-generated cleanup logs with lustre internal
4367 cleanup routines. Eliminates the need for client-cleanup and
4370 Severity : enhancement
4372 Description: add support for EAs (user and system) on lustre filesystems
4373 Details : it is now possible to store extended attributes in the Lustre
4374 client filesystem, and with the user_xattr mount option it
4375 is possible to allow users to store EAs on their files also
4377 Severity : enhancement
4379 Description: Add possibility (config option) to show minimal available OST free
4381 Details : When compiled with --enable-mindf configure option, statfs(2)
4382 (and so, df) will return least minimal free space available from
4383 all OSTs as amount of free space on FS, instead of summary of
4384 free spaces of all OSTs.
4386 Severity : enhancement
4388 Description: do not expand extent locks acquired on OST-side
4389 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4390 by server: they are not cached anyway.
4393 Frequency : when mmap is used/binaries executed from Lustre
4395 Description: Unmmap pages before throwing them away from read cache.
4396 Details : llap_shrink cache now attempts to unmap pages before discarding
4397 them (if unmapping failed - do not discard). SLES9 kernel has
4398 extra checks that trigger if this unmapping is not done first.
4403 Description: lconf didn't resolve symlinks before checking to see whether a
4404 given mountpoint was already in use
4407 Frequency : when migrating failover services
4408 Bugzilla : 6395, 9514
4409 Description: When migrating a subset of services from a node (e.g. failback
4410 from a failover service node) the remaining services would
4411 time out and evict clients.
4412 Details : lconf --force (implied by --failover) sets the global obd_timeout
4413 to 5 seconds in order to quickly disconnect, but this caused
4414 other RPCs to time out too quickly. Do not change the global
4415 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4417 Severity : enhancement
4418 Frequency : if MDS is started with down OST
4419 Bugzilla : 9439,5706
4420 Description: Allow startup/shutdown of an MDS without depending on the
4421 availability of the OSTs.
4422 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4423 Add appropriate locking and lov-osc refcounts for safe
4424 cleaning. Add osc abort_inflight calls in case the
4425 synchronize never started.
4428 Frequency : occasional (Cray XT3 only)
4430 Description: root not authorized to access files in CRAY_PORTALS environment
4431 Details : The client process capabilities were not honoured on the MDS in
4432 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4433 been accessed by an authorized user then root was able to access
4434 the file on the local client also. The root user capabilities
4435 are now allowed on the MDS, as this environment has secure UID.
4438 Frequency : occasional
4440 Description: ldiskfs "too long searching" message happens too often
4441 Details : A debugging message (otherwise harmless) prints too often on
4442 the OST console. This has been reduced to only happen when
4443 there are fragmentation problems on the filesystem.
4448 Description: Division by zero in statfs when all OSCs are inactive
4449 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4450 lov_statfs divides by value returned by lov_get_stripecnt().
4454 Bugzilla : 9489, 3273
4455 Description: First write from each client to each OST was only 4kB in size,
4456 to initialize client writeback cache, which caused sub-optimal
4457 RPCs and poor layout on disk for the first writen file.
4458 Details : Clients now request an initial cache grant at (re)connect time
4459 and so that they can start streaming writes to the cache right
4460 away and always do full-sized RPCs if there is enough data.
4461 If the OST is rebooted the client also re-establishes its grant
4462 so that client cached writes will be honoured under the grant.
4467 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4468 Details : Now I/O RPCs go to different portal number and (presumably) fast
4469 lock requests (and glimses) and other RPCs get their own service
4470 threads pool that should be able to service those RPCs
4473 Severity : enhancement
4475 Description: Ability to exchange lustre version between client and servers and
4476 issue warnings at client side if client is too old. Also for
4477 liblustre clients there is ability to refuse connection of too old
4479 Details : New 'version' field is added to connect data structure that is
4480 filled with version info. That info is later checked by server and
4484 Frequency : rare, liblustre only.
4485 Bugzilla : 9296, 9581
4486 Description: Two simultaneous writes from liblustre at offset within same page
4487 might proceed at the same time overwriting eachother with stale
4489 Details : I/O lock withing llu_file_prwv was released too early, before data
4490 actually was hitting the wire. Extended lock-holding time until
4491 server acknowledges receiving data.
4494 Frequency : extremely rare. Never observed in practice.
4496 Description: avoid generating lustre_handle cookie of 0.
4497 Details : class_handle_hash() generates handle cookies by incrementing
4498 global counter, and can hit 0 occasionaly (this is unlikely, but
4499 not impossible, because initial value of cookie counter is
4500 selected randonly). Value of 0 is used as a sentinel meaning
4501 "unassigned handle" --- avoid it. Also coalesce two critical
4502 sections in this function into one.
4504 Severity : enhancement
4506 Description: allow liblustre clients to delegate truncate locking to OST
4507 Details : To avoid overhead of locking, liblustre client instructs OST to
4508 take extent lock in ost_punch() on client's behalf. New connection
4509 flag is added to handle backward compatibility.
4511 Severity : enhancement
4512 Bugzilla : 4928, 7341, 9758
4513 Description: allow number of OST service threads to be specified
4514 Details : a module parameter allows the number of OST service threads
4515 to be specified via "options ost ost_num_threads={N}" in the
4516 OSS's /etc/modules.conf or /etc/modprobe.conf.
4520 Bugzilla : 6146, 9635, 9895
4521 Description: servers crash with bad pointer in target_handle_connect()
4522 Details : In rare cases when a client is reconnecting it was possible that
4523 the connection request was the last reference for that export.
4524 We would temporarily drop the export reference and get a new
4525 one, but this may have been the last reference and the export
4526 was just destroyed. Get new reference before dropping old one.
4528 Severity : enhancement
4529 Frequency : if client is started with failover MDS
4531 Description: Allow multiple MDS hostnames in the mount command
4532 Details : Try to read the configuration from all specified MDS
4533 hostnames during a client mount in case the "primary"
4536 Severity : enhancement
4538 Description: Stop sending data to evicted clients as soon as possible.
4539 Details : Check if the client we are about to send or are sending data to
4540 was evicted already. (Check is done every second of waiting,
4541 for which l_wait_event interface was extended to allow checking
4542 of exit condition at specified intervals).
4545 Frequency : rare, normally only when NFS exporting is done from client
4547 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4549 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4552 Frequency : always, if srand() is called before liblustre initialization
4554 Description: Liblustre uses system PRNG disturbing its usage by user application
4555 Details : Introduce internal to lustre fast and high-quality PRNG for
4556 lustre usage and make liblustre and some other places in generic
4557 lustre code to use it.
4559 Severity : enhancement
4560 Bugzilla : 9477, 9557, 9870
4561 Description: Verify that the MDS configuration logs are updated when xml is
4562 Details : Check if the .xml configuration logs are newer than the config
4563 logs stored on the MDS and report an error if this is the case.
4564 Request --write-conf, or allow starting with --old_conf.
4566 Severity : enhancement
4568 Description: Handle symlinks in the path when checking if Lustre is mounted.
4569 Details : Resolve intermediate symlinks when checking if a client has
4570 mounted a filesystem to avoid duplicate client mounts.
4575 Description: lconf can hit an error exception but still return success.
4576 Details : The lconf command catches the Command error exception at the top
4577 level script context and will exit with the associated exit
4578 status, but doesn't ensure that this exit status is non-zero.
4583 Description: failure of ptlrpc thread startup can cause oops
4584 Details : Starting a ptlrpc service thread can fail if there are a large
4585 number of threads or the server memory is very fragmented.
4586 Handle this without oopsing.
4589 Frequency : always, only if liblustre and non-default acceptor port was used
4591 Description: liblustre cannot connect to servers with non-default acceptor port
4592 Details : tcpnal_set_default_params() was not called and was therefore
4593 ignoring the environment varaible TCPNAL_PORT, as well as other
4594 TCPNAL_ environment variables
4599 Description: two objects could be created on the same OST for a single file
4600 Details : If an OST is down, in some cases it was possible to create two
4601 objects on a single OST for a single file. No problems other
4602 than potential performance impact and spurious error messages.
4606 Bugzilla : 5681, 9562
4607 Description: Client may oops in ll_unhash_aliases
4608 Details : Client dcache may become inconsistent in race condition.
4609 In some cases "getcwd" can fail if the current directory is
4615 Description: Inode refcounting problems in NFS export code
4616 Details : link_raw functions used to call d_instantiate without obtaining
4617 extra inode reference first.
4621 Bugzilla : 9942, 9903
4622 Description: Referencing freed requests leading to crash, memleaks with NFS.
4623 Details : We used to require that call to ll_revalidate_it was always
4624 followed by ll_lookup_it. Also with revalidate_special() it is
4625 possible to call ll_revalidate_it() twice for the same dentry
4626 even if first occurence returned success. This fix changes semantic
4627 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4628 reference on a request referred from the intent.
4629 ll_intent_release() then releases such a request.
4632 Frequency : rare, normally benchmark loads only
4634 Description: unlinked inodes were kept in memory on the client
4635 Details : If a client is repeatedly creating and unlinking files it
4636 can accumulate a lot of stale inodes in the inode slab cache.
4637 If there is no other client load running this can cause the
4638 client node to run out of memory. Instead flush old inodes
4639 from client cache that have the same inode number as a new inode.
4642 Frequency : SLES9 2.6.5 kernel and long filenames only
4643 Bugzilla : 9969, 10379
4644 Description: utime reports stale NFS file handle
4645 Details : SLES9 uses out-of-dentry names in some cases, which confused
4646 the lustre dentry revalidation. Change it to always use the
4650 Frequency : rare, unless heavy write-truncate concurrency is continuous
4651 Bugzilla : 4180, 6984, 7171, 9963, 9331
4652 Description: OST becomes very slow and/or deadlocked during object unlink
4653 Details : filter_destroy() was holding onto the parent directory lock
4654 while truncating+unlinking objects. For very large objects this
4655 may block other threads for a long time and slow overall OST
4656 responsiveness. It may also be possible to get a lock ordering
4657 deadlock in this case, or run out of journal credits because of
4658 the combined truncate+unlink. Solution is to do object truncate
4659 first in one transaction without parent lock, and then do the
4660 final unlink in a new transaction with the parent lock. This
4661 reduces the lock hold time dramatically.
4664 Frequency : rare, 2.4 kernels only
4666 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4667 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4668 before unmounting their filesystem in order to avoid deadlock.
4669 The kernel_locked() function in 2.4 kernels only checks whether
4670 the kernel lock is held, not whether it is this process that is
4671 holding it as 2.6 kernels do.
4676 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4677 Details : The client ptlrpc code may be trying to reconnect to a down
4678 server before a previous connection attempt has timed out.
4679 Increase the reconnect interval to be longer than the connection
4680 timeout interval to avoid sending duplicate connections to
4684 Frequency : echo_client brw_test command
4686 Description: fix echo_client to work with OST preallocated code
4687 Details : OST preallocation code (5137) didn't take echo_client IO path
4688 into account: echo_client calls filter methods outside of any
4689 OST thread and, hence, there is no per-thread preallocated
4690 pages and buffers to use. Solution: hijack pga pages for IO. As
4691 a byproduct, this avoids unnecessary data copying.
4695 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4696 Description: Client can oops in mdc_commit_close() after open replay
4697 Details : It was possible for the MDS to return an open request with no
4698 transaction number in mds_finish_transno() if the client was
4699 evicted, but without actually returning an error. Clients
4700 would later try to replay that open and may trip an assertion
4701 Simplify the client close codepath, and always return an error
4702 from the MDS in case the open is not successful.
4705 Frequency : rare, 2.6 OSTs only
4707 Description: OST may deadlock under high load on fragmented files
4708 Details : If there was a heavy load and highly-fragmented OST filesystems
4709 it was possible to have all the OST threads deadlock waiting on
4710 allocation of biovecs, because the biovecs were not released
4711 until the entire RPC IO was completed. Instead, release biovecs
4712 as soon as they are complete to ensure forward IO progress.
4714 Severity : enhancement
4716 Description: Support for specifying external journal device at mount
4717 Details : If an OST or MDS device is formatted with an external journal
4718 device, this device major/minor is stored in the ext3 superblock
4719 and may not be valid for failover. Allow detecting and
4720 specifying the external journal at mount time.
4725 Description: Mounting an MDS with pending unlinked files may cause oops
4726 Details : target_finish_recovery() calls mds_postrecov() which returned
4727 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4728 considers this an error and immediately begins cleaning up the
4729 lov, just after starting the mds_lov process
4731 Severity : enhancement
4733 Description: Implement 'lfs df' to report actual free space on per-OST basis
4734 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4735 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4736 usage of objects; '-h' to report in human readable format.
4738 ------------------------------------------------------------------------------
4740 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4747 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4748 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4749 at startup during orphan unlinking. Instead, skip buddy update
4750 before setup, as it will be regenerated after recovery is complete.
4755 Description: If an OST is inactive, its locks might reference stale inodes.
4756 Details : lov_change_cbdata() must iterate over all namespaces, even if
4757 they are inactive to clear inode references from the lock.
4759 Severity : enhancement
4760 Frequency : occasional, if non-standard max_dirty_mb used
4762 Description: Client will block write RPCs if not enough grant
4763 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4764 then the client will block writes while waiting for another RPC
4765 to complete instead of consuming its dirty limit. With change
4766 we get improved performance when max_dirty_mb is small.
4768 Severity : enhancement
4769 Bugzilla : 3389, 6253
4770 Description: Add support for supplementary groups on the MDS.
4771 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4772 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4773 lookups for user supplementary groups into a cache.
4777 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4778 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4779 for liblustre O_EXCL is enforced by MDS.
4783 Description: readdir never returns NULL in liblustre.
4784 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4785 dentry in struct dirent.
4789 Frequency : liblustre only, depends on application IO pattern
4790 Description: liblustre clients evicted if not contacting servers
4791 Details : Don't put liblustre clients into the ping_evictor list, so
4792 they will not be evicted by the pinger ever.
4794 Severity : enhancement
4796 Description: Add ability to evict clients by NID from MDS.
4797 Details : By echoing "nid:$NID" string into
4798 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4799 $NID would be instantly evicted from this MDS and from all active
4800 OSTs connected to it.
4804 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4805 Details : lookup_it_finish() used to query file size from OSTs that was not
4810 Description: service threads change working directory to that of init
4811 Details : Starting lustre service threads may pin the working directory
4812 of the parent thread, making that filesystem busy. Threads
4813 now change to the working directory of init to avoid this.
4817 Frequency : during shutdown only
4818 Description: shutdown with a failed MDS or OST can cause unmount to hang
4819 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4824 Frequency : 2.6 only
4825 Description: chmod/chown may include an extra supplementary group
4826 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4827 supplementary group and if none is specified this is used.
4830 Bugzilla : 5479 (6816)
4832 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4833 Details : If lookup is in progress on a file that is unlinked we might try
4834 to revalidate the inode and fail in revalidate after lookup is
4835 complete and ll_file_open() enqueues the open again but
4836 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4839 Frequency : always, if lconf --abort_recovery used
4841 Description: lconf --abort_recovery fails with 'Operation not supported'
4842 Details : lconf was attempting to abort recovery on the MDT device and not
4845 ------------------------------------------------------------------------------
4847 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4852 Frequency : rare (only unsupported configurations with a node running as an
4854 Bugzilla : 6514, 5137
4855 Description: Mounting a Lustre file system on a node running as an OST could
4857 Details : OSTs now preallocates memory needed to write out data at
4858 startup, instead of when needed, to avoid having to
4859 allocate memory in possibly low memory situations.
4860 Specifically, if the file system is mounted on on OST,
4861 memory pressure could force it to try to write out data,
4862 which it needed to allocate memory to do. Due to the low
4863 memory, it would be unable to do so and the node would
4864 become unresponsive.
4866 Severity : enhancement
4868 Description: Addition of lconf --service command line option
4869 Details : lconf now accepts a '--service <arg>' option, which is
4870 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4872 Severity : enhancement
4874 Description: Failover mode is now the default for OSTs.
4875 Details : By default, OSTs will now run in failover mode. To return to
4876 the old behaviour, add '--failout' to the lmc line for OSTs.
4878 Severity : enhancement
4880 Description: Health checks are now provided for MDS and OSTs
4881 Details : Additional detailed health check information on MSD and OSTs
4882 is now provided through the procfs health_check value.
4885 Frequency : occasional, depends on IO load
4887 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4888 after numerous create/delete cycles
4889 Details : The ext3 inode allocation policy would not allocate new inodes
4890 very well on the OSTs because there are no new directories
4891 being created. Instead we look for groups with free space if
4892 the parent directories are nearly full.
4897 Description: Network or server problems during mount may cause partially
4898 mounted clients instead of returning an error.
4899 Details : The config llog parsing code may overwrite the error return
4900 code during mount error handling, returning success instead
4906 Description: MDS can fail to allocate large reply buffers
4907 Details : After long uptimes the MDS can fail to allocate large reply
4908 buffers (e.g. zconf client mount config records) due to memory
4909 fragmentation or consumption by the buffer cache. Preallocate
4910 some large reply buffers so that these replies can be sent even
4911 under memory pressure.
4915 Frequency : rare (liblustre)
4916 Description: fsx running with liblustre complained that using truncate() to
4917 extend the file doesn't work. This patch corrects that issue.
4918 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4919 ATTR_SIZE and lsm use in llu_setattr_raw.
4923 Frequency : rare, only 2.6 kernels
4924 Description: Unusual file access patterns on the MDS may result in inode
4925 data being lost in very rare circumstances.
4926 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4927 code in the 2.6 kernel caused the inode and/or EA data not to
4928 be read from disk, causing single-file corruption.
4932 Frequency : rare, only 2.6 filesystems using extents
4933 Description: Heavy concurrent write and delete load may cause data corruption.
4934 Details : It was possible under high-load situations to have an extent
4935 metadata block in the block device cache from a just-unlinked
4936 file overwrite a newly-allocated data block. We now unmap any
4937 metadata buffers that alias just-allocated data blocks.
4941 Frequency : filesystems with default stripe_count larger than 77
4942 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4943 Details : lconf specifies an inode size of 4096 bytes when the default
4944 stripe_count is larger than 77. This conflicts with the default
4945 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4946 this case to avoid pinning too much memory for large EAs.
4948 ------------------------------------------------------------------------------
4950 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4955 Frequency : rare (extremely heavy IO load with hundreds of clients)
4957 Description: Client is evicted, gets IO error writing to file
4958 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4959 caused clients to be evicted for AST timeouts. The fixes in
4960 bug 5192 mean we no longer need to have such short AST timeouts
4961 so ldlm_timeout has been increased.
4964 Frequency : occasional during --force or --failover shutdown under load
4965 Bugzilla : 5949, 4834
4966 Description: Server oops/LBUG if stopped with --force or --failover under load
4967 Details : a collection of import/export refcount and cleanup ordering
4968 issues fixed for safer force cleanup
4971 Frequency : only filesystems larger than 120 OSTs
4972 Bugzilla : 5990, 6223
4973 Description: lfs getstripe would oops on a very large filesystem
4974 Details : lov_getconfig used kfree on vmalloc'd memory
4977 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4978 Bugzilla : 6242, 6243
4979 Description: reading from files that had been truncated to a non-zero size
4980 but never opened returned no data
4981 Details : ll_file_read() reads zeros from no-object files to EOF
4986 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4987 Details : There's little harm in aborting MDS/OSS recovery and letting it
4988 try again, so I removed the LASSERT and return an error instead.
4990 Severity : enhancement
4992 Description: New debugging infrastructure for tracking down data corruption
4993 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4994 (b) cover more of the client-side code path, and (c) try to narrow
4995 down where problems occurred
4999 Bugzilla : 3819, 4364, 4397, 6313
5000 Description: Racing close and eviction MDS could cause assertion in mds_close
5001 Details : It was possible to get multiple mfd references during close and
5002 client eviction, leading to one thread referencing a freed mfd.
5004 Severity: : enhancement
5005 Bugzilla : 3262, 6359
5006 Description: Attempts to reconnect to servers are now more aggressive.
5007 Details : This builds on the enhanced upcall-less recovery that was added
5008 in 1.4.2. When trying to reconnect to servers, clients will
5009 now try each server in the failover group every 10 seconds. By
5010 default, clients would previously try one server every 25 seconds.
5015 Description: After recovery, certain operations trigger a failed
5016 assertion on a client.
5017 Details : Failing over an mds, using lconf -d --failover, while a
5018 client was doing a readdir() call would cause the client to
5019 LBUG after recovery completed and the readdir() was resent.
5021 Severity : enhancement
5023 Description: Default groups are now added by lconf
5024 Details : You can now run lconf --group <servicename> without having to
5025 manually add groups with lmc.
5028 Frequency : occasional
5030 Description: Nodes with an elan id of 0 trigger a failed assertion
5033 Frequency : always when accessing e.g. tty/console device nodes
5035 Description: tty and some other devices nodes cannot be used on lustre
5036 Details : file's private_data field is used by device data and lustre
5037 values in there got lost. New field was added to struct file to
5038 store fs-specific private data.
5041 Frequency : when exporting Lustre via NFS
5043 Description: NFSD failed occasionally when looking up a path component
5044 Details : NFSD is looking up ".." which was broken in ext3 directories
5045 that had grown large enough to become hashed.
5048 Frequency : Clusters with multiple interfaces not on the same subnet
5050 Description: Nodes will repeatedly try to reconnect to an interface which it
5051 cannot reach and report an error to the log.
5052 Details : Extra peer list entries will be created by lconf with some peers
5053 unreachable. lconf now validates the peer before adding it.
5056 Frequency : Only if a default stripe is set on the filesystem root.
5058 Description: Setting a default stripe on the filesystem root prevented the
5059 filesystem from being remounted.
5060 Details : The client was sending extra request flags in the root getattr
5061 request and did not allocate a reply buffer for the dir EA.
5064 Frequency : occasional, higher if lots of files are accessed by one client
5065 Bugzilla : 6159, 6097
5066 Description: Client trips assertion regarding lsm mismatch/magic
5067 Details : While revalidating inodes the VFS looks up inodes with ifind()
5068 and in rare cases can find an inode that is being freed.
5069 The ll_test_inode() code will free the lsm during ifind()
5070 when it finds an existing inode and then the VFS later attaches
5071 this free lsm to a new inode.
5075 Bugzilla : 6422, 7030
5076 Description: MDS deadlock between mkdir and client eviction
5077 Details : Creating a new file via mkdir or mknod (starting a transaction
5078 and getting the ns lock) can deadlock with client eviction
5079 (gets ns lock and trying to finish a synchronous transaction).
5082 Frequency : occasional
5083 Description: While starting a server, the fsfilt_ext3 module could not be
5085 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
5086 kernels. Previously, lconf would still use the ext3 name
5087 when trying to load modules. Now, it will correctly use
5088 ext3 on 2.4 and ldiskfs on 2.6.
5090 Severity : enhancement
5091 Description: The default stripe count has been changed to 1
5092 Details : The interpretation of the default stripe count (0, to lfs
5093 or lmc) has been changed to mean striping across a single
5094 OST, rather than all available. For general usage we have
5095 found a stripe count of 1 or 2 works best.
5097 Severity : enhancement
5098 Description: Add support for compiling against Cray portals.
5099 Details : Conditional compiling for some areas that are different
5103 Frequency : occasional
5104 Bugzilla : 6409, 6834
5105 Description: Creating files with an explicit stripe count may lead to
5106 a failed assertion on the MDS
5107 Details : If some OSTs are full or unavailable, creating files may
5108 trigger a failed assertion on the MDS. Now, Lustre will
5109 try to use other servers or return an error to the
5113 Frequency : occasional
5115 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
5116 caused lock timeout during readahead (since 1.4.2).
5117 Details : Processes doing readahead might match a lock that hasn't been
5118 granted yet if there are overlapping and conflicting lock
5119 requests. The readahead process waits on ungranted lock
5120 (original lock is CBPENDING), while OST waits for that process
5121 to cancel CBPENDING read lock and eventually evicts client.
5123 Severity : enhancement
5125 Description: Initial enabling of flock support for clients
5126 Details : Implements fcntl advisory locking and file status functions.
5127 This feature is provided as an optional mount flag (default
5128 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
5129 locking are implemented yet, and those that are are not guaranteed
5130 to be completely correct in production environments.
5131 mount -t lustre -o [flock|noflock] ...
5134 Frequency : occasional
5136 Description: OSTs running 2.4 kernels but with extents enabled might trip an
5137 assertion in the ext3 JBD (journaling) layer.
5138 Details : The b_committed_data struct is protected by the big kernel lock
5139 in 2.4 kernels, serializing journal_commit_transaction() and
5140 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
5141 access to this struct. In 2.6 kernels there is finer grained
5142 locking to improve SMP performance of the JBD layer.
5146 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
5151 Description: merge and cleanup kernel patches.
5154 -----------------------------------------------------------------------------
5156 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
5158 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
5159 versions of Lustre. Please update all servers and clients to
5160 version 1.4.2 or later at the same time. You must also run
5161 "lconf --write-conf {config}.xml" on the MDS while it is stopped
5162 to update the configuration logs.
5164 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
5165 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
5166 - swab llog records of type '0' so we get proper header size/idx (5861)
5167 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
5168 - fix rename of one directory over another leaking an inode (5953)
5169 - avoid SetPageDirty on 2.6 (5981)
5170 - don't re-add just-being-destroyed locks to the waiting list (5653)
5171 - when creating new directories, inherit the parent's custom
5172 striping settings if present parent (3048)
5173 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
5174 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
5175 - don't hold client locks on temporary worklist from l_lru (5666)
5176 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
5177 - automatically evict dead clients (5921)
5178 - Update file size properly in create+truncate+fstat case (6196)
5179 - Do not unhash mountpoint dentries, do not allow removal of
5181 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
5182 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
5183 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
5184 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
5185 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
5186 - Correctly handle reads of files with no objects (6243)
5187 - lctl recover will also mark a device active if deactivate used (5933)
5189 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
5190 - allow --write-conf on an MDS with different nettype than client (5619)
5191 - don't write config llogs to MDS for mounts not from that MDS (5617)
5192 - lconf should create multiple TCP connections from a client (5201)
5193 - init scripts are now turned off by default; run chkconfig --on
5194 lustre and chkconfig --on lustrefs to use them
5195 - upcalls are no longer needed for clients to recover to failover
5197 - add --abort-recovery option to lconf to abort recovery on device
5199 - add support for an arbitrary number of OSTs (3026)
5200 - Quota support protocol changes.
5201 - forward compatibility changes to wire structs (6007)
5202 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
5203 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
5204 - improved "lustre status" script
5205 - initialize blocksize for non-regular files (6062)
5206 - added --disable-server and --disable-client configure options (5782)
5207 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
5208 - Vanilla 2.4.29 support
5209 - increase maximum number of obd devices to 520 (6242)
5210 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
5211 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
5212 SLES 9 2.6.5 kernels
5213 - limit stripes per file to 160 (the maximum EA size) (6093)
5215 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5218 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
5219 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
5220 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
5221 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
5222 - small ext3 extents cleanups and fixes (5733)
5223 - improved mballoc code, several small races and bugs fixed (5733, 5638)
5224 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
5225 - avoid needless client->OST connect, fix handle mismatch (5317)
5226 - fix DLM error path that led to out-of-sync client, long delays (5779)
5227 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
5228 - fix several locking issues related to i_size (5492,5624,5654,5672)
5229 - don't move pending lock onto export if it is already evicted (5683)
5230 - fix kernel oops when creating .foo in unlinked directory (5548)
5231 - fix deadlock in obdfilter statistics vs. object create (5811)
5232 - use time_{before,after} to avoid timer jiffies wrap (5882)
5233 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
5234 - Do not leak request if server was not able to process it (5154)
5235 - If mds_open unable to find parent dir, make that negative lookup(5154)
5236 - don't create new directories with extent-mapping (5909, 5936)
5238 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
5239 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
5240 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
5241 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
5242 - support for RHEL4 kernel on i686 (5773)
5243 - provide error messages when incompatible logs are encountered (5898)
5245 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
5246 * version 1.4.0.10 (1.4.1 release candidate 1)
5248 - don't keep a lock reference when lock is not granted (4238)
5249 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5250 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
5251 - drop import inflight refcount on signal_completed_replay error (5255)
5252 - unlock page after async write error during send (3677)
5253 - handle missing objects in filter_preprw_read properly (5265)
5254 - no transno return for symlink open, don't save no-trasno open (3440)
5255 - don't try to complete elan receive that already failed (4012)
5256 - free RPC server reply state on error (5406)
5257 - clean up thread from ptlrpc_start_thread() on error (5160)
5258 - readahead could read extra page into cache that wasn't ejected (5388)
5259 - prevent races in class_attach/setup/cleanup/detach (5260)
5260 - don't dereference de->d_inode after l_dput of de (5458)
5261 - use "int" for stripe value returned from lock_to_stripe (5544)
5262 - mballoc allocation and error-checking fixes in 2.6 (5504)
5263 - block device patches to fix I/O request sizes in 2.6 (5482)
5264 - look up hostnames for IB nals (5602)
5265 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
5266 - don't start multiple acceptors for the same port (5277)
5267 - fix incorrect LASSERT in mds_getattr_name (5635)
5268 - export a proc file for general "ping" checking (5628)
5269 - fix "lfs check" to not block when the MDS is down (5628)
5271 - service request history (4965)
5272 - put {ll,lov,osc}_async_page structs in a single slab (4699)
5273 - create an "evict_client" /proc entry on OSTs, like the MDS has
5274 - fix mount usage message, return errors per mount(8) (5168)
5275 - change grep [] to grep "[]" in tests so they work in more UMLs
5276 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
5277 - remove ancient LOV_MAGIC_V0 EA support (5047)
5278 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
5279 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
5280 - allow readahead allocations to fail when low on memory (5383)
5281 - mmap locking landed again, after considerable improvement (2828)
5282 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
5284 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
5287 - send OST transaction number in read/write reply to free req (4966)
5288 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5289 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5290 - the watchdog thread now runs as interruptible (5246)
5291 - flock/lockf fixes (but it's still disabled, pending 5135)
5292 - don't use EXT3 constants in llite code (5094)
5293 - memory shortage at startup could cause assertion (5176)
5295 - reorganization of lov code
5296 - single portals codebase
5298 - add extents/mballoc support (5025)
5299 - direct I/O reads in the obdfilter (4048)
5300 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5302 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5305 - send OST transaction number in read/write reply to free req (4966)
5306 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5307 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5308 - the watchdog thread now runs as interruptible (5246)
5309 - handle missing objects in filter_preprw_read properly (5265)
5310 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5311 - drop import inflight refcount on signal_completed_replay error (5255)
5312 - unlock page after async write error during send (3677)
5313 - return original error code on reconstructed replies (3761)
5314 - no transno return for symlink open, don't save no-trasno open (3440)
5316 - add pid to ldlm debugging output (4922)
5317 - bump the watchdog timeouts -- we can't handle 30sec yet
5318 - extra debugging for orphan dentry/inode bug (5259)
5320 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5323 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5324 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5325 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5326 - fix NULL dereference in /proc/sys/portals/routes (4827)
5327 - allow failed mdc_close() operations to be interrupted (4561)
5328 - stop precreate on OST before MDS would time out on it (4778)
5329 - don't send partial-page writes before EOF from client (4410)
5330 - discard client grant for sub-page writes on large-page clients (4520)
5331 - don't free dentries not owned by NFS code, check generation (4806)
5332 - fix lsm leak if mds_create_objects() fails (4801)
5333 - limit debug_daemon file size, always print CERROR messages (4789)
5334 - use transno after validating reply (3892)
5335 - process timed out requests if import state changes (3754)
5336 - update mtime on OST during writes, return in glimpse (4829)
5337 - add mkfsoptions to LDAP (4679)
5338 - use ->max_readahead method instead of zapping global ra (5039)
5339 - don't interrupt __l_wait_event() during strace
5341 - add software watchdogs to catch hung threads quickly (4941)
5342 - make lustrefs init script start after nfs is mounted
5343 - fix CWARN/ERROR duplication (4930)
5344 - return async write errors to application if possible (2248)
5345 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5346 - print NAL number in %x format (4645)
5347 - update barely-supported suse-2.4.21-171 series (4842)
5348 - support for sles 9 %post scripts
5349 - support for building 2.6 kernel-source packages
5350 - support for sles km_* packages
5352 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5355 - ignore -ENOENT errors in osc_destroy (3639)
5356 - notify osc create thread that OSC is being cleaned up (4600)
5357 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5358 - reconstruct ost_handle() like mds_handle() (4657)
5359 - create a new thread to do import eviction to avoid deadlock (3969)
5360 - let lconf resolve symlinked-to devices (4629)
5361 - don't unlink "objects" from directory with default EA (4554)
5362 - hold socknal file ref over connect in case target is down (4394)
5363 - allow more than 32000 subdirectories in a single directory (3244)
5364 - fix blocks count for O_DIRECT writes (3751)
5365 - OST returns ENOSPC from object create when no space left (4539)
5366 - don't send truncate RPC if file size isn't changing (4410)
5367 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5368 - bind to privileged port in socknal and tcpnal (3689)
5370 - rate limit CERROR/CWARN console message to avoid overload (4519)
5371 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5372 - basic mmap support (3918)
5373 - kernel patch series update from b1_4 (4711)
5375 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5378 - avoid crash during MDS cleanup with OST shut down (2775)
5379 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5380 - don't use bad inodes on the MDS (3744)
5381 - dynamic object preallocation to improve recovery speed (4236)
5382 - don't hold spinlock over lock dumping or change debug flags (4401)
5383 - don't zero obd_dev when it is force cleaned (3651)
5384 - print grants to console if they go negative (4431)
5385 - "lctl deactivate" will stop automatic recovery attempts (3406)
5386 - look for existing locks in ldlm_handle_enqueue() (3764)
5387 - don't resolve lock handle twice in recovery avoiding race (4401)
5388 - revalidate should check working dir is a directory (4134)
5390 - don't always mark "slow" obdfilter messages as errors (4418)
5392 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5395 - don't close LustreDB during write_conf until it is done (3860)
5396 - fix typo in lconf for_each_profile (3821)
5397 - allow dumping logs from multiple threads at one time (3820)
5398 - don't allow multiple threads in OSC recovery (3812)
5399 - fix debug_size parameters (3864)
5400 - fix mds_postrecov to initialize import for llog ctxt (3121)
5401 - replace config semaphore with spinlock (3306)
5402 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5403 - don't allow enqueue to complete on a destroyed export (3822)
5404 - down write_lock before checking llog header bitmap (3825)
5405 - recover from lock replay timeout (3764)
5406 - up llog sem before sending rpc (3652)
5407 - reduce ns lock hold times when setting kms (3267)
5408 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5409 - fix NULL deref and obd_dev leak on setup error (3312)
5410 - replace some LBUG about llog ops with error handling (3841)
5411 - don't match INVALID dentries from d_lookup and spin (3784)
5412 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5413 - fix invalid assertion in ptlrpc_set_wait (3880)
5415 - add libwrap support for the TCP acceptor (3996)
5416 - add /proc/sys/portals/routes for non-root route listing (3994)
5417 - allow setting MDS UUID in .xml (2580)
5418 - print the stack of a process that LBUGs (4228)
5420 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5423 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5424 - make sure to unset replay flag from failed open requests (3440)
5425 - if default stripe count is 0, use OST count for inode size (3636)
5426 - update parent mtime/ctime on client for create/unlink (2611)
5427 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5428 - free recovery state on server during a forced cleanup (3571)
5429 - unregister_reply for resent reqs (3063)
5430 - loop back devices mounting and status check on 2.6 (3563)
5431 - fix resource-creation race that can provoke i_size == 0 (3513)
5432 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5433 - more debugging for page-accounting assertion (3746)
5434 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5435 - avoid deadlock after precreation failure (3758)
5436 - fix race and lock order deadlock in orphan handling (3450, 3750)
5437 - add validity checks when grabbing inodes from l_ast_data (3599)
5439 - add /proc/.../recovery_status to obdfilter (3428)
5440 - lightweight CDEBUG infrastructure, debug daemon (3668)
5441 - change default OSC RPC parameters to be better on small clusters
5442 - turn off OST read cache for files smaller than 32MB
5443 - install man pages and include them in rpms (3100)
5444 - add new init script for (un)mounting lustre filesystems (2593)
5445 - run chkconfig in %post for init scripts (3701)
5446 - drop scimac NAL (unmaintained)
5448 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5451 - clean kiobufs before and after use (3485)
5452 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5453 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5454 - change init script to fail more clearly if not run as root (1528)
5455 - allow clients to reconnect during replay (1742)
5456 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5457 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5458 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5459 - fix cond_resched (3554)
5460 - only evict unfinished clients after recovery (3515)
5461 - allow bulk resend, prevent data loss (3570)
5462 - dynamic ptlrpc request buffer allocation (2102)
5463 - don't allow unlinking open directory if it isn't empty (2904)
5464 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5465 - remove extraneous obd dereference causing LASSERT failure (3334)
5466 - don't use get_cycles() when creating temp. files on the mds (3156)
5467 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5468 - handle EEXIST for set-stripe, set proper directory name (3336)
5470 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5471 - fix ksocknal_fmb_callback() error messages (2918)
5473 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5476 - don't copy lvb into (possibly NULL) reply on error (2983)
5477 - don't deref dentry after dput, don't free lvb on error (2922)
5478 - use the kms to determine writeback rpc length (2947)
5479 - increment oti_logcookies when osc is inactive (2948)
5480 - update client's i_blocks count via lvb messages (2543)
5481 - handle intent open/close of special files properly (1557)
5482 - mount MDS with errors=remount-ro, like obdfilter (2009)
5483 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5484 - don't use cancelling-locks' kms values (2947)
5485 - use highest lock extent for kms, not last one (2925)
5486 - don't dereference ERR_PTR() dentry in error handling path (3107)
5487 - fix thread race in portals_debug_dumplog() (3122)
5488 - create lprocfs device entries at setup instead of at attach (1519)
5489 - common AST error handler, don't evict client on completion race (3145)
5490 - zero nameidata in detach_mnt in 2.6 (3118)
5491 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5492 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5493 - initialize RPC timeout timer earlier for 2.6 (3219)
5494 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5495 - print nal/nid for unknown nid (3258)
5496 - additional checks for oscc recovery before doing precreate (3284)
5497 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5498 - don't crash in mdc_close for bad permissions on open (3285)
5499 - zero i_rdev for non-device files (3147)
5500 - clear page->private before handing to FS, better assertion (3119)
5501 - tune the read pipeline (3236)
5502 - fix incorrect decref of invalidated dentry (2350)
5503 - provide read-ahead stats and refine rpc in flight stats (3328)
5504 - don't hold journal transaction open across create RPC (3313)
5505 - update atime on MDS at close time (3265)
5506 - close LDAP connection when recovering to avoid server load (3315)
5507 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5508 - don't leak open file on MDS after open resend (3325)
5509 - serialize filter_precreate and filter_destroy_precreated (3329)
5510 - loop device shouldn't call sync_dev() for nul device (3092)
5511 - clear page cache after eviction (2766)
5512 - resynchronize MDS->OST in background (2824)
5513 - refuse to mount the same filesystem twice on same mountpoint (3394)
5514 - allow llmount to create routes for mounting behind routers (3320)
5515 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5516 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5517 - fix rare problem with rename on htree directories (3417)
5519 - allow default OST striping configuration per directory (1414)
5520 - fix compilation for qswnal for 2.6 kernels (3125)
5521 - increase maximum number of MDS request buffers for large systems
5522 - change liblustreapi to be useful for external progs like lfsck (3098)
5523 - increase local configuration timeout for slow disks (3353)
5524 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5526 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5529 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5530 - don't overwrite extent policy data in reply if lock was blocked (2901)
5531 - drop filter export grants atomically with removal from device (2663)
5532 - del obd_self_export from work_list in class_disconnect_exports (2908)
5533 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5534 - swab reply message in mdc_close, other PPC fixes (2464)
5535 - fix destroying of named logs (2325)
5536 - overwrite old logs when running lconf --write_conf (2264)
5537 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5538 - fix race in target_handle_connect (2898)
5539 - mds_reint_create() should take same inode create lock (2926)
5540 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5541 - don't close files for self_export to avoid uninitialized obd (2936)
5542 - allow MDS with the same name as client node (2939)
5543 - hold dentry reference for closed log files for unlink (2325)
5544 - reserve space for all logs during transactions (2059)
5545 - don't evict page beyond end of stripe extent (2925)
5546 - don't oops on a deleted current working directory (2399)
5547 - handle hard links to targets without a parent properly (2517)
5548 - don't dereference NULL lock when racing during eviction (2867)
5549 - don't grow lock extents when lots of conflicting locks (2919)
5551 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5554 - account for cache space usage on clients to avoid data loss (974)
5555 - lfsck support in lustre kernel code (2349)
5556 - reduce journal credits needed for BRW writes (2370)
5557 - orphan handling to avoid losing space on client/server crashes
5558 - ptlrpcd can be blocked, stopping ALL progress (2477)
5559 - use lock value blocks to assist in proper KMS, faster stat (1021)
5560 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5561 - recovery for initial connections (2355)
5562 - fixes for mds_cleanup_orphans (1934)
5563 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5564 - block all file creations until orphan recovery completes (1901)
5565 - client remove rq_connection from request struct (2423)
5566 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5567 - recovery timer race (2670)
5568 - mdc_close recovey bug (2532)
5569 - ptlrpc cleanup bug (2710)
5570 - mds timeout on local locks (2588)
5571 - namespace lock held during RPCs (2431)
5572 - handle interrupted sync write properly (2503)
5573 - don't try to handle a message that hasn't been replied to (2699)
5574 - client assert failure during cleanup after abort recovery (2701)
5575 - leak mdc device after failed mount (2712)
5576 - ptlrpc_check_set allows timedout requests to complete (2714)
5577 - wait for inflight reqs when ptlrpcd finishes (2710)
5578 - make sure unregistered services are removed from the srv_list
5579 - reset bulk XID's when resending them (caught by 1138 test)
5580 - unregister_bulk after timeout
5581 - fix lconf error (2694)
5582 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5583 - readahead locks pages, leaves pending causing memory pressure (2673)
5584 - increase OST request buffers to 4096 on large machines (2729)
5585 - fix up permission of existing directories in simple_mkdir (2661)
5586 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5587 - don't assign transno to errored transactions (2742)
5588 - don't delete objects on OST if given a bogus objid from MDS (2751)
5589 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5590 - if rq_no_resend, then timeout request after recovery (2432)
5591 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5592 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5593 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5594 - MDS recovery shouldn't delete objects that it has given out (2730)
5595 - if enqueue arrives after completion, don't clobber LVB (2819)
5596 - don't unlock pages twice when trigger_group_io returns error (2814)
5597 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5598 - don't write pages to disk if there was an error (1450)
5599 - don't ping imports that have recovery disabled (2676)
5600 - take buffered bytes into account when balancing socknal conn (2817)
5601 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5602 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5603 - remove little-endian swabbing of llog records (1987)
5604 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5605 - retry reposting request buffers if they fail (1191)
5606 - grow extent at grant time to avoid granting a revoked lock (2809)
5607 - lock revoke doesn't evict page if covered by a second lock (2765)
5608 - disable VM readahead to avoid reading outside lock extents (2805)
5610 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5611 - updated kernel patches for hp-2.4.20 kernel (2681)
5613 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5616 - fix truncated write corruption (2366)
5617 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5619 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5620 - ensure len(uuid) < 37 in lmc (1171)
5621 - fix ia64 OOPS in llog_test (2255)
5622 - zero end of page at obdfilter for partial page writes (2648)
5623 - don't leave stale dentries around after renames (bug 2428)
5624 - fix timeouts when evicting a client with a single lock held (2642)
5625 - set deadline for the initial HELLO message to drain (2634)
5626 - print out dotted-quad IP addresses in the socknal (2302)
5628 - additional debugging for MDS client eviction problem (2443)
5629 - fix mkfsoptions support for osts (2603, 2604)
5631 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5634 - add series for the vanilla 2.6.0 kernel
5635 - add series for the vanilla 2.4.24 kernel
5636 - add series for a cray x86/64 UL kernel drop
5637 - fix xattr patches for the vanilla 2.4.19 series
5639 - generate true UUIDs in lmc (1171)
5640 - have portals stack dumping break in UML (2466)
5641 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5642 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5643 - wake callback waiting threads on client eviction (2460)
5644 - Add --ptldebug and --subsystem to lmc (1719)
5645 - update assertion to allow safe interrupt allocation
5646 - set rq_no_resend for cancel requests (2432)
5647 - recalculate ptlrpcd timeout after resend (2494)
5648 - call vfs_rmdir when removing pending directories (2368)
5649 - fix renaming a file to itself (2429)
5650 - lmc creates a default one-stripe lov (2454)
5651 - expand procfs space to handle large clusters (2326)
5652 - increase UML stack to avoid overflow
5653 - update lconf's list of debug and subsystem masks
5654 - fix lfs find --obd (2510)
5655 - /proc tunable for disabling filter read caching (2591)
5656 - stop rpm packages from altering slapd.conf (2301)
5657 - disable nagle in the socknal under 0conf (2578)
5658 - choose mds inode size based on stripe count (2572)
5659 - fix kernel-source rpm problems (2516)
5661 - add --disable-doc to avoid pdf generation (2421)
5662 - update documentation, tests, type-os, comments
5663 - avoid format warnings on ia64
5664 - remove the TOE NAL
5665 - tiny code cleanups by removing unused fields
5667 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5670 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5671 - remove the most blatant lies from BUILDING file (2371)
5672 - change default debug level to reasonable production setting
5673 - reduce client side cache size to reduce cache flush time
5674 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5675 - make TCP zerocopy and pinger support enabled by default (2476)
5676 - sync writes completed after process exits caused crashes (2319)
5677 - maintain correct mount count on the MDS (2356)
5678 - backout 1557, because 2316 wasn't really fixed
5679 - better file I/O statistics gathering in /proc
5680 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5681 - another kernel patch to fix zero-copy TCP function export
5682 - don't take duplicate lock when processing re-sent getattr (2420)
5683 - lctl uses obd_self_export instead of creating new conn (2353)
5684 - MDS/OST recovery case which requires object creation asserted (2425)
5685 - move lfs from /usr/sbin to /usr/bin in packages
5686 - fix race between mds_client_add and mds_client_free (2417)
5687 - use kmalloc instead of slabs in portals (2430)
5688 - don't create duplicate records when a failover MDS is present (2442)
5689 - remove unnecessary mount age check (2332)
5690 - don't remove directory inodes from locks prematurely (2451)
5691 - don't break if MDS service name is the same as hostname (2103)
5692 - fix races in client write RPC generation when cache full (2482)
5694 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5697 - remove now-unused request->rq_obd (278)
5698 - if an allocation fails, print out how much memory we've used (1933)
5699 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5700 - add the "configurable stack size" patch to most series files (1256)
5701 - ability to write large log records, for 100+ OST configs (2306)
5702 - fix NULL deref when filter_prep fails (2314)
5703 - fix operator precedence error in filter_sync
5704 - dynamic allocation of socknal TX descriptors (2315)
5705 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5706 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5707 - ensure that CWARN messages in Portals always get to the syslog
5708 - __init/__exit are not for prototype decls (ldlm_init/exit)
5709 - x86-64 compile warning fixes
5710 - fix gateway LMC keyword conflict (2318)
5711 - fix MDS lock inversions in getattr/reint paths (1844)
5712 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5713 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5714 - DLM race condition prevented some lock evictions (2328)
5715 - ENOMEM detection and retry on socknal sends (2230)
5716 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5717 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5718 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5719 - better error messages when a client is rejected during recovery (1505)
5720 - avoid cancelling locks which were never granted, after failure (2330)
5721 - fix i_sem/journal inversion in mds_client_add (2333)
5722 - fix truncate/getattr lock cycle deadlock (2334)
5723 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5724 - fix two rare exit paths which could leak an l_lock() ref (2321)
5725 - fencepost error in MDS/OST orphan recovery (2226)
5726 - make log record alignment 8 bytes (1988)
5727 - lstripe now fails when requested offset > ost_count (2237)
5728 - ensure that all kernel series have a complete list.h (1607)
5729 - fix crashes in special-file operations (2316)
5730 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5732 - allow configurable automake binary, for testing new versions
5733 - small update to the lfs documentation
5735 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5737 * fix negative export reference count in fsfilt_sync (2312)
5739 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5740 * release candidate 0.9.1
5742 - orphans are moved into the PENDING directory for possible recovery
5743 - replayed opens now open by fid for orphan/rename safety (1042)
5744 - last close of an orphan inode generates a transno (683)
5745 - chdir() and mount() now pin the directory entry (1020)
5746 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5747 - discard very old requests without processing them (1502)
5748 - remove some common, well-understood CERRORs (1505)
5749 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5750 - clear "grant" flags in OST replies until OST grant code lands (1644)
5751 - fix read performance by not clobbering i_blksize on client (1598)
5752 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5753 - make LRU size a /proc tunable, clears locks when reduced (707)
5754 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5755 - server locks take references on exports now (1558)
5756 - build fixes for 2.4.20-rh trees (1663)
5757 - return an error from lov_create if all OSCs are inactive (1751)
5758 - fix import levels when a reconnect happens without a timeout (1597)
5759 - exit early from mds_open if we get a lookup error (1749)
5760 - partial page read at EOF wouldn't wait for disk before sending (1642)
5761 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5762 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5763 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5764 - write/truncate lock inversion (1639)
5765 - Don't auto-load obdclass, portals modules during cleanup (1495)
5766 - fix timestamps from jumping to "now" (1763)
5767 - extra journal assertions (1648)
5768 - add an extra multiunlink test (1771)
5769 - fix read_record/write_record API (1776)
5770 - fix leak of offset_extent, possible incorrect i_size later (1772)
5771 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5772 - Debugging for the kqswnal_get_idle_tx problems (1820)
5773 - Allow recovery to be attempted multiple times (1536)
5774 - Write out MDS last_rcvd file after it is first created (1600)
5775 - Fix tx_descriptor leak in failed transmit situations (1827)
5776 - ext3 journaling fixes for assertion failure after IO error (1871)
5777 - class_export_put() on freed export after completion AST error (1896)
5778 - Fix revalidate looping in VFS (1322)
5779 - Don't access a freed export during MDS_REINT timeout (1521)
5780 - Add open-unlink recovery support on the MDS (1673,1764)
5781 - Return an error if no MDS data was read from last_rcvd (1946)
5782 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5783 - Don't setup a disk device that doesn't match exported UUID (317)
5784 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5785 - avoid committing NULL handle in force close
5786 - local.sh is now a one-stripe LOV configuration
5787 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5788 - fix ext3 htree duplicate directory entry corruption (1516)
5789 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5790 - update to new LOV EA format (2097)
5791 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5793 2003-06-15 Phil Schwan <phil@clusterfs.com>
5796 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5797 - per-import/export recovery handling (958, 931, 959)
5798 - multiple last-rcvd slots, for serving multiple FSes (949)
5799 - connections are again shared between multiple imp/exports (963, 964)
5800 - "umount -f" would hang if any requests needed to be sent (393, 978)
5801 - avoid pinning large req buffer by copying for queued messages (989)
5802 - add "uuid" to "lctl device" command to help upcalls (991)
5803 - "open" RPCs with transnos would confuse recovery counters (1037)
5804 - do proper endian conversion of all wire messages (288, 340, 891)
5805 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5806 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5807 - fix ptlrpc_connection leak in target_handle_connect (1174)
5808 - fix import refcounting bug in OST and MDS cleanup (1134)
5809 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5810 - fix very unlikely obd_types race condition (501)
5811 - remove osc_open hack for echo_client (1187)
5812 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5813 - a failure in read_inode2 leads to deadlock (1139)
5814 - cancel ack-locks as soon as transaction is committed (1072)
5815 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5816 - make sure to commitrw after any preprw to avoid deadlock (1162)
5817 - failing to execute a file in a lustre FS would lock inode (1203)
5818 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5819 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5820 - obd_uuid2tgt would walk off the end of the list (1255)
5821 - on IA64 the osc would give portals incorrect bulk size (1258)
5822 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5823 - fix lock inversion caused by new llite matching code (1282)
5824 - limit the number of dirty pages on a client to 10MB (1286)
5825 - timed out locks were not being corrected cancelled (1289)
5826 - fix O_DIRECT above 4GB on IA-32 (1292)
5827 * major user-visible changes
5828 - fail out/fail over policy now controlled by the upcall (993)
5830 - add OBD_PING to check server availability and failure (954)
5831 - lustre messages are now sent in sending host order (288, 340, 891)
5832 - add eadatalen to MDS getattr reply (340)
5833 - OST read replies may contain second buffer, with per-page status (593)
5835 2003-03-11 Phil Schwan <phil@clusterfs.com>
5838 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5839 - failed open()s could cause deadlock; fixed (867, 869)
5840 - stop cancelling OST locks when files are closed (481)
5841 - overlapping XID spaces caused network corruption (851, 853)
5842 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5843 - setattr_raw added, to do single-RPC, server-side setattrs
5844 - lmc/lconf syntax change for OST UUIDs
5845 - fix crashy race condition between ptlrpc_free_req and osc_close
5846 - don't use request in mdc_enqueue if we hit a timeout (889)
5847 - don't set the inode i_size for regular files from the MDS (896)
5848 - handle out of order completion AST (842)
5849 - don't LBUG if a lock request times out after receiving AST (913)
5850 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5851 - if a bad lock AST arrives, send an error instead of dropping entirely
5852 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5853 - fix leak in bulk IO when only partially completed (899, 900, 926)
5854 - fix O_DIRECT for ia64 (55)
5855 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5856 - C-z after timeout could hang a process forever; fixed (977)
5858 - client-side I/O cache (678, 924, 929, 941, 970)
5860 - READPAGE and SETATTRs which don't take server-side locks get
5863 2003-02-11 Phil Schwan <phil@clusterfs.com>
5866 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5867 - Fix fsfilt_extN_readpage() to read a full page of directory
5868 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5869 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5870 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5871 - Most intent operations were removed, in favour of a new RPC mode
5872 that does a single RPC to the server and bypasses most of the VFS
5873 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5874 - Aggressively cancel local locks on DLM servers
5875 - mds_reint_unlink sends EA to the client if it's the last nlink.
5876 client uses that EA to unlink OST objects.
5877 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5878 - recursive symlinks were fixed (439)
5879 - fixed NULL deref in DEBUG_REQ
5880 - filter_update_lastobjid no longer calls sync, which annoyed extN
5881 - fixed multi-client small-writes to a single file problem (445)
5882 - fixed mtime updates during file writes (607)
5883 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5884 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5885 - fixed Linux OST size reporting problem (444, 656)
5886 - OST now updates object mtime with writes or setattr (607, 619)
5887 - client verifies file size before zeroing page past EOF (445)
5888 - OST now writes last allocated objid to disk with allocation (108)
5889 - LOV on echo now works (409)
5891 - mds_reint_unlink sends a new buffer, with the EA included. this
5892 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5893 set if a regular file was being unlinked, and it was the last link
5894 - use PtlGet from the target for bulk writes (315)
5895 - OST now updates object mtime with writes or setattr (607, 619)
5896 - LDLM now has a grant-time callback to revalidate locked items, if
5898 - Many MDS operations were reorganized to combat race conditions
5900 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5901 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5903 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5906 - Fully reactivate OST imports after reconnection (512, others)
5907 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5908 - More graceful error handling for truncating on dead OST (515)
5909 - Don't error out unless we're actually accessing dead stripes (474)
5910 - Fix garbage sizes when stripes are missing (410)
5911 - LRU counters were broken, causing constant lock purge (433, 432)
5912 - garbage on read from stripes with failed OSTs (441)
5913 - mark OSCs as active before reconnecting during recovery (438)
5914 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5915 - lfind did not preserve OST order in output (443)
5916 - symlinks cause hung clients, incorrect data (439)
5917 - stop dereferencing request after dropping refcount (457)
5918 - don't LASSERT(spin_is_locked) on non-SMP (455)
5919 - fixes for many rename() bugs
5920 - fstat didn't correctly synchronize attributes (399)
5921 - server must handle lock cancellation during blocking AST prep (487)
5922 - bulk descriptors were free()d too soon (511)
5923 - fix paths in lconf, which would load incorrect modules (451, 507)
5924 - fix confusing lconf 'host not found' error message (386)
5925 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5926 - fix race condition in mdc_blocking_ast() for inode access (526)
5927 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5928 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5929 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5930 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5931 - fix lstripe in several ways (kernel side) (527)
5932 - fix request leak in ldlm_cli_enqueue (262)
5933 - incorrect OSC was marked inactive after OST failure
5934 - call mds_fs_cleanup before unmounting filesystem (524)
5935 - fix races between taking ns_lock and ldlm_lock_change_resource
5936 - fix races updating LOV export open file list
5937 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5938 - fix recovery NULL deref in ldlm_cli_cancel_unused
5939 - fix some DLM races by using new hash table for lock handles (419)
5940 - permit the client to specify desired inodes, at replay
5941 - duplicate requests when we queue them for replay reintegration
5942 - fix last_rcvd offset calculation
5943 - sync after each recovered transaction, so we always make progress
5944 - never, not always, ERESTART requests without transnos
5945 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5947 - skip replay if the MDS says that the client is already connected
5948 - don't check for a recovery-enabled export to match lctl's UUID
5949 - don't INC_USE_COUNT for phantom exports
5950 - don't crash when cleaning up phantom exports (567)
5951 - don't double-finish or set replay data for errored mdc_open requests
5952 - abort requests when they time out, so we don't get old replies
5953 - send/receive replies for AST messages again
5954 - if the client says that it doesn't have the lock, cancel it on the
5956 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5957 mark it as destroyed, it will all work out when decref is called
5958 - fix module use counts (22, 581)
5960 - ASTs now expect a reply (server cancels lock on error reply)
5962 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5965 - fix many simultaneous client startup (392)
5966 - fix dentry->d_it clobbering
5967 - credentials weren't being shipped for readdir/getattr operations
5968 - remove invalid assertions triggered during some concurrent MD
5970 - proper Lustre versions added (336, 389)
5971 - fix memory leak for create error case (398)
5972 - fix LOV locking bug that would get cli/srv out of sync
5973 - fix echo client over LOV (409)
5974 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5975 - fix double-O_EXCL intent crash (424)
5976 - avoid sending multiple lock CANCELs (352)
5978 - MDS can do multi-client recovery (modulo bugs in new code)
5980 - many updates, edits, cleanups
5982 2002-11-18 Phil Schwan <phil@clusterfs.com>
5985 - fix null d_it dereference (346)
5986 - fix full OST/dbench hang (333)
5987 - fix permission problem with file removal (286)
5988 - fix removal of OSCs from LOV when they fail
5989 - fix NULL deref during bulk timeout (214)
5990 - fix problems related to multiple filesystems on one MDS (241)
5991 - fixed serious subtle metadata locking bugs
5992 - free locks on clients when inodes are removed due to memory
5994 - fix inode pointer in lock data (285)
5995 - partial support for multiple MDS on a single host (241)
5996 - data locks weren't cancelled at clear_inode time (290, 311)
5997 - intent locks could lead to unbounded lock growth (205)
5998 - added a maximum lock count, an LRU list, and a flusher
5999 - fix multiple rename (365)
6000 - properly abstracted the echo client
6001 - OSC locked 1 byte too many; fixed
6002 - rewrote brw callback code:
6003 - fixed recovery bugs related to LOVs (306)
6004 - fixed too-many-pages-in-one-write crash (191)
6005 - fixed (again) crash in sync_io_timeout (214)
6006 - probably fixed callback-related race (385)
6008 - Add capability to MDS protocol
6009 - LDLM cancellations and callbacks on different portals
6011 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
6014 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
6015 - defer open object destruction to close time (601981, 138)
6016 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
6017 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
6018 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
6019 - don't call obd_setattr in ll_file_release for destroyed objects
6020 * protocol change to lustre_msg: move |version| and add |flags|
6021 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
6022 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
6023 * added replay of create, unlink, link and rename operations during
6024 MDS failover; recovery should be much more robust now
6025 * remove failed OSCs from LOVs (only lov_create uses this so far)
6026 * the lustre-HOWTO was brought (more) up to date (582544)
6028 2002-10-23 Phil Schwan <phil@clusterfs.com>
6031 - in-use dentries weren't being reused properly (617851)
6032 - prevent multiple LDLM setup (599178)
6033 - fix LOV size calculations for truncate (617853)
6034 - fix client handling of MDS intent errors (POSIX)
6035 - fix permission bug in lovstripe.c test (624321)
6036 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
6037 - truncate past end of file could corrupt data
6038 - proper cleanup after timeouts, crashes, etc (592524, 550815)
6039 - a race in recovery could return ETIMEDOUT to apps (623947)
6040 - building outside the source directory was fixed
6041 * the lustre-HOWTO was brought (more) up to date (582544)
6042 * major progress was made on recovery functionality
6044 2002-10-10 Phil Schwan <phil@clusterfs.com>
6047 - recovery deadlock fix
6048 - rm -rf causes LBUG fix (617817)
6049 - file open by multiple tasks fix (618962)
6050 - directory permissions bugs (602707 and 620007)
6051 - journal_stop fixed with locking (611313)
6052 - O_APPEND failures resolved (618273, perhaps 614459)
6053 - lconf PATH fix (619770)
6054 - IA64 build fix (621450)
6055 - RPC buffer sizes scale with amount of memory
6057 2002-10-01 Phil Schwan <phil@clusterfs.com>
6060 - locks would be cancelled without throwing away data pages,
6061 resulting in inconsistent data (605627)
6062 - inode attributes were not always being refreshed (605627, 612449)
6063 - lconf now continues to cleanup after lctl reports an error
6064 - MDS now enforces user permissions (602707)
6065 - lprocfs cleanup fixed, but not yet enabled (614157)
6066 - fixed infinite server hang, should a client not respond to an AST
6067 - avoid going into recovery if user calls readlink() with a buffer
6068 that's too small (613941)
6069 - AST RPCs no longer require replies (614867) -- this may be changed
6070 - don't crash server if client sends an IOV that's too big (611336)
6071 - fixed lock conversion deadlock (611892)
6072 - fixed the following of symlinks (614622)
6073 * recovery: the server can remove locks from a client that dies, other
6074 clients can make progress
6075 * more extN patch fixes
6076 * compile-time configurable ptlrpc buffer allocations
6078 - collaborative read cache document
6079 - Lustre Lite Performance CDR document-in-progress
6081 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6084 - fix typo in patch-2.4.18
6086 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6089 - clear ptlrpc request each time in handle_incoming_request()
6090 - unlink of files now destroys the object on the OST
6092 2002-09-19 Peter Braam <braam@clusterfs.com>
6094 * add hard link support
6095 * change obdfile creation method
6096 * kernel patch changed
6098 2002-09-19 Peter Braam <braam@clusterfs.com>
6101 - stack overflow bug in extN fixed
6103 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
6105 * documentation updates
6106 - add man pages for config tools
6107 - update tests/README to describe testing with new config tools
6108 - finish metadata API descriptions
6109 * bug fixes and cleanups
6110 - statfs workaround for 16TB limit
6111 - LOV stripe allocation improved, can stripe on subset of OSTs
6112 - LOV file size/IO offset was wrong for files > 4GB in size
6113 - object EA data was being dropped, caused files to be unreadable
6114 - memory overflow with non-LOV OST caused memory corruption
6115 - fixed regression tests to work with new config tools, obdfilter
6116 - fixed bug when directory size became larger than 1 block
6117 - fixed bug (for single client case) when PWD was deleted
6118 - invalidate local directory pages when doing intent-based ops
6119 - avoid LDLM oops when lock callback contained bad data
6121 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
6123 * documentation updates
6124 * bug fixes and cleanups
6125 - configuration tools
6128 - 64-bit compile warnings
6129 - 64-bit internal statfs data
6131 * test_brw on persistent OST devices
6133 * lprocfs (disabled)
6135 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
6137 * documentation updates
6138 * bug fixes and cleanups
6139 * configuration tools
6141 2002-08-30 Peter J. Braam <braam@clusterfs.com>
6144 * many small fixes to 0_5_4
6145 * io/network handling
6146 * thinkos in MDS operations
6148 2002-08-24 Peter J. Braam <braam@clusterfs.com>
6151 * crucial basic fixes to 0.5.3
6152 * IOR, Iozone work over Elan
6155 2002-08-07 Phil Schwan <phil@clusterfs.com>
6156 * version 0_5_3, our first alpha
6157 * we use the new Portals iovs
6158 * documentation updates
6159 * bug fixes and cleanups
6160 * small changes in the DLM wire protocol
6162 2002-07-25 Peter J. Braam <braam@clusterfs.com>
6163 * version 0_5_1 with some initial stability,
6164 * locking on MD and file I/O.
6165 * documentation updates
6166 * several bug fixes since 0.5.0
6167 * small changes in wire protocol
6169 2002-07-18 Phil Schwan <phil@clusterfs.com>
6171 * delivered as Lustre Light Alpha
6172 * fixed a crash after handling invalid MDS requests
6173 * fixed directory pages for architectures with non-4k pages sizes
6175 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
6176 * release version v0_4_4
6177 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
6179 * Fixes a number of interruption problems with OST operations.
6180 * Update documentation for portals header changes
6181 * Move all wire protocol structs/defines to lustre_idl.h
6182 * Fixes symlink length bug.
6183 * Add tcpdump to repository.
6185 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
6186 * release version v0_4_3
6187 * Fixes statfs for inodes on extN.
6188 * Fixes bug in runtests which would delete /etc/hosts.
6189 * Use 64-bit object IDs wherever possible (not into VFS though)
6190 Remove ost_get_info, which is unused by lustre, and out of date.
6192 2002-07-03 Peter Braam <braam@clusterfs.com>
6193 * release version v0_4_2 Fixes a lookup error (type not passed)
6194 * move forward to head of Portals
6195 * move forward to latest Lustre kernel
6197 2002-06-25 Peter Braam <braam@clusterfs.com>
6198 * release version v0_4_1. Hopefully stable on single node use.