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.
18 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and
19 OEL5.4 kernel to 2.6.18-164.11.1.0.1.el5.
23 Description: New acc-sm script and various test-framework enhancments.
27 Description: Previously landed patch was reverted.
31 Description: Update SLES10 kernel to 2.6.16.60-0.42.8 (patchless)
35 Description: Add ldiskfs maxdirsize mount option.
39 Description: Fix mkfs.lustre support for 16TB LUNs.
43 Description: Add cascading_rw to lustre mpi tests.
47 Description: Store SOM data in EA.
48 Details : Also several SOM related cleanups.
52 Description: Add refcount to md_open_data
56 Description: Use refcount on lov_request_set to prevent memory corruption.
60 Description: Statahead: use parent dir's i_mutex for synchronization.
64 Description: loadgen improvements.
68 Description: Enable peer health detection in ksocklnd and ko2iblnd.
72 Description: Control DCACHE_LUSTRE_INVALID flag with MDS_INODE_LOOKCUP lock
76 Description: Fix race on obd_devcie:md_stats
80 Description: Fix SLES 11 client builds.
81 Details : Resolved uninitalized pointer gcc warnings.
85 Description: Allow non-root access for "lfs check".
86 Details : Added a check in obd_class_ioctl() for OBD_IOC_PING_TARGET.
90 Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
92 Severity : enhancement
94 Description: Update kernel to RHEL5.4 2.6.18-164.6.1.el5 and
95 OEL5 2.6.18-164.6.1.0.1.el5(Both in-kernel OFED enabled).
99 Description: Sanity test_27p fails.
100 Details : OBD_FAIL_OST_ENOSPC has been moved to filter_statfs().
101 Sanity tests 27 were cleaned up and improved.
105 Description: Unify req format on client/servers
106 Details : Use new req_capsule API [almost] everywhere instead of old PTLRPC
107 buffers and swabbers approach..
112 Description: drop join file stuff.
114 Severity : enhancement
116 Description: Add 2.6.27 and SLES11 (patchless client) support
120 Description: Fix in ptlrpc_expire_one_request() to print the signed time
123 Severity : enhancement
125 Description: Build kernels (RHEL5, OEL5 and SLES10/11) using the vendor's own
128 Severity : enhancement
130 Description: Remove set_info(KEY_UNLINKED) from MDS/OSC
132 Severity : enhancement
134 Description: correctly handle big reply message.
135 Details : send LNet event if reply is bigger then buffer and adjust this buffer
140 Description: Drop unnecessary __GFP_NOMEMALLOC flag from filter_get_page()
142 Severity : enhancement
144 Description: Add support for OEL5.
148 Description: Conf-sanity.sh 50g test - deactivated OST should not cause a panic.
152 Description: client could not reconnect to OST because of an active request.
153 Details : The client now retries to reconnect to the same server, if a
154 connect request failed with EBUSY or -EAGAIN. Test case 35b
155 conf-sanity.sh has been created.
159 Description: Don't allow make backward step on assignin osc next id.
160 Details : race between allocation next id and ll_sync thread can be cause
161 of set wrong osc next id and can be kill valid ost objects.
165 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
169 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
173 Description: Error handling in osc_statfs_interpret() has been improved.
174 Details : Check in osc_statfs_interpret() for EBADR.
178 Description: Do not send statfs() requests to OSTs disabled by administrator.
179 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
181 Severity : enhancement
183 Description: Added sanity-quota test to for lfs(1) to work with numeric uid/gid.
185 Severity : enhancement
187 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
189 Severity : enhancement
191 Description: Enhancement for lfs(1) command to use numeric uid/gid.
195 Description: Improved cleanup procedure for conf-sanity test_46a.
199 Description: Do not start bulk transfer, if client tried to reconnect.
200 Details : Do not start bulk transfer, if client tried to reconnect,
201 but failed due to active rpcs. Check for exp_abort_active_req
202 has been implemented.
206 Description: Prevent inconsistences between linux and lustre mount structures.
207 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
208 Make the sleep interruptible.
210 Severity : enhancement
212 Description: provide server to client comms path
213 Details : server to client communications path via new LDLM_SET_INFO rpc
217 Description: truncate starts GFP_FS allocation under transaction causing deadlock
218 Details : ldiskfs_truncate calls grab_cache_page which may start page
219 allocation under an open transaction. This may lead to
220 calling prune_icache with consequent lustre reentrance.
224 Description: send correctly lsm on open replay
225 Details : MDS is trust to LSM size on replay open, but client can set wrong size
230 Description: enable client interface failover
231 Details : When a child reconnects from another NID, properly update export
232 nid hash position and ldlm reverse import.
234 Severity : enhancement
236 Description: Communicate OST degraded/readonly state via statfs to MDS
237 Details : Flags in the statfs returned from OSTs indicate whether the
238 OST is in a degraded RAID state, or if the filesystem has
239 turned read-only after a filesystem error is detected.
244 Description: don't panic if EPROTO was hit when reading symlink
245 Details : correctly handling request reference in error cases.
247 Severity : enhancement
249 Description: Add LustreNetLink, a kernel-userspace communcation path.
251 Severity : enhancement
253 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
258 Description: access to llog context before init.
259 Details : move handling CATALOGS file at osc layer and forbid access to llog
263 Frequency : rare, only when client is writing to OST on local node
265 Description: Avoid deadlock for local client writes
266 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
267 memory freeing context. This allows OST threads to set the
268 PF_MEMALLOC flag on task structures in order to allocate memory
269 from reserved pools and complete IO.
270 Use GFP_HIGHUSER for OST allocations for non-local client writes,
271 so that the OST threads generate memory pressure and allow
272 inactive pages to be reclaimed.
274 Severity : enhancement
276 Description: Update kernel to RHEL5.3 2.6.18-128.1.14.el5.
281 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
282 Details : move ldlm namespace creation in setup phase to avoid grab
283 _lprocfs_lock with cli_sem held.
287 Description: Temporarily disable grant shrink.
288 Details : Disable the feature for debugging.
291 Frequency : only when formatting test filesystems
293 Description: Unable to run several mkfs.lustre on loop devices at the same time
294 Details : mkfs.lustre returns error 256 on the concurrent loop devices
295 formatting. The solution is to proper handle the error.
297 Severity : enhancement
299 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
301 Severity : enhancement
303 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
308 Description: resolve race between obd_disconnect and class_disconnect_exports
309 Details : if obd_disconnect will be called to already disconnected export he
310 forget release one reference and osc module can't unloaded.
312 Severity : enhancement
314 Description: Allow tuning service thread via /proc
315 Details : For each service a new
316 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
317 created that can be used to set min/max thread counts, and get the
318 current number of running threads.
323 Descriptoin: don't return error if have partially created objects for file.
324 Details : lov_update_create_set uses set->set_success as index for created
325 objects, so if some requests will be failed, they will have hole
326 at end of array and qos_shrink_lsm for allocate correct lsm.
328 Severity : enhancement
330 Description: Update OFED support to 1.4.1
335 Description: Reduce small size read RPC
336 Details : Set read-ahead limit for every file and only do read-ahead when
337 available read-ahead pages are bigger than 1M to avoid small size
340 Severity : enhancement
342 Description: move AT tunable parameters for more consistent usage
343 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
345 Severity : enhancement
347 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
348 Details : allow skip disconnected ost for send statfs request and hide error
354 Description: fix racy locking of mballoc block bitmaps causing BUG
355 Details : The locking of the mballoc buddy bitmap and the in-memory
356 block bitmap was using two different spin locks in some
357 cases. This made it possible to incorrectly access the
358 mballoc bitmap while another process was modifying it,
359 causing a sanity assertion to fail. While no on-disk corruption
360 was reported, there was some risk of this happening.
363 Frequency : rare, on llog test 6
365 Descriptoin: don't allow connect to already connected import
366 Details : allowing connect to already connected import is hide connecting problem.
371 Descriptoin: allow kill process which wait statahead result
372 Details : for some reasons 'ls' can stick in waiting result from statahead,
373 in this case need way for kill this process.
375 Severity : enhancement
377 Description: Add state history info file, enhance import info file
378 Details : Track import connection state changes in a new osc/mdc proc file;
379 add overview-type data to the osc/mdc import proc file.
381 Severity : enhancement
383 Description: MDS create should not wait for statfs RPC while holding DLM lock.
385 Severity : enhancement
387 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
392 Descriptoin: OST grant too much space to client even there are not enough space.
393 Details : Client will shrink its grant cache to OST if there are no write
394 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
395 this grant cache if there are already not enough avaible space
396 (left_space < total_clients * 32M).
399 Frequency : start MDS on uncleanly shutdowned MDS device
401 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
402 Details : stay in waiting mds<>ost recovery finished produce random bugs
403 due race between two ll_sync thread for one lov target. send
404 ACTIVATE event only if connect realy finished and import have
408 Frequency : rare, connect and disconnect target at same time
410 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
411 Details : don't call obd_disconnect under lov_lock. this long time
412 operation and can block ptlrpcd which answer to connect request.
417 Descriptoin: don't lose wakeup for imp_recovery_waitq
418 Details : recover_import_no_retry or invalidate_import and import_close can
419 both sleep on imp_recovery_waitq, but we was send only one wakeup
423 Frequency : always with long access acl
425 Descriptoin: mds can't pack reply with long acl.
426 Details : mds don't control size of acl but they limited by reint/getattr
429 Severity : enhancement
431 Description: Update to SLES10 kernel-2.6.16.60-0.33.
433 Severity : enhancement
435 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
438 Frequency : start MDS on uncleanly shutdowned MDS device
440 Descriptoin: aborting recovery hang on MDS
441 Details : don't throttle destroy RPCs for the MDT.
444 Frequency : on remount
446 Description: external journal device not working after the remount
447 Details : clear dev_rdonly flag for external journal devices in
453 Description: shutdown vs evict race
454 Details : client_disconnect_export vs connect request race.
455 if client will evicted at this time - we start invalidate
456 thread without referece to import and import can be freed
460 Frequency : rare, need acl's on inode.
462 Description: client can't handle ost additional correctly
463 Details : if ost was added after client connected to mds client can have
464 hit lnet_try_match_md ... to big messages to wide striped files.
465 in this case need teach client to handle config events about add
466 lov target and update client max ea size at that event.
468 Severity : enhancement
470 Description: Changelogs
471 Details : Changelogs are a lightweight mechanism to track filesystem
472 metadata and namespace changes. The changelog is recorded
473 permanently on the MDTs, and is periodically "consumed" / purged
474 when records are no longer needed.
476 Severity : enhancement
478 Description: compact fld format with extents
479 Details : Store range of seq rather than every seq in FLD. Seq
480 controller update FLD rather than clients. In Case of CMD, mdt0
481 has FLD, all other metadata server act as non persistent proxy
482 for FLD queries and cache fld entries in fld cache.
487 Description: don't skip ost target if they assigned to file
488 Details : Drop slow OSCs if we can, but not for requested start idx.
489 This means "if OSC is slow and it is not the requested
490 start OST, then it can be skipped, otherwise skip it only
491 if it is inactive/recovering/out-of-space.
495 Description: more cleanup in mds_lov
496 Details : not send LOV EA under replay, we can't know about they size at this
497 time. Don't allow client connect to mds before any ost connected,
498 for avoid problems with LOV EA size and returning EIO to client.
500 Severity : enhancement
502 Description: Interoperability at server side (Disk interoperability)
504 Severity : enhancement
506 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
508 Severity : enhancement
510 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
512 Severity : enhancement
514 Description: New client IO stack (CLIO).
516 Severity : enhancement
518 Description: Commit on sharing. Eliminate inter-client dependencies between
519 uncommitted transactions by doing transaction commits.
520 Thereby clients may recovery independently.
523 Frequency : Create a symlink file with a very long name
525 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
526 Details : If there is no extra space in the request for early cancels,
527 ldlm_req_handles_avail() returns 0 instead of a negative value.
529 Severity : enhancement
531 Description: Add /proc entry for import status
532 Details : The mdc, osc, and mgc import directories now have
533 an import directory that contains useful import data for debugging
536 Severity : enhancement
538 Description: Re-disable certain /proc logging
539 Details : Enable and disable client's offset_stats, extents_stats and
540 extents_stats_per_process stats logging on the fly.
543 Frequency : Only on FC kernels 2.6.22+
545 Description: oops in statahead
546 Details : Do not drop reference count for the dentry from VFS when lookup,
547 VFS will do that by itself.
549 Severity : enhancement
551 Description: Generic /proc file permissions
552 Details : Set /Proc file permissions in a more generic way to enable non-
553 root users operate on some /proc files.
557 Description: Hitting mdc_commit_close() ASSERTION
558 Details : Properly handle request reference release in
559 ll_release_openhandle().
563 Description: quota recovery deadlock during mds failover
564 Details : This patch includes att18982, att18236, att18237 in bz14840.
566 1. fix osts hang when mds does failover with quotaon
567 2. prevent watchdog storm when osts threads wait for the
572 Frequency : only patchless client
573 Description: add workaround for race between add/remove dentry from hash
575 Severity : enhancement
577 Description: Allow OST glimpses to return PW locks
581 Description: LBUG when llog conf file is full
582 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
586 Description: Prevent import from entering FULL state when server in recovery
590 Description: service mount cannot take device name with ":"
591 Details : Only when device name contains ":/" will mount treat it as
597 Description: replace ptlrpcd with the statahead thread to interpret the async
598 statahead RPC callback
602 Frequency : on recovery
603 Description: I/O failures after umount during fail back
604 Details : if client reconnected to restarted server we need join to recovery
605 instead of find server handler is changed and process self
606 eviction with cancel all locks.
608 Severity : enhancement
610 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
612 Severity : enhancement
614 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
616 Severity : enhancement
618 Description: Upcall on Lustre log has been dumped
619 Details : Allow for a user mode script to be called once a Lustre log has
620 been dumped. It passes the filename of the dumped log to the
621 script, the location of the script can be specified via
622 /proc/sys/lnet/debug_log_upcall.
627 Description: avoid idr_remove called for id which is not allocated.
628 Details : Move assigment s_dev for clustered nfs to end of initialization,
629 to avoid problem with error handling.
634 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
635 Details : When connection is reused this not moved from CONN_UNUSED_HASH
636 into CONN_USED_HASH and this prodice warning when put connection
637 again in unused hash.
640 Severity : enhancement
642 Description: File striping can now be set to use an arbitrary pool of OSTs.
644 Severity : enhancement
646 Description: Export bytes_read/bytes_write count on OSC/OST.
650 Description: Early reply size mismatch, MGC loses connection
651 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
652 the connect flags are properly negotiated.
657 Description: quotas are not honored with O_DIRECT
658 Details : all writes with the flag O_DIRECT will use grants which leads to
659 this problem. Now using OBD_BRW_SYNC to guard this.
663 Description: add quota statistics
664 Details : 1. sort out quota proc entries and proc code.
665 2. add quota statistics
667 Severity : enhancement
669 Description: enable quota support for HEAD.
673 Description: Properly propagate oinfo flags from lov to osc for statfs
674 Details : restore missing copy oi_flags to lov requests.
676 Severity : enhancement
678 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
679 : l_getgroups(8), lst(8), routerstat(8)
680 Details : included man pages for llobdstat(8), llstat(8),
681 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
683 Severity : enhancement
685 Description: configure's --enable-quota should check the
686 : kernel .config for CONFIG_QUOTA
687 Details : configure is terminated if --enable-quota is passed but
688 : no quota support is in kernel
692 Description: Remove portals compatibility
693 Details : Remove portals compatibility, not interoperable with releases
698 Description: Resolve device initialization race
699 Details : Prevent proc handler from accessing devices added to the
700 obd_devs array but yet be intialized.
702 Severity : enhancement
704 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
706 Severity : enhancement
708 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
713 Description: Using wrong pointer in osc_brw_prep_request
714 Details : Access to array[-1] can produce panic if kernel compiled with
715 CONFIG_PAGE_ALLOC enabled
719 Description: Client runs out of low memory
720 Details : Consider only lowmem when counting initial number of llap pages
724 Description: *optional* service tags registration
725 Details : if the "service tags" package is installed on a Lustre node
726 When the filesystem is mounted, a local-node service tag will
727 be created. See http://inventory.sun.com/ for more information
728 about the Service Tags asset management system.
732 Description: Kernel BUG tries to release flock
733 Details : Lustre does not destroy flock lock before last reference goes
734 away. So always drop flock locks when client is evicted and
735 perform unlock regardless of successfulness of speaking to MDS.
739 Description: add refcount protection for osc callbacks, avoid panic on shutdown
743 Description: sanity test 65a fails if stripecount of -1 is set
744 Details : handle -1 striping on filesystem in ll_dirstripe_verify
749 Description: ASSERTION(CheckWriteback(page,cmd)) failed
750 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
753 Severity : enhancement
755 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
759 Description: do not process already freed flock
760 Details : flock can possibly be freed by another thread before it reaches
761 to ldlm_flock_completion_ast.
765 Description: LBUG during stress test
766 Details : Need properly lock accesses the flock deadlock detection list.
770 Description: oops in page fault handler
771 Details : kernel page fault handler can return two special 'pages' in error
772 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
776 Description: timeout with invalidate import.
777 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
778 should be handled by ptlrpcd. This produce long age waiting and
779 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
781 Severity : enhancement
783 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
787 Description: enable MGS and MDT services start separately
788 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
789 a co-located MGS without starting the MGS, which is a complement
790 to 'nosvc' mount option.
794 Frequency : after recovery
795 Description: precreate to many object's after del orphan.
796 Details : del orphan st in oscc last_id == next_id and this triger growing
797 count of precreated objects. Set flag LOW to skip increase count
798 of precreated objects.
802 Frequency : rare, on clear nid stats
803 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
804 Details : when clean nid stats sometimes try destroy live entry,
805 and this produce panic in free.
809 Description: Stack overflow during MDS log replay
810 ease stack pressure by using a thread dealing llog_process.
814 Description: wait until IO finished before start new when do lock cancel.
815 Details : VM protocol want old IO finished before start new, in this case
816 need wait until PG_writeback is cleared until check dirty flag and
817 call writepages in lock cancel callback.
819 Severity : enhancement
821 Description: using special macro for print time and cleanup in includes.
825 Description: mds_mfd_close() ASSERTION(rc == 0)
826 Details : In mds_mfd_close(), we need protect inode's writecount change
827 within its orphan write semaphore to prevent possible races.
831 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
835 Frequency : rare, on shutdown ost
836 Description: don't hit live lock with umount ost.
837 Details : shrink_dcache_parent can be in long loop with destroy dentries,
838 use shrink_dcache_sb instead.
842 Description: don't panic with use echo client
843 Details : echo client pass NULL as client nid pointer and this produce null
848 Description: fix build on ppc32
849 Details : compile code with -m64 flag produce wrong object file for ppc32.
853 Description: add message levels for liblustreapi
857 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
858 Details : Move the 'good_osts' check before the 'total_bavail' check. This
859 will result in an -EAGAIN and in the exit call path we call
860 alloc_rr() which will with increasing aggressiveness attempt to
861 aquire precreated objects on the minimum number of required OSCs.
865 Description: Use old size assignment to avoid deadlock
866 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
867 the scheduling while holding a spinlock. We do not need locking
868 for size in ll_update_inode() because size is only updated from
869 the MDS for directories or files without objects, so there is no
870 other place to do the update, and concurrent access to such inodes
871 are protected by the inode lock.
875 Description: resolve "_IOWR redefined" build error on SLES10
879 Description: dump the memory debugging after all modules are unloaded to
880 suppress false negative in conf_sanity test 39
882 Severity : enhancement
884 Description: build kernel-ib packages for OFED 1.3 in our release cycle
889 Description: fix SLES kernel versioning
890 Details : the kernel version for our SLES 10 kernel did not include a "-"
891 before the "smp" at the end. while this was not a problem in
892 general, it did mean that software trying to use the kernel
893 version to try to detect a vendor specific kernel would fail.
894 this was most evident by the OFED build scripts.
898 Description: Don't update lov_desc members until making sure they are valid
899 Details : When updating lov_desc members via proc fs, need fix their
900 validities before doing the real update.
904 Description: don't put request into delay list while invalidate in flight.
905 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
906 import in flight. this produce timeout for invalidate and sometimes
907 can cause stale data.
911 Frequency : on ppc only
912 Description: not convert ost objects for directory because it's not exist.
913 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
917 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
919 Severity : enhancement
921 Description: Update to support 2.6.22.14 vanilla kernel.
925 Frequency : rare, on recovery
926 Description: read procfs can produce deadlock in some situation
927 Details : Holding lprocfs lock which send rpc can produce block for destroy
928 obd objects and this also block reconnect with -EALREADY. This
929 isn't fix all lprocfs bugs - but make it rare.
931 Severity : enhancement
933 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
936 Frequency : frequent on X2 node
938 Description: mdc_set_open_replay_data LBUG
939 Details : Set replay data for requests that are eligible for replay.
943 Description: lustre_mgs: operation 101 on unconnected MGS
944 Details : When MGC is disconnected from MGS long enough, MGS will evict the
945 MGC, and late on MGC cannot successfully connect to MGS and a lot
946 of the error messages complaining that MGS is not connected.
950 Frequency : on network error
951 Description: panic with double free request if network error
952 Details : mdc_finish_enqueue is finish request if any network error ocuring,
953 but it's true only for synchronus enqueue, for async enqueue
954 (via ptlrpcd) this incorrect and ptlrpcd want finish request
957 Severity : enhancement
959 Description: client-side metadata stat-ahead during readdir(directory readahead)
960 Details : perform client-side metadata stat-ahead when the client detects
961 readdir and sequential stat of dir entries therein
964 Frequency : on start mds
966 Description: Implement get_info(last_id) in obdfilter.
969 Frequency : occasional
971 Description: Correctly check stale fid, not start epoch if ost not support SOM
972 Details : open with flag O_CREATE need set old fid in op_fid3 because
973 op_fid2 was overwritten with new generated fid, but mds can answer
974 with one of these two fids and both is not stale. Setattr
975 incorrectly started an epoch and assume will be called
976 done_writing, but without SOM done_writing ever being called.
979 Frequency : rare, depends on device drivers and load
981 Description: MDS or OSS nodes crash due to stack overflow
982 Details : Code changes in 1.8.0 increased the stack usage of some functions.
983 In some cases, in conjunction with device drivers that use a lot
984 of stack the MDS (or possibly OSS) service threads could overflow
985 the stack. One change which was identified to consume additional
986 stack has been reworked to avoid the extra stack usage.
989 Frequency : occasional
991 Description: Do not fail import if osc_interpret_create gets -EAGAIN
992 Details : If osc_interpret_create got -EAGAIN it immediately exits and
993 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
994 oscc_has_objects and see OSC has no objests and call
995 oscc_internal_create to resend create request.
997 Severity : enhancement
999 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
1001 Severity : enhancement
1003 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
1006 Frequency : very rare
1008 Description: Fix replay if there is an un-replied request and open
1009 Details : In some cases, older replay request will revert the
1010 mcd->mcd_last_xid on MDS which is used to record the client's
1011 latest sent request.
1013 Severity : enhancement
1015 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
1017 Severity : enhancement
1019 Description: Add rhel5 support to HEAD.
1021 Serverity : enhancement
1023 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
1028 Description: Don't allow skipping OSTs if index has been specified.
1029 Details : Don't allow skipping OSTs if index has been specified, make
1030 locking in internal create lots better.
1034 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
1035 Details : access to bitfield in structure is always rounded to long
1036 and this produce problem with not atomic change any bit.
1040 Description: Lustre make rpms failed.
1041 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
1042 builds Lustre rpms from tarball.
1045 Frequency : rare on shutdown ost
1047 Description: If llog cancel was not send before clean_exports phase, this can
1048 produce deadlock in llog code.
1049 Details : If llog thread has last reference to obd and call class_import_put
1050 this produce deadlock because llog_cleanup_commit_master wait when
1051 last llog_commit_thread exited, but this never success because was
1052 called from llog_commit_thread.
1056 Description: allow userland application know is lost one of stripes.
1057 Details : fill lvb_blocks with error code on ost and return it to
1058 application if error flag found.
1062 Description: NULL lov_tgts causing MDS oops
1063 Details : more safe checks for NULL lov_tgts for avoid oops.
1065 Severity : enhancement
1067 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
1071 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
1072 Details : If an OST is down the MDS will hang indefinitely in
1073 obd_statfs_rqset() waiting for the statfs data. While for
1074 MDS QOS usage of statfs, it should not stuck in waiting.
1076 Severity : enhancement
1078 Description: remote_acl support
1079 Details : Support ACL-based permission check for remote user.
1080 Support setfacl/getfacl for remote user with the utils
1081 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
1082 the system "{s,g}etfacl" utils.
1084 Severity : enhancement
1086 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
1088 Severity : enhancement
1090 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
1094 Description: make mgs_setparam() handle fsname containing dash
1095 Details : fsname containing a dash does not work with lctl conf_param
1097 Severity : enhancement
1099 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
1101 Severity : enhancement
1103 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
1107 Frequency : rare, at shutdown
1108 Description: access already free / zero obd_namespace.
1109 Details : if client_disconnect_export was called without force flag set,
1110 and exist connect request in flight, this can produce access to
1111 NULL pointer (or already free pointer) when connect_interpret
1112 store ocd flags in obd_namespace.
1116 Frequency : only at startup
1117 Description: not alloc memory with spinlock held.
1118 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
1119 if any spinlock held.
1121 Severity : enhancement
1123 Description: make lustre randomly fail allocating memory
1124 Details : Make lustre randomly failed allocating memory for testing purpose.
1126 Severity : enhancement
1128 Description: lost problems with lov objid file
1129 Details : Fixes some scability and access to not inited memory problems
1130 in work with lov objdid file.
1135 Description: lfs find does not continue on file error
1136 Details : Continue other files processing when a file/dir is absent.
1140 Description: Inconsistent usage of lustre_pack_reply()
1141 Details : Standardize the usage of lustre_pack_reply() such that it
1142 always generate a CERROR on failure.
1145 Frequency : occasional
1147 Description: MDS hang or stay in waiting lock
1148 Details : If client receive lock with CBPENDING flag ldlm need send lock
1149 cancel as separate rpc, to avoid situation when cancel request
1150 can't processed due all i/o threads stay in wait lock.
1154 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
1158 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
1162 Description: Update to SLES9 kernel-2.6.5-7.287.3.
1166 Description: Update to RHEL4 latest kernel.
1168 Severity : enhancement
1170 Description: Build SLES10 patchless client fails
1171 Details : The configure was broken by run ./configure with
1172 --with-linux-obj=.... argument for patchless client. When the
1173 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
1174 can't search header adequately. Use absolute path such as
1175 -I($LINUX)/include instead.
1179 Description: interrupt oig_wait produce painc on resend.
1180 Details : brw_redo_request can be used for resend requests from ptlrpcd and
1181 private set, and this produce situation when rq_ptlrpcd_data not
1182 copyed to new allocated request and triggered LBUG on assert
1183 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
1184 ptlrpcd set if request is changed and can be safety changed to use
1189 Description: LASSERT_{REQ,REP}SWAB macros are buggy
1190 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
1191 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
1192 these macros and replace them with their difinitions instead.
1196 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
1197 Details : Update which_patch & target file for SLES10 latest kernel.
1200 Description: Remove client patches from SLES 10 kernel.
1201 Details : This causes SLES 10 clients to behave as patchless clients
1202 even on a Lustre-patched (server) kernel.
1204 Severity : enhancement
1206 Description: self-adjustable client's lru lists
1207 Details : use adaptive algorithm for managing client cached locks lru
1208 lists according to current server load, other client's work
1209 pattern, memory activities, etc. Both, server and client
1210 side namespaces provide number of proc tunables for controlling
1213 Severity : enhancement
1215 Description: light-weight GSS support
1216 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
1217 and reduce performance overhead.
1219 Severity : enhancement
1221 Description: Linux keyring support
1222 Details : Support using service of Linux keyring for Lustre GSS internal
1223 context refresh/cache mechanism.
1227 Description: Fix errors in lfs documentation
1228 Details : Fixes man pages
1232 Description: don't use GFP_* in generic Lustre code.
1233 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
1237 Description: obdclass is limited by single OBD_ALLOC(idarray)
1238 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
1242 Description: warnings with build patchless client with vanila 2.6.19 and up
1243 Details : change old ctl_table style and replace ctl_table/ctl_table_header
1244 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1248 Frequency : Only for SLES
1249 Description: sanity_quota fail test_1
1250 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1251 file, which makes TSTID[2] inunique.
1255 Frequency : Only for SLES9
1256 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1257 Details : Update target/ChangeLog/which_patch .
1261 Frequency : Only for test_10 in sanity-quota.sh
1262 Description: a bug in quota test code
1263 Details : checking a test flag in a wrong place
1267 Frequency : Only for test_13 in sanity-quota.sh
1268 Description: a bug in quota test script
1269 Details : shouldn't check overfull allocated quota
1273 Description: jbd statistics
1274 Details : Port older jbd statistics patch for sles10
1276 Severity : enhancement
1278 Description: Add printing inode info into message about error in writepage.
1283 Description: reply_lock_interpret crash due to race with it and lock cancel.
1284 Details : Do not replay locks that are being cancelled. Do not reference
1285 locks by their address during replay, just by their handle.
1289 Frequency : When flocks are used.
1290 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1291 Details : Flock locks might destroy just granted lock if it could be merged
1292 with another existing flock, this is done in completion handler,
1293 so teach ldlm_cli_enquque_fini that this is a valid case for
1299 Description: Oops in read and write path when failing to allocate lock.
1300 Details : Check if lock allocation failed and return error back.
1304 Description: lstripe command fails for valid OST index
1305 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1306 instead of lov->desc.ld_active_tgt_count.
1310 Description: sanity.sh failed test 103
1311 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1316 Description: df doesn't work properly if diskfs blocksize != 4K
1317 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1321 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1322 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1324 Severity : enhancement
1326 Description: Adaptive timeouts
1327 Details : RPC timeouts adapt to changing server load and network
1328 conditions to reduce resend attempts and improve recovery time.
1332 Description: llapi_file_create() does not allow some changes
1333 Details : add llapi_file_open() that allows specifying the mode and
1334 open flags, and also returns an open file handle.
1338 Description: merge and cleanup kernel patches.
1339 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1343 Description: Add journal checksum support.(Kernel part)
1344 Details : The journal checksum feature adds two new flags i.e
1345 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1346 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1347 indicates that the commit block contains the checksum for
1348 the blocks described by the descriptor blocks. Now commit
1349 record can be sent to disk without waiting for descriptor
1350 blocks to be written to disk. This behavior is controlled
1351 using JBD2_FEATURE_ASYNC_COMMIT flag.
1355 Description: OSS needs mutliple precreate threads
1356 Details : Add ability to start more than one create thread per OSS.
1360 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1361 Details : execute lfs setstripe on client
1365 Description: mds_obd_create error creating tmp object
1366 Details : When the user sets quota on root, llog will be affected and can't
1367 create files and write files.
1370 Frequency : Always on ia64 patchless client, and possibly others.
1372 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1373 Details : This allows the patchless client to be loaded on architectures
1374 without this export.
1378 Description: RedHat Update kernel for RHEL5
1379 Details : Add the kernel config file for RHEL5.
1383 Description: RedHat Update kernel for RHEL5
1384 Details : Modify the kernel config file more closer RHEL5.
1388 Description: Build failure against Centos5 (RHEL5)
1389 Details : Define PAGE_SIZE when it isn't present.
1393 Description: test_7 on sanity-quota.sh can't be used on separate machines
1394 Details : add facet to handle it
1398 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1399 Details : with NFS, the anon dentry's parent was set to itself in
1400 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1401 corresponding dentry other than use rec->ur_name.
1403 Severity : enhancement
1405 Description: lfs setstripe enhancement
1406 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1410 Description: enable data checksumming by default
1411 Details : enable checksum by default, allow --disable-checksum
1412 configure option and "-o nochecksum" mount option. Checksums
1413 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1417 Description: lustre support for RHEL5
1418 Details : Add support for RHEL5.
1422 Description: Client eviction due to failover config
1423 Details : after a connection loss, the lustre client should attempt to
1424 reconnect to the last active server first before trying the
1425 other potential connections.
1429 Description: when mds and osts use different quota unit(32bit and 64bit),
1430 quota will be released repeatly.
1431 Details : void sending multiple quota reqs to mds, which will keep the
1432 status between the reqs.
1436 Description: osts not allocated evenly to files
1437 Details : change the condition to increase offset_idx
1440 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1442 Description: Data corruption for OSTs that are formatted larger than 2TB
1444 Details : When generating the bio request for lustre file writes the
1445 sector number would overflow a temporary variable before being
1446 used for the IO. The data reads correctly from Lustre (which
1447 will overflow in a similar manner) but other file data or
1448 filesystem metadata may be corrupted in some cases.
1452 Description: Tune the kernel for good SCSI performance.
1453 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1454 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1459 Description: rewrite ext2-derived code in obdclass/uuid.c
1460 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1461 from scratch preserving functionality.
1465 Description: "lfs find -obd UUID" prints directories
1466 Details : "lfs find -obd UUID" will return all directory names instead
1467 of just file names. It is incorrect because the directories
1468 do not reside on the OSTs.
1472 Description: use i_size_read and i_size_write in 2.6 port
1473 Details : replace inode->i_size access with i_size_read/write()
1476 Frequency : when removing large files
1478 Description: scheduling issue during removal of large Lustre files
1479 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1480 It causes scheduling issues when removing large files (17TB in the
1484 Frequency : only with liblustre clients on XT3
1486 Description: evictions taking too long
1487 Details : allow llrd to evict clients directly on OSTs
1490 Frequency : only on ppc
1492 Description: /proc/fs/lustre/devices broken on ppc
1493 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1494 We should make sure the type of 'index' is loff_t and then cast
1495 explicitly as needed below. Do not assign an explicitly cast
1499 Frequency : only for rhel5
1501 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1502 Details : Modify the target file & which_kernel.
1504 Severity : enhancement
1506 Description: omit set fsid for export NFS
1507 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1508 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1512 Frequency : Always, for kernels after 2.6.16
1513 Description: Fix warning idr_remove called for id=.. which is not allocated.
1514 Details : Last kernels save old s_dev before kill super and not allow
1515 to restore from callback - restore it before call kill_anon_super.
1519 Description: buffer overruns could theoretically occur
1520 Details : llapi_semantic_traverse() modifies the "path" argument by
1521 appending values to the end of the origin string, and a
1522 overrun may occur. Adding buffer overrun check in liblustreapi.
1526 Description: Fix error on 'ls .' at the top of the Lustre mount.
1527 Details : Don't revalidate dentry if it is a root dentry.
1531 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1532 Details : Modify vm-tunables-rhel4.patch.
1536 Description: Kernel config for 2.6.18-vanilla.
1537 Details : Modify targets/2.6-vanilla.target.in.
1538 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1539 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1540 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1541 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1545 Description: improve handling recoverable errors
1546 Details : if request processig with error which can be recoverable on server
1547 request should be resend, otherwise page released from cache and
1552 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1553 Details : Modify target file & which_patch.
1554 A flaw was found in the IA32 system call emulation provided
1555 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1556 value could be stored in the %RAX register, which could trigger an
1557 out-of-bounds system call table access. An untrusted local user
1558 could exploit this flaw to run code in the kernel
1559 (ie a root privilege escalation). (CVE-2007-4573).
1563 Description: O_DIRECT bypasses client statistics.
1564 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1565 read_ahead_stats were updated. copied the stats section from
1566 osc_send_oap_rpc() into async_internal().
1570 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1574 Description: change order of libsysio includes
1575 Details : '#include sysio.h' should always come before '#include xtio.h'
1579 Description: handle "serious error: objid * already exists" more gracefully
1580 Details : If LAST_ID value on disk is smaller than the objects existing in
1581 the O/0/d* directories, it indicates disk corruption and causes an
1582 LBUG(). If the object is 0-length, then we should use the existing
1583 object. This will help to avoid a full fsck in most cases.
1585 Severity : enhancement
1587 Description: adapt the lustre_config script to support the upgrade case
1588 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1589 server targets to 1.6 in parallel.
1593 Description: To avoid grant space > avaible space when the disk is almost
1594 full. Without this patch you might see the error "grant XXXX >
1595 available" or some LBUG about grant, when the disk is almost
1597 Details : In filter_check_grant, for non_grant cache write, we should
1598 check the left space by if (*left > ungranted + bytes), instead
1599 of (*left > ungranted), because only we are sure the left space
1600 is enough for another "bytes", then the ungrant space should be
1601 increase. In client, we should update cl_avail_grant only there
1602 is OBD_MD_FLGRANT in the reply.
1606 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1607 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1608 described in CVE-2007-4573. This problem could allow untrusted
1609 local users to gain root access.
1612 Frequency : when using O_DIRECT and quotas
1614 Description: Incorrect file ownership on O_DIRECT output files
1615 Details : block usage reported by 'lfs quota' does not take into account
1616 files that have been written with O_DIRECT.
1621 Description: touch file failed when fs is not full
1622 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1623 otherwise we can get ENOSP while fs is not full.
1627 Description: parallel lock callbacks
1628 Details : Instead of sending blocking and completion callbacks as separated
1629 requests, adding them to a set and sending in parallel.
1632 Frequency : only for Cray XT3
1633 Bugzilla : 12829/13455
1634 Description: Changing primary group doesn't change the group lustre assigns to
1636 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1637 overridden with the primary group provided by the group upcall,
1638 whereas the supplied fsgid can be trusted if it is in the list of
1639 supplementary groups returned by the group upcall.
1641 Severity : enhancement
1643 Description: Allow masking D_WARNING, D_ERROR messages from console
1644 Details : Console messages can now be disabled via lnet.printk.
1648 Description: User code with malformed file open parameter crashes client node
1649 Details : Before packing join_file req, all the related reference should be
1650 checked carefully in case some malformed flags cause fake
1651 join_file req on client.
1655 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1656 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1657 mode is equal to the granted mode. However, it should also check
1658 that LDLM_FL_CP_REQD is not set.
1662 Description: Heavy nfs access might result in deadlocks
1663 Details : After ELC code landed, it is now improper to enqueue any mds
1664 locks under och_sem, because enqueue might want to decide to
1665 cancel open locks for same inode we are holding och_sem for.
1669 Description: Client eviction while running blogbench
1670 Details : A lot of unlink operations with concurrent I/O can lead to a
1671 deadlock causing evictions. To address the problem, the number of
1672 oustanding OST_DESTROY requests is now throttled to
1673 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1674 callbacks are processed in priority.
1678 Description: enable ACLs on MDS by default
1679 Details : ACLs must be enabled on MDS by default.
1682 Frequency : PPC/PPC64 only
1684 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1685 Details : fix duplicated definitions between asm-ppc64/types.h and
1686 lustre_types.h on PPC.
1689 Frequency : PPC/PPC64 only
1691 Description: asm-ppc/segment.h does not exist
1692 Details : fix compile issue on PPC.
1696 Description: better handle error messages in extents code
1699 Frequency : RHEL4 only
1701 Description: mkfs is very slow on IA64/RHEL4
1702 Details : A performance regression has been discovered in the MPT Fusion
1703 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1704 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1705 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1707 Severity : enhancement
1709 Description: SNMP support enhancement
1710 Details : Adding total number of sampled request for an MDS node in snmp
1713 Severity : enhancement
1715 Description: Optimize ldlm waiting list processing for PR extent locks
1716 Details : When processing waiting list for read extent lock and meeting read
1717 lock that is same or wider to it that is not contended, skip
1718 processing rest of the list and immediatelly return current
1719 status of conflictness, since we are guaranteed there are no
1720 conflicting locks in the rest of the list.
1724 Description: Time out and refuse to reconnect
1725 Details : When the failover node is the primary node, it is possible
1726 to have two identical connections in imp_conn_list. We must
1727 compare not conn's pointers but NIDs, otherwise we can defeat
1728 connection throttling.
1732 Description: port llog fixes from b1_6 into HEAD
1733 Details : Port llog reference couting and some llog cleanups from b1_6
1734 (bug 10800) into HEAD, for protect from panic and access to already
1735 free llog structures.
1739 Description: Detect stride IO mode in read-ahead
1740 Details : When a client does stride read, read-ahead should detect that and
1741 read-ahead pages according to the detected stride pattern.
1745 Description: data checksumming impacts single node performance
1746 Details : add support for several checksum algorithm. Currently, only CRC32
1747 and Adler-32 are supported. The checksum type can be changed on
1748 the fly via /proc/fs/lustre/osc/*/checksum_type.
1752 Description: use adler32 for page checksums
1753 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1758 Description: build for x2 fails
1759 Details : fix compile issue on Cray systems.
1763 Description: Properly match for duplicate locks
1764 Details : Due to different lock order from skiplists code, we need to
1765 traverse entire list for now
1768 Frequency : only on PPC/SLES10
1770 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1771 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1772 be defined. Add a hack in mkfs_lustre.c to work around this compile
1777 Description: LASSERT on MDS when client holding flock lock dies
1778 Details : ldlm pool logic depends on number of granted locks equal to
1779 number of released locks which is not true for flock locks, so
1780 just exclude such locks from consideration.
1784 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1785 Details : Use fsfilt_sync() for both the whole filesystem sync and
1786 individual file sync to eliminate dangerous inode locking
1787 with I_LOCK that can lead to a deadlock.
1791 Description: performance in 1.6.3
1792 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1793 sure that 1MB requests are not fragmented by the block layer.
1795 Severity : enhancement
1797 Description: organize the server-side client stats on per-nid basis
1798 Details : Change the structure of stats under obdfilter and mds to
1806 The "uuid"s file would list the uuids of _active_ exports.
1807 And the clear entry is to clear all stats and stale nids.
1809 Severity : enhancement
1811 Description: eliminate client locks in face of contention
1812 Details : file contention detection and lockless i/o implementation
1813 for contended files.
1817 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1820 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1822 Description: MDS may lose file striping (and hence file data) in some cases
1823 Details : If there are additional extended attributes stored on the MDS,
1824 in particular ACLs, SELinux, or user attributes (if user_xattr
1825 is specified for the client mount options) then there is a risk
1826 of attribute loss. Additionally, the Lustre file striping
1827 needs to be larger than default (e.g. striped over all OSTs),
1828 and an additional attribute must be stored initially in the
1829 inode and then increase in size enough to be moved to the
1830 external attribute block (e.g. ACL growing in size) for file
1835 Description: skiplist implementation simplification
1836 Details : skiplists are used to group compatible locks on granted list
1837 that was implemented as tracking first and last lock of each
1838 lock group the patch changes that to using doubly linked lists
1842 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1843 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1844 because LASSERT on both the data supplied by a client, and the
1845 data on disk is dangerous and incorrect.
1847 Severity : enhancement
1849 Description: Slow truncate/writes to huge files at high offsets.
1850 Details : Directly associate cached pages to lock that protect those pages,
1851 this allows us to quickly find what pages to write and remove
1852 once lock callback is received.
1856 Description: more ldlm soft lockups
1857 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1858 starve other threads from the resource lock for a long time in
1859 case of long waiting queue, so change the debug level from
1860 D_OTHER to the less frequently used D_INFO.
1862 Severity : enhancement
1864 Description: add -gid, -group, -uid, -user options to lfs find
1868 Description: Hung threads in invalidate_inode_pages2_range
1869 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1870 once one is completed. As a result, some RPCs are stuck in the
1871 queue and are never sent.
1875 Description: filter threads hungs on waiting journal commit
1876 Details : Cleanup filter group llog code, then only filter group llog will
1877 be only created in the MDS/OST syncing process.
1881 Description: Procfs and llog threads access destoryed import sometimes.
1882 Details : Sync the import destoryed process with procfs and llog threads by
1883 the import refcount and semaphore.
1885 Severity : enhancement
1887 Description: openlock cache of b1_6 port to HEAD
1892 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1893 Details : If insertion of an extent fails, then discard the inode
1894 preallocation and free data blocks else it can lead to duplicate
1899 Description: don't always update ctime in ext3_xattr_set_handle()
1900 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1901 In some cases the ctime should not be updated, for example for
1902 2.0->1.8 compatibility it is necessary to delete an xattr and it
1903 should not update the ctime.
1907 Bugzilla : 15713/16362
1908 Description: Assertion in iopen_connect_dentry in 1.6.3
1909 Details : looking up an inode via iopen with the wrong generation number can
1910 populate the dcache with a disconneced dentry while the inode
1911 number is in the process of being reallocated. This causes an
1912 assertion failure in iopen since the inode's dentry list contains
1913 both a connected and disconnected dentry.
1917 Description: assertion failure in ldlm_handle2lock()
1918 Details : fix a race between class_handle_unhash() and class_handle2object()
1919 introduced in lustre 1.6.5 by bug 13622.
1924 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1925 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1930 Description: oops in sd_iostats_seq_show()
1931 Details : unloading/reloading the scsi low level driver triggers a kernel
1932 bug when trying to access the sd iostat file.
1937 Description: Kernel panics during QLogic driver reload
1938 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1939 patch, causing memory corruption.
1944 Description: journal_dev option does not work in b1_6
1945 Details : pass mount option during pre-mount.
1947 Severity : enhancement
1949 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1950 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1951 extent information of a file. It can be used to map logical blocks
1952 in a file to physical blocks in the block device.
1956 Description: LDLM soft lockups - improvement
1957 Details : It is be possible to send the lock handle along with each read
1958 or write request because the client is already doing a lock match
1959 itself so there isn't any reason the OST should have to re-do that
1963 Frequency : only on Cray X2
1965 Description: X2 build failures
1966 Details : fix build failures on Cray X2.
1970 Description: xid & resent requests
1971 Details : Initialize RPC XID from clock at startup (randomly if clock is
1974 Severity : enhancement
1976 Description: Add lustre_start utility to start or stop multiple Lustre servers
1981 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1982 Details : In case of memory pressure, list_del() can be called twice on
1983 req->rq_history_list, causing a kernel oops.
1987 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1988 Details : fix stack overflow in the distributed lock manager by defering
1989 export eviction after a failed AST to the elt thread instead of
1990 handling it in the dlm interpret routine.
1994 Description: Convert some comments to new format.
1995 Details : Update documenting comments to match doxygen conventions.
1999 Description: Grammar fixes.
2000 Details : A couple of trivial spelling fixes.
2004 Description: OSD_COUNTERS-mandatory
2005 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
2006 to simplify the code.
2010 Description: simplify cmm_device freeing logic.
2011 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
2015 Description: Add lockdep support to dt_object_operations locking interface.
2016 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
2017 parameter indicating lock ordering. Update mdd code to use new
2022 Description: Introduce failloc constants for lockless IO tests.
2023 Details : Add two new failloc constants to test lockless IO. Only one of
2024 them in implemented---another is checked in yet to be landed
2029 Description: Add lockdep support for inode mutex.
2030 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
2031 in the situations where Lustre has to lock more than one inode
2036 Description: Add optional invariants checking support.
2037 Details : Add new LINVRNT() macro, optional on new --enable-invariants
2038 configure switch. This macro is to be used for consistency and
2039 sanity checks that are too expensive to be left in `production'
2044 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
2045 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
2050 Description: Add lockdep support for ldlm_lock and ldlm_resource.
2051 Details : Use spin_lock_nested() in (the only) situation where more than
2052 one ldlm_lock is locked simultaneously. Also, fix possible
2053 dead-lock in ldlm_lock_change_resource() by enforcing particular
2058 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
2059 Details : Instead of specifying each ldlm_lock call-back through separate
2060 parameter, wrap them into struct ldlm_callback_suite.
2064 Description: Kill join_lru obd method and its callers.
2065 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
2066 in memory---a requirement implemented through ->o_join_lru() obd
2067 method in HEAD. Remove this method and its users.
2071 Description: Add asynchronous ldlm ENQUEUE completion handler.
2072 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
2073 case that stock ldlm_completion_ast() cannot handle as it waits
2074 until lock is granted. Introduce new ldlm_completion_ast_async()
2075 for this. Also comment ldlm_completion_ast().
2079 Description: ldlm_error <-> errno conversion.
2080 Details : Add functions to map (rather arbitrary) between LDLM error codes
2081 and standard errno values. CLIO needs this to prevent LDLM specific
2082 constants from escaping ldlm and osc.
2086 Description: Kill unused ldlm_handle2lock_ns() function.
2087 Details : Kill unused ldlm_handle2lock_ns() function.
2091 Description: Add lu_ref support to ldlm_lock
2092 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
2093 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
2094 and ldlm_resource. LDLM interface has to be changed, because code that
2095 releases a reference on a lock, has to "know" what reference this is.
2096 In the most frequent case
2098 lock = ldlm_handle2lock(handle);
2100 LDLM_LOCK_PUT(lock);
2102 no changes are required. When any other reference (received _not_ from
2103 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
2104 instead of LDLM_LOCK_PUT().
2106 Arguably, changes are pervasive, and interface requires some discipline
2107 for proper use. On the other hand, it was very instrumental in finding
2108 a few leaked lock references.
2112 Description: Add ldlm_lock_addref_try().
2113 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
2114 attempts to addref a lock that might be being canceled
2119 Description: Add ldlm_weigh_callback().
2120 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
2121 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
2122 hard-coded `number of pages' logic.
2126 Description: Add lockdep annotations to llog code.
2127 Details : Use appropriately tagged _nested() locking calls in the places
2128 where llog takes more than one ->lgh_lock lock.
2132 Description: Add loi_kms_set().
2133 Details : Wrap kms updates into a helper function.
2137 Description: Constify instances of struct lsm_operations.
2138 Details : Constify instances of struct lsm_operations.
2142 Description: lu_conf support.
2143 Details : On a server, a file system object is uniquely identified
2144 by a fid, which is sufficient to locate and load all object
2145 state (inode). On a client, on the other hand, more data are
2146 necessary instantiate an object. Change lu_object_find() and
2147 friends to take additional `lu_conf' argument describing object.
2148 Typically this includes layout information.
2152 Description: lu_context fixes.
2153 Details : Introduce new lu_context functions that are needed on the client
2154 side, where some system threads (ptlrpcd) are shared by multiple
2155 modules, and so cannot be stopped during module shutdown.
2159 Description: Add start and stop methods to lu_device_type_operations.
2160 Details : Introduce two new methods in lu_device_type_operations, that are
2161 invoked when first instance of a given type is created and last one
2162 is destroyed respectively. This is need by CLIO.
2166 Description: Add lu_ref support to struct lu_device.
2167 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
2168 track leaked references.
2172 Description: Introduce lu_kmem_descr.
2173 Details : lu_kmem_descr and its companion interface allow to create
2174 and destroy a number of kmem caches at once.
2178 Description: Fix lu_object finalization race.
2179 Details : Fix a race between lu_object_find() finding an object and its
2180 concurrent finalization. This race is (most likely) not possible
2181 on the server, but might happen on the client.
2185 Description: Introduce lu_ref interface.
2186 Details : lu_ref is a debugging module allowing to track references to
2187 a given object. It is quite cpu expensive, and has to be
2188 explicitly enabled with --enable-lu_ref. See usage description
2193 Description: Factor lu_site procfs stats into a separate function.
2194 Details : Separate lu_site stats printing code into a separate function
2195 that can be reused on a client.
2199 Description: Constify instances of struct {lu,dt,md}_device_operations.
2200 Details : Constify instances of struct {lu,dt,md}_device_operations.
2204 Description: Introduce struct md_site and move meta-data specific parts of
2205 struct lu_site here.
2206 Details : Move md-specific fields out of struct lu_site into special struct
2207 md_site, so that lu_site can be used on a client.
2211 Description: Kill mdd_lov_destroy().
2212 Details : Remove unused mdd code.
2216 Description: Add st_block checking to multistat.c.
2217 Details : Add st_block checking to multistat.c.
2221 Description: Add lu_ref support to struct obd_device.
2222 Details : Add lu_ref tracking to obd_device.
2226 Description: Kill obd_set_fail_loc().
2227 Details : Remove unused code.
2231 Description: Add special type for ptlrpc_request interpret functions.
2232 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
2233 there. Actual usage will be in CLIO.
2237 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
2238 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
2239 doesn't work with lockdep.
2243 Description: Add rwv.c test program.
2244 Details : New testing program exercising readv(2) and writev(2) (Qian).
2248 Description: Add sendfile.c test program.
2249 Details : New testing program exercising sendfile(2) (Jay).
2253 Description: Ratelimit a message that can be very frequent.
2254 Details : Ratelimit a memory allocation failure message that can
2259 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2260 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2264 Description: Liblustre build fixes.
2265 Details : Liblustre build fixes.
2269 Description: libcfs: add cfs_{need,cond}_resched() interface.
2270 Details : libcfs: add cfs_{need,cond}_resched() definition and
2271 implementations for Linux, NT, and liblustre.
2273 Severity : enhancement
2275 Description: More exported tunables for mballoc
2276 Details : Add support for tunable preallocation window and new tunables for
2277 large/small requests
2281 Description: Detect corruption of block bitmap and checking for preallocations
2282 Details : Checks validity of on-disk block bitmap. Also it does better
2283 checking of number of applied preallocations. When corruption is
2284 found, it turns filesystem readonly to prevent further corruptions.
2288 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2289 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2290 window to 0. If it does hit cache miss, and read-pattern is still
2291 stride-io mode, does not reset the stride window, but also does not
2292 increase the stride window length in this case.
2296 Frequency : only for big-endian servers
2297 Description: Check if system is big-endian while mounting fs with extents feature
2298 Details : Mounting a filesystem with extents feature will fail on big-endian
2299 systems since ext3-based ldiskfs is not supported on big-endian
2300 systems. This can be over-riden with "bigendian_extents" mount option.
2302 Severity : enhancement
2304 Description: The root squash functionality
2305 Details : A security feature, which is to prevent users from being able
2306 to mount lustre on their desktop, run as root, and delete
2307 all of the files in the filesystem. The goal is accomplished by
2308 remapping user id (UID) and group id (GID) of the root user to
2309 a UID and GID specified by the system administartor via Lustre
2310 configuration management server (MGS). The functionality also
2311 allows to specify sets of clients for which the remapping does
2316 Description: Excessive recovery window
2317 Details : With AT enabled, the recovery window can be excessively long (6000+
2318 seconds). To address this problem, we no longer use
2319 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2320 timeout no longer depends on the service time, it is set to
2321 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2322 time via pb_service_time.
2326 Description: Watchdog triggered on MDS failover
2327 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2328 the OSTs know that the MDS "UUID" can be reused for the same export
2329 from a different NID, so we do not need to wait for the export to be
2333 Frequency : rare, only if using MMP with Linux RAID
2335 Description: MMP doesn't work with Linux RAID
2336 Details : While using HA for Lustre servers with Linux RAID, it is possible
2337 that MMP will not detect multiple mounts. To make this work we
2338 need to unplug the device queue in RAID when the MMP block is being
2339 written. Also while reading the MMP block, we should read it from
2340 disk and not the cached one.
2342 Severity : enhancement
2344 Description: open file using fid
2345 Details : A file can be opened using just its fid, like
2346 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2349 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2352 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2353 Details : add proper locking for creation of jbd_4k slab cache
2357 Description: MMP check in ext3_remount() fails without displaying any error
2358 Details : When multiple mount protection fails during remount, proper error
2361 Severity : enhancement
2363 Description: Allow stripe size to be up to 4G-64k
2364 Details : Fix math logic to allow large stripe sizes.
2368 Description: add check for >8TB ldiskfs filesystems
2369 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2370 Don't allow >8TB ldiskfs filesystems to be mounted without
2371 force_over_8tb mount option
2376 Descriptoin: OSS complains "too many blocks" and crashes
2377 Details : In filter_commitrw_write():
2378 1. lquota_chkquota();
2379 2. obdo_from_inode();
2380 3. lquota_pending_commit();
2381 "2." has changed uid and gid in this case, then caused
2382 lquota_chkquota() and lquota_pending_commit() to use
2383 different uid and gid. That is the root of the bug.
2387 Descriptoin: getstripe doesn't report stripe count = -1 sanely
2388 Details : Detect and print out the 'stripe count = -1' case properly.
2392 Description: Out or order replies might be lost on replay
2393 Details : In ptlrpc_retain_replayable_request if we cannot find retained
2394 request with tid smaller then one currently being added, add it
2395 to the start, not end of the list.
2397 --------------------------------------------------------------------------------
2399 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2401 * Support for kernels:
2402 2.6.5-7.283 (SLES 9),
2403 2.6.9-55.EL (RHEL 4),
2404 2.6.16.46-0.14 (SLES 10),
2405 2.6.18.8 vanilla (kernel.org)
2406 * Client support for unpatched kernels:
2407 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2408 2.6.16 - 2.6.22 vanilla (kernel.org)
2409 * Due to recently discovered recovery problems, we do not recommend
2410 using patchless RHEL 4 clients with this or any earlier release.
2411 * Recommended e2fsprogs version: 1.39.cfs8
2412 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2413 * Starting with this release, the ldiskfs backing filesystem required
2414 by Lustre is now in its own package, lustre-ldiskfs. This package
2415 should be installed. It is versioned separately from Lustre and
2416 may be released separately in future.
2420 Description: block reactivating mgc import until all deactivates complete
2421 Details : Fix race when failing back MDT/MGS to itself (testing)
2424 Frequency : at statup only
2426 Description: mds_lov_synchronize race leads to various problems
2427 Details : simultaneous MDT->OST connections at startup can cause the
2428 sync to abort, leaving the OSC in a bad state.
2430 Severity : enhancement
2432 Description: add optional extra BUILD_VERSION info
2433 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2434 to override the lustre version.
2437 Frequency : 2.6.18 servers only
2439 Description: ll_kern_mount() doesn't release the module reference
2440 Details : The ldiskfs module reference count never drops down to 0
2441 because ll_kern_mount() doesn't release the module reference.
2446 Description: server LBUG when using old ost_num_threads parameter
2447 Details : Accept the old ost_num_threads parameter but warn that it
2448 is deprecated, and fix an off-by-one error that caused an LBUG.
2453 Description: Transient SCSI error results in persistent IO issue
2454 Details : iobuf->dr_error is not reinitialized to 0 between two
2458 Frequency : sometimes when underlying device returns I/O errors
2460 Description: OSTs not going read-only during write failures
2461 Details : OSTs are not remounted read-only when the journal commit threads
2462 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2463 instead of the ext3 wrappers.
2467 Description: poor connect scaling with increasing client count
2468 Details : Don't run filter_grant_sanity_check for more than 100 exports
2469 to improve scaling for large numbers of clients.
2472 Frequency : SLES10 only
2474 Description: sanity-quota.sh quotacheck failed: rc = -22
2475 Details : Quotas cannot be enabled on SLES10.
2478 Frequency : liblustre clients only
2480 Description: getdirentries does not give error when run on compute nodes
2481 Details : getdirentries does not fail when the size specified as an argument
2482 is too small to contain at least one entry
2484 Severity : enhancement
2486 Description: Add LNET router traceability for debug purposes
2487 Details : If a checksum failure occurs with a router as part of the
2488 IO path, the NID of the last router that forwarded the bulk data
2489 is printed so it can be identified.
2494 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2495 Details : Due to a race condition, liblustre clients were occasionally
2496 evicted incorrectly.
2498 Severity : enhancement
2500 Description: lfs setstripe use optional parameters instead of postional
2503 Severity : enhancement
2505 Description: Nanosecond timestamp support for ldiskfs
2506 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2507 resolution timestamps. There is not yet support for this at
2508 the Lustre filesystem level.
2511 Frequency : during server recovery
2513 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2514 Details : request with rq_no_resend flag not awake l_wait_event if they get
2518 Frequency : nfs export on patchless client
2520 Description: connectathon hang when test nfs export over patchless client
2521 Details : Disconnected dentry cannot be found with lookup, so we do not need
2522 to unhash it or make it invalid
2525 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2526 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2527 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2532 Description: open req refcounting wrong on reconnect
2533 Details : If reconnect happened between getting open reply from server and
2534 call to mdc_set_replay_data in ll_file_open, we will schedule
2535 replay for unreferenced request that we are about to free.
2536 Subsequent close will crash in variety of ways.
2537 Check that request is still eligible for replay in
2538 mdc_set_replay_data().
2543 Description: disable writes to filesystem when reading health_check file
2544 Details : the default for reading the health_check proc file has changed
2545 to NOT do a journal transaction and write to disk, because this
2546 can cause reads of the /proc file to hang and block HA state
2547 checking on a healthy but otherwise heavily loaded system. It
2548 is possible to return to the previous behaviour during configure
2549 with --enable-health-write.
2551 Severity : enhancement
2553 Description: 64-bit inode version
2554 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2555 to the inode. This will be required for version-based recovery.
2560 Description: MDS fails to start if a duplicate client export is detected
2561 Details : in some rare cases it was possible for a client to connect to
2562 an MDS multiple times. Upon recovery the MDS would detect this
2563 and fail during startup. Handle this more gracefully.
2565 Severity : enhancement
2567 Description: Add -o localflock option to simulate old noflock
2569 Details : This will achieve local-only flock/fcntl locks
2575 Description: log_commit_thread vs filter_destroy race leads to crash
2576 Details : Take import reference before releasing llog record semaphore
2581 Description: Wrong request locking in request set processing
2582 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2583 imp_delayed_list, in this place should be used imp_lock.
2586 Frequency : when reconnection
2588 Description: Grant Leak when osc reconnect to OST
2589 Details : When osc reconnect ost, OST(filter) should check whether it
2590 should grant more space to client by comparing fed_grant and
2591 cl_avail_grant, and return the granted space to client instead
2592 of "new granted" space, because client will call osc_init_grant
2593 to update the client grant space info.
2596 Frequency : when client reconnect to OST
2598 Description: Grant Leak when osc do resend and replay bulk write
2599 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2600 bulk write request, because the grant info while be sync between
2601 OSC and OST when reconnect, and we should ignore the grant info
2602 these of resend/replay write req.
2607 Description: Grant space more than avaiable left space sometimes.
2608 Details : When then OST is about to be full, if two bulk writing from
2609 different clients came to OST. Accord the avaliable space of the
2610 OST, the first req should be permitted, and the second one
2611 should be denied by ENOSPC. But if the seconde arrived before
2612 the first one is commited. The OST might wrongly permit second
2613 writing, which will cause grant space > avaiable space.
2616 Frequency : when client is evicted
2618 Description: Grant might be wrongly erased when osc is evicted by OST
2619 Details : when the import is evicted by server, it will fork another
2620 thread ptlrpc_invalidate_import_thread to invalidate the
2621 import, where the grant will be set to 0. While the original
2622 thread will update the grant it got when connecting. So if
2623 the former happened latter, the grant will be wrongly errased
2624 because of this race.
2629 Description: Checking Stale with correct fid
2630 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2631 to check whether it is stale, because sometimes, we want the
2632 enqueue happened anyway, and op_data.fid2 will not be initialized.
2634 Severity : enhancement
2636 Description: update patchless client
2637 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2640 Frequency : only with 2.4 kernel
2642 Description: random memory corruption
2643 Details : size of struct ll_inode_info is to big for union inode.u and this
2644 can be cause of random memory corruption.
2649 Description: Memory leak in recovery
2650 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2651 It should also check obd_fail before fsfilt_start, otherwise if
2652 fsfilt_start return -EROFS,(failover mds during mds recovery).
2653 then the req will return with repmsg->transno = 0 and rc = EROFS.
2654 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2655 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2656 be freed no matter whether fsfilt_commit success or not.
2659 Frequency : only with huge count clients
2661 Description: Prevents from taking the superblock lock in llap_from_page for
2663 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2664 for llap_from_page call in ll_removepage prevents from taking the
2665 superblock lock for a soon died page.
2670 Description: Not check open intent error before release open handle
2671 Details : in some rare cases, the open intent error is not checked before
2672 release open handle, which may cause
2673 ASSERTION(open_req->rq_transno != 0), because it tries to release
2674 the failed open handle.
2679 Description: Set cat log bitmap only after create log success.
2680 Details : in some rare cases, the cat log bitmap is set too early. and it
2681 should be set only after create log success.
2685 Description: Accessing a block bevice can re-enable I/O when Lustre is
2686 tearing down a device.
2687 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2692 Description: service threads may hog cpus when there are a lot of requests
2693 Details : Insert cond_resched to give other threads a chance to use some CPU
2698 Description: the cat log was not initialized in recovery
2699 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2700 unlink log on mds will not be initialized until mds post
2701 recovery. And also in mds post recovery, the unlink log will
2702 initialization will be done asynchronausly, so there will be race
2703 between add unlink log and unlink log initialization.
2707 Description: brw_stats were being printed incorrectly
2708 Details : brw_stats were being printed as log2 but all of them were not
2709 recorded as log2. Also remove some code duplication arising from
2710 filter_tally_{read,write}.
2714 Frequency : rare, only in recovery.
2715 Description: ASSERTION(req->rq_type != LI_POISON) failed
2716 Details : imp_lock should be held while iterating over imp_sending_list for
2717 prevent destroy request after get timeout in ptlrpc_queue_wait.
2721 Description: replay-single.sh test 52 fails
2722 Details : A lock's skiplist need to be cleanup when it being unlinked
2723 from its resource list.
2727 Description: Short directio read returns full requested size rather than
2729 Details : Direct I/O operations should return actual amount of bytes
2730 transferred rather than requested size.
2732 Severity : enhancement
2734 Description: metadata RPC reduction (e.g. for rm performance)
2735 Details : decrease the amount of synchronous RPC between clients and servers
2736 by canceling conflicing lock before the operation on the client
2737 and packing thier handles into the main operation RPC to server.
2739 Severity : enhancement
2741 Description: Async OSC create to avoid the blocking unnecessarily.
2742 Details : If a OST has no remain object, system will block on the creating
2743 when need to create a new object on this OST. Now, ways use
2744 pre-created objects when available, instead of blocking on an
2745 empty osc while others are not empty. If we must block, we block
2746 for the shortest possible period of time.
2750 Description: improve handling recoverable errors
2751 Details : if request processig with error which can be recoverable on server
2752 request should be resend, otherwise page released from cache and
2755 Severity : enhancement
2757 Description: refine locking for avoid write wrong info into lov_objid file
2758 Details : fix possible races with add new target and write/update data in
2761 --------------------------------------------------------------------------------
2763 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2768 Frequency : on some architectures
2770 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2771 Details : Uninitialized flags sometimes cause configuration commands to
2775 Frequency : patchless clients only
2777 Description: missing __iget() symbol export
2778 Details : The __iget() symbol export is missing. To avoid the need for
2779 this on patchless clients the deathrow inode reaper is turned
2780 off, and we depend on the VM to clean up old inodes. This
2781 dependency was during via the fix for bug 12181.
2783 --------------------------------------------------------------------------------
2785 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2787 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2788 INTEROPERATE with older versions automatically. In many cases a
2789 special upgrade step is needed. Please read the
2790 user documentation before upgrading any part of a live system.
2791 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2792 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2793 * WARNING: Lustre configuration and startup changes are required with
2794 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2798 Severity : enhancement
2800 Description: MountConf
2801 Details : Lustre configuration is now managed via mkfs and mount
2802 commands instead of lmc and lconf. New obd types (MGS, MGC)
2803 are added for dynamic configuration management. See
2804 https://mail.clusterfs.com/wikis/lustre/MountConf for
2807 Severity : enhancement
2809 Description: dynamic OST addition
2810 Details : OSTs can now be added to a live filesystem
2812 Severity : enhancement
2814 Description: startup order invariance
2815 Details : MDTs and OSTs can be started in any order. Clients only
2816 require the MDT to complete startup.
2818 Severity : enhancement
2820 Description: parallel, asynchronous orphan cleanup
2821 Details : orphan cleanup is now performed in separate threads for each
2822 OST, allowing parallel non-blocking operation.
2824 Severity : enhancement
2826 Description: optimized stripe assignment
2827 Details : stripe assignments are now made based on ost space available,
2828 ost previous usage, and OSS previous usage, in order to try
2829 to optimize storage space and networking resources.
2831 Severity : enhancement
2833 Description: Permanently set tunables
2834 Details : All writable /proc/fs/lustre tunables can now be permanently
2835 set on a per-server basis, at mkfs time or on a live system.
2837 Severity : enhancement
2839 Description: Lustre message v2
2840 Details : Add lustre message format v2.
2842 Severity : enhancement
2844 Description: client OST exclusion list
2845 Details : Clients can be started with a list of OSTs that should be
2846 declared "inactive" for known non-responsive OSTs.
2850 Description: SPEC SFS validation failure on NFS v2 over lustre.
2851 Details : Changes the blocksize for regular files to be 2x RPC size,
2852 and not depend on stripe size.
2854 Severity : enhancement
2856 Description: Multiple MD RPCs in flight.
2857 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2858 To avoid overly-overloading MDS, introduce a limit on number of
2859 MDS RPCs in flight for a single client and add /proc controls
2860 to adjust this limit.
2862 Severity : enhancement
2864 Description: client read/write statistics
2865 Details : Add client read/write call usage stats for performance
2866 analysis of user processes.
2867 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2868 file access. extents_stats shows chunk size distribution.
2869 extents_stats_per_process show chunk size distribution per
2872 Severity : enhancement
2874 Description: mds statistics
2875 Details : Add detailed mds operations statistics in
2876 /proc/fs/lustre/mds/*/stats.
2880 Description: Failure of copying files with lustre special EAs.
2881 Details : Client side always return success for setxattr call for lustre
2882 special xattr (currently only "trusted.lov").
2887 Description: Refcount LNET uuids
2888 Details : The global LNET uuid list grew linearly with every startup;
2889 refcount repeated list entries instead of always adding to
2892 Severity : enhancement
2894 Description: Dynamic service threads
2895 Details : Within a small range, start extra service threads
2896 automatically when the request queue builds up.
2899 Frequency : mixed-endian client/server environments
2901 Description: mixed-endian crashes
2902 Details : The new msg_v2 system had some failures in mixed-endian
2905 Severity : enhancement
2907 Description: Easy OST removal
2908 Details : OSTs can be permanently deactivated with e.g. 'lctl
2909 conf_param lustre-OST0001.osc.active=0'
2911 Severity : enhancement
2913 Description: MGS proc entries
2914 Details : Added basic proc entries for the MGS showing what filesystems
2917 Severity : enhancement
2919 Description: provide MGS failover
2920 Details : Added config lock reacquisition after MGS server failover.
2922 Severity : enhancement
2924 Description: add Linux 2.4 support
2925 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2929 Description: a large application tries to do I/O to the same resource and dies
2930 in the middle of it.
2931 Details : Check the req->rq_arrival time after the call to
2932 ost_brw_lock_get(), but before we do anything about
2933 processing it & sending the BULK transfer request. This
2934 should help move old stale pending locks off the queue as
2935 quickly as obd_timeout.
2938 Frequency : when an incorrect nid is specified during startup
2940 Description: ptlrpc connect to non-existant node causes kernel crash
2941 Details : LNET can't be re-entered from an event callback, which
2942 happened when we expire a message after the export has been
2943 cleaned up. Instead, hand the zombie cleanup off to another
2946 Severity : enhancement
2948 Description: plain/inodebits lock performance improvement
2949 Details : Grouping plain/inodebits in granted list by their request modes
2950 and bits policy, thus improving the performance of search through
2954 Frequency : only if OST filesystem is corrupted
2956 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2957 Details : for a short time RPCs with bulk IO are in the replay list,
2958 but replay of bulk IOs is unimplemented. If the OST filesystem
2959 is corrupted due to disk cache incoherency and then replay is
2960 started it is possible to trip an assertion. Avoid putting
2961 committed RPCs into the replay list at all to avoid this issue.
2964 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2966 Description: System hang on startup
2967 Details : This bug allowed the liblustre (e.g. catamount) client to
2968 return to the app before handling all startup RPCs. This
2969 could leave the node unresponsive to lustre network traffic
2970 and manifested as a server ptllnd timeout.
2972 Severity : enhancement
2974 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2975 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2976 boolean to control whether to print peer debug info when a
2977 client's RPC times out.
2980 Frequency : only for kernels with patches from Lustre below 1.4.3
2982 Description: Remove old rdonly API
2983 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2986 Frequency : only for devices with external journals
2988 Description: Set external device read-only also
2989 Details : During a commanded failover stop, we set the disk device
2990 read-only while the server shuts down. We now also set any
2991 external journal device read-only at the same time.
2994 Frequency : when upgrading from 1.4 while trying to change parameters
2996 Description: The wrong (new) MDC name was used when setting parameters for
2997 upgraded MDT's. Also allows changing of OSC (and MDC)
2998 parameters if --writeconf is specified at tunefs upgrade time.
3001 Frequency : when setting specific ost indicies
3003 Description: QOS code breaks on skipped indicies
3004 Details : Add checks for missing OST indicies in the QOS code, so OSTs
3005 created with --index need not be sequential.
3007 Severity : enhancement
3009 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3010 Details : The uninit_groups feature works in conjunction with the kernel
3011 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3012 up the pass1 processing of e2fsck. This is a read-only feature
3013 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3014 mount filesystems that have had this feature enabled.
3016 Severity : enhancement
3018 Description: Improve multi-block allocation algorithm to avoid fragmentation
3019 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3020 allocation locality and avoid filesystem fragmentation.
3022 ------------------------------------------------------------------------------
3024 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
3026 * Support for kernels:
3027 2.4.21-47.0.1.EL (RHEL 3)
3028 2.6.5-7.283 (SLES 9)
3029 2.6.9-42.0.10.EL (RHEL 4)
3030 2.6.12.6 vanilla (kernel.org)
3031 2.6.16.27-0.9 (SLES 10)
3032 * Recommended e2fsprogs version: 1.39.cfs5
3034 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3038 Frequency : occasional, depends on client load and configuration
3039 Bugzilla : 12181, 12203
3040 Description: data loss for recently-modified files
3042 Details : In some cases it is possible that recently written or created
3043 files may not be written to disk in a timely manner (this should
3044 normally be within 30s unless client IO load is very high).
3045 The problem appears as zero-length files or files that are a
3046 multiple of 1MB in size after a client crash or client eviction
3047 that are missing data at the end of the file.
3049 This problem is more likely to be hit on clients where files are
3050 repeatedly created and unlinked in the same directory, clients
3051 have a large amount of RAM, have many CPUs, the filesystem has
3052 many OSTs, the clients are rebooted frequently, and/or the files
3053 are not accessed by other nodes after being written.
3055 The presence of the problem can be detected by looking at
3056 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
3057 smaller than the second (nr_unused) then dirty files will not
3058 be flushed automatically to disk. "sync; sleep 10" should be
3059 run several times on the node before unmounting it to update
3060 Lustre (this is also safe to run on nodes without this problem).
3062 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
3063 that can cause this same problem, so customers using that kernel
3064 also need to update the kernel in addition to Lustre. In order
3065 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
3067 It is normal that files written just before a client crash (less
3068 than 30s) may not yet have been flushed to disk, even for local
3072 Frequency : frequent on thin XT3 nodes
3074 Description: UUID collision on thin XT3 Linux nodes
3075 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
3076 randomly, since we relied on an insufficiently-seeded PRNG.
3081 Description: OSS hangs after "All ost request buffers busy"
3082 Details : A deadlock between quota and journal operations caused OSS
3083 hangs after printing "All ost request buffers busy."
3086 Frequency : always on liblustre builds
3088 Description: Cleanup compiler warnings on liblustre
3091 Frequency : always on liblustre builds on XT3
3093 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
3097 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
3098 > 32000 subdirectories
3099 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
3100 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
3101 subdirectory count crosses 32000. This will aid e2fsck to
3102 correctly handle more than 32000 subdirectories.
3105 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3107 Description: System hang on startup
3108 Details : This bug allowed the liblustre (e.g. catamount) client to
3109 return to the app before handling all startup RPCs. This
3110 could leave the node unresponsive to lustre network traffic
3111 and manifested as a server ptllnd timeout.
3113 Severity : enhancement
3115 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3116 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
3117 boolean to control whether to print peer debug info when a
3118 client's RPC times out.
3123 Description: make O_SYNC working on 2.6 kernels
3124 Details : 2.6 kernels use different method for mark pages for write,
3125 so need add a code to lustre for O_SYNC work.
3130 Description: Failure to close file and release space on NFS
3131 Details : Put inode details into lock acquired in ll_intent_file_open.
3132 Use mdc_intent_lock in ll_intent_open to properly
3133 detect all kind of errors unhandled by mdc_enqueue.
3138 Description: proc file read during shutdown sometimes raced obd removal,
3140 Details : Add lock to prevent obd access after proc file removal.
3143 Frequency : Only for files larger than 4GB on 32-bit clients.
3145 Description: improperly doing page alignment of locks
3146 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
3147 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
3152 Description: RPCs being resent when they shouldn't be
3153 Details : Some RPCs that should not be resent are being resent. This
3154 can cause inconsistencies in the RPC state machine. Do not
3155 resend such requests.
3158 Frequency : rare, only with NFS export
3160 Description: Crash on NFS re-export node
3161 Details : under very unusual load conditions an assertion is hit in
3162 ll_intent_file_open()
3165 Frequency : only if OST filesystem is corrupted
3167 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3168 Details : for a short time RPCs with bulk IO are in the replay list,
3169 but replay of bulk IOs is unimplemented. If the OST filesystem
3170 is corrupted due to disk cache incoherency and then replay is
3171 started it is possible to trip an assertion. Avoid putting
3172 committed RPCs into the replay list at all to avoid this issue.
3177 Description: large O_DIRECT requests fail under memory pressure/fragmentation
3178 Details : Large single O_DIRECT read and write calls can fail to allocate
3179 a sufficiently large buffer to process the request. In case of
3180 allocation failure the allocation is retried with a smaller
3181 buffer and broken into smaller requests.
3183 Severity : enhancement
3185 Description: Add -o localflock option to simulate old noflock behaviour.
3186 Details : This will achieve local-only flock/fcntl locks coherentness.
3191 Description: versioning check is incomplete
3192 Details : Checking the version difference of client vs. server, report
3193 error if the gap is too big.
3198 Description: add support PG_writeback bit
3199 Details : add support for PG_writeback bit for Lustre, for more carefull
3200 work with page cache in 2.6 kernel. This also fix some deadlocks
3201 and remove hack for work O_SYNC with 2.6 kernel.
3203 Severity : enhancement
3205 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3206 Details : The uninit_groups feature works in conjunction with the kernel
3207 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3208 up the pass1 processing of e2fsck. This is a read-only feature
3209 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3210 mount filesystems that have had this feature enabled.
3212 Severity : enhancement
3214 Description: Improve multi-block allocation algorithm to avoid fragmentation
3215 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3216 allocation locality and avoid filesystem fragmentation.
3218 ------------------------------------------------------------------------------
3220 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
3222 * Support for kernels:
3223 2.6.9-42.0.3.EL (RHEL 4)
3224 2.6.5-7.276 (SLES 9)
3225 2.4.21-47.0.1.EL (RHEL 3)
3226 2.6.12.6 vanilla (kernel.org)
3227 2.6.16.21-0.8 (SLES10)
3228 * Recommended e2fsprogs version: 1.39.cfs2-0
3230 * The backwards-compatible /proc/sys/portals symlink has been removed
3231 in this release. Before upgrading, please ensure that you change
3232 any configuration scripts or /etc/sysctl.conf files that access
3233 /proc/sys/portals/* or sysctl portals.* to use the corresponding
3234 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
3235 in advance of the upgrade on any system running Lustre 1.4.6 or
3236 newer, since /proc/sys/lnet was added in that version.
3237 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3241 Frequency : only when quota is used
3243 Description: avoid scanning export list for quota master
3244 Details : Change the algorithms to avoid scanning export list in order
3245 to improve the efficiency.
3248 Frequency : MDS failover only, very rarely
3250 Description: "went back in time" messages on mds failover
3251 Details : The greatest transno may be lost when the current operation
3252 finishes with an error (transno==0) and the client's last_rcvd
3253 record is over-written. Save the greatest transno in the
3254 mds_last_transno for this case.
3257 Frequency : always for specific kernels and striping counts
3259 Description: client may get "Matching packet too big" without ACL support
3260 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
3261 when trying to access files in certain configurations. The
3262 clients should in fact be denied when mounting because they do
3263 not understand ACLs.
3266 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3268 Description: many clients connecting with IO in progress causes connect timeouts
3269 Details : Avoid synchronous journal commits to avoid delays caused by many
3270 clients connecting/disconnecting when bulk IO is in progress.
3271 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3272 OST_IO_PORTAL to avoid delays behind potentially many pending
3276 Frequency : occasionally with multiple writers to a single file
3278 Description: shared writes to file may result in wrong size reported by stat()
3279 Details : Allow growing of kms when extent lock is cancelled
3282 Frequency : always with random mmap IO to multi-striped file
3284 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3285 Details : Only if the hole is at the end of OST object so that kms is too
3286 small. Fix is to increase kms accordingly in ll_nopage.
3289 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3291 Description: writes to a missing object would leak memory on the OST
3292 Details : If there is an inconsistency between the MDS and OST filesystems,
3293 such that the MDS references an object that doesn't exist, writes
3294 to that object will leak memory due to incorrect cleanup in the
3295 error handling path, eventually running out of memory on the OST.
3300 Description: Creating too long symlink causes lustre errors
3301 Details : Check symlink and name lengths before sending requests to MDS.
3304 Frequency : only if flock is enabled (not on by default)
3306 Description: posix locks not released on fd closure on 2.6.9+
3307 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3308 kernels, this caused such locks not to be released on fd close and
3309 then assertions on fs unmount about still used locks.
3312 Frequency : MDS failover only, very rarely
3314 Description: clients may get ASSERTION(granted_lock != NULL)
3315 Details : When request was taking a long time, and a client was resending
3316 a getattr by name lock request. The were multiple lock requests
3317 with the same client lock handle and
3318 mds_getattr_name->fixup_handle_for_resent_request found one of the
3319 lock handles but later failed with ASSERTION(granted_lock != NULL).
3324 Description: handle->h_buffer_credits > 0, assertion failure
3325 Details : h_buffer_credits is zero after truncate, causing assertion
3326 failure. This patch extends the transaction or creates a new
3330 Frequency : NFS re-export or patchless client
3331 Bugzilla : 11179, 10796
3332 Description: Crash on NFS re-export node (__d_move)
3333 Details : We do not want to hash the dentry if we don't have a lock.
3334 But if this dentry is later used in d_move, we'd hit uninitialised
3335 list head d_hash, so we just do this to init d_hash field but
3336 leave dentry unhashed.
3339 Frequency : NFS re-export or patchless client
3341 Description: NFS exports has problem with symbolic link
3342 Details : lustre client didn't properly install dentry when re-exported
3343 to NFS or running patchless client.
3346 Frequency : NFS re-export or patchless client
3348 Description: Various nfs/patchless fixes.
3349 Details : fixes reuse disconected alias for lookup process - this fixes
3350 warning "find_exported_dentry: npd != pd",
3351 fix permission error with open files at nfs.
3352 fix apply umask when do revalidate.
3355 Frequency : occasional
3357 Description: Crash on NFS re-export node
3358 Details : calling clear_page() on the wrong pointer triggered oops in
3359 generic_mapping_read().
3362 Frequency : rarely, using O_DIRECT IO
3364 Description: unaligned directio crashes client with LASSERT
3365 Details : check for unaligned buffers before trying any requests.
3368 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3370 Description: stale data returned from RAID cache
3371 Details : If only a small amount of IO is done to the RAID device before
3372 reading it again it is possible to get stale data from the RAID
3373 cache instead of reading it from disk.
3376 Frequency : always for sles10 kernel
3378 Description: sles10 support
3379 Details : ll_follow_link: compile fixes and using of nd_set_link
3380 under newer kernels.
3383 Frequency : depends on arch, kernel and compiler version, always on sles10
3386 Description: recursive or deep enough symlinks cause stack overflow
3387 Details : getting rid of large stack-allocated variable in
3391 Frequency : depends on hardware
3393 Description: lustre write performance loss in the SLES10 kernel
3394 Details : the performance loss is caused by using of write barriers in the
3395 ext3 code. The SLES10 kernel turns barrier support on by
3396 default. The fix is to undo that change for ldiskfs.
3398 ------------------------------------------------------------------------------
3400 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3402 * Support for kernels:
3403 2.6.9-42.0.3EL (RHEL 4)
3404 2.6.5-7.276 (SLES 9)
3405 2.4.21-47.0.1.EL (RHEL 3)
3406 2.6.12.6 vanilla (kernel.org)
3410 Frequency : quota enabled and large files being deleted
3412 Description: releasing more than 4GB of quota at once hangs OST
3413 Details : If a user deletes more than 4GB of files on a single OST it
3414 will cause the OST to spin in an infinite loop. Release
3415 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3420 Description: statfs data retrieved from /proc may be stale or zero
3421 Details : When reading per-device statfs data from /proc, in the
3422 {kbytes,files}_{total,free,avail} files, it may appear
3423 as zero or be out of date.
3426 Frequency : systems with MD RAID1 external journal devices
3428 Description: lconf's call to blkid is confused by RAID1 journal devices
3429 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3430 of returning all block devices that match the journal UUID.
3433 Frequency : always, for aggregate stripe size over 4GB
3435 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3436 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3437 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3440 Frequency : always if "lfs find" used on a local file/directory
3442 Description: "lfs find" segfaults if used on a local file/directory
3443 Details : The case where a directory component was not specified wasn't
3444 handled correctly. Handle this properly.
3447 Frequency : always on ppc64
3449 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3450 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3451 perform endianness conversion.
3454 Frequency : rarely (truncate to non-zero file size after write under load)
3455 Bugzilla : 10730, 10687
3456 Description: Files padded with zeros to next 4K multiple
3457 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3458 it is possible that files that are truncated to a non-zero size
3459 immediately after being written are filled with zero bytes beyond
3460 the truncated size. No file data is lost.
3462 Severity : enhancement
3464 Description: Allow recovery/failover for liblustre clients.
3465 Details : liblustre clients were unaware of failover configurations until
3468 Severity : enhancement
3470 Description: user file locks should fail when not mounting with flock option
3471 Details : Set up an error-returning stub in ll_file_operations.lock field
3472 to prevent incorrect behaviour when client is mounted without
3473 flock option. Also, set up properly f_op->flock field for
3477 Frequency : always on ia64
3479 Description: "lfs df" loops on printing out MDS statfs information
3480 Details : The obd_ioctl_data was not initialized and in some systems
3481 this caused a failure during the ioctl that did not return
3482 an error. Initialize the struct and return an error on failure.
3485 Frequency : SLES 9 only
3487 Description: Error of copying files with lustre special EAs as root
3488 Details : Client side always return success for setxattr call for lustre
3489 special xattr (currently only "trusted.lov").
3492 Frequency : rarely on clusters with both ia64+i386 clients
3494 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3495 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3496 clients are doing concurrent writes to the same file it is
3497 possible that the ia64 clients may LASSERT because the OST
3498 extent locks are not PAGE_SIZE aligned. Ensure that grown
3499 locks are always aligned on the request boundary.
3502 Frequency : specific use, occasional
3504 Description: Overwriting in use executable truncates on-disk binary image
3505 Details : If one node attempts to overwrite an executable in use by
3506 another node, we now correctly return ETXTBSY instead of
3507 truncating the file.
3509 Severity : enhancement
3511 Description: Async OSC create to avoid the blocking unnecessarily.
3512 Details : If a OST has no remain object, system will block on the creating
3513 when need to create a new object on this OST. Now, ways use
3514 pre-created objects when available, instead of blocking on an
3515 empty osc while others are not empty. If we must block, we block
3516 for the shortest possible period of time.
3521 Description: chmod on Lustre root is propagated to other clients
3522 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3523 invalid by the follow_mount time.
3526 Frequency : liblustre clients only
3528 Description: Race in 'instant cancel' lock handling could lead to such locks
3529 never to be granted in case of SMP MDS
3530 Details : Do not destroy not yet granted but cbpending locks in
3534 Frequency : replay/resend of open
3536 Description: non null lock assetion failure in mds_intent_policy
3537 Details : Trying to replay/resend lockless open requests resulted in
3538 mds_open() returning 0 with no lock. Now it sets a flag if
3539 a lock is going to be returned.
3541 Severity : enhancement
3543 Description: Checksum enhancements
3544 Details : New checksum enhancements allow for resending RPCs that failed
3547 Severity : enhancement
3549 Description: Tunables on number of dirty pages in cacche
3550 Details : Allow to set limit on number of dirty pages cached.
3555 Description: client crash on unmount - lock still has references
3556 Details : In some error handling cases it was possible to leak a lock
3557 reference on a client while accessing a file. This was not
3558 harmful to the client during operation, but would cause the
3559 client to crash when the filesystem is unmounted.
3562 Frequency : specific case, rare
3564 Description: ETXTBSY on mds though file not in use by client
3565 Details : ETXTBSY is no longer incorrectly returned when attempting to
3566 chmod or chown a directory that the user previously tried to
3567 execute or a currently-executing binary.
3570 Frequency : extremely rare except on liblustre-based clients
3572 Description: Lustre space not freed when files are deleted
3573 Details : Clean up open-unlinked files after client eviction. Previously
3574 the unlink was skipped and the files remained as orphans.
3579 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3580 Details : In some cases it was possible to send a blocking callback to
3581 a client doing a glimpse, even though that client didn't get
3582 a lock granted. When the glimpse lock is cancelled on the OST
3583 the freed lock is left on the waiting list and corrupted the list.
3586 Frequency : all core dumps
3588 Description: Broke core dumps to lustre
3589 Details : Negative dentry may be unhashed if parent does not have UPDATE
3590 lock, but some callers, e.g. do_coredump, expect dentry to be
3591 hashed after successful create, hash it in ll_create_it.
3593 ------------------------------------------------------------------------------
3595 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3597 * Support for kernels:
3598 2.6.9-42.0.2.EL (RHEL 4)
3599 2.6.5-7.276 (SLES 9)
3600 2.4.21-40.EL (RHEL 3)
3601 2.6.12.6 vanilla (kernel.org)
3605 Frequency : always on RHEL 3
3607 Description: Number of open files grows over time
3608 Details : The number of open files grows over time, whether or not
3609 Lustre is started. This was due to a filp leak introduced
3610 by one of our kernel patches.
3612 ------------------------------------------------------------------------------
3614 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3616 * Support for kernels:
3617 2.6.9-42.EL (RHEL 4)
3618 2.6.5-7.276 (SLES 9)
3619 2.4.21-40.EL (RHEL 3)
3620 2.6.12.6 vanilla (kernel.org)
3625 Bugzilla : 5719, 9635, 9792, 9684
3626 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3627 Details : If a server is under heavy load and cannot reply to new
3628 connection requests before the client resends the (re)connect,
3629 the connection handling code can behave badly if two service
3630 threads are concurrently handing separate (re)connections from
3631 the same client. Add better locking to the connection handling
3632 code, and ensure that only a single connection will be processed
3633 for a given client UUID, even if the lock is dropped.
3635 Severity : enhancement
3637 Description: add TCP zero-copy support to kernel
3638 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3639 sends if the hardware supports scatter-gather and checksumming.
3640 This allows socklnd to do client-write and server-read more
3641 efficiently and reduce CPU utilization from skbuf copying.
3644 Frequency : only if NFS exporting from client
3646 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3647 Details : If setting attributes on a file created under NFS that had
3648 never been opened it would be possible to oops the client
3649 if the file had no objects.
3652 Frequency : always for liblustre
3654 Description: liblustre client does MDS+OSTs setattr RPC for each write
3655 Details : When doing a write from a liblustre client, the client
3656 incorrectly issued an RPC to the MDS and each OST the file was
3657 striped over in order to update the timestamps. When writing
3658 with small chunks and many clients this could overwhelm the MDS
3659 with RPCs. In all cases it would slow down the write because
3660 these RPCs are unnecessary.
3662 Severity : enhancement
3664 Description: allow number of MDS service threads to be changed at module load
3665 Details : It is now possible to change the number of MDS service threads
3666 running. Adding "options mds mds_num_threads={N}" to the MDS's
3667 /etc/modprobe.conf will set the number of threads for the next
3668 time Lustre is restarted (assuming the "mds" module is also
3669 reloaded at that time). The default number of threads will
3670 stay the same, 32 for most systems.
3675 Description: OST crash if filesystem is unformatted or corrupt
3676 Details : If an OST is started on a device that has never been formatted
3677 or if the filesystem is corrupt and cannot even mount then the
3678 error handling cleanup routines would dereference a NULL pointer.
3683 Description: NULL pointer deref in llap_from_page.
3684 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3685 incorrect) mapping if the page was truncated/reclaimed while it was
3686 searched for. Check for this condition and skip such pages when
3687 doing readahead. Introduce extra check to llap_from_page() to
3688 verify page->mapping->host is non-NULL (so page is not anonymous).
3691 Frequency : Sometimes when using sys_sendfile
3693 Description: "page not covered by a lock" warnings from ll_readpage
3694 Details : sendfile called ll_readpage without right page locks present.
3695 Now we introduced ll_file_sendfile that does necessary locking
3696 around call to generic_file_sendfile() much like we do in
3700 Frequency : with certain MDS communication failures at client mount time
3702 Description: NULL pointer deref after failed client mount
3703 Details : a client connection request may delayed by the network layer
3704 and not be sent until after the PTLRPC layer has timed out the
3705 request. If the client fails the mount immediately it will try
3706 to clean up before the network times out the request. Add a
3707 reference from the request import to the obd device and delay
3708 the cleanup until the network drops the request.
3711 Frequency : occasionally during client (re)connect
3713 Description: assertion failure during client (re)connect
3714 Details : processing a client connection request may be delayed by the
3715 client or server longer than the client connect timeout. This
3716 causes the client to resend the connection request. If the
3717 original connection request is replied in this interval, the
3718 client may trip an assertion failure in ptlrpc_connect_interpret()
3719 which thought it would be the only running connect process.
3722 Frequency : only with obd_echo servers and clients that are rebooted
3724 Description: kernel BUG accessing uninitialized data structure
3725 Details : When running an obd_echo server it did not start the ping_evictor
3726 thread, and when a client was evicted an uninitialized data
3727 structure was accessed. Start the ping_evictor in the RPC
3728 service startup instead of the OBD startup.
3730 Severity : enhancement
3731 Bugzilla : 10193 (patchless)
3732 Description: Remove dependency on various unexported kernel interfaces.
3733 Details : No longer need reparent_to_init, exit_mm, exit_files,
3734 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3737 Frequency : rare (only users of deprecated and unsupported LDAP config)
3739 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3740 Details : LDAP apparently contains 'lustreName' attributes instead of
3741 'name'. A simple remapping of the name is sufficient.
3744 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3746 Description: waiting_locks_callback trips kernel BUG if client is evicted
3747 Details : Running with the dump_on_timeout debug flag turned on makes
3748 it possible that the waiting_locks_callback() can try to dump
3749 the Lustre kernel debug logs from an interrupt handler. Defer
3750 this log dumping to the expired_lock_main() thread.
3752 Severity : enhancement
3754 Description: Support NFS exporting on 2.6 kernels.
3755 Details : Implement non-rawops metadata methods for NFS server to use without
3756 changing NFS server code.
3759 Frequency : very rare (synthetic metadata workload only)
3761 Description: two racing renames might cause an MDS thread to deadlock
3762 Details : Running the "racer" program may cause one MDS thread to rename
3763 a file from being the source of a rename to being the target of
3764 a rename at exactly the same time that another thread is doing
3765 so, and the second thread has already enqueued these locks after
3766 doing a lookup of the target and is trying to relock them in
3767 order. Ensure that we don't try to re-lock the same resource.
3770 Frequency : only very large systems with liblustre clients
3772 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3773 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3774 all OSTs in parallel. This allows the eviction of stale liblustre
3775 clients to proceed much faster than if they were done in series,
3776 and also offers similar improvements for other set_info RPCs.
3781 Description: excessive CPU usage during initial read phase on client
3782 Details : During the initial read phase on a client, it would agressively
3783 retry readahead on the file, consuming too much CPU and impacting
3784 performance (since 1.4.5.8). Improve the readahead algorithm
3785 to avoid this, and also improve some other common cases (read
3786 of small files in particular, where "small" is files smaller than
3787 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3792 Description: MDS crash when receiving packet with unknown intent.
3793 Details : Do not LBUG in unknown intent case, just return -EFAULT
3795 Severity : enhancement
3796 Bugzilla : 9293, 9385
3797 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3798 Details : Do not serialize getattr (non-intent version) and statfs.
3801 Frequency : occasional, when OST network is overloaded/intermittent
3803 Description: client evicted by OST after bulk IO timeout
3804 Details : If a client sends a bulk IO request (read or write) the OST
3805 may evict the client if it is unresposive to its data GET/PUT
3806 request. This is incorrect if the network is overloaded (takes
3807 too long to transfer the RPC data) or dropped the OST GET/PUT
3808 request. There is no need to evict the client at all, since
3809 the pinger and/or lock callbacks will handle this, and the
3810 client can restart the bulk request.
3813 Frequency : Always when mmapping file with no objects
3815 Description: client crashes when mmapping file with no objects
3816 Details : Check that we actually have objects in a file before doing any
3817 operations on objects in ll_vm_open, ll_vm_close and
3823 Description: Request leak when working with deleted CWD
3824 Details : Introduce advanced request refcount tracking for requests
3825 referenced from lustre intent.
3827 Severity : Enhancement
3829 Description: Cache open file handles on client.
3830 Details : MDS now will return special lock along with openhandle, if
3831 requested and client is allowed to hold openhandle, even if unused,
3832 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3833 closing files for every read/write openration.
3835 Severity : Enhancement
3837 Description: Cache open negative dentries on client when possible.
3838 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3839 negative dentries on lock revocation.
3844 Description: Remounting a client read-only wasn't possible with a zconf mount
3845 Details : It wasn't possible to remount a client read-only with llmount.
3847 Severity : enhancement
3848 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3849 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3850 MPICH is included to add Lustre-specific ADIO interfaces.
3851 This is based closely on the UFS ADIO layer and only differs
3852 in file creation, in order to allow the OST striping to be set.
3853 This is user-contributed code and not supported by CFS.
3858 Description: extended inode attributes (immutable, append-only) work improperly
3859 when 2.4 and 2.6 kernels are used on client/server or vice versa
3860 Details : Introduce kernel-independent values for these flags.
3862 Severity : enhancement
3865 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3866 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3867 granularity. Now, Fractional MB granularity is be supported,
3868 this is very useful for low memory system.
3870 Severity : enhancement
3872 Description: Getattr by fid
3873 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3874 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3875 have locks and avoids getattr rpc requests for it on every stat.
3878 Frequency : Always, for filesystems larger than 2TB
3880 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3881 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3882 filesystems larger than 2TB to be mounted with mballoc enabled.
3885 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3887 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3888 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3889 filesystems larger than 2TB is mounted then the kernel will
3890 silently corrupt the start of the filesystem. CONFIG_LBD is
3891 enabled for all CFS-supported kernels, but the possibility of
3892 this happening with a modified kernel config exists.
3894 Severity : enhancement
3896 Description: add client O_DIRECT support for 2.6 kernels
3897 Details : It is now possible to do O_DIRECT reads and writes to files
3898 in the Lustre client mountpoint on 2.6 kernel clients.
3900 Severity : enhancement
3902 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3903 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3904 parallel, not waiting for response from every OST before sending
3905 a rpc to the next OST.
3910 Description: setattr vs write race when updating file timestamps
3911 Details : Client processes that update a file timestamp into the past
3912 right after writing to the file (e.g. tar) it is possible that
3913 the updated file modification time can be reset to the current
3914 time due to a race between processing the setattr and write RPC.
3916 Severity : enhancement
3918 Description: Bring 'lfs find' closer in line with regular Linux find.
3919 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3920 -print0 options and obtains all the needed info through the lustre
3923 Severity : enhancement
3925 Description: support up to 1024 configured devices on one node
3926 Details : change obd_dev array from statically allocated to dynamically
3927 allocated structs as they are first used to reduce memory usage
3932 Description: Flush dirty partially truncated pages during truncate
3933 Details : Immediatelly flush partially truncated pages in filter_setattr,
3934 this way we completely avoid having any pages in page cache on OST
3935 and can retire ugly workarounds during writes to flush such pages.
3940 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3941 Details : protect inode from truncation within vfs_unlink() context
3942 just take a reference before calling vfs_unlink() and release it
3943 when parent's i_sem is free.
3948 Description: last_id value checked outside lock on OST caused LASSERT failure
3949 Details : If there were multiple MDS->OST object precreate requests in
3950 flight, it was possible that the OST's last object id was checked
3951 outside a lock and incorrectly tripped an assertion. Move checks
3952 inside locks, and discard old precreate requests.
3955 Frequency : always, if extents are used on OSTs
3957 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3958 Details : OSTs using the extents format would not zero the high 16 bits of
3959 the index physical block number. This is not a problem for any
3960 OST filesystems smaller than 16TB, and no kernels support ext3
3961 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3962 new/modified files) and can be fixed for existing filesystems
3963 with e2fsprogs-1.39-cfs1.
3968 Description: import connection selection may be incorrect if timer wraps
3969 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3970 import connections to be ignored if the 32-bit jiffies counter
3971 wraps. Use a 64-bit jiffies counter.
3974 Frequency : very large clusters immediately after boot
3976 Description: LNET request buffers exhausted under heavy short-term load
3977 Details : If a large number of client requests are generated on a service
3978 that has previously never seen so many requests it is possible
3979 that the request buffer growth cannot keep up with the spike in
3980 demand. Instead of dropping incoming requests, they are held in
3981 the LND until the RPC service can accept more requests.
3984 Frequency : Sometimes during replay
3986 Description: Assertion failure in ll_local_open after replay.
3987 Details : If replay happened on an open request reply before we were able
3988 to set replay handler, reply will become not swabbed tripping the
3989 assertion in ll_local_open. Now we set the handler right after
3990 recognising of open request
3993 Frequency : very rare
3995 Description: kernel reports "badness in vsnprintf"
3996 Details : Reading from the "recovery_status" /proc file in small chunks
3997 may cause a negative length in lprocfs_obd_rd_recovery_status()
3998 call to vsnprintf() (which is otherwise harmless). Exit early
3999 if there is no more space in the output buffer.
4001 Severity : enhancement
4003 Description: clear OBD RPC statistics by writing to them
4004 Details : It is now possible to clear the OBD RPC statistics by writing
4005 to the "stats" file.
4010 Description: Client mtime is not the same on different clients after utimes
4011 Details : In some cases, the client was using the utimes() syscall on
4012 a file cached on another node. The clients now validate the
4013 ctime from the MDS + OSTs to determine which one is right.
4018 Description: Inability to activate failout mode
4019 Details : lconf script incorrectly assumed that in pythong string's numeric
4020 value is used in comparisons.
4023 Frequency : always with multiple stripes per file
4025 Description: Inefficient object allocation for mutli-stripe files
4026 Details : When selecting which OSTs to stripe files over, for files with
4027 a stripe count that divides evenly into the number of OSTs,
4028 the MDS is always picking the same starting OST for each file.
4029 Return the OST selection heuristic to the original design.
4034 Description: mount failures may take full timeout to return an error
4035 Details : Under some heavy load conditions it is possible that a
4036 failed mount can wait for the full obd_timeout interval,
4037 possibly several minutes, before reporting an error.
4038 Instead return an error as soon as the status is known.
4040 Frequency : quota enabled and large files being deleted
4042 Description: releasing more than 4GB of quota at once hangs OST
4043 Details : If a user deletes more than 4GB of files on a single OST it
4044 will cause the OST to spin in an infinite loop. Release
4045 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
4050 Description: statfs data retrieved from /proc may be stale or zero
4051 Details : When reading per-device statfs data from /proc, in the
4052 {kbytes,files}_{total,free,avail} files, it may appear
4053 as zero or be out of date.
4056 Frequency : systems with MD RAID1 external journal devices
4058 Description: lconf's call to blkid is confused by RAID1 journal devices
4059 Details : Use the "blkid -l" flag to locate the MD RAID device instead
4060 of returning all block devices that match the journal UUID.
4063 Frequency : always, for aggregate stripe size over 4GB
4065 Description: assertion fails when trying to use 4GB stripe size
4066 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
4067 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
4070 Frequency : always on ppc64
4072 Description: the first write on an ext3 filesystem with mballoc got stuck
4073 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
4074 perform endianness conversion.
4076 ------------------------------------------------------------------------------
4078 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
4080 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
4081 INTEROPERATE with older versions automatically. Please read the
4082 user documentation before upgrading any part of a live system.
4083 * WARNING: Lustre networking configuration changes are required with
4084 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
4087 * Support for kernels:
4088 2.6.9-22.0.2.EL (RHEL 4)
4089 2.6.5-7.244 (SLES 9)
4090 2.6.12.6 vanilla (kernel.org)
4093 Severity : enhancement
4094 Bugzilla : 7981/8208
4095 Description: Introduced Lustre Networking (LNET)
4096 Details : LNET is new networking infrastructure for Lustre, it includes
4097 a reorganized network configuration mode (see the user
4098 documentation for full details) as well as support for routing
4099 between different network fabrics. Lustre Networking Devices
4100 (LNDs) for the supported network fabrics have also been
4101 created for this new infrastructure.
4103 Severity : enhancement
4104 Description: Introduced Access control lists
4105 Details : clients can set ACLs on files and directories in order to have
4106 more fine-grained permissions than the standard Unix UGO+RWX.
4107 The MDS must be started with the "-o acl" mount option.
4109 Severity : enhancement
4110 Description: Introduced filesystem quotas
4111 Details : Administrators may now establish per-user quotas on the
4114 Severity : enhancement
4116 Description: Configuration change for the XT3
4117 The PTLLND is now used to run Lustre over Portals on the XT3
4118 The configure option(s) --with-cray-portals are no longer used.
4119 Rather --with-portals=<path-to-portals-includes> is used to
4120 enable building on the XT3. In addition to enable XT3 specific
4121 features the option --enable-cray-xt3 must be used.
4126 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
4127 Details : A race between allocating a new llcd and re-getting the llcd_lock
4128 allowed another thread to grab newly-allocated llcd.
4130 Severity : enhancement
4132 Description: 2.6 OST async journal commit and locking fix to improve performance
4133 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
4134 2.6 kernels are now async as they already were in 2.4 kernels so
4135 that they can commit concurrently with the network bulk transfer.
4136 For block-allocated files the filter allocation semaphore is held
4137 to avoid filesystem fragmentation during allocation. BKL lock
4138 removed for 2.6 xattr operations where it is no longer needed.
4143 Description: lconf incorrectly determined whether two IP networks could talk
4144 Details : In some more complicated routing and multiple-network
4145 configurations, lconf will avoid trying to make a network
4146 connection to a disjoint part of the IP space. It was doing the
4147 math incorrectly for one set of cases.
4152 Description: Fix for potential infinite loop processing records in an llog.
4153 Details : If an llog record is corrupted/zeroed, it is possible to loop
4154 forever in llog_process(). Validate the llog record length
4155 and skip the remainder of the block on error.
4158 Frequency : occasional (liblustre only)
4160 Description: liblustre could not open files whose last component is a symlink
4161 Details : sysio_path_walk() would incorrectly pass the open intent to
4162 intermediate path components.
4165 Frequency : rare (liblustre only with non-standard tuning)
4166 Bugzilla : 7201 (7350)
4167 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
4168 Details : llu_lookup_finish_locks() tries to set lock data on a lock
4169 after it has been released, only do this for referenced locks
4171 Severity : enhancement
4173 Description: specifying an (invalid) directory default stripe_size of -1
4174 would reset the directory default striping
4175 Details : stripe_size -1 was used internally to signal directory stripe
4176 removal, now use "all default" to signal dir stripe removal
4177 as a directory striping of "all default" is not useful
4180 Frequency : common for large clusters running liblustre clients
4182 Description: doing an ls when liblustre clients are running is slow
4183 Details : sending a glimpse AST to a liblustre client waits for every AST
4184 to time out, as liblustre clients will not respond. Since they
4185 cannot cache data we refresh the OST lock LVB from disk instead.
4187 Severity : enhancement
4189 Description: doing an ls at the same time as file IO can be slow
4190 Details : enqueue and other "small" requests can be blocked behind many
4191 large IO requests. Create a new OST IO portal for non-IO
4192 requests so they can be processed faster.
4195 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
4197 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
4198 created file with mode 000
4199 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
4200 allow an ftruncate() as a non-root user to a file with mode 000.
4201 Lustre now respects this flag to disable mode checks when
4202 truncating a file owned by the user
4205 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
4208 Description: Revoking locks from clients that went dead or catatonic might take
4210 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
4211 cancellation of such locks instant on servers without waiting for
4212 any reply from clients. Clients drops these locks when cancel
4213 notification from server is received without replying.
4216 Frequency : liblustre-only, when liblustre client dies or becomes busy
4218 Description: Doing ls on Linux clients can take a long time with active
4220 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
4221 granting such locks to it in the first place if possible. Locks
4222 are taken by proxy on the OST during the read or write and
4223 dropped immediately afterward. Add connect flags handling, do
4224 not grant locks to liblustre clients for glimpse ASTs.
4226 Severity : enhancement
4228 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
4229 Details : Existing read-ahead algorithm is tuned for the case of streamlined
4230 sequential reads and behaves badly with applications doing random
4231 reads. Improve it by reading ahead at least read region, and
4232 avoiding excessive large RPC for small reads.
4234 Severity : enhancement
4236 Description: Creating more than 1000 files for a single job may cause a load
4237 imbalance on the OSTs if there are also a large number of OSTs.
4238 Details : qos_prep_create() uses an OST index reseed value that is an
4239 even multiple of the number of available OSTs so that if the
4240 reseed happens in the middle of the object allocation it will
4241 still utilize the OSTs as uniformly as possible.
4246 Description: OST or MDS may oops in ping_evictor_main()
4247 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
4248 but doesn't restart at beginning of obd_exports_timed list
4251 Severity : enhancement
4253 Description: improve by-nid export eviction on the MDS and OST
4254 Details : allow multiple exports with the same NID to be evicted at one
4255 time without re-searching the exports list.
4258 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
4260 Description: MDS may oops in groups_free()
4261 Details : in rare race conditions a newly allocated group_info struct is
4262 freed again, and this can be NULL. The 2.4 compatibility code
4263 for groups_free() checked for a NULL pointer, but 2.6 did not.
4266 Frequency : common for liblustre clients doing little filesystem IO
4267 Bugzilla : 9352, 7313
4268 Description: server may evict liblustre clients accessing contended locks
4269 Details : if a client is granted a lock or receives a completion AST
4270 with a blocking AST already set it would not reply to the AST
4271 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4274 Frequency : lfs setstripe, only systems with more than 160 OSTs
4276 Description: unable to set striping with a starting offset beyond OST 160
4277 Details : llapi_create_file() incorrectly limited the starting stripe
4278 index to the maximum single-file stripe count.
4281 Frequency : LDAP users only
4283 Description: lconf did not handle in-kernel recovery with LDAP properly
4284 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4286 Severity : enhancement
4288 Description: bind OST threads to NUMA nodes to improve performance
4289 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4290 node and do their allocations there to localize memory access
4292 Severity : enhancement
4294 Description: llmount can determine client NID directly from Myrinet (GM)
4295 Details : the client NID code from gmnalnid was moved directly into
4296 llmount, removing the need to use this or specifying the
4297 client NID explicitly when mounting GM clients with zeroconf
4300 Frequency : if client is started with down MDS
4302 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4303 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4304 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4305 ptlrpc_import_delayed_req() because we didn't check if the
4306 request was interrupted, and we also didn't break out of the
4307 event loop if there was no timeout
4312 Description: data loss during non-page-aligned writes to a single file from
4313 both multiple nodes and multiple threads on one node at same time
4314 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4315 inconsistency led to false short-reads, that were cached and later
4316 used by ->prepare_write() to fill in partially written page,
4317 leading to data loss.
4320 Frequency : always, if lconf --abort_recovery used
4322 Description: lconf --abort_recovery fails with 'Operation not supported'
4323 Details : lconf was attempting to abort recovery on the MDT device and not
4326 Severity : enhancement
4328 Description: remove cleanup logs
4329 Details : replace lconf-generated cleanup logs with lustre internal
4330 cleanup routines. Eliminates the need for client-cleanup and
4333 Severity : enhancement
4335 Description: add support for EAs (user and system) on lustre filesystems
4336 Details : it is now possible to store extended attributes in the Lustre
4337 client filesystem, and with the user_xattr mount option it
4338 is possible to allow users to store EAs on their files also
4340 Severity : enhancement
4342 Description: Add possibility (config option) to show minimal available OST free
4344 Details : When compiled with --enable-mindf configure option, statfs(2)
4345 (and so, df) will return least minimal free space available from
4346 all OSTs as amount of free space on FS, instead of summary of
4347 free spaces of all OSTs.
4349 Severity : enhancement
4351 Description: do not expand extent locks acquired on OST-side
4352 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4353 by server: they are not cached anyway.
4356 Frequency : when mmap is used/binaries executed from Lustre
4358 Description: Unmmap pages before throwing them away from read cache.
4359 Details : llap_shrink cache now attempts to unmap pages before discarding
4360 them (if unmapping failed - do not discard). SLES9 kernel has
4361 extra checks that trigger if this unmapping is not done first.
4366 Description: lconf didn't resolve symlinks before checking to see whether a
4367 given mountpoint was already in use
4370 Frequency : when migrating failover services
4371 Bugzilla : 6395, 9514
4372 Description: When migrating a subset of services from a node (e.g. failback
4373 from a failover service node) the remaining services would
4374 time out and evict clients.
4375 Details : lconf --force (implied by --failover) sets the global obd_timeout
4376 to 5 seconds in order to quickly disconnect, but this caused
4377 other RPCs to time out too quickly. Do not change the global
4378 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4380 Severity : enhancement
4381 Frequency : if MDS is started with down OST
4382 Bugzilla : 9439,5706
4383 Description: Allow startup/shutdown of an MDS without depending on the
4384 availability of the OSTs.
4385 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4386 Add appropriate locking and lov-osc refcounts for safe
4387 cleaning. Add osc abort_inflight calls in case the
4388 synchronize never started.
4391 Frequency : occasional (Cray XT3 only)
4393 Description: root not authorized to access files in CRAY_PORTALS environment
4394 Details : The client process capabilities were not honoured on the MDS in
4395 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4396 been accessed by an authorized user then root was able to access
4397 the file on the local client also. The root user capabilities
4398 are now allowed on the MDS, as this environment has secure UID.
4401 Frequency : occasional
4403 Description: ldiskfs "too long searching" message happens too often
4404 Details : A debugging message (otherwise harmless) prints too often on
4405 the OST console. This has been reduced to only happen when
4406 there are fragmentation problems on the filesystem.
4411 Description: Division by zero in statfs when all OSCs are inactive
4412 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4413 lov_statfs divides by value returned by lov_get_stripecnt().
4417 Bugzilla : 9489, 3273
4418 Description: First write from each client to each OST was only 4kB in size,
4419 to initialize client writeback cache, which caused sub-optimal
4420 RPCs and poor layout on disk for the first writen file.
4421 Details : Clients now request an initial cache grant at (re)connect time
4422 and so that they can start streaming writes to the cache right
4423 away and always do full-sized RPCs if there is enough data.
4424 If the OST is rebooted the client also re-establishes its grant
4425 so that client cached writes will be honoured under the grant.
4430 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4431 Details : Now I/O RPCs go to different portal number and (presumably) fast
4432 lock requests (and glimses) and other RPCs get their own service
4433 threads pool that should be able to service those RPCs
4436 Severity : enhancement
4438 Description: Ability to exchange lustre version between client and servers and
4439 issue warnings at client side if client is too old. Also for
4440 liblustre clients there is ability to refuse connection of too old
4442 Details : New 'version' field is added to connect data structure that is
4443 filled with version info. That info is later checked by server and
4447 Frequency : rare, liblustre only.
4448 Bugzilla : 9296, 9581
4449 Description: Two simultaneous writes from liblustre at offset within same page
4450 might proceed at the same time overwriting eachother with stale
4452 Details : I/O lock withing llu_file_prwv was released too early, before data
4453 actually was hitting the wire. Extended lock-holding time until
4454 server acknowledges receiving data.
4457 Frequency : extremely rare. Never observed in practice.
4459 Description: avoid generating lustre_handle cookie of 0.
4460 Details : class_handle_hash() generates handle cookies by incrementing
4461 global counter, and can hit 0 occasionaly (this is unlikely, but
4462 not impossible, because initial value of cookie counter is
4463 selected randonly). Value of 0 is used as a sentinel meaning
4464 "unassigned handle" --- avoid it. Also coalesce two critical
4465 sections in this function into one.
4467 Severity : enhancement
4469 Description: allow liblustre clients to delegate truncate locking to OST
4470 Details : To avoid overhead of locking, liblustre client instructs OST to
4471 take extent lock in ost_punch() on client's behalf. New connection
4472 flag is added to handle backward compatibility.
4474 Severity : enhancement
4475 Bugzilla : 4928, 7341, 9758
4476 Description: allow number of OST service threads to be specified
4477 Details : a module parameter allows the number of OST service threads
4478 to be specified via "options ost ost_num_threads={N}" in the
4479 OSS's /etc/modules.conf or /etc/modprobe.conf.
4483 Bugzilla : 6146, 9635, 9895
4484 Description: servers crash with bad pointer in target_handle_connect()
4485 Details : In rare cases when a client is reconnecting it was possible that
4486 the connection request was the last reference for that export.
4487 We would temporarily drop the export reference and get a new
4488 one, but this may have been the last reference and the export
4489 was just destroyed. Get new reference before dropping old one.
4491 Severity : enhancement
4492 Frequency : if client is started with failover MDS
4494 Description: Allow multiple MDS hostnames in the mount command
4495 Details : Try to read the configuration from all specified MDS
4496 hostnames during a client mount in case the "primary"
4499 Severity : enhancement
4501 Description: Stop sending data to evicted clients as soon as possible.
4502 Details : Check if the client we are about to send or are sending data to
4503 was evicted already. (Check is done every second of waiting,
4504 for which l_wait_event interface was extended to allow checking
4505 of exit condition at specified intervals).
4508 Frequency : rare, normally only when NFS exporting is done from client
4510 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4512 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4515 Frequency : always, if srand() is called before liblustre initialization
4517 Description: Liblustre uses system PRNG disturbing its usage by user application
4518 Details : Introduce internal to lustre fast and high-quality PRNG for
4519 lustre usage and make liblustre and some other places in generic
4520 lustre code to use it.
4522 Severity : enhancement
4523 Bugzilla : 9477, 9557, 9870
4524 Description: Verify that the MDS configuration logs are updated when xml is
4525 Details : Check if the .xml configuration logs are newer than the config
4526 logs stored on the MDS and report an error if this is the case.
4527 Request --write-conf, or allow starting with --old_conf.
4529 Severity : enhancement
4531 Description: Handle symlinks in the path when checking if Lustre is mounted.
4532 Details : Resolve intermediate symlinks when checking if a client has
4533 mounted a filesystem to avoid duplicate client mounts.
4538 Description: lconf can hit an error exception but still return success.
4539 Details : The lconf command catches the Command error exception at the top
4540 level script context and will exit with the associated exit
4541 status, but doesn't ensure that this exit status is non-zero.
4546 Description: failure of ptlrpc thread startup can cause oops
4547 Details : Starting a ptlrpc service thread can fail if there are a large
4548 number of threads or the server memory is very fragmented.
4549 Handle this without oopsing.
4552 Frequency : always, only if liblustre and non-default acceptor port was used
4554 Description: liblustre cannot connect to servers with non-default acceptor port
4555 Details : tcpnal_set_default_params() was not called and was therefore
4556 ignoring the environment varaible TCPNAL_PORT, as well as other
4557 TCPNAL_ environment variables
4562 Description: two objects could be created on the same OST for a single file
4563 Details : If an OST is down, in some cases it was possible to create two
4564 objects on a single OST for a single file. No problems other
4565 than potential performance impact and spurious error messages.
4569 Bugzilla : 5681, 9562
4570 Description: Client may oops in ll_unhash_aliases
4571 Details : Client dcache may become inconsistent in race condition.
4572 In some cases "getcwd" can fail if the current directory is
4578 Description: Inode refcounting problems in NFS export code
4579 Details : link_raw functions used to call d_instantiate without obtaining
4580 extra inode reference first.
4584 Bugzilla : 9942, 9903
4585 Description: Referencing freed requests leading to crash, memleaks with NFS.
4586 Details : We used to require that call to ll_revalidate_it was always
4587 followed by ll_lookup_it. Also with revalidate_special() it is
4588 possible to call ll_revalidate_it() twice for the same dentry
4589 even if first occurence returned success. This fix changes semantic
4590 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4591 reference on a request referred from the intent.
4592 ll_intent_release() then releases such a request.
4595 Frequency : rare, normally benchmark loads only
4597 Description: unlinked inodes were kept in memory on the client
4598 Details : If a client is repeatedly creating and unlinking files it
4599 can accumulate a lot of stale inodes in the inode slab cache.
4600 If there is no other client load running this can cause the
4601 client node to run out of memory. Instead flush old inodes
4602 from client cache that have the same inode number as a new inode.
4605 Frequency : SLES9 2.6.5 kernel and long filenames only
4606 Bugzilla : 9969, 10379
4607 Description: utime reports stale NFS file handle
4608 Details : SLES9 uses out-of-dentry names in some cases, which confused
4609 the lustre dentry revalidation. Change it to always use the
4613 Frequency : rare, unless heavy write-truncate concurrency is continuous
4614 Bugzilla : 4180, 6984, 7171, 9963, 9331
4615 Description: OST becomes very slow and/or deadlocked during object unlink
4616 Details : filter_destroy() was holding onto the parent directory lock
4617 while truncating+unlinking objects. For very large objects this
4618 may block other threads for a long time and slow overall OST
4619 responsiveness. It may also be possible to get a lock ordering
4620 deadlock in this case, or run out of journal credits because of
4621 the combined truncate+unlink. Solution is to do object truncate
4622 first in one transaction without parent lock, and then do the
4623 final unlink in a new transaction with the parent lock. This
4624 reduces the lock hold time dramatically.
4627 Frequency : rare, 2.4 kernels only
4629 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4630 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4631 before unmounting their filesystem in order to avoid deadlock.
4632 The kernel_locked() function in 2.4 kernels only checks whether
4633 the kernel lock is held, not whether it is this process that is
4634 holding it as 2.6 kernels do.
4639 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4640 Details : The client ptlrpc code may be trying to reconnect to a down
4641 server before a previous connection attempt has timed out.
4642 Increase the reconnect interval to be longer than the connection
4643 timeout interval to avoid sending duplicate connections to
4647 Frequency : echo_client brw_test command
4649 Description: fix echo_client to work with OST preallocated code
4650 Details : OST preallocation code (5137) didn't take echo_client IO path
4651 into account: echo_client calls filter methods outside of any
4652 OST thread and, hence, there is no per-thread preallocated
4653 pages and buffers to use. Solution: hijack pga pages for IO. As
4654 a byproduct, this avoids unnecessary data copying.
4658 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4659 Description: Client can oops in mdc_commit_close() after open replay
4660 Details : It was possible for the MDS to return an open request with no
4661 transaction number in mds_finish_transno() if the client was
4662 evicted, but without actually returning an error. Clients
4663 would later try to replay that open and may trip an assertion
4664 Simplify the client close codepath, and always return an error
4665 from the MDS in case the open is not successful.
4668 Frequency : rare, 2.6 OSTs only
4670 Description: OST may deadlock under high load on fragmented files
4671 Details : If there was a heavy load and highly-fragmented OST filesystems
4672 it was possible to have all the OST threads deadlock waiting on
4673 allocation of biovecs, because the biovecs were not released
4674 until the entire RPC IO was completed. Instead, release biovecs
4675 as soon as they are complete to ensure forward IO progress.
4677 Severity : enhancement
4679 Description: Support for specifying external journal device at mount
4680 Details : If an OST or MDS device is formatted with an external journal
4681 device, this device major/minor is stored in the ext3 superblock
4682 and may not be valid for failover. Allow detecting and
4683 specifying the external journal at mount time.
4688 Description: Mounting an MDS with pending unlinked files may cause oops
4689 Details : target_finish_recovery() calls mds_postrecov() which returned
4690 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4691 considers this an error and immediately begins cleaning up the
4692 lov, just after starting the mds_lov process
4694 Severity : enhancement
4696 Description: Implement 'lfs df' to report actual free space on per-OST basis
4697 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4698 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4699 usage of objects; '-h' to report in human readable format.
4701 ------------------------------------------------------------------------------
4703 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4710 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4711 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4712 at startup during orphan unlinking. Instead, skip buddy update
4713 before setup, as it will be regenerated after recovery is complete.
4718 Description: If an OST is inactive, its locks might reference stale inodes.
4719 Details : lov_change_cbdata() must iterate over all namespaces, even if
4720 they are inactive to clear inode references from the lock.
4722 Severity : enhancement
4723 Frequency : occasional, if non-standard max_dirty_mb used
4725 Description: Client will block write RPCs if not enough grant
4726 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4727 then the client will block writes while waiting for another RPC
4728 to complete instead of consuming its dirty limit. With change
4729 we get improved performance when max_dirty_mb is small.
4731 Severity : enhancement
4732 Bugzilla : 3389, 6253
4733 Description: Add support for supplementary groups on the MDS.
4734 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4735 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4736 lookups for user supplementary groups into a cache.
4740 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4741 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4742 for liblustre O_EXCL is enforced by MDS.
4746 Description: readdir never returns NULL in liblustre.
4747 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4748 dentry in struct dirent.
4752 Frequency : liblustre only, depends on application IO pattern
4753 Description: liblustre clients evicted if not contacting servers
4754 Details : Don't put liblustre clients into the ping_evictor list, so
4755 they will not be evicted by the pinger ever.
4757 Severity : enhancement
4759 Description: Add ability to evict clients by NID from MDS.
4760 Details : By echoing "nid:$NID" string into
4761 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4762 $NID would be instantly evicted from this MDS and from all active
4763 OSTs connected to it.
4767 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4768 Details : lookup_it_finish() used to query file size from OSTs that was not
4773 Description: service threads change working directory to that of init
4774 Details : Starting lustre service threads may pin the working directory
4775 of the parent thread, making that filesystem busy. Threads
4776 now change to the working directory of init to avoid this.
4780 Frequency : during shutdown only
4781 Description: shutdown with a failed MDS or OST can cause unmount to hang
4782 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4787 Frequency : 2.6 only
4788 Description: chmod/chown may include an extra supplementary group
4789 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4790 supplementary group and if none is specified this is used.
4793 Bugzilla : 5479 (6816)
4795 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4796 Details : If lookup is in progress on a file that is unlinked we might try
4797 to revalidate the inode and fail in revalidate after lookup is
4798 complete and ll_file_open() enqueues the open again but
4799 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4802 Frequency : always, if lconf --abort_recovery used
4804 Description: lconf --abort_recovery fails with 'Operation not supported'
4805 Details : lconf was attempting to abort recovery on the MDT device and not
4808 ------------------------------------------------------------------------------
4810 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4815 Frequency : rare (only unsupported configurations with a node running as an
4817 Bugzilla : 6514, 5137
4818 Description: Mounting a Lustre file system on a node running as an OST could
4820 Details : OSTs now preallocates memory needed to write out data at
4821 startup, instead of when needed, to avoid having to
4822 allocate memory in possibly low memory situations.
4823 Specifically, if the file system is mounted on on OST,
4824 memory pressure could force it to try to write out data,
4825 which it needed to allocate memory to do. Due to the low
4826 memory, it would be unable to do so and the node would
4827 become unresponsive.
4829 Severity : enhancement
4831 Description: Addition of lconf --service command line option
4832 Details : lconf now accepts a '--service <arg>' option, which is
4833 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4835 Severity : enhancement
4837 Description: Failover mode is now the default for OSTs.
4838 Details : By default, OSTs will now run in failover mode. To return to
4839 the old behaviour, add '--failout' to the lmc line for OSTs.
4841 Severity : enhancement
4843 Description: Health checks are now provided for MDS and OSTs
4844 Details : Additional detailed health check information on MSD and OSTs
4845 is now provided through the procfs health_check value.
4848 Frequency : occasional, depends on IO load
4850 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4851 after numerous create/delete cycles
4852 Details : The ext3 inode allocation policy would not allocate new inodes
4853 very well on the OSTs because there are no new directories
4854 being created. Instead we look for groups with free space if
4855 the parent directories are nearly full.
4860 Description: Network or server problems during mount may cause partially
4861 mounted clients instead of returning an error.
4862 Details : The config llog parsing code may overwrite the error return
4863 code during mount error handling, returning success instead
4869 Description: MDS can fail to allocate large reply buffers
4870 Details : After long uptimes the MDS can fail to allocate large reply
4871 buffers (e.g. zconf client mount config records) due to memory
4872 fragmentation or consumption by the buffer cache. Preallocate
4873 some large reply buffers so that these replies can be sent even
4874 under memory pressure.
4878 Frequency : rare (liblustre)
4879 Description: fsx running with liblustre complained that using truncate() to
4880 extend the file doesn't work. This patch corrects that issue.
4881 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4882 ATTR_SIZE and lsm use in llu_setattr_raw.
4886 Frequency : rare, only 2.6 kernels
4887 Description: Unusual file access patterns on the MDS may result in inode
4888 data being lost in very rare circumstances.
4889 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4890 code in the 2.6 kernel caused the inode and/or EA data not to
4891 be read from disk, causing single-file corruption.
4895 Frequency : rare, only 2.6 filesystems using extents
4896 Description: Heavy concurrent write and delete load may cause data corruption.
4897 Details : It was possible under high-load situations to have an extent
4898 metadata block in the block device cache from a just-unlinked
4899 file overwrite a newly-allocated data block. We now unmap any
4900 metadata buffers that alias just-allocated data blocks.
4904 Frequency : filesystems with default stripe_count larger than 77
4905 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4906 Details : lconf specifies an inode size of 4096 bytes when the default
4907 stripe_count is larger than 77. This conflicts with the default
4908 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4909 this case to avoid pinning too much memory for large EAs.
4911 ------------------------------------------------------------------------------
4913 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4918 Frequency : rare (extremely heavy IO load with hundreds of clients)
4920 Description: Client is evicted, gets IO error writing to file
4921 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4922 caused clients to be evicted for AST timeouts. The fixes in
4923 bug 5192 mean we no longer need to have such short AST timeouts
4924 so ldlm_timeout has been increased.
4927 Frequency : occasional during --force or --failover shutdown under load
4928 Bugzilla : 5949, 4834
4929 Description: Server oops/LBUG if stopped with --force or --failover under load
4930 Details : a collection of import/export refcount and cleanup ordering
4931 issues fixed for safer force cleanup
4934 Frequency : only filesystems larger than 120 OSTs
4935 Bugzilla : 5990, 6223
4936 Description: lfs getstripe would oops on a very large filesystem
4937 Details : lov_getconfig used kfree on vmalloc'd memory
4940 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4941 Bugzilla : 6242, 6243
4942 Description: reading from files that had been truncated to a non-zero size
4943 but never opened returned no data
4944 Details : ll_file_read() reads zeros from no-object files to EOF
4949 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4950 Details : There's little harm in aborting MDS/OSS recovery and letting it
4951 try again, so I removed the LASSERT and return an error instead.
4953 Severity : enhancement
4955 Description: New debugging infrastructure for tracking down data corruption
4956 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4957 (b) cover more of the client-side code path, and (c) try to narrow
4958 down where problems occurred
4962 Bugzilla : 3819, 4364, 4397, 6313
4963 Description: Racing close and eviction MDS could cause assertion in mds_close
4964 Details : It was possible to get multiple mfd references during close and
4965 client eviction, leading to one thread referencing a freed mfd.
4967 Severity: : enhancement
4968 Bugzilla : 3262, 6359
4969 Description: Attempts to reconnect to servers are now more aggressive.
4970 Details : This builds on the enhanced upcall-less recovery that was added
4971 in 1.4.2. When trying to reconnect to servers, clients will
4972 now try each server in the failover group every 10 seconds. By
4973 default, clients would previously try one server every 25 seconds.
4978 Description: After recovery, certain operations trigger a failed
4979 assertion on a client.
4980 Details : Failing over an mds, using lconf -d --failover, while a
4981 client was doing a readdir() call would cause the client to
4982 LBUG after recovery completed and the readdir() was resent.
4984 Severity : enhancement
4986 Description: Default groups are now added by lconf
4987 Details : You can now run lconf --group <servicename> without having to
4988 manually add groups with lmc.
4991 Frequency : occasional
4993 Description: Nodes with an elan id of 0 trigger a failed assertion
4996 Frequency : always when accessing e.g. tty/console device nodes
4998 Description: tty and some other devices nodes cannot be used on lustre
4999 Details : file's private_data field is used by device data and lustre
5000 values in there got lost. New field was added to struct file to
5001 store fs-specific private data.
5004 Frequency : when exporting Lustre via NFS
5006 Description: NFSD failed occasionally when looking up a path component
5007 Details : NFSD is looking up ".." which was broken in ext3 directories
5008 that had grown large enough to become hashed.
5011 Frequency : Clusters with multiple interfaces not on the same subnet
5013 Description: Nodes will repeatedly try to reconnect to an interface which it
5014 cannot reach and report an error to the log.
5015 Details : Extra peer list entries will be created by lconf with some peers
5016 unreachable. lconf now validates the peer before adding it.
5019 Frequency : Only if a default stripe is set on the filesystem root.
5021 Description: Setting a default stripe on the filesystem root prevented the
5022 filesystem from being remounted.
5023 Details : The client was sending extra request flags in the root getattr
5024 request and did not allocate a reply buffer for the dir EA.
5027 Frequency : occasional, higher if lots of files are accessed by one client
5028 Bugzilla : 6159, 6097
5029 Description: Client trips assertion regarding lsm mismatch/magic
5030 Details : While revalidating inodes the VFS looks up inodes with ifind()
5031 and in rare cases can find an inode that is being freed.
5032 The ll_test_inode() code will free the lsm during ifind()
5033 when it finds an existing inode and then the VFS later attaches
5034 this free lsm to a new inode.
5038 Bugzilla : 6422, 7030
5039 Description: MDS deadlock between mkdir and client eviction
5040 Details : Creating a new file via mkdir or mknod (starting a transaction
5041 and getting the ns lock) can deadlock with client eviction
5042 (gets ns lock and trying to finish a synchronous transaction).
5045 Frequency : occasional
5046 Description: While starting a server, the fsfilt_ext3 module could not be
5048 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
5049 kernels. Previously, lconf would still use the ext3 name
5050 when trying to load modules. Now, it will correctly use
5051 ext3 on 2.4 and ldiskfs on 2.6.
5053 Severity : enhancement
5054 Description: The default stripe count has been changed to 1
5055 Details : The interpretation of the default stripe count (0, to lfs
5056 or lmc) has been changed to mean striping across a single
5057 OST, rather than all available. For general usage we have
5058 found a stripe count of 1 or 2 works best.
5060 Severity : enhancement
5061 Description: Add support for compiling against Cray portals.
5062 Details : Conditional compiling for some areas that are different
5066 Frequency : occasional
5067 Bugzilla : 6409, 6834
5068 Description: Creating files with an explicit stripe count may lead to
5069 a failed assertion on the MDS
5070 Details : If some OSTs are full or unavailable, creating files may
5071 trigger a failed assertion on the MDS. Now, Lustre will
5072 try to use other servers or return an error to the
5076 Frequency : occasional
5078 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
5079 caused lock timeout during readahead (since 1.4.2).
5080 Details : Processes doing readahead might match a lock that hasn't been
5081 granted yet if there are overlapping and conflicting lock
5082 requests. The readahead process waits on ungranted lock
5083 (original lock is CBPENDING), while OST waits for that process
5084 to cancel CBPENDING read lock and eventually evicts client.
5086 Severity : enhancement
5088 Description: Initial enabling of flock support for clients
5089 Details : Implements fcntl advisory locking and file status functions.
5090 This feature is provided as an optional mount flag (default
5091 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
5092 locking are implemented yet, and those that are are not guaranteed
5093 to be completely correct in production environments.
5094 mount -t lustre -o [flock|noflock] ...
5097 Frequency : occasional
5099 Description: OSTs running 2.4 kernels but with extents enabled might trip an
5100 assertion in the ext3 JBD (journaling) layer.
5101 Details : The b_committed_data struct is protected by the big kernel lock
5102 in 2.4 kernels, serializing journal_commit_transaction() and
5103 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
5104 access to this struct. In 2.6 kernels there is finer grained
5105 locking to improve SMP performance of the JBD layer.
5109 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
5114 Description: merge and cleanup kernel patches.
5117 -----------------------------------------------------------------------------
5119 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
5121 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
5122 versions of Lustre. Please update all servers and clients to
5123 version 1.4.2 or later at the same time. You must also run
5124 "lconf --write-conf {config}.xml" on the MDS while it is stopped
5125 to update the configuration logs.
5127 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
5128 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
5129 - swab llog records of type '0' so we get proper header size/idx (5861)
5130 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
5131 - fix rename of one directory over another leaking an inode (5953)
5132 - avoid SetPageDirty on 2.6 (5981)
5133 - don't re-add just-being-destroyed locks to the waiting list (5653)
5134 - when creating new directories, inherit the parent's custom
5135 striping settings if present parent (3048)
5136 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
5137 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
5138 - don't hold client locks on temporary worklist from l_lru (5666)
5139 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
5140 - automatically evict dead clients (5921)
5141 - Update file size properly in create+truncate+fstat case (6196)
5142 - Do not unhash mountpoint dentries, do not allow removal of
5144 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
5145 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
5146 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
5147 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
5148 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
5149 - Correctly handle reads of files with no objects (6243)
5150 - lctl recover will also mark a device active if deactivate used (5933)
5152 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
5153 - allow --write-conf on an MDS with different nettype than client (5619)
5154 - don't write config llogs to MDS for mounts not from that MDS (5617)
5155 - lconf should create multiple TCP connections from a client (5201)
5156 - init scripts are now turned off by default; run chkconfig --on
5157 lustre and chkconfig --on lustrefs to use them
5158 - upcalls are no longer needed for clients to recover to failover
5160 - add --abort-recovery option to lconf to abort recovery on device
5162 - add support for an arbitrary number of OSTs (3026)
5163 - Quota support protocol changes.
5164 - forward compatibility changes to wire structs (6007)
5165 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
5166 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
5167 - improved "lustre status" script
5168 - initialize blocksize for non-regular files (6062)
5169 - added --disable-server and --disable-client configure options (5782)
5170 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
5171 - Vanilla 2.4.29 support
5172 - increase maximum number of obd devices to 520 (6242)
5173 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
5174 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
5175 SLES 9 2.6.5 kernels
5176 - limit stripes per file to 160 (the maximum EA size) (6093)
5178 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5181 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
5182 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
5183 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
5184 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
5185 - small ext3 extents cleanups and fixes (5733)
5186 - improved mballoc code, several small races and bugs fixed (5733, 5638)
5187 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
5188 - avoid needless client->OST connect, fix handle mismatch (5317)
5189 - fix DLM error path that led to out-of-sync client, long delays (5779)
5190 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
5191 - fix several locking issues related to i_size (5492,5624,5654,5672)
5192 - don't move pending lock onto export if it is already evicted (5683)
5193 - fix kernel oops when creating .foo in unlinked directory (5548)
5194 - fix deadlock in obdfilter statistics vs. object create (5811)
5195 - use time_{before,after} to avoid timer jiffies wrap (5882)
5196 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
5197 - Do not leak request if server was not able to process it (5154)
5198 - If mds_open unable to find parent dir, make that negative lookup(5154)
5199 - don't create new directories with extent-mapping (5909, 5936)
5201 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
5202 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
5203 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
5204 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
5205 - support for RHEL4 kernel on i686 (5773)
5206 - provide error messages when incompatible logs are encountered (5898)
5208 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
5209 * version 1.4.0.10 (1.4.1 release candidate 1)
5211 - don't keep a lock reference when lock is not granted (4238)
5212 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5213 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
5214 - drop import inflight refcount on signal_completed_replay error (5255)
5215 - unlock page after async write error during send (3677)
5216 - handle missing objects in filter_preprw_read properly (5265)
5217 - no transno return for symlink open, don't save no-trasno open (3440)
5218 - don't try to complete elan receive that already failed (4012)
5219 - free RPC server reply state on error (5406)
5220 - clean up thread from ptlrpc_start_thread() on error (5160)
5221 - readahead could read extra page into cache that wasn't ejected (5388)
5222 - prevent races in class_attach/setup/cleanup/detach (5260)
5223 - don't dereference de->d_inode after l_dput of de (5458)
5224 - use "int" for stripe value returned from lock_to_stripe (5544)
5225 - mballoc allocation and error-checking fixes in 2.6 (5504)
5226 - block device patches to fix I/O request sizes in 2.6 (5482)
5227 - look up hostnames for IB nals (5602)
5228 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
5229 - don't start multiple acceptors for the same port (5277)
5230 - fix incorrect LASSERT in mds_getattr_name (5635)
5231 - export a proc file for general "ping" checking (5628)
5232 - fix "lfs check" to not block when the MDS is down (5628)
5234 - service request history (4965)
5235 - put {ll,lov,osc}_async_page structs in a single slab (4699)
5236 - create an "evict_client" /proc entry on OSTs, like the MDS has
5237 - fix mount usage message, return errors per mount(8) (5168)
5238 - change grep [] to grep "[]" in tests so they work in more UMLs
5239 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
5240 - remove ancient LOV_MAGIC_V0 EA support (5047)
5241 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
5242 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
5243 - allow readahead allocations to fail when low on memory (5383)
5244 - mmap locking landed again, after considerable improvement (2828)
5245 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
5247 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
5250 - send OST transaction number in read/write reply to free req (4966)
5251 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5252 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5253 - the watchdog thread now runs as interruptible (5246)
5254 - flock/lockf fixes (but it's still disabled, pending 5135)
5255 - don't use EXT3 constants in llite code (5094)
5256 - memory shortage at startup could cause assertion (5176)
5258 - reorganization of lov code
5259 - single portals codebase
5261 - add extents/mballoc support (5025)
5262 - direct I/O reads in the obdfilter (4048)
5263 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5265 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5268 - send OST transaction number in read/write reply to free req (4966)
5269 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5270 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5271 - the watchdog thread now runs as interruptible (5246)
5272 - handle missing objects in filter_preprw_read properly (5265)
5273 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5274 - drop import inflight refcount on signal_completed_replay error (5255)
5275 - unlock page after async write error during send (3677)
5276 - return original error code on reconstructed replies (3761)
5277 - no transno return for symlink open, don't save no-trasno open (3440)
5279 - add pid to ldlm debugging output (4922)
5280 - bump the watchdog timeouts -- we can't handle 30sec yet
5281 - extra debugging for orphan dentry/inode bug (5259)
5283 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5286 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5287 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5288 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5289 - fix NULL dereference in /proc/sys/portals/routes (4827)
5290 - allow failed mdc_close() operations to be interrupted (4561)
5291 - stop precreate on OST before MDS would time out on it (4778)
5292 - don't send partial-page writes before EOF from client (4410)
5293 - discard client grant for sub-page writes on large-page clients (4520)
5294 - don't free dentries not owned by NFS code, check generation (4806)
5295 - fix lsm leak if mds_create_objects() fails (4801)
5296 - limit debug_daemon file size, always print CERROR messages (4789)
5297 - use transno after validating reply (3892)
5298 - process timed out requests if import state changes (3754)
5299 - update mtime on OST during writes, return in glimpse (4829)
5300 - add mkfsoptions to LDAP (4679)
5301 - use ->max_readahead method instead of zapping global ra (5039)
5302 - don't interrupt __l_wait_event() during strace
5304 - add software watchdogs to catch hung threads quickly (4941)
5305 - make lustrefs init script start after nfs is mounted
5306 - fix CWARN/ERROR duplication (4930)
5307 - return async write errors to application if possible (2248)
5308 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5309 - print NAL number in %x format (4645)
5310 - update barely-supported suse-2.4.21-171 series (4842)
5311 - support for sles 9 %post scripts
5312 - support for building 2.6 kernel-source packages
5313 - support for sles km_* packages
5315 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5318 - ignore -ENOENT errors in osc_destroy (3639)
5319 - notify osc create thread that OSC is being cleaned up (4600)
5320 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5321 - reconstruct ost_handle() like mds_handle() (4657)
5322 - create a new thread to do import eviction to avoid deadlock (3969)
5323 - let lconf resolve symlinked-to devices (4629)
5324 - don't unlink "objects" from directory with default EA (4554)
5325 - hold socknal file ref over connect in case target is down (4394)
5326 - allow more than 32000 subdirectories in a single directory (3244)
5327 - fix blocks count for O_DIRECT writes (3751)
5328 - OST returns ENOSPC from object create when no space left (4539)
5329 - don't send truncate RPC if file size isn't changing (4410)
5330 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5331 - bind to privileged port in socknal and tcpnal (3689)
5333 - rate limit CERROR/CWARN console message to avoid overload (4519)
5334 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5335 - basic mmap support (3918)
5336 - kernel patch series update from b1_4 (4711)
5338 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5341 - avoid crash during MDS cleanup with OST shut down (2775)
5342 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5343 - don't use bad inodes on the MDS (3744)
5344 - dynamic object preallocation to improve recovery speed (4236)
5345 - don't hold spinlock over lock dumping or change debug flags (4401)
5346 - don't zero obd_dev when it is force cleaned (3651)
5347 - print grants to console if they go negative (4431)
5348 - "lctl deactivate" will stop automatic recovery attempts (3406)
5349 - look for existing locks in ldlm_handle_enqueue() (3764)
5350 - don't resolve lock handle twice in recovery avoiding race (4401)
5351 - revalidate should check working dir is a directory (4134)
5353 - don't always mark "slow" obdfilter messages as errors (4418)
5355 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5358 - don't close LustreDB during write_conf until it is done (3860)
5359 - fix typo in lconf for_each_profile (3821)
5360 - allow dumping logs from multiple threads at one time (3820)
5361 - don't allow multiple threads in OSC recovery (3812)
5362 - fix debug_size parameters (3864)
5363 - fix mds_postrecov to initialize import for llog ctxt (3121)
5364 - replace config semaphore with spinlock (3306)
5365 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5366 - don't allow enqueue to complete on a destroyed export (3822)
5367 - down write_lock before checking llog header bitmap (3825)
5368 - recover from lock replay timeout (3764)
5369 - up llog sem before sending rpc (3652)
5370 - reduce ns lock hold times when setting kms (3267)
5371 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5372 - fix NULL deref and obd_dev leak on setup error (3312)
5373 - replace some LBUG about llog ops with error handling (3841)
5374 - don't match INVALID dentries from d_lookup and spin (3784)
5375 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5376 - fix invalid assertion in ptlrpc_set_wait (3880)
5378 - add libwrap support for the TCP acceptor (3996)
5379 - add /proc/sys/portals/routes for non-root route listing (3994)
5380 - allow setting MDS UUID in .xml (2580)
5381 - print the stack of a process that LBUGs (4228)
5383 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5386 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5387 - make sure to unset replay flag from failed open requests (3440)
5388 - if default stripe count is 0, use OST count for inode size (3636)
5389 - update parent mtime/ctime on client for create/unlink (2611)
5390 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5391 - free recovery state on server during a forced cleanup (3571)
5392 - unregister_reply for resent reqs (3063)
5393 - loop back devices mounting and status check on 2.6 (3563)
5394 - fix resource-creation race that can provoke i_size == 0 (3513)
5395 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5396 - more debugging for page-accounting assertion (3746)
5397 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5398 - avoid deadlock after precreation failure (3758)
5399 - fix race and lock order deadlock in orphan handling (3450, 3750)
5400 - add validity checks when grabbing inodes from l_ast_data (3599)
5402 - add /proc/.../recovery_status to obdfilter (3428)
5403 - lightweight CDEBUG infrastructure, debug daemon (3668)
5404 - change default OSC RPC parameters to be better on small clusters
5405 - turn off OST read cache for files smaller than 32MB
5406 - install man pages and include them in rpms (3100)
5407 - add new init script for (un)mounting lustre filesystems (2593)
5408 - run chkconfig in %post for init scripts (3701)
5409 - drop scimac NAL (unmaintained)
5411 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5414 - clean kiobufs before and after use (3485)
5415 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5416 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5417 - change init script to fail more clearly if not run as root (1528)
5418 - allow clients to reconnect during replay (1742)
5419 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5420 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5421 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5422 - fix cond_resched (3554)
5423 - only evict unfinished clients after recovery (3515)
5424 - allow bulk resend, prevent data loss (3570)
5425 - dynamic ptlrpc request buffer allocation (2102)
5426 - don't allow unlinking open directory if it isn't empty (2904)
5427 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5428 - remove extraneous obd dereference causing LASSERT failure (3334)
5429 - don't use get_cycles() when creating temp. files on the mds (3156)
5430 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5431 - handle EEXIST for set-stripe, set proper directory name (3336)
5433 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5434 - fix ksocknal_fmb_callback() error messages (2918)
5436 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5439 - don't copy lvb into (possibly NULL) reply on error (2983)
5440 - don't deref dentry after dput, don't free lvb on error (2922)
5441 - use the kms to determine writeback rpc length (2947)
5442 - increment oti_logcookies when osc is inactive (2948)
5443 - update client's i_blocks count via lvb messages (2543)
5444 - handle intent open/close of special files properly (1557)
5445 - mount MDS with errors=remount-ro, like obdfilter (2009)
5446 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5447 - don't use cancelling-locks' kms values (2947)
5448 - use highest lock extent for kms, not last one (2925)
5449 - don't dereference ERR_PTR() dentry in error handling path (3107)
5450 - fix thread race in portals_debug_dumplog() (3122)
5451 - create lprocfs device entries at setup instead of at attach (1519)
5452 - common AST error handler, don't evict client on completion race (3145)
5453 - zero nameidata in detach_mnt in 2.6 (3118)
5454 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5455 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5456 - initialize RPC timeout timer earlier for 2.6 (3219)
5457 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5458 - print nal/nid for unknown nid (3258)
5459 - additional checks for oscc recovery before doing precreate (3284)
5460 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5461 - don't crash in mdc_close for bad permissions on open (3285)
5462 - zero i_rdev for non-device files (3147)
5463 - clear page->private before handing to FS, better assertion (3119)
5464 - tune the read pipeline (3236)
5465 - fix incorrect decref of invalidated dentry (2350)
5466 - provide read-ahead stats and refine rpc in flight stats (3328)
5467 - don't hold journal transaction open across create RPC (3313)
5468 - update atime on MDS at close time (3265)
5469 - close LDAP connection when recovering to avoid server load (3315)
5470 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5471 - don't leak open file on MDS after open resend (3325)
5472 - serialize filter_precreate and filter_destroy_precreated (3329)
5473 - loop device shouldn't call sync_dev() for nul device (3092)
5474 - clear page cache after eviction (2766)
5475 - resynchronize MDS->OST in background (2824)
5476 - refuse to mount the same filesystem twice on same mountpoint (3394)
5477 - allow llmount to create routes for mounting behind routers (3320)
5478 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5479 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5480 - fix rare problem with rename on htree directories (3417)
5482 - allow default OST striping configuration per directory (1414)
5483 - fix compilation for qswnal for 2.6 kernels (3125)
5484 - increase maximum number of MDS request buffers for large systems
5485 - change liblustreapi to be useful for external progs like lfsck (3098)
5486 - increase local configuration timeout for slow disks (3353)
5487 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5489 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5492 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5493 - don't overwrite extent policy data in reply if lock was blocked (2901)
5494 - drop filter export grants atomically with removal from device (2663)
5495 - del obd_self_export from work_list in class_disconnect_exports (2908)
5496 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5497 - swab reply message in mdc_close, other PPC fixes (2464)
5498 - fix destroying of named logs (2325)
5499 - overwrite old logs when running lconf --write_conf (2264)
5500 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5501 - fix race in target_handle_connect (2898)
5502 - mds_reint_create() should take same inode create lock (2926)
5503 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5504 - don't close files for self_export to avoid uninitialized obd (2936)
5505 - allow MDS with the same name as client node (2939)
5506 - hold dentry reference for closed log files for unlink (2325)
5507 - reserve space for all logs during transactions (2059)
5508 - don't evict page beyond end of stripe extent (2925)
5509 - don't oops on a deleted current working directory (2399)
5510 - handle hard links to targets without a parent properly (2517)
5511 - don't dereference NULL lock when racing during eviction (2867)
5512 - don't grow lock extents when lots of conflicting locks (2919)
5514 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5517 - account for cache space usage on clients to avoid data loss (974)
5518 - lfsck support in lustre kernel code (2349)
5519 - reduce journal credits needed for BRW writes (2370)
5520 - orphan handling to avoid losing space on client/server crashes
5521 - ptlrpcd can be blocked, stopping ALL progress (2477)
5522 - use lock value blocks to assist in proper KMS, faster stat (1021)
5523 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5524 - recovery for initial connections (2355)
5525 - fixes for mds_cleanup_orphans (1934)
5526 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5527 - block all file creations until orphan recovery completes (1901)
5528 - client remove rq_connection from request struct (2423)
5529 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5530 - recovery timer race (2670)
5531 - mdc_close recovey bug (2532)
5532 - ptlrpc cleanup bug (2710)
5533 - mds timeout on local locks (2588)
5534 - namespace lock held during RPCs (2431)
5535 - handle interrupted sync write properly (2503)
5536 - don't try to handle a message that hasn't been replied to (2699)
5537 - client assert failure during cleanup after abort recovery (2701)
5538 - leak mdc device after failed mount (2712)
5539 - ptlrpc_check_set allows timedout requests to complete (2714)
5540 - wait for inflight reqs when ptlrpcd finishes (2710)
5541 - make sure unregistered services are removed from the srv_list
5542 - reset bulk XID's when resending them (caught by 1138 test)
5543 - unregister_bulk after timeout
5544 - fix lconf error (2694)
5545 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5546 - readahead locks pages, leaves pending causing memory pressure (2673)
5547 - increase OST request buffers to 4096 on large machines (2729)
5548 - fix up permission of existing directories in simple_mkdir (2661)
5549 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5550 - don't assign transno to errored transactions (2742)
5551 - don't delete objects on OST if given a bogus objid from MDS (2751)
5552 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5553 - if rq_no_resend, then timeout request after recovery (2432)
5554 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5555 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5556 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5557 - MDS recovery shouldn't delete objects that it has given out (2730)
5558 - if enqueue arrives after completion, don't clobber LVB (2819)
5559 - don't unlock pages twice when trigger_group_io returns error (2814)
5560 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5561 - don't write pages to disk if there was an error (1450)
5562 - don't ping imports that have recovery disabled (2676)
5563 - take buffered bytes into account when balancing socknal conn (2817)
5564 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5565 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5566 - remove little-endian swabbing of llog records (1987)
5567 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5568 - retry reposting request buffers if they fail (1191)
5569 - grow extent at grant time to avoid granting a revoked lock (2809)
5570 - lock revoke doesn't evict page if covered by a second lock (2765)
5571 - disable VM readahead to avoid reading outside lock extents (2805)
5573 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5574 - updated kernel patches for hp-2.4.20 kernel (2681)
5576 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5579 - fix truncated write corruption (2366)
5580 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5582 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5583 - ensure len(uuid) < 37 in lmc (1171)
5584 - fix ia64 OOPS in llog_test (2255)
5585 - zero end of page at obdfilter for partial page writes (2648)
5586 - don't leave stale dentries around after renames (bug 2428)
5587 - fix timeouts when evicting a client with a single lock held (2642)
5588 - set deadline for the initial HELLO message to drain (2634)
5589 - print out dotted-quad IP addresses in the socknal (2302)
5591 - additional debugging for MDS client eviction problem (2443)
5592 - fix mkfsoptions support for osts (2603, 2604)
5594 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5597 - add series for the vanilla 2.6.0 kernel
5598 - add series for the vanilla 2.4.24 kernel
5599 - add series for a cray x86/64 UL kernel drop
5600 - fix xattr patches for the vanilla 2.4.19 series
5602 - generate true UUIDs in lmc (1171)
5603 - have portals stack dumping break in UML (2466)
5604 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5605 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5606 - wake callback waiting threads on client eviction (2460)
5607 - Add --ptldebug and --subsystem to lmc (1719)
5608 - update assertion to allow safe interrupt allocation
5609 - set rq_no_resend for cancel requests (2432)
5610 - recalculate ptlrpcd timeout after resend (2494)
5611 - call vfs_rmdir when removing pending directories (2368)
5612 - fix renaming a file to itself (2429)
5613 - lmc creates a default one-stripe lov (2454)
5614 - expand procfs space to handle large clusters (2326)
5615 - increase UML stack to avoid overflow
5616 - update lconf's list of debug and subsystem masks
5617 - fix lfs find --obd (2510)
5618 - /proc tunable for disabling filter read caching (2591)
5619 - stop rpm packages from altering slapd.conf (2301)
5620 - disable nagle in the socknal under 0conf (2578)
5621 - choose mds inode size based on stripe count (2572)
5622 - fix kernel-source rpm problems (2516)
5624 - add --disable-doc to avoid pdf generation (2421)
5625 - update documentation, tests, type-os, comments
5626 - avoid format warnings on ia64
5627 - remove the TOE NAL
5628 - tiny code cleanups by removing unused fields
5630 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5633 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5634 - remove the most blatant lies from BUILDING file (2371)
5635 - change default debug level to reasonable production setting
5636 - reduce client side cache size to reduce cache flush time
5637 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5638 - make TCP zerocopy and pinger support enabled by default (2476)
5639 - sync writes completed after process exits caused crashes (2319)
5640 - maintain correct mount count on the MDS (2356)
5641 - backout 1557, because 2316 wasn't really fixed
5642 - better file I/O statistics gathering in /proc
5643 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5644 - another kernel patch to fix zero-copy TCP function export
5645 - don't take duplicate lock when processing re-sent getattr (2420)
5646 - lctl uses obd_self_export instead of creating new conn (2353)
5647 - MDS/OST recovery case which requires object creation asserted (2425)
5648 - move lfs from /usr/sbin to /usr/bin in packages
5649 - fix race between mds_client_add and mds_client_free (2417)
5650 - use kmalloc instead of slabs in portals (2430)
5651 - don't create duplicate records when a failover MDS is present (2442)
5652 - remove unnecessary mount age check (2332)
5653 - don't remove directory inodes from locks prematurely (2451)
5654 - don't break if MDS service name is the same as hostname (2103)
5655 - fix races in client write RPC generation when cache full (2482)
5657 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5660 - remove now-unused request->rq_obd (278)
5661 - if an allocation fails, print out how much memory we've used (1933)
5662 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5663 - add the "configurable stack size" patch to most series files (1256)
5664 - ability to write large log records, for 100+ OST configs (2306)
5665 - fix NULL deref when filter_prep fails (2314)
5666 - fix operator precedence error in filter_sync
5667 - dynamic allocation of socknal TX descriptors (2315)
5668 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5669 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5670 - ensure that CWARN messages in Portals always get to the syslog
5671 - __init/__exit are not for prototype decls (ldlm_init/exit)
5672 - x86-64 compile warning fixes
5673 - fix gateway LMC keyword conflict (2318)
5674 - fix MDS lock inversions in getattr/reint paths (1844)
5675 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5676 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5677 - DLM race condition prevented some lock evictions (2328)
5678 - ENOMEM detection and retry on socknal sends (2230)
5679 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5680 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5681 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5682 - better error messages when a client is rejected during recovery (1505)
5683 - avoid cancelling locks which were never granted, after failure (2330)
5684 - fix i_sem/journal inversion in mds_client_add (2333)
5685 - fix truncate/getattr lock cycle deadlock (2334)
5686 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5687 - fix two rare exit paths which could leak an l_lock() ref (2321)
5688 - fencepost error in MDS/OST orphan recovery (2226)
5689 - make log record alignment 8 bytes (1988)
5690 - lstripe now fails when requested offset > ost_count (2237)
5691 - ensure that all kernel series have a complete list.h (1607)
5692 - fix crashes in special-file operations (2316)
5693 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5695 - allow configurable automake binary, for testing new versions
5696 - small update to the lfs documentation
5698 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5700 * fix negative export reference count in fsfilt_sync (2312)
5702 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5703 * release candidate 0.9.1
5705 - orphans are moved into the PENDING directory for possible recovery
5706 - replayed opens now open by fid for orphan/rename safety (1042)
5707 - last close of an orphan inode generates a transno (683)
5708 - chdir() and mount() now pin the directory entry (1020)
5709 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5710 - discard very old requests without processing them (1502)
5711 - remove some common, well-understood CERRORs (1505)
5712 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5713 - clear "grant" flags in OST replies until OST grant code lands (1644)
5714 - fix read performance by not clobbering i_blksize on client (1598)
5715 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5716 - make LRU size a /proc tunable, clears locks when reduced (707)
5717 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5718 - server locks take references on exports now (1558)
5719 - build fixes for 2.4.20-rh trees (1663)
5720 - return an error from lov_create if all OSCs are inactive (1751)
5721 - fix import levels when a reconnect happens without a timeout (1597)
5722 - exit early from mds_open if we get a lookup error (1749)
5723 - partial page read at EOF wouldn't wait for disk before sending (1642)
5724 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5725 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5726 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5727 - write/truncate lock inversion (1639)
5728 - Don't auto-load obdclass, portals modules during cleanup (1495)
5729 - fix timestamps from jumping to "now" (1763)
5730 - extra journal assertions (1648)
5731 - add an extra multiunlink test (1771)
5732 - fix read_record/write_record API (1776)
5733 - fix leak of offset_extent, possible incorrect i_size later (1772)
5734 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5735 - Debugging for the kqswnal_get_idle_tx problems (1820)
5736 - Allow recovery to be attempted multiple times (1536)
5737 - Write out MDS last_rcvd file after it is first created (1600)
5738 - Fix tx_descriptor leak in failed transmit situations (1827)
5739 - ext3 journaling fixes for assertion failure after IO error (1871)
5740 - class_export_put() on freed export after completion AST error (1896)
5741 - Fix revalidate looping in VFS (1322)
5742 - Don't access a freed export during MDS_REINT timeout (1521)
5743 - Add open-unlink recovery support on the MDS (1673,1764)
5744 - Return an error if no MDS data was read from last_rcvd (1946)
5745 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5746 - Don't setup a disk device that doesn't match exported UUID (317)
5747 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5748 - avoid committing NULL handle in force close
5749 - local.sh is now a one-stripe LOV configuration
5750 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5751 - fix ext3 htree duplicate directory entry corruption (1516)
5752 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5753 - update to new LOV EA format (2097)
5754 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5756 2003-06-15 Phil Schwan <phil@clusterfs.com>
5759 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5760 - per-import/export recovery handling (958, 931, 959)
5761 - multiple last-rcvd slots, for serving multiple FSes (949)
5762 - connections are again shared between multiple imp/exports (963, 964)
5763 - "umount -f" would hang if any requests needed to be sent (393, 978)
5764 - avoid pinning large req buffer by copying for queued messages (989)
5765 - add "uuid" to "lctl device" command to help upcalls (991)
5766 - "open" RPCs with transnos would confuse recovery counters (1037)
5767 - do proper endian conversion of all wire messages (288, 340, 891)
5768 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5769 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5770 - fix ptlrpc_connection leak in target_handle_connect (1174)
5771 - fix import refcounting bug in OST and MDS cleanup (1134)
5772 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5773 - fix very unlikely obd_types race condition (501)
5774 - remove osc_open hack for echo_client (1187)
5775 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5776 - a failure in read_inode2 leads to deadlock (1139)
5777 - cancel ack-locks as soon as transaction is committed (1072)
5778 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5779 - make sure to commitrw after any preprw to avoid deadlock (1162)
5780 - failing to execute a file in a lustre FS would lock inode (1203)
5781 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5782 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5783 - obd_uuid2tgt would walk off the end of the list (1255)
5784 - on IA64 the osc would give portals incorrect bulk size (1258)
5785 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5786 - fix lock inversion caused by new llite matching code (1282)
5787 - limit the number of dirty pages on a client to 10MB (1286)
5788 - timed out locks were not being corrected cancelled (1289)
5789 - fix O_DIRECT above 4GB on IA-32 (1292)
5790 * major user-visible changes
5791 - fail out/fail over policy now controlled by the upcall (993)
5793 - add OBD_PING to check server availability and failure (954)
5794 - lustre messages are now sent in sending host order (288, 340, 891)
5795 - add eadatalen to MDS getattr reply (340)
5796 - OST read replies may contain second buffer, with per-page status (593)
5798 2003-03-11 Phil Schwan <phil@clusterfs.com>
5801 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5802 - failed open()s could cause deadlock; fixed (867, 869)
5803 - stop cancelling OST locks when files are closed (481)
5804 - overlapping XID spaces caused network corruption (851, 853)
5805 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5806 - setattr_raw added, to do single-RPC, server-side setattrs
5807 - lmc/lconf syntax change for OST UUIDs
5808 - fix crashy race condition between ptlrpc_free_req and osc_close
5809 - don't use request in mdc_enqueue if we hit a timeout (889)
5810 - don't set the inode i_size for regular files from the MDS (896)
5811 - handle out of order completion AST (842)
5812 - don't LBUG if a lock request times out after receiving AST (913)
5813 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5814 - if a bad lock AST arrives, send an error instead of dropping entirely
5815 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5816 - fix leak in bulk IO when only partially completed (899, 900, 926)
5817 - fix O_DIRECT for ia64 (55)
5818 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5819 - C-z after timeout could hang a process forever; fixed (977)
5821 - client-side I/O cache (678, 924, 929, 941, 970)
5823 - READPAGE and SETATTRs which don't take server-side locks get
5826 2003-02-11 Phil Schwan <phil@clusterfs.com>
5829 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5830 - Fix fsfilt_extN_readpage() to read a full page of directory
5831 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5832 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5833 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5834 - Most intent operations were removed, in favour of a new RPC mode
5835 that does a single RPC to the server and bypasses most of the VFS
5836 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5837 - Aggressively cancel local locks on DLM servers
5838 - mds_reint_unlink sends EA to the client if it's the last nlink.
5839 client uses that EA to unlink OST objects.
5840 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5841 - recursive symlinks were fixed (439)
5842 - fixed NULL deref in DEBUG_REQ
5843 - filter_update_lastobjid no longer calls sync, which annoyed extN
5844 - fixed multi-client small-writes to a single file problem (445)
5845 - fixed mtime updates during file writes (607)
5846 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5847 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5848 - fixed Linux OST size reporting problem (444, 656)
5849 - OST now updates object mtime with writes or setattr (607, 619)
5850 - client verifies file size before zeroing page past EOF (445)
5851 - OST now writes last allocated objid to disk with allocation (108)
5852 - LOV on echo now works (409)
5854 - mds_reint_unlink sends a new buffer, with the EA included. this
5855 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5856 set if a regular file was being unlinked, and it was the last link
5857 - use PtlGet from the target for bulk writes (315)
5858 - OST now updates object mtime with writes or setattr (607, 619)
5859 - LDLM now has a grant-time callback to revalidate locked items, if
5861 - Many MDS operations were reorganized to combat race conditions
5863 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5864 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5866 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5869 - Fully reactivate OST imports after reconnection (512, others)
5870 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5871 - More graceful error handling for truncating on dead OST (515)
5872 - Don't error out unless we're actually accessing dead stripes (474)
5873 - Fix garbage sizes when stripes are missing (410)
5874 - LRU counters were broken, causing constant lock purge (433, 432)
5875 - garbage on read from stripes with failed OSTs (441)
5876 - mark OSCs as active before reconnecting during recovery (438)
5877 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5878 - lfind did not preserve OST order in output (443)
5879 - symlinks cause hung clients, incorrect data (439)
5880 - stop dereferencing request after dropping refcount (457)
5881 - don't LASSERT(spin_is_locked) on non-SMP (455)
5882 - fixes for many rename() bugs
5883 - fstat didn't correctly synchronize attributes (399)
5884 - server must handle lock cancellation during blocking AST prep (487)
5885 - bulk descriptors were free()d too soon (511)
5886 - fix paths in lconf, which would load incorrect modules (451, 507)
5887 - fix confusing lconf 'host not found' error message (386)
5888 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5889 - fix race condition in mdc_blocking_ast() for inode access (526)
5890 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5891 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5892 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5893 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5894 - fix lstripe in several ways (kernel side) (527)
5895 - fix request leak in ldlm_cli_enqueue (262)
5896 - incorrect OSC was marked inactive after OST failure
5897 - call mds_fs_cleanup before unmounting filesystem (524)
5898 - fix races between taking ns_lock and ldlm_lock_change_resource
5899 - fix races updating LOV export open file list
5900 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5901 - fix recovery NULL deref in ldlm_cli_cancel_unused
5902 - fix some DLM races by using new hash table for lock handles (419)
5903 - permit the client to specify desired inodes, at replay
5904 - duplicate requests when we queue them for replay reintegration
5905 - fix last_rcvd offset calculation
5906 - sync after each recovered transaction, so we always make progress
5907 - never, not always, ERESTART requests without transnos
5908 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5910 - skip replay if the MDS says that the client is already connected
5911 - don't check for a recovery-enabled export to match lctl's UUID
5912 - don't INC_USE_COUNT for phantom exports
5913 - don't crash when cleaning up phantom exports (567)
5914 - don't double-finish or set replay data for errored mdc_open requests
5915 - abort requests when they time out, so we don't get old replies
5916 - send/receive replies for AST messages again
5917 - if the client says that it doesn't have the lock, cancel it on the
5919 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5920 mark it as destroyed, it will all work out when decref is called
5921 - fix module use counts (22, 581)
5923 - ASTs now expect a reply (server cancels lock on error reply)
5925 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5928 - fix many simultaneous client startup (392)
5929 - fix dentry->d_it clobbering
5930 - credentials weren't being shipped for readdir/getattr operations
5931 - remove invalid assertions triggered during some concurrent MD
5933 - proper Lustre versions added (336, 389)
5934 - fix memory leak for create error case (398)
5935 - fix LOV locking bug that would get cli/srv out of sync
5936 - fix echo client over LOV (409)
5937 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5938 - fix double-O_EXCL intent crash (424)
5939 - avoid sending multiple lock CANCELs (352)
5941 - MDS can do multi-client recovery (modulo bugs in new code)
5943 - many updates, edits, cleanups
5945 2002-11-18 Phil Schwan <phil@clusterfs.com>
5948 - fix null d_it dereference (346)
5949 - fix full OST/dbench hang (333)
5950 - fix permission problem with file removal (286)
5951 - fix removal of OSCs from LOV when they fail
5952 - fix NULL deref during bulk timeout (214)
5953 - fix problems related to multiple filesystems on one MDS (241)
5954 - fixed serious subtle metadata locking bugs
5955 - free locks on clients when inodes are removed due to memory
5957 - fix inode pointer in lock data (285)
5958 - partial support for multiple MDS on a single host (241)
5959 - data locks weren't cancelled at clear_inode time (290, 311)
5960 - intent locks could lead to unbounded lock growth (205)
5961 - added a maximum lock count, an LRU list, and a flusher
5962 - fix multiple rename (365)
5963 - properly abstracted the echo client
5964 - OSC locked 1 byte too many; fixed
5965 - rewrote brw callback code:
5966 - fixed recovery bugs related to LOVs (306)
5967 - fixed too-many-pages-in-one-write crash (191)
5968 - fixed (again) crash in sync_io_timeout (214)
5969 - probably fixed callback-related race (385)
5971 - Add capability to MDS protocol
5972 - LDLM cancellations and callbacks on different portals
5974 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5977 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5978 - defer open object destruction to close time (601981, 138)
5979 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5980 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5981 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5982 - don't call obd_setattr in ll_file_release for destroyed objects
5983 * protocol change to lustre_msg: move |version| and add |flags|
5984 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5985 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5986 * added replay of create, unlink, link and rename operations during
5987 MDS failover; recovery should be much more robust now
5988 * remove failed OSCs from LOVs (only lov_create uses this so far)
5989 * the lustre-HOWTO was brought (more) up to date (582544)
5991 2002-10-23 Phil Schwan <phil@clusterfs.com>
5994 - in-use dentries weren't being reused properly (617851)
5995 - prevent multiple LDLM setup (599178)
5996 - fix LOV size calculations for truncate (617853)
5997 - fix client handling of MDS intent errors (POSIX)
5998 - fix permission bug in lovstripe.c test (624321)
5999 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
6000 - truncate past end of file could corrupt data
6001 - proper cleanup after timeouts, crashes, etc (592524, 550815)
6002 - a race in recovery could return ETIMEDOUT to apps (623947)
6003 - building outside the source directory was fixed
6004 * the lustre-HOWTO was brought (more) up to date (582544)
6005 * major progress was made on recovery functionality
6007 2002-10-10 Phil Schwan <phil@clusterfs.com>
6010 - recovery deadlock fix
6011 - rm -rf causes LBUG fix (617817)
6012 - file open by multiple tasks fix (618962)
6013 - directory permissions bugs (602707 and 620007)
6014 - journal_stop fixed with locking (611313)
6015 - O_APPEND failures resolved (618273, perhaps 614459)
6016 - lconf PATH fix (619770)
6017 - IA64 build fix (621450)
6018 - RPC buffer sizes scale with amount of memory
6020 2002-10-01 Phil Schwan <phil@clusterfs.com>
6023 - locks would be cancelled without throwing away data pages,
6024 resulting in inconsistent data (605627)
6025 - inode attributes were not always being refreshed (605627, 612449)
6026 - lconf now continues to cleanup after lctl reports an error
6027 - MDS now enforces user permissions (602707)
6028 - lprocfs cleanup fixed, but not yet enabled (614157)
6029 - fixed infinite server hang, should a client not respond to an AST
6030 - avoid going into recovery if user calls readlink() with a buffer
6031 that's too small (613941)
6032 - AST RPCs no longer require replies (614867) -- this may be changed
6033 - don't crash server if client sends an IOV that's too big (611336)
6034 - fixed lock conversion deadlock (611892)
6035 - fixed the following of symlinks (614622)
6036 * recovery: the server can remove locks from a client that dies, other
6037 clients can make progress
6038 * more extN patch fixes
6039 * compile-time configurable ptlrpc buffer allocations
6041 - collaborative read cache document
6042 - Lustre Lite Performance CDR document-in-progress
6044 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6047 - fix typo in patch-2.4.18
6049 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6052 - clear ptlrpc request each time in handle_incoming_request()
6053 - unlink of files now destroys the object on the OST
6055 2002-09-19 Peter Braam <braam@clusterfs.com>
6057 * add hard link support
6058 * change obdfile creation method
6059 * kernel patch changed
6061 2002-09-19 Peter Braam <braam@clusterfs.com>
6064 - stack overflow bug in extN fixed
6066 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
6068 * documentation updates
6069 - add man pages for config tools
6070 - update tests/README to describe testing with new config tools
6071 - finish metadata API descriptions
6072 * bug fixes and cleanups
6073 - statfs workaround for 16TB limit
6074 - LOV stripe allocation improved, can stripe on subset of OSTs
6075 - LOV file size/IO offset was wrong for files > 4GB in size
6076 - object EA data was being dropped, caused files to be unreadable
6077 - memory overflow with non-LOV OST caused memory corruption
6078 - fixed regression tests to work with new config tools, obdfilter
6079 - fixed bug when directory size became larger than 1 block
6080 - fixed bug (for single client case) when PWD was deleted
6081 - invalidate local directory pages when doing intent-based ops
6082 - avoid LDLM oops when lock callback contained bad data
6084 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
6086 * documentation updates
6087 * bug fixes and cleanups
6088 - configuration tools
6091 - 64-bit compile warnings
6092 - 64-bit internal statfs data
6094 * test_brw on persistent OST devices
6096 * lprocfs (disabled)
6098 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
6100 * documentation updates
6101 * bug fixes and cleanups
6102 * configuration tools
6104 2002-08-30 Peter J. Braam <braam@clusterfs.com>
6107 * many small fixes to 0_5_4
6108 * io/network handling
6109 * thinkos in MDS operations
6111 2002-08-24 Peter J. Braam <braam@clusterfs.com>
6114 * crucial basic fixes to 0.5.3
6115 * IOR, Iozone work over Elan
6118 2002-08-07 Phil Schwan <phil@clusterfs.com>
6119 * version 0_5_3, our first alpha
6120 * we use the new Portals iovs
6121 * documentation updates
6122 * bug fixes and cleanups
6123 * small changes in the DLM wire protocol
6125 2002-07-25 Peter J. Braam <braam@clusterfs.com>
6126 * version 0_5_1 with some initial stability,
6127 * locking on MD and file I/O.
6128 * documentation updates
6129 * several bug fixes since 0.5.0
6130 * small changes in wire protocol
6132 2002-07-18 Phil Schwan <phil@clusterfs.com>
6134 * delivered as Lustre Light Alpha
6135 * fixed a crash after handling invalid MDS requests
6136 * fixed directory pages for architectures with non-4k pages sizes
6138 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
6139 * release version v0_4_4
6140 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
6142 * Fixes a number of interruption problems with OST operations.
6143 * Update documentation for portals header changes
6144 * Move all wire protocol structs/defines to lustre_idl.h
6145 * Fixes symlink length bug.
6146 * Add tcpdump to repository.
6148 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
6149 * release version v0_4_3
6150 * Fixes statfs for inodes on extN.
6151 * Fixes bug in runtests which would delete /etc/hosts.
6152 * Use 64-bit object IDs wherever possible (not into VFS though)
6153 Remove ost_get_info, which is unused by lustre, and out of date.
6155 2002-07-03 Peter Braam <braam@clusterfs.com>
6156 * release version v0_4_2 Fixes a lookup error (type not passed)
6157 * move forward to head of Portals
6158 * move forward to latest Lustre kernel
6160 2002-06-25 Peter Braam <braam@clusterfs.com>
6161 * release version v0_4_1. Hopefully stable on single node use.