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 2.6.29.4-167.fc11 (FC11)
13 * Recommended e2fsprogs version: 1.41.10-sun2
16 Description: Lustre does not do 1MB IOs to HW RAID
17 Details : Bump MAX_PHYS/HW_SEGMENTS and SG_ALL to 256 in the RHEL5 kernel.
19 Severity : enhancement
21 Description: lfs quota output is not very convenient for awk/sed-parsing
22 Details : Some positions in lfs quota output table could be empty or
23 non-empty which made it hard to parse it with scripts, now
24 a dash is put instead of space where there is not supposed
29 Description: Update to OFED 1.5.1
30 Details : OFED 1.5.1 will be used in environments that don't use
35 Description: Remove lnet. prefix from lctl params display
36 Details : Also renamed lnet.memused to lnet_memused to
41 Description: Use conf_param -d to remove permanant settings.
45 Description: Add list_param -R
46 Details : lctl list_param now supports recursive mode.
49 Bugzilla : 21829, 22312
50 Description: Fix llobdstat.
51 Details : 1. Now updating the write_bytes counter
52 2. Port script fixes from 1.8.
56 Description: decrease the usage of memory on clients.
57 Details : 1. On clients, recycle dentries and inodes unused.
58 2. Delete the code related to ll_deathrow(att 6215 in bug 1443). It
63 Description: llobdstat fix and enhancement
64 Details : add a counter to set a limit to how many samples will be returned
65 fix a wildcard in the path to limit to obdfilter stats only
69 Description: lfs check only allowed for root
70 Details : Code cleanups and added a regresion test of non-root use of
75 Description: MDS Oopses in ptlrpc_dispatch_difficult_reply()/__list_add()
76 Details : Fix hr->hr_index overload.
80 Description: Fix FIEMAP functionality.
84 Description: FC11 patchless client support added
88 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and
89 OEL5.4 kernel to 2.6.18-164.11.1.0.1.el5.
93 Description: New acc-sm script and various test-framework enhancments.
97 Description: Previously landed patch was reverted.
101 Description: Update SLES10 kernel to 2.6.16.60-0.42.8 (patchless)
105 Description: Add ldiskfs maxdirsize mount option.
109 Description: Fix mkfs.lustre support for 16TB LUNs.
113 Description: Print mmp_check_interval
114 Details : print mmp_check_interval and make it possible to abort
115 mount operation in case it takes too long.
119 Description: Add cascading_rw to lustre mpi tests.
123 Description: Store SOM data in EA.
124 Details : Also several SOM related cleanups.
128 Description: Add refcount to md_open_data
132 Description: Use refcount on lov_request_set to prevent memory corruption.
136 Description: Statahead: use parent dir's i_mutex for synchronization.
140 Description: loadgen improvements.
144 Description: Enable peer health detection in ksocklnd and ko2iblnd.
148 Description: Control DCACHE_LUSTRE_INVALID flag with MDS_INODE_LOOKCUP lock
152 Description: Fix race on obd_devcie:md_stats
156 Description: Fix SLES 11 client builds.
157 Details : Resolved uninitalized pointer gcc warnings.
161 Description: Allow non-root access for "lfs check".
162 Details : Added a check in obd_class_ioctl() for OBD_IOC_PING_TARGET.
166 Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
168 Severity : enhancement
170 Description: Update kernel to RHEL5.4 2.6.18-164.6.1.el5 and
171 OEL5 2.6.18-164.6.1.0.1.el5(Both in-kernel OFED enabled).
175 Description: Sanity test_27p fails.
176 Details : OBD_FAIL_OST_ENOSPC has been moved to filter_statfs().
177 Sanity tests 27 were cleaned up and improved.
181 Description: Unify req format on client/servers
182 Details : Use new req_capsule API [almost] everywhere instead of old PTLRPC
183 buffers and swabbers approach..
188 Description: drop join file stuff.
190 Severity : enhancement
192 Description: Add 2.6.27 and SLES11 (patchless client) support
196 Description: Fix in ptlrpc_expire_one_request() to print the signed time
199 Severity : enhancement
201 Description: Build kernels (RHEL5, OEL5 and SLES10/11) using the vendor's own
204 Severity : enhancement
206 Description: Remove set_info(KEY_UNLINKED) from MDS/OSC
208 Severity : enhancement
210 Description: correctly handle big reply message.
211 Details : send LNet event if reply is bigger then buffer and adjust this buffer
216 Description: Drop unnecessary __GFP_NOMEMALLOC flag from filter_get_page()
218 Severity : enhancement
220 Description: Add support for OEL5.
224 Description: Conf-sanity.sh 50g test - deactivated OST should not cause a panic.
228 Description: client could not reconnect to OST because of an active request.
229 Details : The client now retries to reconnect to the same server, if a
230 connect request failed with EBUSY or -EAGAIN. Test case 35b
231 conf-sanity.sh has been created.
235 Description: Don't allow make backward step on assignin osc next id.
236 Details : race between allocation next id and ll_sync thread can be cause
237 of set wrong osc next id and can be kill valid ost objects.
241 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
245 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
249 Description: Error handling in osc_statfs_interpret() has been improved.
250 Details : Check in osc_statfs_interpret() for EBADR.
254 Description: Do not send statfs() requests to OSTs disabled by administrator.
255 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
257 Severity : enhancement
259 Description: Added sanity-quota test to for lfs(1) to work with numeric uid/gid.
261 Severity : enhancement
263 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
265 Severity : enhancement
267 Description: Enhancement for lfs(1) command to use numeric uid/gid.
271 Description: Improved cleanup procedure for conf-sanity test_46a.
275 Description: Do not start bulk transfer, if client tried to reconnect.
276 Details : Do not start bulk transfer, if client tried to reconnect,
277 but failed due to active rpcs. Check for exp_abort_active_req
278 has been implemented.
282 Description: Prevent inconsistences between linux and lustre mount structures.
283 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
284 Make the sleep interruptible.
286 Severity : enhancement
288 Description: provide server to client comms path
289 Details : server to client communications path via new LDLM_SET_INFO rpc
293 Description: truncate starts GFP_FS allocation under transaction causing deadlock
294 Details : ldiskfs_truncate calls grab_cache_page which may start page
295 allocation under an open transaction. This may lead to
296 calling prune_icache with consequent lustre reentrance.
300 Description: send correctly lsm on open replay
301 Details : MDS is trust to LSM size on replay open, but client can set wrong size
306 Description: enable client interface failover
307 Details : When a child reconnects from another NID, properly update export
308 nid hash position and ldlm reverse import.
310 Severity : enhancement
312 Description: Communicate OST degraded/readonly state via statfs to MDS
313 Details : Flags in the statfs returned from OSTs indicate whether the
314 OST is in a degraded RAID state, or if the filesystem has
315 turned read-only after a filesystem error is detected.
320 Description: don't panic if EPROTO was hit when reading symlink
321 Details : correctly handling request reference in error cases.
323 Severity : enhancement
325 Description: Add LustreNetLink, a kernel-userspace communcation path.
327 Severity : enhancement
329 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
334 Description: access to llog context before init.
335 Details : move handling CATALOGS file at osc layer and forbid access to llog
339 Frequency : rare, only when client is writing to OST on local node
341 Description: Avoid deadlock for local client writes
342 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
343 memory freeing context. This allows OST threads to set the
344 PF_MEMALLOC flag on task structures in order to allocate memory
345 from reserved pools and complete IO.
346 Use GFP_HIGHUSER for OST allocations for non-local client writes,
347 so that the OST threads generate memory pressure and allow
348 inactive pages to be reclaimed.
350 Severity : enhancement
352 Description: Update kernel to RHEL5.3 2.6.18-128.1.14.el5.
357 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
358 Details : move ldlm namespace creation in setup phase to avoid grab
359 _lprocfs_lock with cli_sem held.
363 Description: Temporarily disable grant shrink.
364 Details : Disable the feature for debugging.
367 Frequency : only when formatting test filesystems
369 Description: Unable to run several mkfs.lustre on loop devices at the same time
370 Details : mkfs.lustre returns error 256 on the concurrent loop devices
371 formatting. The solution is to proper handle the error.
373 Severity : enhancement
375 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
377 Severity : enhancement
379 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
384 Description: resolve race between obd_disconnect and class_disconnect_exports
385 Details : if obd_disconnect will be called to already disconnected export he
386 forget release one reference and osc module can't unloaded.
388 Severity : enhancement
390 Description: Allow tuning service thread via /proc
391 Details : For each service a new
392 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
393 created that can be used to set min/max thread counts, and get the
394 current number of running threads.
399 Descriptoin: don't return error if have partially created objects for file.
400 Details : lov_update_create_set uses set->set_success as index for created
401 objects, so if some requests will be failed, they will have hole
402 at end of array and qos_shrink_lsm for allocate correct lsm.
404 Severity : enhancement
406 Description: Update OFED support to 1.4.1
411 Description: Reduce small size read RPC
412 Details : Set read-ahead limit for every file and only do read-ahead when
413 available read-ahead pages are bigger than 1M to avoid small size
416 Severity : enhancement
418 Description: move AT tunable parameters for more consistent usage
419 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
421 Severity : enhancement
423 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
424 Details : allow skip disconnected ost for send statfs request and hide error
430 Description: fix racy locking of mballoc block bitmaps causing BUG
431 Details : The locking of the mballoc buddy bitmap and the in-memory
432 block bitmap was using two different spin locks in some
433 cases. This made it possible to incorrectly access the
434 mballoc bitmap while another process was modifying it,
435 causing a sanity assertion to fail. While no on-disk corruption
436 was reported, there was some risk of this happening.
439 Frequency : rare, on llog test 6
441 Descriptoin: don't allow connect to already connected import
442 Details : allowing connect to already connected import is hide connecting problem.
447 Descriptoin: allow kill process which wait statahead result
448 Details : for some reasons 'ls' can stick in waiting result from statahead,
449 in this case need way for kill this process.
451 Severity : enhancement
453 Description: Add state history info file, enhance import info file
454 Details : Track import connection state changes in a new osc/mdc proc file;
455 add overview-type data to the osc/mdc import proc file.
457 Severity : enhancement
459 Description: MDS create should not wait for statfs RPC while holding DLM lock.
461 Severity : enhancement
463 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
468 Descriptoin: OST grant too much space to client even there are not enough space.
469 Details : Client will shrink its grant cache to OST if there are no write
470 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
471 this grant cache if there are already not enough avaible space
472 (left_space < total_clients * 32M).
475 Frequency : start MDS on uncleanly shutdowned MDS device
477 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
478 Details : stay in waiting mds<>ost recovery finished produce random bugs
479 due race between two ll_sync thread for one lov target. send
480 ACTIVATE event only if connect realy finished and import have
484 Frequency : rare, connect and disconnect target at same time
486 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
487 Details : don't call obd_disconnect under lov_lock. this long time
488 operation and can block ptlrpcd which answer to connect request.
493 Descriptoin: don't lose wakeup for imp_recovery_waitq
494 Details : recover_import_no_retry or invalidate_import and import_close can
495 both sleep on imp_recovery_waitq, but we was send only one wakeup
499 Frequency : always with long access acl
501 Descriptoin: mds can't pack reply with long acl.
502 Details : mds don't control size of acl but they limited by reint/getattr
505 Severity : enhancement
507 Description: Update to SLES10 kernel-2.6.16.60-0.33.
509 Severity : enhancement
511 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
514 Frequency : start MDS on uncleanly shutdowned MDS device
516 Descriptoin: aborting recovery hang on MDS
517 Details : don't throttle destroy RPCs for the MDT.
520 Frequency : on remount
522 Description: external journal device not working after the remount
523 Details : clear dev_rdonly flag for external journal devices in
529 Description: shutdown vs evict race
530 Details : client_disconnect_export vs connect request race.
531 if client will evicted at this time - we start invalidate
532 thread without referece to import and import can be freed
536 Frequency : rare, need acl's on inode.
538 Description: client can't handle ost additional correctly
539 Details : if ost was added after client connected to mds client can have
540 hit lnet_try_match_md ... to big messages to wide striped files.
541 in this case need teach client to handle config events about add
542 lov target and update client max ea size at that event.
544 Severity : enhancement
546 Description: Changelogs
547 Details : Changelogs are a lightweight mechanism to track filesystem
548 metadata and namespace changes. The changelog is recorded
549 permanently on the MDTs, and is periodically "consumed" / purged
550 when records are no longer needed.
552 Severity : enhancement
554 Description: compact fld format with extents
555 Details : Store range of seq rather than every seq in FLD. Seq
556 controller update FLD rather than clients. In Case of CMD, mdt0
557 has FLD, all other metadata server act as non persistent proxy
558 for FLD queries and cache fld entries in fld cache.
563 Description: don't skip ost target if they assigned to file
564 Details : Drop slow OSCs if we can, but not for requested start idx.
565 This means "if OSC is slow and it is not the requested
566 start OST, then it can be skipped, otherwise skip it only
567 if it is inactive/recovering/out-of-space.
571 Description: more cleanup in mds_lov
572 Details : not send LOV EA under replay, we can't know about they size at this
573 time. Don't allow client connect to mds before any ost connected,
574 for avoid problems with LOV EA size and returning EIO to client.
576 Severity : enhancement
578 Description: Interoperability at server side (Disk interoperability)
580 Severity : enhancement
582 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
584 Severity : enhancement
586 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
588 Severity : enhancement
590 Description: New client IO stack (CLIO).
592 Severity : enhancement
594 Description: Commit on sharing. Eliminate inter-client dependencies between
595 uncommitted transactions by doing transaction commits.
596 Thereby clients may recovery independently.
599 Frequency : Create a symlink file with a very long name
601 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
602 Details : If there is no extra space in the request for early cancels,
603 ldlm_req_handles_avail() returns 0 instead of a negative value.
605 Severity : enhancement
607 Description: Add /proc entry for import status
608 Details : The mdc, osc, and mgc import directories now have
609 an import directory that contains useful import data for debugging
612 Severity : enhancement
614 Description: Re-disable certain /proc logging
615 Details : Enable and disable client's offset_stats, extents_stats and
616 extents_stats_per_process stats logging on the fly.
619 Frequency : Only on FC kernels 2.6.22+
621 Description: oops in statahead
622 Details : Do not drop reference count for the dentry from VFS when lookup,
623 VFS will do that by itself.
625 Severity : enhancement
627 Description: Generic /proc file permissions
628 Details : Set /Proc file permissions in a more generic way to enable non-
629 root users operate on some /proc files.
633 Description: Hitting mdc_commit_close() ASSERTION
634 Details : Properly handle request reference release in
635 ll_release_openhandle().
639 Description: quota recovery deadlock during mds failover
640 Details : This patch includes att18982, att18236, att18237 in bz14840.
642 1. fix osts hang when mds does failover with quotaon
643 2. prevent watchdog storm when osts threads wait for the
648 Frequency : only patchless client
649 Description: add workaround for race between add/remove dentry from hash
651 Severity : enhancement
653 Description: Allow OST glimpses to return PW locks
657 Description: LBUG when llog conf file is full
658 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
662 Description: Prevent import from entering FULL state when server in recovery
666 Description: service mount cannot take device name with ":"
667 Details : Only when device name contains ":/" will mount treat it as
673 Description: replace ptlrpcd with the statahead thread to interpret the async
674 statahead RPC callback
678 Frequency : on recovery
679 Description: I/O failures after umount during fail back
680 Details : if client reconnected to restarted server we need join to recovery
681 instead of find server handler is changed and process self
682 eviction with cancel all locks.
684 Severity : enhancement
686 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
688 Severity : enhancement
690 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
692 Severity : enhancement
694 Description: Upcall on Lustre log has been dumped
695 Details : Allow for a user mode script to be called once a Lustre log has
696 been dumped. It passes the filename of the dumped log to the
697 script, the location of the script can be specified via
698 /proc/sys/lnet/debug_log_upcall.
703 Description: avoid idr_remove called for id which is not allocated.
704 Details : Move assigment s_dev for clustered nfs to end of initialization,
705 to avoid problem with error handling.
710 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
711 Details : When connection is reused this not moved from CONN_UNUSED_HASH
712 into CONN_USED_HASH and this prodice warning when put connection
713 again in unused hash.
716 Severity : enhancement
718 Description: File striping can now be set to use an arbitrary pool of OSTs.
720 Severity : enhancement
722 Description: Export bytes_read/bytes_write count on OSC/OST.
726 Description: Early reply size mismatch, MGC loses connection
727 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
728 the connect flags are properly negotiated.
733 Description: quotas are not honored with O_DIRECT
734 Details : all writes with the flag O_DIRECT will use grants which leads to
735 this problem. Now using OBD_BRW_SYNC to guard this.
739 Description: add quota statistics
740 Details : 1. sort out quota proc entries and proc code.
741 2. add quota statistics
743 Severity : enhancement
745 Description: enable quota support for HEAD.
749 Description: Properly propagate oinfo flags from lov to osc for statfs
750 Details : restore missing copy oi_flags to lov requests.
752 Severity : enhancement
754 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
755 : l_getgroups(8), lst(8), routerstat(8)
756 Details : included man pages for llobdstat(8), llstat(8),
757 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
759 Severity : enhancement
761 Description: configure's --enable-quota should check the
762 : kernel .config for CONFIG_QUOTA
763 Details : configure is terminated if --enable-quota is passed but
764 : no quota support is in kernel
768 Description: Remove portals compatibility
769 Details : Remove portals compatibility, not interoperable with releases
774 Description: Resolve device initialization race
775 Details : Prevent proc handler from accessing devices added to the
776 obd_devs array but yet be intialized.
778 Severity : enhancement
780 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
782 Severity : enhancement
784 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
789 Description: Using wrong pointer in osc_brw_prep_request
790 Details : Access to array[-1] can produce panic if kernel compiled with
791 CONFIG_PAGE_ALLOC enabled
795 Description: Client runs out of low memory
796 Details : Consider only lowmem when counting initial number of llap pages
800 Description: *optional* service tags registration
801 Details : if the "service tags" package is installed on a Lustre node
802 When the filesystem is mounted, a local-node service tag will
803 be created. See http://inventory.sun.com/ for more information
804 about the Service Tags asset management system.
808 Description: Kernel BUG tries to release flock
809 Details : Lustre does not destroy flock lock before last reference goes
810 away. So always drop flock locks when client is evicted and
811 perform unlock regardless of successfulness of speaking to MDS.
815 Description: add refcount protection for osc callbacks, avoid panic on shutdown
819 Description: sanity test 65a fails if stripecount of -1 is set
820 Details : handle -1 striping on filesystem in ll_dirstripe_verify
825 Description: ASSERTION(CheckWriteback(page,cmd)) failed
826 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
829 Severity : enhancement
831 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
835 Description: do not process already freed flock
836 Details : flock can possibly be freed by another thread before it reaches
837 to ldlm_flock_completion_ast.
841 Description: LBUG during stress test
842 Details : Need properly lock accesses the flock deadlock detection list.
846 Description: oops in page fault handler
847 Details : kernel page fault handler can return two special 'pages' in error
848 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
852 Description: timeout with invalidate import.
853 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
854 should be handled by ptlrpcd. This produce long age waiting and
855 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
857 Severity : enhancement
859 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
863 Description: enable MGS and MDT services start separately
864 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
865 a co-located MGS without starting the MGS, which is a complement
866 to 'nosvc' mount option.
870 Frequency : after recovery
871 Description: precreate to many object's after del orphan.
872 Details : del orphan st in oscc last_id == next_id and this triger growing
873 count of precreated objects. Set flag LOW to skip increase count
874 of precreated objects.
878 Frequency : rare, on clear nid stats
879 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
880 Details : when clean nid stats sometimes try destroy live entry,
881 and this produce panic in free.
885 Description: Stack overflow during MDS log replay
886 ease stack pressure by using a thread dealing llog_process.
890 Description: wait until IO finished before start new when do lock cancel.
891 Details : VM protocol want old IO finished before start new, in this case
892 need wait until PG_writeback is cleared until check dirty flag and
893 call writepages in lock cancel callback.
895 Severity : enhancement
897 Description: using special macro for print time and cleanup in includes.
901 Description: mds_mfd_close() ASSERTION(rc == 0)
902 Details : In mds_mfd_close(), we need protect inode's writecount change
903 within its orphan write semaphore to prevent possible races.
907 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
911 Frequency : rare, on shutdown ost
912 Description: don't hit live lock with umount ost.
913 Details : shrink_dcache_parent can be in long loop with destroy dentries,
914 use shrink_dcache_sb instead.
918 Description: don't panic with use echo client
919 Details : echo client pass NULL as client nid pointer and this produce null
924 Description: fix build on ppc32
925 Details : compile code with -m64 flag produce wrong object file for ppc32.
929 Description: add message levels for liblustreapi
933 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
934 Details : Move the 'good_osts' check before the 'total_bavail' check. This
935 will result in an -EAGAIN and in the exit call path we call
936 alloc_rr() which will with increasing aggressiveness attempt to
937 aquire precreated objects on the minimum number of required OSCs.
941 Description: Use old size assignment to avoid deadlock
942 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
943 the scheduling while holding a spinlock. We do not need locking
944 for size in ll_update_inode() because size is only updated from
945 the MDS for directories or files without objects, so there is no
946 other place to do the update, and concurrent access to such inodes
947 are protected by the inode lock.
951 Description: resolve "_IOWR redefined" build error on SLES10
955 Description: dump the memory debugging after all modules are unloaded to
956 suppress false negative in conf_sanity test 39
958 Severity : enhancement
960 Description: build kernel-ib packages for OFED 1.3 in our release cycle
965 Description: fix SLES kernel versioning
966 Details : the kernel version for our SLES 10 kernel did not include a "-"
967 before the "smp" at the end. while this was not a problem in
968 general, it did mean that software trying to use the kernel
969 version to try to detect a vendor specific kernel would fail.
970 this was most evident by the OFED build scripts.
974 Description: Don't update lov_desc members until making sure they are valid
975 Details : When updating lov_desc members via proc fs, need fix their
976 validities before doing the real update.
980 Description: don't put request into delay list while invalidate in flight.
981 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
982 import in flight. this produce timeout for invalidate and sometimes
983 can cause stale data.
987 Frequency : on ppc only
988 Description: not convert ost objects for directory because it's not exist.
989 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
993 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
995 Severity : enhancement
997 Description: Update to support 2.6.22.14 vanilla kernel.
1001 Frequency : rare, on recovery
1002 Description: read procfs can produce deadlock in some situation
1003 Details : Holding lprocfs lock which send rpc can produce block for destroy
1004 obd objects and this also block reconnect with -EALREADY. This
1005 isn't fix all lprocfs bugs - but make it rare.
1007 Severity : enhancement
1009 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
1012 Frequency : frequent on X2 node
1014 Description: mdc_set_open_replay_data LBUG
1015 Details : Set replay data for requests that are eligible for replay.
1019 Description: lustre_mgs: operation 101 on unconnected MGS
1020 Details : When MGC is disconnected from MGS long enough, MGS will evict the
1021 MGC, and late on MGC cannot successfully connect to MGS and a lot
1022 of the error messages complaining that MGS is not connected.
1026 Frequency : on network error
1027 Description: panic with double free request if network error
1028 Details : mdc_finish_enqueue is finish request if any network error ocuring,
1029 but it's true only for synchronus enqueue, for async enqueue
1030 (via ptlrpcd) this incorrect and ptlrpcd want finish request
1033 Severity : enhancement
1035 Description: client-side metadata stat-ahead during readdir(directory readahead)
1036 Details : perform client-side metadata stat-ahead when the client detects
1037 readdir and sequential stat of dir entries therein
1040 Frequency : on start mds
1042 Description: Implement get_info(last_id) in obdfilter.
1045 Frequency : occasional
1047 Description: Correctly check stale fid, not start epoch if ost not support SOM
1048 Details : open with flag O_CREATE need set old fid in op_fid3 because
1049 op_fid2 was overwritten with new generated fid, but mds can answer
1050 with one of these two fids and both is not stale. Setattr
1051 incorrectly started an epoch and assume will be called
1052 done_writing, but without SOM done_writing ever being called.
1055 Frequency : rare, depends on device drivers and load
1057 Description: MDS or OSS nodes crash due to stack overflow
1058 Details : Code changes in 1.8.0 increased the stack usage of some functions.
1059 In some cases, in conjunction with device drivers that use a lot
1060 of stack the MDS (or possibly OSS) service threads could overflow
1061 the stack. One change which was identified to consume additional
1062 stack has been reworked to avoid the extra stack usage.
1065 Frequency : occasional
1067 Description: Do not fail import if osc_interpret_create gets -EAGAIN
1068 Details : If osc_interpret_create got -EAGAIN it immediately exits and
1069 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
1070 oscc_has_objects and see OSC has no objests and call
1071 oscc_internal_create to resend create request.
1073 Severity : enhancement
1075 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
1077 Severity : enhancement
1079 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
1082 Frequency : very rare
1084 Description: Fix replay if there is an un-replied request and open
1085 Details : In some cases, older replay request will revert the
1086 mcd->mcd_last_xid on MDS which is used to record the client's
1087 latest sent request.
1089 Severity : enhancement
1091 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
1093 Severity : enhancement
1095 Description: Add rhel5 support to HEAD.
1097 Serverity : enhancement
1099 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
1104 Description: Don't allow skipping OSTs if index has been specified.
1105 Details : Don't allow skipping OSTs if index has been specified, make
1106 locking in internal create lots better.
1110 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
1111 Details : access to bitfield in structure is always rounded to long
1112 and this produce problem with not atomic change any bit.
1116 Description: Lustre make rpms failed.
1117 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
1118 builds Lustre rpms from tarball.
1121 Frequency : rare on shutdown ost
1123 Description: If llog cancel was not send before clean_exports phase, this can
1124 produce deadlock in llog code.
1125 Details : If llog thread has last reference to obd and call class_import_put
1126 this produce deadlock because llog_cleanup_commit_master wait when
1127 last llog_commit_thread exited, but this never success because was
1128 called from llog_commit_thread.
1132 Description: allow userland application know is lost one of stripes.
1133 Details : fill lvb_blocks with error code on ost and return it to
1134 application if error flag found.
1138 Description: NULL lov_tgts causing MDS oops
1139 Details : more safe checks for NULL lov_tgts for avoid oops.
1141 Severity : enhancement
1143 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
1147 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
1148 Details : If an OST is down the MDS will hang indefinitely in
1149 obd_statfs_rqset() waiting for the statfs data. While for
1150 MDS QOS usage of statfs, it should not stuck in waiting.
1152 Severity : enhancement
1154 Description: remote_acl support
1155 Details : Support ACL-based permission check for remote user.
1156 Support setfacl/getfacl for remote user with the utils
1157 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
1158 the system "{s,g}etfacl" utils.
1160 Severity : enhancement
1162 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
1164 Severity : enhancement
1166 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
1170 Description: make mgs_setparam() handle fsname containing dash
1171 Details : fsname containing a dash does not work with lctl conf_param
1173 Severity : enhancement
1175 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
1177 Severity : enhancement
1179 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
1183 Frequency : rare, at shutdown
1184 Description: access already free / zero obd_namespace.
1185 Details : if client_disconnect_export was called without force flag set,
1186 and exist connect request in flight, this can produce access to
1187 NULL pointer (or already free pointer) when connect_interpret
1188 store ocd flags in obd_namespace.
1192 Frequency : only at startup
1193 Description: not alloc memory with spinlock held.
1194 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
1195 if any spinlock held.
1197 Severity : enhancement
1199 Description: make lustre randomly fail allocating memory
1200 Details : Make lustre randomly failed allocating memory for testing purpose.
1202 Severity : enhancement
1204 Description: lost problems with lov objid file
1205 Details : Fixes some scability and access to not inited memory problems
1206 in work with lov objdid file.
1211 Description: lfs find does not continue on file error
1212 Details : Continue other files processing when a file/dir is absent.
1216 Description: Inconsistent usage of lustre_pack_reply()
1217 Details : Standardize the usage of lustre_pack_reply() such that it
1218 always generate a CERROR on failure.
1221 Frequency : occasional
1223 Description: MDS hang or stay in waiting lock
1224 Details : If client receive lock with CBPENDING flag ldlm need send lock
1225 cancel as separate rpc, to avoid situation when cancel request
1226 can't processed due all i/o threads stay in wait lock.
1230 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
1234 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
1238 Description: Update to SLES9 kernel-2.6.5-7.287.3.
1242 Description: Update to RHEL4 latest kernel.
1244 Severity : enhancement
1246 Description: Build SLES10 patchless client fails
1247 Details : The configure was broken by run ./configure with
1248 --with-linux-obj=.... argument for patchless client. When the
1249 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
1250 can't search header adequately. Use absolute path such as
1251 -I($LINUX)/include instead.
1255 Description: interrupt oig_wait produce painc on resend.
1256 Details : brw_redo_request can be used for resend requests from ptlrpcd and
1257 private set, and this produce situation when rq_ptlrpcd_data not
1258 copyed to new allocated request and triggered LBUG on assert
1259 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
1260 ptlrpcd set if request is changed and can be safety changed to use
1265 Description: LASSERT_{REQ,REP}SWAB macros are buggy
1266 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
1267 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
1268 these macros and replace them with their difinitions instead.
1272 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
1273 Details : Update which_patch & target file for SLES10 latest kernel.
1276 Description: Remove client patches from SLES 10 kernel.
1277 Details : This causes SLES 10 clients to behave as patchless clients
1278 even on a Lustre-patched (server) kernel.
1280 Severity : enhancement
1282 Description: self-adjustable client's lru lists
1283 Details : use adaptive algorithm for managing client cached locks lru
1284 lists according to current server load, other client's work
1285 pattern, memory activities, etc. Both, server and client
1286 side namespaces provide number of proc tunables for controlling
1289 Severity : enhancement
1291 Description: light-weight GSS support
1292 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
1293 and reduce performance overhead.
1295 Severity : enhancement
1297 Description: Linux keyring support
1298 Details : Support using service of Linux keyring for Lustre GSS internal
1299 context refresh/cache mechanism.
1303 Description: Fix errors in lfs documentation
1304 Details : Fixes man pages
1308 Description: don't use GFP_* in generic Lustre code.
1309 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
1313 Description: obdclass is limited by single OBD_ALLOC(idarray)
1314 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
1318 Description: warnings with build patchless client with vanila 2.6.19 and up
1319 Details : change old ctl_table style and replace ctl_table/ctl_table_header
1320 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1324 Frequency : Only for SLES
1325 Description: sanity_quota fail test_1
1326 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1327 file, which makes TSTID[2] inunique.
1331 Frequency : Only for SLES9
1332 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1333 Details : Update target/ChangeLog/which_patch .
1337 Frequency : Only for test_10 in sanity-quota.sh
1338 Description: a bug in quota test code
1339 Details : checking a test flag in a wrong place
1343 Frequency : Only for test_13 in sanity-quota.sh
1344 Description: a bug in quota test script
1345 Details : shouldn't check overfull allocated quota
1349 Description: jbd statistics
1350 Details : Port older jbd statistics patch for sles10
1352 Severity : enhancement
1354 Description: Add printing inode info into message about error in writepage.
1359 Description: reply_lock_interpret crash due to race with it and lock cancel.
1360 Details : Do not replay locks that are being cancelled. Do not reference
1361 locks by their address during replay, just by their handle.
1365 Frequency : When flocks are used.
1366 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1367 Details : Flock locks might destroy just granted lock if it could be merged
1368 with another existing flock, this is done in completion handler,
1369 so teach ldlm_cli_enquque_fini that this is a valid case for
1375 Description: Oops in read and write path when failing to allocate lock.
1376 Details : Check if lock allocation failed and return error back.
1380 Description: lstripe command fails for valid OST index
1381 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1382 instead of lov->desc.ld_active_tgt_count.
1386 Description: sanity.sh failed test 103
1387 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1392 Description: df doesn't work properly if diskfs blocksize != 4K
1393 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1397 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1398 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1400 Severity : enhancement
1402 Description: Adaptive timeouts
1403 Details : RPC timeouts adapt to changing server load and network
1404 conditions to reduce resend attempts and improve recovery time.
1408 Description: llapi_file_create() does not allow some changes
1409 Details : add llapi_file_open() that allows specifying the mode and
1410 open flags, and also returns an open file handle.
1414 Description: merge and cleanup kernel patches.
1415 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1419 Description: Add journal checksum support.(Kernel part)
1420 Details : The journal checksum feature adds two new flags i.e
1421 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1422 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1423 indicates that the commit block contains the checksum for
1424 the blocks described by the descriptor blocks. Now commit
1425 record can be sent to disk without waiting for descriptor
1426 blocks to be written to disk. This behavior is controlled
1427 using JBD2_FEATURE_ASYNC_COMMIT flag.
1431 Description: OSS needs mutliple precreate threads
1432 Details : Add ability to start more than one create thread per OSS.
1436 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1437 Details : execute lfs setstripe on client
1441 Description: mds_obd_create error creating tmp object
1442 Details : When the user sets quota on root, llog will be affected and can't
1443 create files and write files.
1446 Frequency : Always on ia64 patchless client, and possibly others.
1448 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1449 Details : This allows the patchless client to be loaded on architectures
1450 without this export.
1454 Description: RedHat Update kernel for RHEL5
1455 Details : Add the kernel config file for RHEL5.
1459 Description: RedHat Update kernel for RHEL5
1460 Details : Modify the kernel config file more closer RHEL5.
1464 Description: Build failure against Centos5 (RHEL5)
1465 Details : Define PAGE_SIZE when it isn't present.
1469 Description: test_7 on sanity-quota.sh can't be used on separate machines
1470 Details : add facet to handle it
1474 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1475 Details : with NFS, the anon dentry's parent was set to itself in
1476 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1477 corresponding dentry other than use rec->ur_name.
1479 Severity : enhancement
1481 Description: lfs setstripe enhancement
1482 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1486 Description: enable data checksumming by default
1487 Details : enable checksum by default, allow --disable-checksum
1488 configure option and "-o nochecksum" mount option. Checksums
1489 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1493 Description: lustre support for RHEL5
1494 Details : Add support for RHEL5.
1498 Description: Client eviction due to failover config
1499 Details : after a connection loss, the lustre client should attempt to
1500 reconnect to the last active server first before trying the
1501 other potential connections.
1505 Description: when mds and osts use different quota unit(32bit and 64bit),
1506 quota will be released repeatly.
1507 Details : void sending multiple quota reqs to mds, which will keep the
1508 status between the reqs.
1512 Description: osts not allocated evenly to files
1513 Details : change the condition to increase offset_idx
1516 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1518 Description: Data corruption for OSTs that are formatted larger than 2TB
1520 Details : When generating the bio request for lustre file writes the
1521 sector number would overflow a temporary variable before being
1522 used for the IO. The data reads correctly from Lustre (which
1523 will overflow in a similar manner) but other file data or
1524 filesystem metadata may be corrupted in some cases.
1528 Description: Tune the kernel for good SCSI performance.
1529 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1530 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1535 Description: rewrite ext2-derived code in obdclass/uuid.c
1536 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1537 from scratch preserving functionality.
1541 Description: "lfs find -obd UUID" prints directories
1542 Details : "lfs find -obd UUID" will return all directory names instead
1543 of just file names. It is incorrect because the directories
1544 do not reside on the OSTs.
1548 Description: use i_size_read and i_size_write in 2.6 port
1549 Details : replace inode->i_size access with i_size_read/write()
1552 Frequency : when removing large files
1554 Description: scheduling issue during removal of large Lustre files
1555 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1556 It causes scheduling issues when removing large files (17TB in the
1560 Frequency : only with liblustre clients on XT3
1562 Description: evictions taking too long
1563 Details : allow llrd to evict clients directly on OSTs
1566 Frequency : only on ppc
1568 Description: /proc/fs/lustre/devices broken on ppc
1569 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1570 We should make sure the type of 'index' is loff_t and then cast
1571 explicitly as needed below. Do not assign an explicitly cast
1575 Frequency : only for rhel5
1577 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1578 Details : Modify the target file & which_kernel.
1580 Severity : enhancement
1582 Description: omit set fsid for export NFS
1583 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1584 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1588 Frequency : Always, for kernels after 2.6.16
1589 Description: Fix warning idr_remove called for id=.. which is not allocated.
1590 Details : Last kernels save old s_dev before kill super and not allow
1591 to restore from callback - restore it before call kill_anon_super.
1595 Description: buffer overruns could theoretically occur
1596 Details : llapi_semantic_traverse() modifies the "path" argument by
1597 appending values to the end of the origin string, and a
1598 overrun may occur. Adding buffer overrun check in liblustreapi.
1602 Description: Fix error on 'ls .' at the top of the Lustre mount.
1603 Details : Don't revalidate dentry if it is a root dentry.
1607 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1608 Details : Modify vm-tunables-rhel4.patch.
1612 Description: Kernel config for 2.6.18-vanilla.
1613 Details : Modify targets/2.6-vanilla.target.in.
1614 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1615 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1616 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1617 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1621 Description: improve handling recoverable errors
1622 Details : if request processig with error which can be recoverable on server
1623 request should be resend, otherwise page released from cache and
1628 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1629 Details : Modify target file & which_patch.
1630 A flaw was found in the IA32 system call emulation provided
1631 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1632 value could be stored in the %RAX register, which could trigger an
1633 out-of-bounds system call table access. An untrusted local user
1634 could exploit this flaw to run code in the kernel
1635 (ie a root privilege escalation). (CVE-2007-4573).
1639 Description: O_DIRECT bypasses client statistics.
1640 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1641 read_ahead_stats were updated. copied the stats section from
1642 osc_send_oap_rpc() into async_internal().
1646 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1650 Description: change order of libsysio includes
1651 Details : '#include sysio.h' should always come before '#include xtio.h'
1655 Description: handle "serious error: objid * already exists" more gracefully
1656 Details : If LAST_ID value on disk is smaller than the objects existing in
1657 the O/0/d* directories, it indicates disk corruption and causes an
1658 LBUG(). If the object is 0-length, then we should use the existing
1659 object. This will help to avoid a full fsck in most cases.
1661 Severity : enhancement
1663 Description: adapt the lustre_config script to support the upgrade case
1664 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1665 server targets to 1.6 in parallel.
1669 Description: To avoid grant space > avaible space when the disk is almost
1670 full. Without this patch you might see the error "grant XXXX >
1671 available" or some LBUG about grant, when the disk is almost
1673 Details : In filter_check_grant, for non_grant cache write, we should
1674 check the left space by if (*left > ungranted + bytes), instead
1675 of (*left > ungranted), because only we are sure the left space
1676 is enough for another "bytes", then the ungrant space should be
1677 increase. In client, we should update cl_avail_grant only there
1678 is OBD_MD_FLGRANT in the reply.
1682 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1683 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1684 described in CVE-2007-4573. This problem could allow untrusted
1685 local users to gain root access.
1688 Frequency : when using O_DIRECT and quotas
1690 Description: Incorrect file ownership on O_DIRECT output files
1691 Details : block usage reported by 'lfs quota' does not take into account
1692 files that have been written with O_DIRECT.
1697 Description: touch file failed when fs is not full
1698 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1699 otherwise we can get ENOSP while fs is not full.
1703 Description: parallel lock callbacks
1704 Details : Instead of sending blocking and completion callbacks as separated
1705 requests, adding them to a set and sending in parallel.
1708 Frequency : only for Cray XT3
1709 Bugzilla : 12829/13455
1710 Description: Changing primary group doesn't change the group lustre assigns to
1712 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1713 overridden with the primary group provided by the group upcall,
1714 whereas the supplied fsgid can be trusted if it is in the list of
1715 supplementary groups returned by the group upcall.
1717 Severity : enhancement
1719 Description: Allow masking D_WARNING, D_ERROR messages from console
1720 Details : Console messages can now be disabled via lnet.printk.
1724 Description: User code with malformed file open parameter crashes client node
1725 Details : Before packing join_file req, all the related reference should be
1726 checked carefully in case some malformed flags cause fake
1727 join_file req on client.
1731 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1732 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1733 mode is equal to the granted mode. However, it should also check
1734 that LDLM_FL_CP_REQD is not set.
1738 Description: Heavy nfs access might result in deadlocks
1739 Details : After ELC code landed, it is now improper to enqueue any mds
1740 locks under och_sem, because enqueue might want to decide to
1741 cancel open locks for same inode we are holding och_sem for.
1745 Description: Client eviction while running blogbench
1746 Details : A lot of unlink operations with concurrent I/O can lead to a
1747 deadlock causing evictions. To address the problem, the number of
1748 oustanding OST_DESTROY requests is now throttled to
1749 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1750 callbacks are processed in priority.
1754 Description: enable ACLs on MDS by default
1755 Details : ACLs must be enabled on MDS by default.
1758 Frequency : PPC/PPC64 only
1760 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1761 Details : fix duplicated definitions between asm-ppc64/types.h and
1762 lustre_types.h on PPC.
1765 Frequency : PPC/PPC64 only
1767 Description: asm-ppc/segment.h does not exist
1768 Details : fix compile issue on PPC.
1772 Description: better handle error messages in extents code
1775 Frequency : RHEL4 only
1777 Description: mkfs is very slow on IA64/RHEL4
1778 Details : A performance regression has been discovered in the MPT Fusion
1779 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1780 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1781 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1783 Severity : enhancement
1785 Description: SNMP support enhancement
1786 Details : Adding total number of sampled request for an MDS node in snmp
1789 Severity : enhancement
1791 Description: Optimize ldlm waiting list processing for PR extent locks
1792 Details : When processing waiting list for read extent lock and meeting read
1793 lock that is same or wider to it that is not contended, skip
1794 processing rest of the list and immediatelly return current
1795 status of conflictness, since we are guaranteed there are no
1796 conflicting locks in the rest of the list.
1800 Description: Time out and refuse to reconnect
1801 Details : When the failover node is the primary node, it is possible
1802 to have two identical connections in imp_conn_list. We must
1803 compare not conn's pointers but NIDs, otherwise we can defeat
1804 connection throttling.
1808 Description: port llog fixes from b1_6 into HEAD
1809 Details : Port llog reference couting and some llog cleanups from b1_6
1810 (bug 10800) into HEAD, for protect from panic and access to already
1811 free llog structures.
1815 Description: Detect stride IO mode in read-ahead
1816 Details : When a client does stride read, read-ahead should detect that and
1817 read-ahead pages according to the detected stride pattern.
1821 Description: data checksumming impacts single node performance
1822 Details : add support for several checksum algorithm. Currently, only CRC32
1823 and Adler-32 are supported. The checksum type can be changed on
1824 the fly via /proc/fs/lustre/osc/*/checksum_type.
1828 Description: use adler32 for page checksums
1829 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1834 Description: build for x2 fails
1835 Details : fix compile issue on Cray systems.
1839 Description: Properly match for duplicate locks
1840 Details : Due to different lock order from skiplists code, we need to
1841 traverse entire list for now
1844 Frequency : only on PPC/SLES10
1846 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1847 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1848 be defined. Add a hack in mkfs_lustre.c to work around this compile
1853 Description: LASSERT on MDS when client holding flock lock dies
1854 Details : ldlm pool logic depends on number of granted locks equal to
1855 number of released locks which is not true for flock locks, so
1856 just exclude such locks from consideration.
1860 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1861 Details : Use fsfilt_sync() for both the whole filesystem sync and
1862 individual file sync to eliminate dangerous inode locking
1863 with I_LOCK that can lead to a deadlock.
1867 Description: performance in 1.6.3
1868 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1869 sure that 1MB requests are not fragmented by the block layer.
1871 Severity : enhancement
1873 Description: organize the server-side client stats on per-nid basis
1874 Details : Change the structure of stats under obdfilter and mds to
1882 The "uuid"s file would list the uuids of _active_ exports.
1883 And the clear entry is to clear all stats and stale nids.
1885 Severity : enhancement
1887 Description: eliminate client locks in face of contention
1888 Details : file contention detection and lockless i/o implementation
1889 for contended files.
1893 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1896 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1898 Description: MDS may lose file striping (and hence file data) in some cases
1899 Details : If there are additional extended attributes stored on the MDS,
1900 in particular ACLs, SELinux, or user attributes (if user_xattr
1901 is specified for the client mount options) then there is a risk
1902 of attribute loss. Additionally, the Lustre file striping
1903 needs to be larger than default (e.g. striped over all OSTs),
1904 and an additional attribute must be stored initially in the
1905 inode and then increase in size enough to be moved to the
1906 external attribute block (e.g. ACL growing in size) for file
1911 Description: skiplist implementation simplification
1912 Details : skiplists are used to group compatible locks on granted list
1913 that was implemented as tracking first and last lock of each
1914 lock group the patch changes that to using doubly linked lists
1918 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1919 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1920 because LASSERT on both the data supplied by a client, and the
1921 data on disk is dangerous and incorrect.
1923 Severity : enhancement
1925 Description: Slow truncate/writes to huge files at high offsets.
1926 Details : Directly associate cached pages to lock that protect those pages,
1927 this allows us to quickly find what pages to write and remove
1928 once lock callback is received.
1932 Description: more ldlm soft lockups
1933 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1934 starve other threads from the resource lock for a long time in
1935 case of long waiting queue, so change the debug level from
1936 D_OTHER to the less frequently used D_INFO.
1938 Severity : enhancement
1940 Description: add -gid, -group, -uid, -user options to lfs find
1944 Description: Hung threads in invalidate_inode_pages2_range
1945 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1946 once one is completed. As a result, some RPCs are stuck in the
1947 queue and are never sent.
1951 Description: filter threads hungs on waiting journal commit
1952 Details : Cleanup filter group llog code, then only filter group llog will
1953 be only created in the MDS/OST syncing process.
1957 Description: Procfs and llog threads access destoryed import sometimes.
1958 Details : Sync the import destoryed process with procfs and llog threads by
1959 the import refcount and semaphore.
1961 Severity : enhancement
1963 Description: openlock cache of b1_6 port to HEAD
1968 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1969 Details : If insertion of an extent fails, then discard the inode
1970 preallocation and free data blocks else it can lead to duplicate
1975 Description: don't always update ctime in ext3_xattr_set_handle()
1976 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1977 In some cases the ctime should not be updated, for example for
1978 2.0->1.8 compatibility it is necessary to delete an xattr and it
1979 should not update the ctime.
1983 Bugzilla : 15713/16362
1984 Description: Assertion in iopen_connect_dentry in 1.6.3
1985 Details : looking up an inode via iopen with the wrong generation number can
1986 populate the dcache with a disconneced dentry while the inode
1987 number is in the process of being reallocated. This causes an
1988 assertion failure in iopen since the inode's dentry list contains
1989 both a connected and disconnected dentry.
1993 Description: assertion failure in ldlm_handle2lock()
1994 Details : fix a race between class_handle_unhash() and class_handle2object()
1995 introduced in lustre 1.6.5 by bug 13622.
2000 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
2001 Details : remove the limit of 256 scsi disks in the sd_iostat patch
2006 Description: oops in sd_iostats_seq_show()
2007 Details : unloading/reloading the scsi low level driver triggers a kernel
2008 bug when trying to access the sd iostat file.
2013 Description: Kernel panics during QLogic driver reload
2014 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
2015 patch, causing memory corruption.
2020 Description: journal_dev option does not work in b1_6
2021 Details : pass mount option during pre-mount.
2023 Severity : enhancement
2025 Description: Add a FIEMAP(FIle Extent MAP) ioctl
2026 Details : FIEMAP ioctl will allow an application to efficiently fetch the
2027 extent information of a file. It can be used to map logical blocks
2028 in a file to physical blocks in the block device.
2032 Description: LDLM soft lockups - improvement
2033 Details : It is be possible to send the lock handle along with each read
2034 or write request because the client is already doing a lock match
2035 itself so there isn't any reason the OST should have to re-do that
2039 Frequency : only on Cray X2
2041 Description: X2 build failures
2042 Details : fix build failures on Cray X2.
2046 Description: xid & resent requests
2047 Details : Initialize RPC XID from clock at startup (randomly if clock is
2050 Severity : enhancement
2052 Description: Add lustre_start utility to start or stop multiple Lustre servers
2057 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
2058 Details : In case of memory pressure, list_del() can be called twice on
2059 req->rq_history_list, causing a kernel oops.
2063 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
2064 Details : fix stack overflow in the distributed lock manager by defering
2065 export eviction after a failed AST to the elt thread instead of
2066 handling it in the dlm interpret routine.
2070 Description: Convert some comments to new format.
2071 Details : Update documenting comments to match doxygen conventions.
2075 Description: Grammar fixes.
2076 Details : A couple of trivial spelling fixes.
2080 Description: OSD_COUNTERS-mandatory
2081 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
2082 to simplify the code.
2086 Description: simplify cmm_device freeing logic.
2087 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
2091 Description: Add lockdep support to dt_object_operations locking interface.
2092 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
2093 parameter indicating lock ordering. Update mdd code to use new
2098 Description: Introduce failloc constants for lockless IO tests.
2099 Details : Add two new failloc constants to test lockless IO. Only one of
2100 them in implemented---another is checked in yet to be landed
2105 Description: Add lockdep support for inode mutex.
2106 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
2107 in the situations where Lustre has to lock more than one inode
2112 Description: Add optional invariants checking support.
2113 Details : Add new LINVRNT() macro, optional on new --enable-invariants
2114 configure switch. This macro is to be used for consistency and
2115 sanity checks that are too expensive to be left in `production'
2120 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
2121 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
2126 Description: Add lockdep support for ldlm_lock and ldlm_resource.
2127 Details : Use spin_lock_nested() in (the only) situation where more than
2128 one ldlm_lock is locked simultaneously. Also, fix possible
2129 dead-lock in ldlm_lock_change_resource() by enforcing particular
2134 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
2135 Details : Instead of specifying each ldlm_lock call-back through separate
2136 parameter, wrap them into struct ldlm_callback_suite.
2140 Description: Kill join_lru obd method and its callers.
2141 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
2142 in memory---a requirement implemented through ->o_join_lru() obd
2143 method in HEAD. Remove this method and its users.
2147 Description: Add asynchronous ldlm ENQUEUE completion handler.
2148 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
2149 case that stock ldlm_completion_ast() cannot handle as it waits
2150 until lock is granted. Introduce new ldlm_completion_ast_async()
2151 for this. Also comment ldlm_completion_ast().
2155 Description: ldlm_error <-> errno conversion.
2156 Details : Add functions to map (rather arbitrary) between LDLM error codes
2157 and standard errno values. CLIO needs this to prevent LDLM specific
2158 constants from escaping ldlm and osc.
2162 Description: Kill unused ldlm_handle2lock_ns() function.
2163 Details : Kill unused ldlm_handle2lock_ns() function.
2167 Description: Add lu_ref support to ldlm_lock
2168 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
2169 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
2170 and ldlm_resource. LDLM interface has to be changed, because code that
2171 releases a reference on a lock, has to "know" what reference this is.
2172 In the most frequent case
2174 lock = ldlm_handle2lock(handle);
2176 LDLM_LOCK_PUT(lock);
2178 no changes are required. When any other reference (received _not_ from
2179 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
2180 instead of LDLM_LOCK_PUT().
2182 Arguably, changes are pervasive, and interface requires some discipline
2183 for proper use. On the other hand, it was very instrumental in finding
2184 a few leaked lock references.
2188 Description: Add ldlm_lock_addref_try().
2189 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
2190 attempts to addref a lock that might be being canceled
2195 Description: Add ldlm_weigh_callback().
2196 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
2197 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
2198 hard-coded `number of pages' logic.
2202 Description: Add lockdep annotations to llog code.
2203 Details : Use appropriately tagged _nested() locking calls in the places
2204 where llog takes more than one ->lgh_lock lock.
2208 Description: Add loi_kms_set().
2209 Details : Wrap kms updates into a helper function.
2213 Description: Constify instances of struct lsm_operations.
2214 Details : Constify instances of struct lsm_operations.
2218 Description: lu_conf support.
2219 Details : On a server, a file system object is uniquely identified
2220 by a fid, which is sufficient to locate and load all object
2221 state (inode). On a client, on the other hand, more data are
2222 necessary instantiate an object. Change lu_object_find() and
2223 friends to take additional `lu_conf' argument describing object.
2224 Typically this includes layout information.
2228 Description: lu_context fixes.
2229 Details : Introduce new lu_context functions that are needed on the client
2230 side, where some system threads (ptlrpcd) are shared by multiple
2231 modules, and so cannot be stopped during module shutdown.
2235 Description: Add start and stop methods to lu_device_type_operations.
2236 Details : Introduce two new methods in lu_device_type_operations, that are
2237 invoked when first instance of a given type is created and last one
2238 is destroyed respectively. This is need by CLIO.
2242 Description: Add lu_ref support to struct lu_device.
2243 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
2244 track leaked references.
2248 Description: Introduce lu_kmem_descr.
2249 Details : lu_kmem_descr and its companion interface allow to create
2250 and destroy a number of kmem caches at once.
2254 Description: Fix lu_object finalization race.
2255 Details : Fix a race between lu_object_find() finding an object and its
2256 concurrent finalization. This race is (most likely) not possible
2257 on the server, but might happen on the client.
2261 Description: Introduce lu_ref interface.
2262 Details : lu_ref is a debugging module allowing to track references to
2263 a given object. It is quite cpu expensive, and has to be
2264 explicitly enabled with --enable-lu_ref. See usage description
2269 Description: Factor lu_site procfs stats into a separate function.
2270 Details : Separate lu_site stats printing code into a separate function
2271 that can be reused on a client.
2275 Description: Constify instances of struct {lu,dt,md}_device_operations.
2276 Details : Constify instances of struct {lu,dt,md}_device_operations.
2280 Description: Introduce struct md_site and move meta-data specific parts of
2281 struct lu_site here.
2282 Details : Move md-specific fields out of struct lu_site into special struct
2283 md_site, so that lu_site can be used on a client.
2287 Description: Kill mdd_lov_destroy().
2288 Details : Remove unused mdd code.
2292 Description: Add st_block checking to multistat.c.
2293 Details : Add st_block checking to multistat.c.
2297 Description: Add lu_ref support to struct obd_device.
2298 Details : Add lu_ref tracking to obd_device.
2302 Description: Kill obd_set_fail_loc().
2303 Details : Remove unused code.
2307 Description: Add special type for ptlrpc_request interpret functions.
2308 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
2309 there. Actual usage will be in CLIO.
2313 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
2314 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
2315 doesn't work with lockdep.
2319 Description: Add rwv.c test program.
2320 Details : New testing program exercising readv(2) and writev(2) (Qian).
2324 Description: Add sendfile.c test program.
2325 Details : New testing program exercising sendfile(2) (Jay).
2329 Description: Ratelimit a message that can be very frequent.
2330 Details : Ratelimit a memory allocation failure message that can
2335 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2336 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2340 Description: Liblustre build fixes.
2341 Details : Liblustre build fixes.
2345 Description: libcfs: add cfs_{need,cond}_resched() interface.
2346 Details : libcfs: add cfs_{need,cond}_resched() definition and
2347 implementations for Linux, NT, and liblustre.
2349 Severity : enhancement
2351 Description: More exported tunables for mballoc
2352 Details : Add support for tunable preallocation window and new tunables for
2353 large/small requests
2357 Description: Detect corruption of block bitmap and checking for preallocations
2358 Details : Checks validity of on-disk block bitmap. Also it does better
2359 checking of number of applied preallocations. When corruption is
2360 found, it turns filesystem readonly to prevent further corruptions.
2364 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2365 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2366 window to 0. If it does hit cache miss, and read-pattern is still
2367 stride-io mode, does not reset the stride window, but also does not
2368 increase the stride window length in this case.
2372 Frequency : only for big-endian servers
2373 Description: Check if system is big-endian while mounting fs with extents feature
2374 Details : Mounting a filesystem with extents feature will fail on big-endian
2375 systems since ext3-based ldiskfs is not supported on big-endian
2376 systems. This can be over-riden with "bigendian_extents" mount option.
2378 Severity : enhancement
2380 Description: The root squash functionality
2381 Details : A security feature, which is to prevent users from being able
2382 to mount lustre on their desktop, run as root, and delete
2383 all of the files in the filesystem. The goal is accomplished by
2384 remapping user id (UID) and group id (GID) of the root user to
2385 a UID and GID specified by the system administartor via Lustre
2386 configuration management server (MGS). The functionality also
2387 allows to specify sets of clients for which the remapping does
2392 Description: Excessive recovery window
2393 Details : With AT enabled, the recovery window can be excessively long (6000+
2394 seconds). To address this problem, we no longer use
2395 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2396 timeout no longer depends on the service time, it is set to
2397 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2398 time via pb_service_time.
2402 Description: Watchdog triggered on MDS failover
2403 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2404 the OSTs know that the MDS "UUID" can be reused for the same export
2405 from a different NID, so we do not need to wait for the export to be
2409 Frequency : rare, only if using MMP with Linux RAID
2411 Description: MMP doesn't work with Linux RAID
2412 Details : While using HA for Lustre servers with Linux RAID, it is possible
2413 that MMP will not detect multiple mounts. To make this work we
2414 need to unplug the device queue in RAID when the MMP block is being
2415 written. Also while reading the MMP block, we should read it from
2416 disk and not the cached one.
2418 Severity : enhancement
2420 Description: open file using fid
2421 Details : A file can be opened using just its fid, like
2422 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2425 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2428 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2429 Details : add proper locking for creation of jbd_4k slab cache
2433 Description: MMP check in ext3_remount() fails without displaying any error
2434 Details : When multiple mount protection fails during remount, proper error
2437 Severity : enhancement
2439 Description: Allow stripe size to be up to 4G-64k
2440 Details : Fix math logic to allow large stripe sizes.
2444 Description: add check for >8TB ldiskfs filesystems
2445 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2446 Don't allow >8TB ldiskfs filesystems to be mounted without
2447 force_over_8tb mount option
2452 Descriptoin: OSS complains "too many blocks" and crashes
2453 Details : In filter_commitrw_write():
2454 1. lquota_chkquota();
2455 2. obdo_from_inode();
2456 3. lquota_pending_commit();
2457 "2." has changed uid and gid in this case, then caused
2458 lquota_chkquota() and lquota_pending_commit() to use
2459 different uid and gid. That is the root of the bug.
2463 Descriptoin: getstripe doesn't report stripe count = -1 sanely
2464 Details : Detect and print out the 'stripe count = -1' case properly.
2468 Description: Out or order replies might be lost on replay
2469 Details : In ptlrpc_retain_replayable_request if we cannot find retained
2470 request with tid smaller then one currently being added, add it
2471 to the start, not end of the list.
2473 --------------------------------------------------------------------------------
2475 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2477 * Support for kernels:
2478 2.6.5-7.283 (SLES 9),
2479 2.6.9-55.EL (RHEL 4),
2480 2.6.16.46-0.14 (SLES 10),
2481 2.6.18.8 vanilla (kernel.org)
2482 * Client support for unpatched kernels:
2483 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2484 2.6.16 - 2.6.22 vanilla (kernel.org)
2485 * Due to recently discovered recovery problems, we do not recommend
2486 using patchless RHEL 4 clients with this or any earlier release.
2487 * Recommended e2fsprogs version: 1.39.cfs8
2488 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2489 * Starting with this release, the ldiskfs backing filesystem required
2490 by Lustre is now in its own package, lustre-ldiskfs. This package
2491 should be installed. It is versioned separately from Lustre and
2492 may be released separately in future.
2496 Description: block reactivating mgc import until all deactivates complete
2497 Details : Fix race when failing back MDT/MGS to itself (testing)
2500 Frequency : at statup only
2502 Description: mds_lov_synchronize race leads to various problems
2503 Details : simultaneous MDT->OST connections at startup can cause the
2504 sync to abort, leaving the OSC in a bad state.
2506 Severity : enhancement
2508 Description: add optional extra BUILD_VERSION info
2509 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2510 to override the lustre version.
2513 Frequency : 2.6.18 servers only
2515 Description: ll_kern_mount() doesn't release the module reference
2516 Details : The ldiskfs module reference count never drops down to 0
2517 because ll_kern_mount() doesn't release the module reference.
2522 Description: server LBUG when using old ost_num_threads parameter
2523 Details : Accept the old ost_num_threads parameter but warn that it
2524 is deprecated, and fix an off-by-one error that caused an LBUG.
2529 Description: Transient SCSI error results in persistent IO issue
2530 Details : iobuf->dr_error is not reinitialized to 0 between two
2534 Frequency : sometimes when underlying device returns I/O errors
2536 Description: OSTs not going read-only during write failures
2537 Details : OSTs are not remounted read-only when the journal commit threads
2538 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2539 instead of the ext3 wrappers.
2543 Description: poor connect scaling with increasing client count
2544 Details : Don't run filter_grant_sanity_check for more than 100 exports
2545 to improve scaling for large numbers of clients.
2548 Frequency : SLES10 only
2550 Description: sanity-quota.sh quotacheck failed: rc = -22
2551 Details : Quotas cannot be enabled on SLES10.
2554 Frequency : liblustre clients only
2556 Description: getdirentries does not give error when run on compute nodes
2557 Details : getdirentries does not fail when the size specified as an argument
2558 is too small to contain at least one entry
2560 Severity : enhancement
2562 Description: Add LNET router traceability for debug purposes
2563 Details : If a checksum failure occurs with a router as part of the
2564 IO path, the NID of the last router that forwarded the bulk data
2565 is printed so it can be identified.
2570 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2571 Details : Due to a race condition, liblustre clients were occasionally
2572 evicted incorrectly.
2574 Severity : enhancement
2576 Description: lfs setstripe use optional parameters instead of postional
2579 Severity : enhancement
2581 Description: Nanosecond timestamp support for ldiskfs
2582 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2583 resolution timestamps. There is not yet support for this at
2584 the Lustre filesystem level.
2587 Frequency : during server recovery
2589 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2590 Details : request with rq_no_resend flag not awake l_wait_event if they get
2594 Frequency : nfs export on patchless client
2596 Description: connectathon hang when test nfs export over patchless client
2597 Details : Disconnected dentry cannot be found with lookup, so we do not need
2598 to unhash it or make it invalid
2601 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2602 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2603 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2608 Description: open req refcounting wrong on reconnect
2609 Details : If reconnect happened between getting open reply from server and
2610 call to mdc_set_replay_data in ll_file_open, we will schedule
2611 replay for unreferenced request that we are about to free.
2612 Subsequent close will crash in variety of ways.
2613 Check that request is still eligible for replay in
2614 mdc_set_replay_data().
2619 Description: disable writes to filesystem when reading health_check file
2620 Details : the default for reading the health_check proc file has changed
2621 to NOT do a journal transaction and write to disk, because this
2622 can cause reads of the /proc file to hang and block HA state
2623 checking on a healthy but otherwise heavily loaded system. It
2624 is possible to return to the previous behaviour during configure
2625 with --enable-health-write.
2627 Severity : enhancement
2629 Description: 64-bit inode version
2630 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2631 to the inode. This will be required for version-based recovery.
2636 Description: MDS fails to start if a duplicate client export is detected
2637 Details : in some rare cases it was possible for a client to connect to
2638 an MDS multiple times. Upon recovery the MDS would detect this
2639 and fail during startup. Handle this more gracefully.
2641 Severity : enhancement
2643 Description: Add -o localflock option to simulate old noflock
2645 Details : This will achieve local-only flock/fcntl locks
2651 Description: log_commit_thread vs filter_destroy race leads to crash
2652 Details : Take import reference before releasing llog record semaphore
2657 Description: Wrong request locking in request set processing
2658 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2659 imp_delayed_list, in this place should be used imp_lock.
2662 Frequency : when reconnection
2664 Description: Grant Leak when osc reconnect to OST
2665 Details : When osc reconnect ost, OST(filter) should check whether it
2666 should grant more space to client by comparing fed_grant and
2667 cl_avail_grant, and return the granted space to client instead
2668 of "new granted" space, because client will call osc_init_grant
2669 to update the client grant space info.
2672 Frequency : when client reconnect to OST
2674 Description: Grant Leak when osc do resend and replay bulk write
2675 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2676 bulk write request, because the grant info while be sync between
2677 OSC and OST when reconnect, and we should ignore the grant info
2678 these of resend/replay write req.
2683 Description: Grant space more than avaiable left space sometimes.
2684 Details : When then OST is about to be full, if two bulk writing from
2685 different clients came to OST. Accord the avaliable space of the
2686 OST, the first req should be permitted, and the second one
2687 should be denied by ENOSPC. But if the seconde arrived before
2688 the first one is commited. The OST might wrongly permit second
2689 writing, which will cause grant space > avaiable space.
2692 Frequency : when client is evicted
2694 Description: Grant might be wrongly erased when osc is evicted by OST
2695 Details : when the import is evicted by server, it will fork another
2696 thread ptlrpc_invalidate_import_thread to invalidate the
2697 import, where the grant will be set to 0. While the original
2698 thread will update the grant it got when connecting. So if
2699 the former happened latter, the grant will be wrongly errased
2700 because of this race.
2705 Description: Checking Stale with correct fid
2706 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2707 to check whether it is stale, because sometimes, we want the
2708 enqueue happened anyway, and op_data.fid2 will not be initialized.
2710 Severity : enhancement
2712 Description: update patchless client
2713 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2716 Frequency : only with 2.4 kernel
2718 Description: random memory corruption
2719 Details : size of struct ll_inode_info is to big for union inode.u and this
2720 can be cause of random memory corruption.
2725 Description: Memory leak in recovery
2726 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2727 It should also check obd_fail before fsfilt_start, otherwise if
2728 fsfilt_start return -EROFS,(failover mds during mds recovery).
2729 then the req will return with repmsg->transno = 0 and rc = EROFS.
2730 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2731 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2732 be freed no matter whether fsfilt_commit success or not.
2735 Frequency : only with huge count clients
2737 Description: Prevents from taking the superblock lock in llap_from_page for
2739 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2740 for llap_from_page call in ll_removepage prevents from taking the
2741 superblock lock for a soon died page.
2746 Description: Not check open intent error before release open handle
2747 Details : in some rare cases, the open intent error is not checked before
2748 release open handle, which may cause
2749 ASSERTION(open_req->rq_transno != 0), because it tries to release
2750 the failed open handle.
2755 Description: Set cat log bitmap only after create log success.
2756 Details : in some rare cases, the cat log bitmap is set too early. and it
2757 should be set only after create log success.
2761 Description: Accessing a block bevice can re-enable I/O when Lustre is
2762 tearing down a device.
2763 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2768 Description: service threads may hog cpus when there are a lot of requests
2769 Details : Insert cond_resched to give other threads a chance to use some CPU
2774 Description: the cat log was not initialized in recovery
2775 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2776 unlink log on mds will not be initialized until mds post
2777 recovery. And also in mds post recovery, the unlink log will
2778 initialization will be done asynchronausly, so there will be race
2779 between add unlink log and unlink log initialization.
2783 Description: brw_stats were being printed incorrectly
2784 Details : brw_stats were being printed as log2 but all of them were not
2785 recorded as log2. Also remove some code duplication arising from
2786 filter_tally_{read,write}.
2790 Frequency : rare, only in recovery.
2791 Description: ASSERTION(req->rq_type != LI_POISON) failed
2792 Details : imp_lock should be held while iterating over imp_sending_list for
2793 prevent destroy request after get timeout in ptlrpc_queue_wait.
2797 Description: replay-single.sh test 52 fails
2798 Details : A lock's skiplist need to be cleanup when it being unlinked
2799 from its resource list.
2803 Description: Short directio read returns full requested size rather than
2805 Details : Direct I/O operations should return actual amount of bytes
2806 transferred rather than requested size.
2808 Severity : enhancement
2810 Description: metadata RPC reduction (e.g. for rm performance)
2811 Details : decrease the amount of synchronous RPC between clients and servers
2812 by canceling conflicing lock before the operation on the client
2813 and packing thier handles into the main operation RPC to server.
2815 Severity : enhancement
2817 Description: Async OSC create to avoid the blocking unnecessarily.
2818 Details : If a OST has no remain object, system will block on the creating
2819 when need to create a new object on this OST. Now, ways use
2820 pre-created objects when available, instead of blocking on an
2821 empty osc while others are not empty. If we must block, we block
2822 for the shortest possible period of time.
2826 Description: improve handling recoverable errors
2827 Details : if request processig with error which can be recoverable on server
2828 request should be resend, otherwise page released from cache and
2831 Severity : enhancement
2833 Description: refine locking for avoid write wrong info into lov_objid file
2834 Details : fix possible races with add new target and write/update data in
2837 --------------------------------------------------------------------------------
2839 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2844 Frequency : on some architectures
2846 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2847 Details : Uninitialized flags sometimes cause configuration commands to
2851 Frequency : patchless clients only
2853 Description: missing __iget() symbol export
2854 Details : The __iget() symbol export is missing. To avoid the need for
2855 this on patchless clients the deathrow inode reaper is turned
2856 off, and we depend on the VM to clean up old inodes. This
2857 dependency was during via the fix for bug 12181.
2859 --------------------------------------------------------------------------------
2861 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2863 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2864 INTEROPERATE with older versions automatically. In many cases a
2865 special upgrade step is needed. Please read the
2866 user documentation before upgrading any part of a live system.
2867 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2868 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2869 * WARNING: Lustre configuration and startup changes are required with
2870 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2874 Severity : enhancement
2876 Description: MountConf
2877 Details : Lustre configuration is now managed via mkfs and mount
2878 commands instead of lmc and lconf. New obd types (MGS, MGC)
2879 are added for dynamic configuration management. See
2880 https://mail.clusterfs.com/wikis/lustre/MountConf for
2883 Severity : enhancement
2885 Description: dynamic OST addition
2886 Details : OSTs can now be added to a live filesystem
2888 Severity : enhancement
2890 Description: startup order invariance
2891 Details : MDTs and OSTs can be started in any order. Clients only
2892 require the MDT to complete startup.
2894 Severity : enhancement
2896 Description: parallel, asynchronous orphan cleanup
2897 Details : orphan cleanup is now performed in separate threads for each
2898 OST, allowing parallel non-blocking operation.
2900 Severity : enhancement
2902 Description: optimized stripe assignment
2903 Details : stripe assignments are now made based on ost space available,
2904 ost previous usage, and OSS previous usage, in order to try
2905 to optimize storage space and networking resources.
2907 Severity : enhancement
2909 Description: Permanently set tunables
2910 Details : All writable /proc/fs/lustre tunables can now be permanently
2911 set on a per-server basis, at mkfs time or on a live system.
2913 Severity : enhancement
2915 Description: Lustre message v2
2916 Details : Add lustre message format v2.
2918 Severity : enhancement
2920 Description: client OST exclusion list
2921 Details : Clients can be started with a list of OSTs that should be
2922 declared "inactive" for known non-responsive OSTs.
2926 Description: SPEC SFS validation failure on NFS v2 over lustre.
2927 Details : Changes the blocksize for regular files to be 2x RPC size,
2928 and not depend on stripe size.
2930 Severity : enhancement
2932 Description: Multiple MD RPCs in flight.
2933 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2934 To avoid overly-overloading MDS, introduce a limit on number of
2935 MDS RPCs in flight for a single client and add /proc controls
2936 to adjust this limit.
2938 Severity : enhancement
2940 Description: client read/write statistics
2941 Details : Add client read/write call usage stats for performance
2942 analysis of user processes.
2943 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2944 file access. extents_stats shows chunk size distribution.
2945 extents_stats_per_process show chunk size distribution per
2948 Severity : enhancement
2950 Description: mds statistics
2951 Details : Add detailed mds operations statistics in
2952 /proc/fs/lustre/mds/*/stats.
2956 Description: Failure of copying files with lustre special EAs.
2957 Details : Client side always return success for setxattr call for lustre
2958 special xattr (currently only "trusted.lov").
2963 Description: Refcount LNET uuids
2964 Details : The global LNET uuid list grew linearly with every startup;
2965 refcount repeated list entries instead of always adding to
2968 Severity : enhancement
2970 Description: Dynamic service threads
2971 Details : Within a small range, start extra service threads
2972 automatically when the request queue builds up.
2975 Frequency : mixed-endian client/server environments
2977 Description: mixed-endian crashes
2978 Details : The new msg_v2 system had some failures in mixed-endian
2981 Severity : enhancement
2983 Description: Easy OST removal
2984 Details : OSTs can be permanently deactivated with e.g. 'lctl
2985 conf_param lustre-OST0001.osc.active=0'
2987 Severity : enhancement
2989 Description: MGS proc entries
2990 Details : Added basic proc entries for the MGS showing what filesystems
2993 Severity : enhancement
2995 Description: provide MGS failover
2996 Details : Added config lock reacquisition after MGS server failover.
2998 Severity : enhancement
3000 Description: add Linux 2.4 support
3001 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
3005 Description: a large application tries to do I/O to the same resource and dies
3006 in the middle of it.
3007 Details : Check the req->rq_arrival time after the call to
3008 ost_brw_lock_get(), but before we do anything about
3009 processing it & sending the BULK transfer request. This
3010 should help move old stale pending locks off the queue as
3011 quickly as obd_timeout.
3014 Frequency : when an incorrect nid is specified during startup
3016 Description: ptlrpc connect to non-existant node causes kernel crash
3017 Details : LNET can't be re-entered from an event callback, which
3018 happened when we expire a message after the export has been
3019 cleaned up. Instead, hand the zombie cleanup off to another
3022 Severity : enhancement
3024 Description: plain/inodebits lock performance improvement
3025 Details : Grouping plain/inodebits in granted list by their request modes
3026 and bits policy, thus improving the performance of search through
3030 Frequency : only if OST filesystem is corrupted
3032 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3033 Details : for a short time RPCs with bulk IO are in the replay list,
3034 but replay of bulk IOs is unimplemented. If the OST filesystem
3035 is corrupted due to disk cache incoherency and then replay is
3036 started it is possible to trip an assertion. Avoid putting
3037 committed RPCs into the replay list at all to avoid this issue.
3040 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3042 Description: System hang on startup
3043 Details : This bug allowed the liblustre (e.g. catamount) client to
3044 return to the app before handling all startup RPCs. This
3045 could leave the node unresponsive to lustre network traffic
3046 and manifested as a server ptllnd timeout.
3048 Severity : enhancement
3050 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3051 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
3052 boolean to control whether to print peer debug info when a
3053 client's RPC times out.
3056 Frequency : only for kernels with patches from Lustre below 1.4.3
3058 Description: Remove old rdonly API
3059 Details : Remove old rdonly API which unused from at least lustre 1.4.3
3062 Frequency : only for devices with external journals
3064 Description: Set external device read-only also
3065 Details : During a commanded failover stop, we set the disk device
3066 read-only while the server shuts down. We now also set any
3067 external journal device read-only at the same time.
3070 Frequency : when upgrading from 1.4 while trying to change parameters
3072 Description: The wrong (new) MDC name was used when setting parameters for
3073 upgraded MDT's. Also allows changing of OSC (and MDC)
3074 parameters if --writeconf is specified at tunefs upgrade time.
3077 Frequency : when setting specific ost indicies
3079 Description: QOS code breaks on skipped indicies
3080 Details : Add checks for missing OST indicies in the QOS code, so OSTs
3081 created with --index need not be sequential.
3083 Severity : enhancement
3085 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3086 Details : The uninit_groups feature works in conjunction with the kernel
3087 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3088 up the pass1 processing of e2fsck. This is a read-only feature
3089 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3090 mount filesystems that have had this feature enabled.
3092 Severity : enhancement
3094 Description: Improve multi-block allocation algorithm to avoid fragmentation
3095 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3096 allocation locality and avoid filesystem fragmentation.
3098 ------------------------------------------------------------------------------
3100 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
3102 * Support for kernels:
3103 2.4.21-47.0.1.EL (RHEL 3)
3104 2.6.5-7.283 (SLES 9)
3105 2.6.9-42.0.10.EL (RHEL 4)
3106 2.6.12.6 vanilla (kernel.org)
3107 2.6.16.27-0.9 (SLES 10)
3108 * Recommended e2fsprogs version: 1.39.cfs5
3110 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3114 Frequency : occasional, depends on client load and configuration
3115 Bugzilla : 12181, 12203
3116 Description: data loss for recently-modified files
3118 Details : In some cases it is possible that recently written or created
3119 files may not be written to disk in a timely manner (this should
3120 normally be within 30s unless client IO load is very high).
3121 The problem appears as zero-length files or files that are a
3122 multiple of 1MB in size after a client crash or client eviction
3123 that are missing data at the end of the file.
3125 This problem is more likely to be hit on clients where files are
3126 repeatedly created and unlinked in the same directory, clients
3127 have a large amount of RAM, have many CPUs, the filesystem has
3128 many OSTs, the clients are rebooted frequently, and/or the files
3129 are not accessed by other nodes after being written.
3131 The presence of the problem can be detected by looking at
3132 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
3133 smaller than the second (nr_unused) then dirty files will not
3134 be flushed automatically to disk. "sync; sleep 10" should be
3135 run several times on the node before unmounting it to update
3136 Lustre (this is also safe to run on nodes without this problem).
3138 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
3139 that can cause this same problem, so customers using that kernel
3140 also need to update the kernel in addition to Lustre. In order
3141 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
3143 It is normal that files written just before a client crash (less
3144 than 30s) may not yet have been flushed to disk, even for local
3148 Frequency : frequent on thin XT3 nodes
3150 Description: UUID collision on thin XT3 Linux nodes
3151 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
3152 randomly, since we relied on an insufficiently-seeded PRNG.
3157 Description: OSS hangs after "All ost request buffers busy"
3158 Details : A deadlock between quota and journal operations caused OSS
3159 hangs after printing "All ost request buffers busy."
3162 Frequency : always on liblustre builds
3164 Description: Cleanup compiler warnings on liblustre
3167 Frequency : always on liblustre builds on XT3
3169 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
3173 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
3174 > 32000 subdirectories
3175 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
3176 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
3177 subdirectory count crosses 32000. This will aid e2fsck to
3178 correctly handle more than 32000 subdirectories.
3181 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3183 Description: System hang on startup
3184 Details : This bug allowed the liblustre (e.g. catamount) client to
3185 return to the app before handling all startup RPCs. This
3186 could leave the node unresponsive to lustre network traffic
3187 and manifested as a server ptllnd timeout.
3189 Severity : enhancement
3191 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3192 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
3193 boolean to control whether to print peer debug info when a
3194 client's RPC times out.
3199 Description: make O_SYNC working on 2.6 kernels
3200 Details : 2.6 kernels use different method for mark pages for write,
3201 so need add a code to lustre for O_SYNC work.
3206 Description: Failure to close file and release space on NFS
3207 Details : Put inode details into lock acquired in ll_intent_file_open.
3208 Use mdc_intent_lock in ll_intent_open to properly
3209 detect all kind of errors unhandled by mdc_enqueue.
3214 Description: proc file read during shutdown sometimes raced obd removal,
3216 Details : Add lock to prevent obd access after proc file removal.
3219 Frequency : Only for files larger than 4GB on 32-bit clients.
3221 Description: improperly doing page alignment of locks
3222 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
3223 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
3228 Description: RPCs being resent when they shouldn't be
3229 Details : Some RPCs that should not be resent are being resent. This
3230 can cause inconsistencies in the RPC state machine. Do not
3231 resend such requests.
3234 Frequency : rare, only with NFS export
3236 Description: Crash on NFS re-export node
3237 Details : under very unusual load conditions an assertion is hit in
3238 ll_intent_file_open()
3241 Frequency : only if OST filesystem is corrupted
3243 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3244 Details : for a short time RPCs with bulk IO are in the replay list,
3245 but replay of bulk IOs is unimplemented. If the OST filesystem
3246 is corrupted due to disk cache incoherency and then replay is
3247 started it is possible to trip an assertion. Avoid putting
3248 committed RPCs into the replay list at all to avoid this issue.
3253 Description: large O_DIRECT requests fail under memory pressure/fragmentation
3254 Details : Large single O_DIRECT read and write calls can fail to allocate
3255 a sufficiently large buffer to process the request. In case of
3256 allocation failure the allocation is retried with a smaller
3257 buffer and broken into smaller requests.
3259 Severity : enhancement
3261 Description: Add -o localflock option to simulate old noflock behaviour.
3262 Details : This will achieve local-only flock/fcntl locks coherentness.
3267 Description: versioning check is incomplete
3268 Details : Checking the version difference of client vs. server, report
3269 error if the gap is too big.
3274 Description: add support PG_writeback bit
3275 Details : add support for PG_writeback bit for Lustre, for more carefull
3276 work with page cache in 2.6 kernel. This also fix some deadlocks
3277 and remove hack for work O_SYNC with 2.6 kernel.
3279 Severity : enhancement
3281 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3282 Details : The uninit_groups feature works in conjunction with the kernel
3283 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3284 up the pass1 processing of e2fsck. This is a read-only feature
3285 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3286 mount filesystems that have had this feature enabled.
3288 Severity : enhancement
3290 Description: Improve multi-block allocation algorithm to avoid fragmentation
3291 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3292 allocation locality and avoid filesystem fragmentation.
3294 ------------------------------------------------------------------------------
3296 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
3298 * Support for kernels:
3299 2.6.9-42.0.3.EL (RHEL 4)
3300 2.6.5-7.276 (SLES 9)
3301 2.4.21-47.0.1.EL (RHEL 3)
3302 2.6.12.6 vanilla (kernel.org)
3303 2.6.16.21-0.8 (SLES10)
3304 * Recommended e2fsprogs version: 1.39.cfs2-0
3306 * The backwards-compatible /proc/sys/portals symlink has been removed
3307 in this release. Before upgrading, please ensure that you change
3308 any configuration scripts or /etc/sysctl.conf files that access
3309 /proc/sys/portals/* or sysctl portals.* to use the corresponding
3310 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
3311 in advance of the upgrade on any system running Lustre 1.4.6 or
3312 newer, since /proc/sys/lnet was added in that version.
3313 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3317 Frequency : only when quota is used
3319 Description: avoid scanning export list for quota master
3320 Details : Change the algorithms to avoid scanning export list in order
3321 to improve the efficiency.
3324 Frequency : MDS failover only, very rarely
3326 Description: "went back in time" messages on mds failover
3327 Details : The greatest transno may be lost when the current operation
3328 finishes with an error (transno==0) and the client's last_rcvd
3329 record is over-written. Save the greatest transno in the
3330 mds_last_transno for this case.
3333 Frequency : always for specific kernels and striping counts
3335 Description: client may get "Matching packet too big" without ACL support
3336 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
3337 when trying to access files in certain configurations. The
3338 clients should in fact be denied when mounting because they do
3339 not understand ACLs.
3342 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3344 Description: many clients connecting with IO in progress causes connect timeouts
3345 Details : Avoid synchronous journal commits to avoid delays caused by many
3346 clients connecting/disconnecting when bulk IO is in progress.
3347 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3348 OST_IO_PORTAL to avoid delays behind potentially many pending
3352 Frequency : occasionally with multiple writers to a single file
3354 Description: shared writes to file may result in wrong size reported by stat()
3355 Details : Allow growing of kms when extent lock is cancelled
3358 Frequency : always with random mmap IO to multi-striped file
3360 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3361 Details : Only if the hole is at the end of OST object so that kms is too
3362 small. Fix is to increase kms accordingly in ll_nopage.
3365 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3367 Description: writes to a missing object would leak memory on the OST
3368 Details : If there is an inconsistency between the MDS and OST filesystems,
3369 such that the MDS references an object that doesn't exist, writes
3370 to that object will leak memory due to incorrect cleanup in the
3371 error handling path, eventually running out of memory on the OST.
3376 Description: Creating too long symlink causes lustre errors
3377 Details : Check symlink and name lengths before sending requests to MDS.
3380 Frequency : only if flock is enabled (not on by default)
3382 Description: posix locks not released on fd closure on 2.6.9+
3383 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3384 kernels, this caused such locks not to be released on fd close and
3385 then assertions on fs unmount about still used locks.
3388 Frequency : MDS failover only, very rarely
3390 Description: clients may get ASSERTION(granted_lock != NULL)
3391 Details : When request was taking a long time, and a client was resending
3392 a getattr by name lock request. The were multiple lock requests
3393 with the same client lock handle and
3394 mds_getattr_name->fixup_handle_for_resent_request found one of the
3395 lock handles but later failed with ASSERTION(granted_lock != NULL).
3400 Description: handle->h_buffer_credits > 0, assertion failure
3401 Details : h_buffer_credits is zero after truncate, causing assertion
3402 failure. This patch extends the transaction or creates a new
3406 Frequency : NFS re-export or patchless client
3407 Bugzilla : 11179, 10796
3408 Description: Crash on NFS re-export node (__d_move)
3409 Details : We do not want to hash the dentry if we don't have a lock.
3410 But if this dentry is later used in d_move, we'd hit uninitialised
3411 list head d_hash, so we just do this to init d_hash field but
3412 leave dentry unhashed.
3415 Frequency : NFS re-export or patchless client
3417 Description: NFS exports has problem with symbolic link
3418 Details : lustre client didn't properly install dentry when re-exported
3419 to NFS or running patchless client.
3422 Frequency : NFS re-export or patchless client
3424 Description: Various nfs/patchless fixes.
3425 Details : fixes reuse disconected alias for lookup process - this fixes
3426 warning "find_exported_dentry: npd != pd",
3427 fix permission error with open files at nfs.
3428 fix apply umask when do revalidate.
3431 Frequency : occasional
3433 Description: Crash on NFS re-export node
3434 Details : calling clear_page() on the wrong pointer triggered oops in
3435 generic_mapping_read().
3438 Frequency : rarely, using O_DIRECT IO
3440 Description: unaligned directio crashes client with LASSERT
3441 Details : check for unaligned buffers before trying any requests.
3444 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3446 Description: stale data returned from RAID cache
3447 Details : If only a small amount of IO is done to the RAID device before
3448 reading it again it is possible to get stale data from the RAID
3449 cache instead of reading it from disk.
3452 Frequency : always for sles10 kernel
3454 Description: sles10 support
3455 Details : ll_follow_link: compile fixes and using of nd_set_link
3456 under newer kernels.
3459 Frequency : depends on arch, kernel and compiler version, always on sles10
3462 Description: recursive or deep enough symlinks cause stack overflow
3463 Details : getting rid of large stack-allocated variable in
3467 Frequency : depends on hardware
3469 Description: lustre write performance loss in the SLES10 kernel
3470 Details : the performance loss is caused by using of write barriers in the
3471 ext3 code. The SLES10 kernel turns barrier support on by
3472 default. The fix is to undo that change for ldiskfs.
3474 ------------------------------------------------------------------------------
3476 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3478 * Support for kernels:
3479 2.6.9-42.0.3EL (RHEL 4)
3480 2.6.5-7.276 (SLES 9)
3481 2.4.21-47.0.1.EL (RHEL 3)
3482 2.6.12.6 vanilla (kernel.org)
3486 Frequency : quota enabled and large files being deleted
3488 Description: releasing more than 4GB of quota at once hangs OST
3489 Details : If a user deletes more than 4GB of files on a single OST it
3490 will cause the OST to spin in an infinite loop. Release
3491 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3496 Description: statfs data retrieved from /proc may be stale or zero
3497 Details : When reading per-device statfs data from /proc, in the
3498 {kbytes,files}_{total,free,avail} files, it may appear
3499 as zero or be out of date.
3502 Frequency : systems with MD RAID1 external journal devices
3504 Description: lconf's call to blkid is confused by RAID1 journal devices
3505 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3506 of returning all block devices that match the journal UUID.
3509 Frequency : always, for aggregate stripe size over 4GB
3511 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3512 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3513 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3516 Frequency : always if "lfs find" used on a local file/directory
3518 Description: "lfs find" segfaults if used on a local file/directory
3519 Details : The case where a directory component was not specified wasn't
3520 handled correctly. Handle this properly.
3523 Frequency : always on ppc64
3525 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3526 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3527 perform endianness conversion.
3530 Frequency : rarely (truncate to non-zero file size after write under load)
3531 Bugzilla : 10730, 10687
3532 Description: Files padded with zeros to next 4K multiple
3533 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3534 it is possible that files that are truncated to a non-zero size
3535 immediately after being written are filled with zero bytes beyond
3536 the truncated size. No file data is lost.
3538 Severity : enhancement
3540 Description: Allow recovery/failover for liblustre clients.
3541 Details : liblustre clients were unaware of failover configurations until
3544 Severity : enhancement
3546 Description: user file locks should fail when not mounting with flock option
3547 Details : Set up an error-returning stub in ll_file_operations.lock field
3548 to prevent incorrect behaviour when client is mounted without
3549 flock option. Also, set up properly f_op->flock field for
3553 Frequency : always on ia64
3555 Description: "lfs df" loops on printing out MDS statfs information
3556 Details : The obd_ioctl_data was not initialized and in some systems
3557 this caused a failure during the ioctl that did not return
3558 an error. Initialize the struct and return an error on failure.
3561 Frequency : SLES 9 only
3563 Description: Error of copying files with lustre special EAs as root
3564 Details : Client side always return success for setxattr call for lustre
3565 special xattr (currently only "trusted.lov").
3568 Frequency : rarely on clusters with both ia64+i386 clients
3570 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3571 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3572 clients are doing concurrent writes to the same file it is
3573 possible that the ia64 clients may LASSERT because the OST
3574 extent locks are not PAGE_SIZE aligned. Ensure that grown
3575 locks are always aligned on the request boundary.
3578 Frequency : specific use, occasional
3580 Description: Overwriting in use executable truncates on-disk binary image
3581 Details : If one node attempts to overwrite an executable in use by
3582 another node, we now correctly return ETXTBSY instead of
3583 truncating the file.
3585 Severity : enhancement
3587 Description: Async OSC create to avoid the blocking unnecessarily.
3588 Details : If a OST has no remain object, system will block on the creating
3589 when need to create a new object on this OST. Now, ways use
3590 pre-created objects when available, instead of blocking on an
3591 empty osc while others are not empty. If we must block, we block
3592 for the shortest possible period of time.
3597 Description: chmod on Lustre root is propagated to other clients
3598 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3599 invalid by the follow_mount time.
3602 Frequency : liblustre clients only
3604 Description: Race in 'instant cancel' lock handling could lead to such locks
3605 never to be granted in case of SMP MDS
3606 Details : Do not destroy not yet granted but cbpending locks in
3610 Frequency : replay/resend of open
3612 Description: non null lock assetion failure in mds_intent_policy
3613 Details : Trying to replay/resend lockless open requests resulted in
3614 mds_open() returning 0 with no lock. Now it sets a flag if
3615 a lock is going to be returned.
3617 Severity : enhancement
3619 Description: Checksum enhancements
3620 Details : New checksum enhancements allow for resending RPCs that failed
3623 Severity : enhancement
3625 Description: Tunables on number of dirty pages in cacche
3626 Details : Allow to set limit on number of dirty pages cached.
3631 Description: client crash on unmount - lock still has references
3632 Details : In some error handling cases it was possible to leak a lock
3633 reference on a client while accessing a file. This was not
3634 harmful to the client during operation, but would cause the
3635 client to crash when the filesystem is unmounted.
3638 Frequency : specific case, rare
3640 Description: ETXTBSY on mds though file not in use by client
3641 Details : ETXTBSY is no longer incorrectly returned when attempting to
3642 chmod or chown a directory that the user previously tried to
3643 execute or a currently-executing binary.
3646 Frequency : extremely rare except on liblustre-based clients
3648 Description: Lustre space not freed when files are deleted
3649 Details : Clean up open-unlinked files after client eviction. Previously
3650 the unlink was skipped and the files remained as orphans.
3655 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3656 Details : In some cases it was possible to send a blocking callback to
3657 a client doing a glimpse, even though that client didn't get
3658 a lock granted. When the glimpse lock is cancelled on the OST
3659 the freed lock is left on the waiting list and corrupted the list.
3662 Frequency : all core dumps
3664 Description: Broke core dumps to lustre
3665 Details : Negative dentry may be unhashed if parent does not have UPDATE
3666 lock, but some callers, e.g. do_coredump, expect dentry to be
3667 hashed after successful create, hash it in ll_create_it.
3669 ------------------------------------------------------------------------------
3671 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3673 * Support for kernels:
3674 2.6.9-42.0.2.EL (RHEL 4)
3675 2.6.5-7.276 (SLES 9)
3676 2.4.21-40.EL (RHEL 3)
3677 2.6.12.6 vanilla (kernel.org)
3681 Frequency : always on RHEL 3
3683 Description: Number of open files grows over time
3684 Details : The number of open files grows over time, whether or not
3685 Lustre is started. This was due to a filp leak introduced
3686 by one of our kernel patches.
3688 ------------------------------------------------------------------------------
3690 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3692 * Support for kernels:
3693 2.6.9-42.EL (RHEL 4)
3694 2.6.5-7.276 (SLES 9)
3695 2.4.21-40.EL (RHEL 3)
3696 2.6.12.6 vanilla (kernel.org)
3701 Bugzilla : 5719, 9635, 9792, 9684
3702 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3703 Details : If a server is under heavy load and cannot reply to new
3704 connection requests before the client resends the (re)connect,
3705 the connection handling code can behave badly if two service
3706 threads are concurrently handing separate (re)connections from
3707 the same client. Add better locking to the connection handling
3708 code, and ensure that only a single connection will be processed
3709 for a given client UUID, even if the lock is dropped.
3711 Severity : enhancement
3713 Description: add TCP zero-copy support to kernel
3714 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3715 sends if the hardware supports scatter-gather and checksumming.
3716 This allows socklnd to do client-write and server-read more
3717 efficiently and reduce CPU utilization from skbuf copying.
3720 Frequency : only if NFS exporting from client
3722 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3723 Details : If setting attributes on a file created under NFS that had
3724 never been opened it would be possible to oops the client
3725 if the file had no objects.
3728 Frequency : always for liblustre
3730 Description: liblustre client does MDS+OSTs setattr RPC for each write
3731 Details : When doing a write from a liblustre client, the client
3732 incorrectly issued an RPC to the MDS and each OST the file was
3733 striped over in order to update the timestamps. When writing
3734 with small chunks and many clients this could overwhelm the MDS
3735 with RPCs. In all cases it would slow down the write because
3736 these RPCs are unnecessary.
3738 Severity : enhancement
3740 Description: allow number of MDS service threads to be changed at module load
3741 Details : It is now possible to change the number of MDS service threads
3742 running. Adding "options mds mds_num_threads={N}" to the MDS's
3743 /etc/modprobe.conf will set the number of threads for the next
3744 time Lustre is restarted (assuming the "mds" module is also
3745 reloaded at that time). The default number of threads will
3746 stay the same, 32 for most systems.
3751 Description: OST crash if filesystem is unformatted or corrupt
3752 Details : If an OST is started on a device that has never been formatted
3753 or if the filesystem is corrupt and cannot even mount then the
3754 error handling cleanup routines would dereference a NULL pointer.
3759 Description: NULL pointer deref in llap_from_page.
3760 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3761 incorrect) mapping if the page was truncated/reclaimed while it was
3762 searched for. Check for this condition and skip such pages when
3763 doing readahead. Introduce extra check to llap_from_page() to
3764 verify page->mapping->host is non-NULL (so page is not anonymous).
3767 Frequency : Sometimes when using sys_sendfile
3769 Description: "page not covered by a lock" warnings from ll_readpage
3770 Details : sendfile called ll_readpage without right page locks present.
3771 Now we introduced ll_file_sendfile that does necessary locking
3772 around call to generic_file_sendfile() much like we do in
3776 Frequency : with certain MDS communication failures at client mount time
3778 Description: NULL pointer deref after failed client mount
3779 Details : a client connection request may delayed by the network layer
3780 and not be sent until after the PTLRPC layer has timed out the
3781 request. If the client fails the mount immediately it will try
3782 to clean up before the network times out the request. Add a
3783 reference from the request import to the obd device and delay
3784 the cleanup until the network drops the request.
3787 Frequency : occasionally during client (re)connect
3789 Description: assertion failure during client (re)connect
3790 Details : processing a client connection request may be delayed by the
3791 client or server longer than the client connect timeout. This
3792 causes the client to resend the connection request. If the
3793 original connection request is replied in this interval, the
3794 client may trip an assertion failure in ptlrpc_connect_interpret()
3795 which thought it would be the only running connect process.
3798 Frequency : only with obd_echo servers and clients that are rebooted
3800 Description: kernel BUG accessing uninitialized data structure
3801 Details : When running an obd_echo server it did not start the ping_evictor
3802 thread, and when a client was evicted an uninitialized data
3803 structure was accessed. Start the ping_evictor in the RPC
3804 service startup instead of the OBD startup.
3806 Severity : enhancement
3807 Bugzilla : 10193 (patchless)
3808 Description: Remove dependency on various unexported kernel interfaces.
3809 Details : No longer need reparent_to_init, exit_mm, exit_files,
3810 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3813 Frequency : rare (only users of deprecated and unsupported LDAP config)
3815 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3816 Details : LDAP apparently contains 'lustreName' attributes instead of
3817 'name'. A simple remapping of the name is sufficient.
3820 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3822 Description: waiting_locks_callback trips kernel BUG if client is evicted
3823 Details : Running with the dump_on_timeout debug flag turned on makes
3824 it possible that the waiting_locks_callback() can try to dump
3825 the Lustre kernel debug logs from an interrupt handler. Defer
3826 this log dumping to the expired_lock_main() thread.
3828 Severity : enhancement
3830 Description: Support NFS exporting on 2.6 kernels.
3831 Details : Implement non-rawops metadata methods for NFS server to use without
3832 changing NFS server code.
3835 Frequency : very rare (synthetic metadata workload only)
3837 Description: two racing renames might cause an MDS thread to deadlock
3838 Details : Running the "racer" program may cause one MDS thread to rename
3839 a file from being the source of a rename to being the target of
3840 a rename at exactly the same time that another thread is doing
3841 so, and the second thread has already enqueued these locks after
3842 doing a lookup of the target and is trying to relock them in
3843 order. Ensure that we don't try to re-lock the same resource.
3846 Frequency : only very large systems with liblustre clients
3848 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3849 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3850 all OSTs in parallel. This allows the eviction of stale liblustre
3851 clients to proceed much faster than if they were done in series,
3852 and also offers similar improvements for other set_info RPCs.
3857 Description: excessive CPU usage during initial read phase on client
3858 Details : During the initial read phase on a client, it would agressively
3859 retry readahead on the file, consuming too much CPU and impacting
3860 performance (since 1.4.5.8). Improve the readahead algorithm
3861 to avoid this, and also improve some other common cases (read
3862 of small files in particular, where "small" is files smaller than
3863 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3868 Description: MDS crash when receiving packet with unknown intent.
3869 Details : Do not LBUG in unknown intent case, just return -EFAULT
3871 Severity : enhancement
3872 Bugzilla : 9293, 9385
3873 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3874 Details : Do not serialize getattr (non-intent version) and statfs.
3877 Frequency : occasional, when OST network is overloaded/intermittent
3879 Description: client evicted by OST after bulk IO timeout
3880 Details : If a client sends a bulk IO request (read or write) the OST
3881 may evict the client if it is unresposive to its data GET/PUT
3882 request. This is incorrect if the network is overloaded (takes
3883 too long to transfer the RPC data) or dropped the OST GET/PUT
3884 request. There is no need to evict the client at all, since
3885 the pinger and/or lock callbacks will handle this, and the
3886 client can restart the bulk request.
3889 Frequency : Always when mmapping file with no objects
3891 Description: client crashes when mmapping file with no objects
3892 Details : Check that we actually have objects in a file before doing any
3893 operations on objects in ll_vm_open, ll_vm_close and
3899 Description: Request leak when working with deleted CWD
3900 Details : Introduce advanced request refcount tracking for requests
3901 referenced from lustre intent.
3903 Severity : Enhancement
3905 Description: Cache open file handles on client.
3906 Details : MDS now will return special lock along with openhandle, if
3907 requested and client is allowed to hold openhandle, even if unused,
3908 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3909 closing files for every read/write openration.
3911 Severity : Enhancement
3913 Description: Cache open negative dentries on client when possible.
3914 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3915 negative dentries on lock revocation.
3920 Description: Remounting a client read-only wasn't possible with a zconf mount
3921 Details : It wasn't possible to remount a client read-only with llmount.
3923 Severity : enhancement
3924 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3925 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3926 MPICH is included to add Lustre-specific ADIO interfaces.
3927 This is based closely on the UFS ADIO layer and only differs
3928 in file creation, in order to allow the OST striping to be set.
3929 This is user-contributed code and not supported by CFS.
3934 Description: extended inode attributes (immutable, append-only) work improperly
3935 when 2.4 and 2.6 kernels are used on client/server or vice versa
3936 Details : Introduce kernel-independent values for these flags.
3938 Severity : enhancement
3941 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3942 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3943 granularity. Now, Fractional MB granularity is be supported,
3944 this is very useful for low memory system.
3946 Severity : enhancement
3948 Description: Getattr by fid
3949 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3950 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3951 have locks and avoids getattr rpc requests for it on every stat.
3954 Frequency : Always, for filesystems larger than 2TB
3956 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3957 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3958 filesystems larger than 2TB to be mounted with mballoc enabled.
3961 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3963 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3964 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3965 filesystems larger than 2TB is mounted then the kernel will
3966 silently corrupt the start of the filesystem. CONFIG_LBD is
3967 enabled for all CFS-supported kernels, but the possibility of
3968 this happening with a modified kernel config exists.
3970 Severity : enhancement
3972 Description: add client O_DIRECT support for 2.6 kernels
3973 Details : It is now possible to do O_DIRECT reads and writes to files
3974 in the Lustre client mountpoint on 2.6 kernel clients.
3976 Severity : enhancement
3978 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3979 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3980 parallel, not waiting for response from every OST before sending
3981 a rpc to the next OST.
3986 Description: setattr vs write race when updating file timestamps
3987 Details : Client processes that update a file timestamp into the past
3988 right after writing to the file (e.g. tar) it is possible that
3989 the updated file modification time can be reset to the current
3990 time due to a race between processing the setattr and write RPC.
3992 Severity : enhancement
3994 Description: Bring 'lfs find' closer in line with regular Linux find.
3995 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3996 -print0 options and obtains all the needed info through the lustre
3999 Severity : enhancement
4001 Description: support up to 1024 configured devices on one node
4002 Details : change obd_dev array from statically allocated to dynamically
4003 allocated structs as they are first used to reduce memory usage
4008 Description: Flush dirty partially truncated pages during truncate
4009 Details : Immediatelly flush partially truncated pages in filter_setattr,
4010 this way we completely avoid having any pages in page cache on OST
4011 and can retire ugly workarounds during writes to flush such pages.
4016 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
4017 Details : protect inode from truncation within vfs_unlink() context
4018 just take a reference before calling vfs_unlink() and release it
4019 when parent's i_sem is free.
4024 Description: last_id value checked outside lock on OST caused LASSERT failure
4025 Details : If there were multiple MDS->OST object precreate requests in
4026 flight, it was possible that the OST's last object id was checked
4027 outside a lock and incorrectly tripped an assertion. Move checks
4028 inside locks, and discard old precreate requests.
4031 Frequency : always, if extents are used on OSTs
4033 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
4034 Details : OSTs using the extents format would not zero the high 16 bits of
4035 the index physical block number. This is not a problem for any
4036 OST filesystems smaller than 16TB, and no kernels support ext3
4037 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
4038 new/modified files) and can be fixed for existing filesystems
4039 with e2fsprogs-1.39-cfs1.
4044 Description: import connection selection may be incorrect if timer wraps
4045 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
4046 import connections to be ignored if the 32-bit jiffies counter
4047 wraps. Use a 64-bit jiffies counter.
4050 Frequency : very large clusters immediately after boot
4052 Description: LNET request buffers exhausted under heavy short-term load
4053 Details : If a large number of client requests are generated on a service
4054 that has previously never seen so many requests it is possible
4055 that the request buffer growth cannot keep up with the spike in
4056 demand. Instead of dropping incoming requests, they are held in
4057 the LND until the RPC service can accept more requests.
4060 Frequency : Sometimes during replay
4062 Description: Assertion failure in ll_local_open after replay.
4063 Details : If replay happened on an open request reply before we were able
4064 to set replay handler, reply will become not swabbed tripping the
4065 assertion in ll_local_open. Now we set the handler right after
4066 recognising of open request
4069 Frequency : very rare
4071 Description: kernel reports "badness in vsnprintf"
4072 Details : Reading from the "recovery_status" /proc file in small chunks
4073 may cause a negative length in lprocfs_obd_rd_recovery_status()
4074 call to vsnprintf() (which is otherwise harmless). Exit early
4075 if there is no more space in the output buffer.
4077 Severity : enhancement
4079 Description: clear OBD RPC statistics by writing to them
4080 Details : It is now possible to clear the OBD RPC statistics by writing
4081 to the "stats" file.
4086 Description: Client mtime is not the same on different clients after utimes
4087 Details : In some cases, the client was using the utimes() syscall on
4088 a file cached on another node. The clients now validate the
4089 ctime from the MDS + OSTs to determine which one is right.
4094 Description: Inability to activate failout mode
4095 Details : lconf script incorrectly assumed that in pythong string's numeric
4096 value is used in comparisons.
4099 Frequency : always with multiple stripes per file
4101 Description: Inefficient object allocation for mutli-stripe files
4102 Details : When selecting which OSTs to stripe files over, for files with
4103 a stripe count that divides evenly into the number of OSTs,
4104 the MDS is always picking the same starting OST for each file.
4105 Return the OST selection heuristic to the original design.
4110 Description: mount failures may take full timeout to return an error
4111 Details : Under some heavy load conditions it is possible that a
4112 failed mount can wait for the full obd_timeout interval,
4113 possibly several minutes, before reporting an error.
4114 Instead return an error as soon as the status is known.
4116 Frequency : quota enabled and large files being deleted
4118 Description: releasing more than 4GB of quota at once hangs OST
4119 Details : If a user deletes more than 4GB of files on a single OST it
4120 will cause the OST to spin in an infinite loop. Release
4121 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
4126 Description: statfs data retrieved from /proc may be stale or zero
4127 Details : When reading per-device statfs data from /proc, in the
4128 {kbytes,files}_{total,free,avail} files, it may appear
4129 as zero or be out of date.
4132 Frequency : systems with MD RAID1 external journal devices
4134 Description: lconf's call to blkid is confused by RAID1 journal devices
4135 Details : Use the "blkid -l" flag to locate the MD RAID device instead
4136 of returning all block devices that match the journal UUID.
4139 Frequency : always, for aggregate stripe size over 4GB
4141 Description: assertion fails when trying to use 4GB stripe size
4142 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
4143 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
4146 Frequency : always on ppc64
4148 Description: the first write on an ext3 filesystem with mballoc got stuck
4149 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
4150 perform endianness conversion.
4152 ------------------------------------------------------------------------------
4154 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
4156 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
4157 INTEROPERATE with older versions automatically. Please read the
4158 user documentation before upgrading any part of a live system.
4159 * WARNING: Lustre networking configuration changes are required with
4160 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
4163 * Support for kernels:
4164 2.6.9-22.0.2.EL (RHEL 4)
4165 2.6.5-7.244 (SLES 9)
4166 2.6.12.6 vanilla (kernel.org)
4169 Severity : enhancement
4170 Bugzilla : 7981/8208
4171 Description: Introduced Lustre Networking (LNET)
4172 Details : LNET is new networking infrastructure for Lustre, it includes
4173 a reorganized network configuration mode (see the user
4174 documentation for full details) as well as support for routing
4175 between different network fabrics. Lustre Networking Devices
4176 (LNDs) for the supported network fabrics have also been
4177 created for this new infrastructure.
4179 Severity : enhancement
4180 Description: Introduced Access control lists
4181 Details : clients can set ACLs on files and directories in order to have
4182 more fine-grained permissions than the standard Unix UGO+RWX.
4183 The MDS must be started with the "-o acl" mount option.
4185 Severity : enhancement
4186 Description: Introduced filesystem quotas
4187 Details : Administrators may now establish per-user quotas on the
4190 Severity : enhancement
4192 Description: Configuration change for the XT3
4193 The PTLLND is now used to run Lustre over Portals on the XT3
4194 The configure option(s) --with-cray-portals are no longer used.
4195 Rather --with-portals=<path-to-portals-includes> is used to
4196 enable building on the XT3. In addition to enable XT3 specific
4197 features the option --enable-cray-xt3 must be used.
4202 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
4203 Details : A race between allocating a new llcd and re-getting the llcd_lock
4204 allowed another thread to grab newly-allocated llcd.
4206 Severity : enhancement
4208 Description: 2.6 OST async journal commit and locking fix to improve performance
4209 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
4210 2.6 kernels are now async as they already were in 2.4 kernels so
4211 that they can commit concurrently with the network bulk transfer.
4212 For block-allocated files the filter allocation semaphore is held
4213 to avoid filesystem fragmentation during allocation. BKL lock
4214 removed for 2.6 xattr operations where it is no longer needed.
4219 Description: lconf incorrectly determined whether two IP networks could talk
4220 Details : In some more complicated routing and multiple-network
4221 configurations, lconf will avoid trying to make a network
4222 connection to a disjoint part of the IP space. It was doing the
4223 math incorrectly for one set of cases.
4228 Description: Fix for potential infinite loop processing records in an llog.
4229 Details : If an llog record is corrupted/zeroed, it is possible to loop
4230 forever in llog_process(). Validate the llog record length
4231 and skip the remainder of the block on error.
4234 Frequency : occasional (liblustre only)
4236 Description: liblustre could not open files whose last component is a symlink
4237 Details : sysio_path_walk() would incorrectly pass the open intent to
4238 intermediate path components.
4241 Frequency : rare (liblustre only with non-standard tuning)
4242 Bugzilla : 7201 (7350)
4243 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
4244 Details : llu_lookup_finish_locks() tries to set lock data on a lock
4245 after it has been released, only do this for referenced locks
4247 Severity : enhancement
4249 Description: specifying an (invalid) directory default stripe_size of -1
4250 would reset the directory default striping
4251 Details : stripe_size -1 was used internally to signal directory stripe
4252 removal, now use "all default" to signal dir stripe removal
4253 as a directory striping of "all default" is not useful
4256 Frequency : common for large clusters running liblustre clients
4258 Description: doing an ls when liblustre clients are running is slow
4259 Details : sending a glimpse AST to a liblustre client waits for every AST
4260 to time out, as liblustre clients will not respond. Since they
4261 cannot cache data we refresh the OST lock LVB from disk instead.
4263 Severity : enhancement
4265 Description: doing an ls at the same time as file IO can be slow
4266 Details : enqueue and other "small" requests can be blocked behind many
4267 large IO requests. Create a new OST IO portal for non-IO
4268 requests so they can be processed faster.
4271 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
4273 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
4274 created file with mode 000
4275 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
4276 allow an ftruncate() as a non-root user to a file with mode 000.
4277 Lustre now respects this flag to disable mode checks when
4278 truncating a file owned by the user
4281 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
4284 Description: Revoking locks from clients that went dead or catatonic might take
4286 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
4287 cancellation of such locks instant on servers without waiting for
4288 any reply from clients. Clients drops these locks when cancel
4289 notification from server is received without replying.
4292 Frequency : liblustre-only, when liblustre client dies or becomes busy
4294 Description: Doing ls on Linux clients can take a long time with active
4296 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
4297 granting such locks to it in the first place if possible. Locks
4298 are taken by proxy on the OST during the read or write and
4299 dropped immediately afterward. Add connect flags handling, do
4300 not grant locks to liblustre clients for glimpse ASTs.
4302 Severity : enhancement
4304 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
4305 Details : Existing read-ahead algorithm is tuned for the case of streamlined
4306 sequential reads and behaves badly with applications doing random
4307 reads. Improve it by reading ahead at least read region, and
4308 avoiding excessive large RPC for small reads.
4310 Severity : enhancement
4312 Description: Creating more than 1000 files for a single job may cause a load
4313 imbalance on the OSTs if there are also a large number of OSTs.
4314 Details : qos_prep_create() uses an OST index reseed value that is an
4315 even multiple of the number of available OSTs so that if the
4316 reseed happens in the middle of the object allocation it will
4317 still utilize the OSTs as uniformly as possible.
4322 Description: OST or MDS may oops in ping_evictor_main()
4323 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
4324 but doesn't restart at beginning of obd_exports_timed list
4327 Severity : enhancement
4329 Description: improve by-nid export eviction on the MDS and OST
4330 Details : allow multiple exports with the same NID to be evicted at one
4331 time without re-searching the exports list.
4334 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
4336 Description: MDS may oops in groups_free()
4337 Details : in rare race conditions a newly allocated group_info struct is
4338 freed again, and this can be NULL. The 2.4 compatibility code
4339 for groups_free() checked for a NULL pointer, but 2.6 did not.
4342 Frequency : common for liblustre clients doing little filesystem IO
4343 Bugzilla : 9352, 7313
4344 Description: server may evict liblustre clients accessing contended locks
4345 Details : if a client is granted a lock or receives a completion AST
4346 with a blocking AST already set it would not reply to the AST
4347 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4350 Frequency : lfs setstripe, only systems with more than 160 OSTs
4352 Description: unable to set striping with a starting offset beyond OST 160
4353 Details : llapi_create_file() incorrectly limited the starting stripe
4354 index to the maximum single-file stripe count.
4357 Frequency : LDAP users only
4359 Description: lconf did not handle in-kernel recovery with LDAP properly
4360 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4362 Severity : enhancement
4364 Description: bind OST threads to NUMA nodes to improve performance
4365 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4366 node and do their allocations there to localize memory access
4368 Severity : enhancement
4370 Description: llmount can determine client NID directly from Myrinet (GM)
4371 Details : the client NID code from gmnalnid was moved directly into
4372 llmount, removing the need to use this or specifying the
4373 client NID explicitly when mounting GM clients with zeroconf
4376 Frequency : if client is started with down MDS
4378 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4379 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4380 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4381 ptlrpc_import_delayed_req() because we didn't check if the
4382 request was interrupted, and we also didn't break out of the
4383 event loop if there was no timeout
4388 Description: data loss during non-page-aligned writes to a single file from
4389 both multiple nodes and multiple threads on one node at same time
4390 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4391 inconsistency led to false short-reads, that were cached and later
4392 used by ->prepare_write() to fill in partially written page,
4393 leading to data loss.
4396 Frequency : always, if lconf --abort_recovery used
4398 Description: lconf --abort_recovery fails with 'Operation not supported'
4399 Details : lconf was attempting to abort recovery on the MDT device and not
4402 Severity : enhancement
4404 Description: remove cleanup logs
4405 Details : replace lconf-generated cleanup logs with lustre internal
4406 cleanup routines. Eliminates the need for client-cleanup and
4409 Severity : enhancement
4411 Description: add support for EAs (user and system) on lustre filesystems
4412 Details : it is now possible to store extended attributes in the Lustre
4413 client filesystem, and with the user_xattr mount option it
4414 is possible to allow users to store EAs on their files also
4416 Severity : enhancement
4418 Description: Add possibility (config option) to show minimal available OST free
4420 Details : When compiled with --enable-mindf configure option, statfs(2)
4421 (and so, df) will return least minimal free space available from
4422 all OSTs as amount of free space on FS, instead of summary of
4423 free spaces of all OSTs.
4425 Severity : enhancement
4427 Description: do not expand extent locks acquired on OST-side
4428 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4429 by server: they are not cached anyway.
4432 Frequency : when mmap is used/binaries executed from Lustre
4434 Description: Unmmap pages before throwing them away from read cache.
4435 Details : llap_shrink cache now attempts to unmap pages before discarding
4436 them (if unmapping failed - do not discard). SLES9 kernel has
4437 extra checks that trigger if this unmapping is not done first.
4442 Description: lconf didn't resolve symlinks before checking to see whether a
4443 given mountpoint was already in use
4446 Frequency : when migrating failover services
4447 Bugzilla : 6395, 9514
4448 Description: When migrating a subset of services from a node (e.g. failback
4449 from a failover service node) the remaining services would
4450 time out and evict clients.
4451 Details : lconf --force (implied by --failover) sets the global obd_timeout
4452 to 5 seconds in order to quickly disconnect, but this caused
4453 other RPCs to time out too quickly. Do not change the global
4454 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4456 Severity : enhancement
4457 Frequency : if MDS is started with down OST
4458 Bugzilla : 9439,5706
4459 Description: Allow startup/shutdown of an MDS without depending on the
4460 availability of the OSTs.
4461 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4462 Add appropriate locking and lov-osc refcounts for safe
4463 cleaning. Add osc abort_inflight calls in case the
4464 synchronize never started.
4467 Frequency : occasional (Cray XT3 only)
4469 Description: root not authorized to access files in CRAY_PORTALS environment
4470 Details : The client process capabilities were not honoured on the MDS in
4471 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4472 been accessed by an authorized user then root was able to access
4473 the file on the local client also. The root user capabilities
4474 are now allowed on the MDS, as this environment has secure UID.
4477 Frequency : occasional
4479 Description: ldiskfs "too long searching" message happens too often
4480 Details : A debugging message (otherwise harmless) prints too often on
4481 the OST console. This has been reduced to only happen when
4482 there are fragmentation problems on the filesystem.
4487 Description: Division by zero in statfs when all OSCs are inactive
4488 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4489 lov_statfs divides by value returned by lov_get_stripecnt().
4493 Bugzilla : 9489, 3273
4494 Description: First write from each client to each OST was only 4kB in size,
4495 to initialize client writeback cache, which caused sub-optimal
4496 RPCs and poor layout on disk for the first writen file.
4497 Details : Clients now request an initial cache grant at (re)connect time
4498 and so that they can start streaming writes to the cache right
4499 away and always do full-sized RPCs if there is enough data.
4500 If the OST is rebooted the client also re-establishes its grant
4501 so that client cached writes will be honoured under the grant.
4506 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4507 Details : Now I/O RPCs go to different portal number and (presumably) fast
4508 lock requests (and glimses) and other RPCs get their own service
4509 threads pool that should be able to service those RPCs
4512 Severity : enhancement
4514 Description: Ability to exchange lustre version between client and servers and
4515 issue warnings at client side if client is too old. Also for
4516 liblustre clients there is ability to refuse connection of too old
4518 Details : New 'version' field is added to connect data structure that is
4519 filled with version info. That info is later checked by server and
4523 Frequency : rare, liblustre only.
4524 Bugzilla : 9296, 9581
4525 Description: Two simultaneous writes from liblustre at offset within same page
4526 might proceed at the same time overwriting eachother with stale
4528 Details : I/O lock withing llu_file_prwv was released too early, before data
4529 actually was hitting the wire. Extended lock-holding time until
4530 server acknowledges receiving data.
4533 Frequency : extremely rare. Never observed in practice.
4535 Description: avoid generating lustre_handle cookie of 0.
4536 Details : class_handle_hash() generates handle cookies by incrementing
4537 global counter, and can hit 0 occasionaly (this is unlikely, but
4538 not impossible, because initial value of cookie counter is
4539 selected randonly). Value of 0 is used as a sentinel meaning
4540 "unassigned handle" --- avoid it. Also coalesce two critical
4541 sections in this function into one.
4543 Severity : enhancement
4545 Description: allow liblustre clients to delegate truncate locking to OST
4546 Details : To avoid overhead of locking, liblustre client instructs OST to
4547 take extent lock in ost_punch() on client's behalf. New connection
4548 flag is added to handle backward compatibility.
4550 Severity : enhancement
4551 Bugzilla : 4928, 7341, 9758
4552 Description: allow number of OST service threads to be specified
4553 Details : a module parameter allows the number of OST service threads
4554 to be specified via "options ost ost_num_threads={N}" in the
4555 OSS's /etc/modules.conf or /etc/modprobe.conf.
4559 Bugzilla : 6146, 9635, 9895
4560 Description: servers crash with bad pointer in target_handle_connect()
4561 Details : In rare cases when a client is reconnecting it was possible that
4562 the connection request was the last reference for that export.
4563 We would temporarily drop the export reference and get a new
4564 one, but this may have been the last reference and the export
4565 was just destroyed. Get new reference before dropping old one.
4567 Severity : enhancement
4568 Frequency : if client is started with failover MDS
4570 Description: Allow multiple MDS hostnames in the mount command
4571 Details : Try to read the configuration from all specified MDS
4572 hostnames during a client mount in case the "primary"
4575 Severity : enhancement
4577 Description: Stop sending data to evicted clients as soon as possible.
4578 Details : Check if the client we are about to send or are sending data to
4579 was evicted already. (Check is done every second of waiting,
4580 for which l_wait_event interface was extended to allow checking
4581 of exit condition at specified intervals).
4584 Frequency : rare, normally only when NFS exporting is done from client
4586 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4588 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4591 Frequency : always, if srand() is called before liblustre initialization
4593 Description: Liblustre uses system PRNG disturbing its usage by user application
4594 Details : Introduce internal to lustre fast and high-quality PRNG for
4595 lustre usage and make liblustre and some other places in generic
4596 lustre code to use it.
4598 Severity : enhancement
4599 Bugzilla : 9477, 9557, 9870
4600 Description: Verify that the MDS configuration logs are updated when xml is
4601 Details : Check if the .xml configuration logs are newer than the config
4602 logs stored on the MDS and report an error if this is the case.
4603 Request --write-conf, or allow starting with --old_conf.
4605 Severity : enhancement
4607 Description: Handle symlinks in the path when checking if Lustre is mounted.
4608 Details : Resolve intermediate symlinks when checking if a client has
4609 mounted a filesystem to avoid duplicate client mounts.
4614 Description: lconf can hit an error exception but still return success.
4615 Details : The lconf command catches the Command error exception at the top
4616 level script context and will exit with the associated exit
4617 status, but doesn't ensure that this exit status is non-zero.
4622 Description: failure of ptlrpc thread startup can cause oops
4623 Details : Starting a ptlrpc service thread can fail if there are a large
4624 number of threads or the server memory is very fragmented.
4625 Handle this without oopsing.
4628 Frequency : always, only if liblustre and non-default acceptor port was used
4630 Description: liblustre cannot connect to servers with non-default acceptor port
4631 Details : tcpnal_set_default_params() was not called and was therefore
4632 ignoring the environment varaible TCPNAL_PORT, as well as other
4633 TCPNAL_ environment variables
4638 Description: two objects could be created on the same OST for a single file
4639 Details : If an OST is down, in some cases it was possible to create two
4640 objects on a single OST for a single file. No problems other
4641 than potential performance impact and spurious error messages.
4645 Bugzilla : 5681, 9562
4646 Description: Client may oops in ll_unhash_aliases
4647 Details : Client dcache may become inconsistent in race condition.
4648 In some cases "getcwd" can fail if the current directory is
4654 Description: Inode refcounting problems in NFS export code
4655 Details : link_raw functions used to call d_instantiate without obtaining
4656 extra inode reference first.
4660 Bugzilla : 9942, 9903
4661 Description: Referencing freed requests leading to crash, memleaks with NFS.
4662 Details : We used to require that call to ll_revalidate_it was always
4663 followed by ll_lookup_it. Also with revalidate_special() it is
4664 possible to call ll_revalidate_it() twice for the same dentry
4665 even if first occurence returned success. This fix changes semantic
4666 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4667 reference on a request referred from the intent.
4668 ll_intent_release() then releases such a request.
4671 Frequency : rare, normally benchmark loads only
4673 Description: unlinked inodes were kept in memory on the client
4674 Details : If a client is repeatedly creating and unlinking files it
4675 can accumulate a lot of stale inodes in the inode slab cache.
4676 If there is no other client load running this can cause the
4677 client node to run out of memory. Instead flush old inodes
4678 from client cache that have the same inode number as a new inode.
4681 Frequency : SLES9 2.6.5 kernel and long filenames only
4682 Bugzilla : 9969, 10379
4683 Description: utime reports stale NFS file handle
4684 Details : SLES9 uses out-of-dentry names in some cases, which confused
4685 the lustre dentry revalidation. Change it to always use the
4689 Frequency : rare, unless heavy write-truncate concurrency is continuous
4690 Bugzilla : 4180, 6984, 7171, 9963, 9331
4691 Description: OST becomes very slow and/or deadlocked during object unlink
4692 Details : filter_destroy() was holding onto the parent directory lock
4693 while truncating+unlinking objects. For very large objects this
4694 may block other threads for a long time and slow overall OST
4695 responsiveness. It may also be possible to get a lock ordering
4696 deadlock in this case, or run out of journal credits because of
4697 the combined truncate+unlink. Solution is to do object truncate
4698 first in one transaction without parent lock, and then do the
4699 final unlink in a new transaction with the parent lock. This
4700 reduces the lock hold time dramatically.
4703 Frequency : rare, 2.4 kernels only
4705 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4706 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4707 before unmounting their filesystem in order to avoid deadlock.
4708 The kernel_locked() function in 2.4 kernels only checks whether
4709 the kernel lock is held, not whether it is this process that is
4710 holding it as 2.6 kernels do.
4715 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4716 Details : The client ptlrpc code may be trying to reconnect to a down
4717 server before a previous connection attempt has timed out.
4718 Increase the reconnect interval to be longer than the connection
4719 timeout interval to avoid sending duplicate connections to
4723 Frequency : echo_client brw_test command
4725 Description: fix echo_client to work with OST preallocated code
4726 Details : OST preallocation code (5137) didn't take echo_client IO path
4727 into account: echo_client calls filter methods outside of any
4728 OST thread and, hence, there is no per-thread preallocated
4729 pages and buffers to use. Solution: hijack pga pages for IO. As
4730 a byproduct, this avoids unnecessary data copying.
4734 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4735 Description: Client can oops in mdc_commit_close() after open replay
4736 Details : It was possible for the MDS to return an open request with no
4737 transaction number in mds_finish_transno() if the client was
4738 evicted, but without actually returning an error. Clients
4739 would later try to replay that open and may trip an assertion
4740 Simplify the client close codepath, and always return an error
4741 from the MDS in case the open is not successful.
4744 Frequency : rare, 2.6 OSTs only
4746 Description: OST may deadlock under high load on fragmented files
4747 Details : If there was a heavy load and highly-fragmented OST filesystems
4748 it was possible to have all the OST threads deadlock waiting on
4749 allocation of biovecs, because the biovecs were not released
4750 until the entire RPC IO was completed. Instead, release biovecs
4751 as soon as they are complete to ensure forward IO progress.
4753 Severity : enhancement
4755 Description: Support for specifying external journal device at mount
4756 Details : If an OST or MDS device is formatted with an external journal
4757 device, this device major/minor is stored in the ext3 superblock
4758 and may not be valid for failover. Allow detecting and
4759 specifying the external journal at mount time.
4764 Description: Mounting an MDS with pending unlinked files may cause oops
4765 Details : target_finish_recovery() calls mds_postrecov() which returned
4766 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4767 considers this an error and immediately begins cleaning up the
4768 lov, just after starting the mds_lov process
4770 Severity : enhancement
4772 Description: Implement 'lfs df' to report actual free space on per-OST basis
4773 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4774 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4775 usage of objects; '-h' to report in human readable format.
4777 ------------------------------------------------------------------------------
4779 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4786 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4787 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4788 at startup during orphan unlinking. Instead, skip buddy update
4789 before setup, as it will be regenerated after recovery is complete.
4794 Description: If an OST is inactive, its locks might reference stale inodes.
4795 Details : lov_change_cbdata() must iterate over all namespaces, even if
4796 they are inactive to clear inode references from the lock.
4798 Severity : enhancement
4799 Frequency : occasional, if non-standard max_dirty_mb used
4801 Description: Client will block write RPCs if not enough grant
4802 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4803 then the client will block writes while waiting for another RPC
4804 to complete instead of consuming its dirty limit. With change
4805 we get improved performance when max_dirty_mb is small.
4807 Severity : enhancement
4808 Bugzilla : 3389, 6253
4809 Description: Add support for supplementary groups on the MDS.
4810 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4811 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4812 lookups for user supplementary groups into a cache.
4816 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4817 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4818 for liblustre O_EXCL is enforced by MDS.
4822 Description: readdir never returns NULL in liblustre.
4823 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4824 dentry in struct dirent.
4828 Frequency : liblustre only, depends on application IO pattern
4829 Description: liblustre clients evicted if not contacting servers
4830 Details : Don't put liblustre clients into the ping_evictor list, so
4831 they will not be evicted by the pinger ever.
4833 Severity : enhancement
4835 Description: Add ability to evict clients by NID from MDS.
4836 Details : By echoing "nid:$NID" string into
4837 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4838 $NID would be instantly evicted from this MDS and from all active
4839 OSTs connected to it.
4843 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4844 Details : lookup_it_finish() used to query file size from OSTs that was not
4849 Description: service threads change working directory to that of init
4850 Details : Starting lustre service threads may pin the working directory
4851 of the parent thread, making that filesystem busy. Threads
4852 now change to the working directory of init to avoid this.
4856 Frequency : during shutdown only
4857 Description: shutdown with a failed MDS or OST can cause unmount to hang
4858 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4863 Frequency : 2.6 only
4864 Description: chmod/chown may include an extra supplementary group
4865 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4866 supplementary group and if none is specified this is used.
4869 Bugzilla : 5479 (6816)
4871 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4872 Details : If lookup is in progress on a file that is unlinked we might try
4873 to revalidate the inode and fail in revalidate after lookup is
4874 complete and ll_file_open() enqueues the open again but
4875 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4878 Frequency : always, if lconf --abort_recovery used
4880 Description: lconf --abort_recovery fails with 'Operation not supported'
4881 Details : lconf was attempting to abort recovery on the MDT device and not
4884 ------------------------------------------------------------------------------
4886 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4891 Frequency : rare (only unsupported configurations with a node running as an
4893 Bugzilla : 6514, 5137
4894 Description: Mounting a Lustre file system on a node running as an OST could
4896 Details : OSTs now preallocates memory needed to write out data at
4897 startup, instead of when needed, to avoid having to
4898 allocate memory in possibly low memory situations.
4899 Specifically, if the file system is mounted on on OST,
4900 memory pressure could force it to try to write out data,
4901 which it needed to allocate memory to do. Due to the low
4902 memory, it would be unable to do so and the node would
4903 become unresponsive.
4905 Severity : enhancement
4907 Description: Addition of lconf --service command line option
4908 Details : lconf now accepts a '--service <arg>' option, which is
4909 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4911 Severity : enhancement
4913 Description: Failover mode is now the default for OSTs.
4914 Details : By default, OSTs will now run in failover mode. To return to
4915 the old behaviour, add '--failout' to the lmc line for OSTs.
4917 Severity : enhancement
4919 Description: Health checks are now provided for MDS and OSTs
4920 Details : Additional detailed health check information on MSD and OSTs
4921 is now provided through the procfs health_check value.
4924 Frequency : occasional, depends on IO load
4926 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4927 after numerous create/delete cycles
4928 Details : The ext3 inode allocation policy would not allocate new inodes
4929 very well on the OSTs because there are no new directories
4930 being created. Instead we look for groups with free space if
4931 the parent directories are nearly full.
4936 Description: Network or server problems during mount may cause partially
4937 mounted clients instead of returning an error.
4938 Details : The config llog parsing code may overwrite the error return
4939 code during mount error handling, returning success instead
4945 Description: MDS can fail to allocate large reply buffers
4946 Details : After long uptimes the MDS can fail to allocate large reply
4947 buffers (e.g. zconf client mount config records) due to memory
4948 fragmentation or consumption by the buffer cache. Preallocate
4949 some large reply buffers so that these replies can be sent even
4950 under memory pressure.
4954 Frequency : rare (liblustre)
4955 Description: fsx running with liblustre complained that using truncate() to
4956 extend the file doesn't work. This patch corrects that issue.
4957 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4958 ATTR_SIZE and lsm use in llu_setattr_raw.
4962 Frequency : rare, only 2.6 kernels
4963 Description: Unusual file access patterns on the MDS may result in inode
4964 data being lost in very rare circumstances.
4965 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4966 code in the 2.6 kernel caused the inode and/or EA data not to
4967 be read from disk, causing single-file corruption.
4971 Frequency : rare, only 2.6 filesystems using extents
4972 Description: Heavy concurrent write and delete load may cause data corruption.
4973 Details : It was possible under high-load situations to have an extent
4974 metadata block in the block device cache from a just-unlinked
4975 file overwrite a newly-allocated data block. We now unmap any
4976 metadata buffers that alias just-allocated data blocks.
4980 Frequency : filesystems with default stripe_count larger than 77
4981 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4982 Details : lconf specifies an inode size of 4096 bytes when the default
4983 stripe_count is larger than 77. This conflicts with the default
4984 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4985 this case to avoid pinning too much memory for large EAs.
4987 ------------------------------------------------------------------------------
4989 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4994 Frequency : rare (extremely heavy IO load with hundreds of clients)
4996 Description: Client is evicted, gets IO error writing to file
4997 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4998 caused clients to be evicted for AST timeouts. The fixes in
4999 bug 5192 mean we no longer need to have such short AST timeouts
5000 so ldlm_timeout has been increased.
5003 Frequency : occasional during --force or --failover shutdown under load
5004 Bugzilla : 5949, 4834
5005 Description: Server oops/LBUG if stopped with --force or --failover under load
5006 Details : a collection of import/export refcount and cleanup ordering
5007 issues fixed for safer force cleanup
5010 Frequency : only filesystems larger than 120 OSTs
5011 Bugzilla : 5990, 6223
5012 Description: lfs getstripe would oops on a very large filesystem
5013 Details : lov_getconfig used kfree on vmalloc'd memory
5016 Frequency : only filesystems exporting via NFS to Solaris 10 clients
5017 Bugzilla : 6242, 6243
5018 Description: reading from files that had been truncated to a non-zero size
5019 but never opened returned no data
5020 Details : ll_file_read() reads zeros from no-object files to EOF
5025 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
5026 Details : There's little harm in aborting MDS/OSS recovery and letting it
5027 try again, so I removed the LASSERT and return an error instead.
5029 Severity : enhancement
5031 Description: New debugging infrastructure for tracking down data corruption
5032 Details : The I/O checksum code was replaced to: (a) control it at runtime,
5033 (b) cover more of the client-side code path, and (c) try to narrow
5034 down where problems occurred
5038 Bugzilla : 3819, 4364, 4397, 6313
5039 Description: Racing close and eviction MDS could cause assertion in mds_close
5040 Details : It was possible to get multiple mfd references during close and
5041 client eviction, leading to one thread referencing a freed mfd.
5043 Severity: : enhancement
5044 Bugzilla : 3262, 6359
5045 Description: Attempts to reconnect to servers are now more aggressive.
5046 Details : This builds on the enhanced upcall-less recovery that was added
5047 in 1.4.2. When trying to reconnect to servers, clients will
5048 now try each server in the failover group every 10 seconds. By
5049 default, clients would previously try one server every 25 seconds.
5054 Description: After recovery, certain operations trigger a failed
5055 assertion on a client.
5056 Details : Failing over an mds, using lconf -d --failover, while a
5057 client was doing a readdir() call would cause the client to
5058 LBUG after recovery completed and the readdir() was resent.
5060 Severity : enhancement
5062 Description: Default groups are now added by lconf
5063 Details : You can now run lconf --group <servicename> without having to
5064 manually add groups with lmc.
5067 Frequency : occasional
5069 Description: Nodes with an elan id of 0 trigger a failed assertion
5072 Frequency : always when accessing e.g. tty/console device nodes
5074 Description: tty and some other devices nodes cannot be used on lustre
5075 Details : file's private_data field is used by device data and lustre
5076 values in there got lost. New field was added to struct file to
5077 store fs-specific private data.
5080 Frequency : when exporting Lustre via NFS
5082 Description: NFSD failed occasionally when looking up a path component
5083 Details : NFSD is looking up ".." which was broken in ext3 directories
5084 that had grown large enough to become hashed.
5087 Frequency : Clusters with multiple interfaces not on the same subnet
5089 Description: Nodes will repeatedly try to reconnect to an interface which it
5090 cannot reach and report an error to the log.
5091 Details : Extra peer list entries will be created by lconf with some peers
5092 unreachable. lconf now validates the peer before adding it.
5095 Frequency : Only if a default stripe is set on the filesystem root.
5097 Description: Setting a default stripe on the filesystem root prevented the
5098 filesystem from being remounted.
5099 Details : The client was sending extra request flags in the root getattr
5100 request and did not allocate a reply buffer for the dir EA.
5103 Frequency : occasional, higher if lots of files are accessed by one client
5104 Bugzilla : 6159, 6097
5105 Description: Client trips assertion regarding lsm mismatch/magic
5106 Details : While revalidating inodes the VFS looks up inodes with ifind()
5107 and in rare cases can find an inode that is being freed.
5108 The ll_test_inode() code will free the lsm during ifind()
5109 when it finds an existing inode and then the VFS later attaches
5110 this free lsm to a new inode.
5114 Bugzilla : 6422, 7030
5115 Description: MDS deadlock between mkdir and client eviction
5116 Details : Creating a new file via mkdir or mknod (starting a transaction
5117 and getting the ns lock) can deadlock with client eviction
5118 (gets ns lock and trying to finish a synchronous transaction).
5121 Frequency : occasional
5122 Description: While starting a server, the fsfilt_ext3 module could not be
5124 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
5125 kernels. Previously, lconf would still use the ext3 name
5126 when trying to load modules. Now, it will correctly use
5127 ext3 on 2.4 and ldiskfs on 2.6.
5129 Severity : enhancement
5130 Description: The default stripe count has been changed to 1
5131 Details : The interpretation of the default stripe count (0, to lfs
5132 or lmc) has been changed to mean striping across a single
5133 OST, rather than all available. For general usage we have
5134 found a stripe count of 1 or 2 works best.
5136 Severity : enhancement
5137 Description: Add support for compiling against Cray portals.
5138 Details : Conditional compiling for some areas that are different
5142 Frequency : occasional
5143 Bugzilla : 6409, 6834
5144 Description: Creating files with an explicit stripe count may lead to
5145 a failed assertion on the MDS
5146 Details : If some OSTs are full or unavailable, creating files may
5147 trigger a failed assertion on the MDS. Now, Lustre will
5148 try to use other servers or return an error to the
5152 Frequency : occasional
5154 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
5155 caused lock timeout during readahead (since 1.4.2).
5156 Details : Processes doing readahead might match a lock that hasn't been
5157 granted yet if there are overlapping and conflicting lock
5158 requests. The readahead process waits on ungranted lock
5159 (original lock is CBPENDING), while OST waits for that process
5160 to cancel CBPENDING read lock and eventually evicts client.
5162 Severity : enhancement
5164 Description: Initial enabling of flock support for clients
5165 Details : Implements fcntl advisory locking and file status functions.
5166 This feature is provided as an optional mount flag (default
5167 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
5168 locking are implemented yet, and those that are are not guaranteed
5169 to be completely correct in production environments.
5170 mount -t lustre -o [flock|noflock] ...
5173 Frequency : occasional
5175 Description: OSTs running 2.4 kernels but with extents enabled might trip an
5176 assertion in the ext3 JBD (journaling) layer.
5177 Details : The b_committed_data struct is protected by the big kernel lock
5178 in 2.4 kernels, serializing journal_commit_transaction() and
5179 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
5180 access to this struct. In 2.6 kernels there is finer grained
5181 locking to improve SMP performance of the JBD layer.
5185 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
5190 Description: merge and cleanup kernel patches.
5193 -----------------------------------------------------------------------------
5195 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
5197 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
5198 versions of Lustre. Please update all servers and clients to
5199 version 1.4.2 or later at the same time. You must also run
5200 "lconf --write-conf {config}.xml" on the MDS while it is stopped
5201 to update the configuration logs.
5203 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
5204 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
5205 - swab llog records of type '0' so we get proper header size/idx (5861)
5206 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
5207 - fix rename of one directory over another leaking an inode (5953)
5208 - avoid SetPageDirty on 2.6 (5981)
5209 - don't re-add just-being-destroyed locks to the waiting list (5653)
5210 - when creating new directories, inherit the parent's custom
5211 striping settings if present parent (3048)
5212 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
5213 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
5214 - don't hold client locks on temporary worklist from l_lru (5666)
5215 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
5216 - automatically evict dead clients (5921)
5217 - Update file size properly in create+truncate+fstat case (6196)
5218 - Do not unhash mountpoint dentries, do not allow removal of
5220 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
5221 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
5222 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
5223 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
5224 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
5225 - Correctly handle reads of files with no objects (6243)
5226 - lctl recover will also mark a device active if deactivate used (5933)
5228 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
5229 - allow --write-conf on an MDS with different nettype than client (5619)
5230 - don't write config llogs to MDS for mounts not from that MDS (5617)
5231 - lconf should create multiple TCP connections from a client (5201)
5232 - init scripts are now turned off by default; run chkconfig --on
5233 lustre and chkconfig --on lustrefs to use them
5234 - upcalls are no longer needed for clients to recover to failover
5236 - add --abort-recovery option to lconf to abort recovery on device
5238 - add support for an arbitrary number of OSTs (3026)
5239 - Quota support protocol changes.
5240 - forward compatibility changes to wire structs (6007)
5241 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
5242 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
5243 - improved "lustre status" script
5244 - initialize blocksize for non-regular files (6062)
5245 - added --disable-server and --disable-client configure options (5782)
5246 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
5247 - Vanilla 2.4.29 support
5248 - increase maximum number of obd devices to 520 (6242)
5249 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
5250 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
5251 SLES 9 2.6.5 kernels
5252 - limit stripes per file to 160 (the maximum EA size) (6093)
5254 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5257 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
5258 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
5259 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
5260 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
5261 - small ext3 extents cleanups and fixes (5733)
5262 - improved mballoc code, several small races and bugs fixed (5733, 5638)
5263 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
5264 - avoid needless client->OST connect, fix handle mismatch (5317)
5265 - fix DLM error path that led to out-of-sync client, long delays (5779)
5266 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
5267 - fix several locking issues related to i_size (5492,5624,5654,5672)
5268 - don't move pending lock onto export if it is already evicted (5683)
5269 - fix kernel oops when creating .foo in unlinked directory (5548)
5270 - fix deadlock in obdfilter statistics vs. object create (5811)
5271 - use time_{before,after} to avoid timer jiffies wrap (5882)
5272 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
5273 - Do not leak request if server was not able to process it (5154)
5274 - If mds_open unable to find parent dir, make that negative lookup(5154)
5275 - don't create new directories with extent-mapping (5909, 5936)
5277 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
5278 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
5279 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
5280 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
5281 - support for RHEL4 kernel on i686 (5773)
5282 - provide error messages when incompatible logs are encountered (5898)
5284 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
5285 * version 1.4.0.10 (1.4.1 release candidate 1)
5287 - don't keep a lock reference when lock is not granted (4238)
5288 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5289 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
5290 - drop import inflight refcount on signal_completed_replay error (5255)
5291 - unlock page after async write error during send (3677)
5292 - handle missing objects in filter_preprw_read properly (5265)
5293 - no transno return for symlink open, don't save no-trasno open (3440)
5294 - don't try to complete elan receive that already failed (4012)
5295 - free RPC server reply state on error (5406)
5296 - clean up thread from ptlrpc_start_thread() on error (5160)
5297 - readahead could read extra page into cache that wasn't ejected (5388)
5298 - prevent races in class_attach/setup/cleanup/detach (5260)
5299 - don't dereference de->d_inode after l_dput of de (5458)
5300 - use "int" for stripe value returned from lock_to_stripe (5544)
5301 - mballoc allocation and error-checking fixes in 2.6 (5504)
5302 - block device patches to fix I/O request sizes in 2.6 (5482)
5303 - look up hostnames for IB nals (5602)
5304 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
5305 - don't start multiple acceptors for the same port (5277)
5306 - fix incorrect LASSERT in mds_getattr_name (5635)
5307 - export a proc file for general "ping" checking (5628)
5308 - fix "lfs check" to not block when the MDS is down (5628)
5310 - service request history (4965)
5311 - put {ll,lov,osc}_async_page structs in a single slab (4699)
5312 - create an "evict_client" /proc entry on OSTs, like the MDS has
5313 - fix mount usage message, return errors per mount(8) (5168)
5314 - change grep [] to grep "[]" in tests so they work in more UMLs
5315 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
5316 - remove ancient LOV_MAGIC_V0 EA support (5047)
5317 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
5318 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
5319 - allow readahead allocations to fail when low on memory (5383)
5320 - mmap locking landed again, after considerable improvement (2828)
5321 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
5323 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
5326 - send OST transaction number in read/write reply to free req (4966)
5327 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5328 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5329 - the watchdog thread now runs as interruptible (5246)
5330 - flock/lockf fixes (but it's still disabled, pending 5135)
5331 - don't use EXT3 constants in llite code (5094)
5332 - memory shortage at startup could cause assertion (5176)
5334 - reorganization of lov code
5335 - single portals codebase
5337 - add extents/mballoc support (5025)
5338 - direct I/O reads in the obdfilter (4048)
5339 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5341 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5344 - send OST transaction number in read/write reply to free req (4966)
5345 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5346 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5347 - the watchdog thread now runs as interruptible (5246)
5348 - handle missing objects in filter_preprw_read properly (5265)
5349 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5350 - drop import inflight refcount on signal_completed_replay error (5255)
5351 - unlock page after async write error during send (3677)
5352 - return original error code on reconstructed replies (3761)
5353 - no transno return for symlink open, don't save no-trasno open (3440)
5355 - add pid to ldlm debugging output (4922)
5356 - bump the watchdog timeouts -- we can't handle 30sec yet
5357 - extra debugging for orphan dentry/inode bug (5259)
5359 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5362 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5363 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5364 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5365 - fix NULL dereference in /proc/sys/portals/routes (4827)
5366 - allow failed mdc_close() operations to be interrupted (4561)
5367 - stop precreate on OST before MDS would time out on it (4778)
5368 - don't send partial-page writes before EOF from client (4410)
5369 - discard client grant for sub-page writes on large-page clients (4520)
5370 - don't free dentries not owned by NFS code, check generation (4806)
5371 - fix lsm leak if mds_create_objects() fails (4801)
5372 - limit debug_daemon file size, always print CERROR messages (4789)
5373 - use transno after validating reply (3892)
5374 - process timed out requests if import state changes (3754)
5375 - update mtime on OST during writes, return in glimpse (4829)
5376 - add mkfsoptions to LDAP (4679)
5377 - use ->max_readahead method instead of zapping global ra (5039)
5378 - don't interrupt __l_wait_event() during strace
5380 - add software watchdogs to catch hung threads quickly (4941)
5381 - make lustrefs init script start after nfs is mounted
5382 - fix CWARN/ERROR duplication (4930)
5383 - return async write errors to application if possible (2248)
5384 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5385 - print NAL number in %x format (4645)
5386 - update barely-supported suse-2.4.21-171 series (4842)
5387 - support for sles 9 %post scripts
5388 - support for building 2.6 kernel-source packages
5389 - support for sles km_* packages
5391 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5394 - ignore -ENOENT errors in osc_destroy (3639)
5395 - notify osc create thread that OSC is being cleaned up (4600)
5396 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5397 - reconstruct ost_handle() like mds_handle() (4657)
5398 - create a new thread to do import eviction to avoid deadlock (3969)
5399 - let lconf resolve symlinked-to devices (4629)
5400 - don't unlink "objects" from directory with default EA (4554)
5401 - hold socknal file ref over connect in case target is down (4394)
5402 - allow more than 32000 subdirectories in a single directory (3244)
5403 - fix blocks count for O_DIRECT writes (3751)
5404 - OST returns ENOSPC from object create when no space left (4539)
5405 - don't send truncate RPC if file size isn't changing (4410)
5406 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5407 - bind to privileged port in socknal and tcpnal (3689)
5409 - rate limit CERROR/CWARN console message to avoid overload (4519)
5410 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5411 - basic mmap support (3918)
5412 - kernel patch series update from b1_4 (4711)
5414 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5417 - avoid crash during MDS cleanup with OST shut down (2775)
5418 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5419 - don't use bad inodes on the MDS (3744)
5420 - dynamic object preallocation to improve recovery speed (4236)
5421 - don't hold spinlock over lock dumping or change debug flags (4401)
5422 - don't zero obd_dev when it is force cleaned (3651)
5423 - print grants to console if they go negative (4431)
5424 - "lctl deactivate" will stop automatic recovery attempts (3406)
5425 - look for existing locks in ldlm_handle_enqueue() (3764)
5426 - don't resolve lock handle twice in recovery avoiding race (4401)
5427 - revalidate should check working dir is a directory (4134)
5429 - don't always mark "slow" obdfilter messages as errors (4418)
5431 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5434 - don't close LustreDB during write_conf until it is done (3860)
5435 - fix typo in lconf for_each_profile (3821)
5436 - allow dumping logs from multiple threads at one time (3820)
5437 - don't allow multiple threads in OSC recovery (3812)
5438 - fix debug_size parameters (3864)
5439 - fix mds_postrecov to initialize import for llog ctxt (3121)
5440 - replace config semaphore with spinlock (3306)
5441 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5442 - don't allow enqueue to complete on a destroyed export (3822)
5443 - down write_lock before checking llog header bitmap (3825)
5444 - recover from lock replay timeout (3764)
5445 - up llog sem before sending rpc (3652)
5446 - reduce ns lock hold times when setting kms (3267)
5447 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5448 - fix NULL deref and obd_dev leak on setup error (3312)
5449 - replace some LBUG about llog ops with error handling (3841)
5450 - don't match INVALID dentries from d_lookup and spin (3784)
5451 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5452 - fix invalid assertion in ptlrpc_set_wait (3880)
5454 - add libwrap support for the TCP acceptor (3996)
5455 - add /proc/sys/portals/routes for non-root route listing (3994)
5456 - allow setting MDS UUID in .xml (2580)
5457 - print the stack of a process that LBUGs (4228)
5459 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5462 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5463 - make sure to unset replay flag from failed open requests (3440)
5464 - if default stripe count is 0, use OST count for inode size (3636)
5465 - update parent mtime/ctime on client for create/unlink (2611)
5466 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5467 - free recovery state on server during a forced cleanup (3571)
5468 - unregister_reply for resent reqs (3063)
5469 - loop back devices mounting and status check on 2.6 (3563)
5470 - fix resource-creation race that can provoke i_size == 0 (3513)
5471 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5472 - more debugging for page-accounting assertion (3746)
5473 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5474 - avoid deadlock after precreation failure (3758)
5475 - fix race and lock order deadlock in orphan handling (3450, 3750)
5476 - add validity checks when grabbing inodes from l_ast_data (3599)
5478 - add /proc/.../recovery_status to obdfilter (3428)
5479 - lightweight CDEBUG infrastructure, debug daemon (3668)
5480 - change default OSC RPC parameters to be better on small clusters
5481 - turn off OST read cache for files smaller than 32MB
5482 - install man pages and include them in rpms (3100)
5483 - add new init script for (un)mounting lustre filesystems (2593)
5484 - run chkconfig in %post for init scripts (3701)
5485 - drop scimac NAL (unmaintained)
5487 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5490 - clean kiobufs before and after use (3485)
5491 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5492 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5493 - change init script to fail more clearly if not run as root (1528)
5494 - allow clients to reconnect during replay (1742)
5495 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5496 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5497 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5498 - fix cond_resched (3554)
5499 - only evict unfinished clients after recovery (3515)
5500 - allow bulk resend, prevent data loss (3570)
5501 - dynamic ptlrpc request buffer allocation (2102)
5502 - don't allow unlinking open directory if it isn't empty (2904)
5503 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5504 - remove extraneous obd dereference causing LASSERT failure (3334)
5505 - don't use get_cycles() when creating temp. files on the mds (3156)
5506 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5507 - handle EEXIST for set-stripe, set proper directory name (3336)
5509 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5510 - fix ksocknal_fmb_callback() error messages (2918)
5512 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5515 - don't copy lvb into (possibly NULL) reply on error (2983)
5516 - don't deref dentry after dput, don't free lvb on error (2922)
5517 - use the kms to determine writeback rpc length (2947)
5518 - increment oti_logcookies when osc is inactive (2948)
5519 - update client's i_blocks count via lvb messages (2543)
5520 - handle intent open/close of special files properly (1557)
5521 - mount MDS with errors=remount-ro, like obdfilter (2009)
5522 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5523 - don't use cancelling-locks' kms values (2947)
5524 - use highest lock extent for kms, not last one (2925)
5525 - don't dereference ERR_PTR() dentry in error handling path (3107)
5526 - fix thread race in portals_debug_dumplog() (3122)
5527 - create lprocfs device entries at setup instead of at attach (1519)
5528 - common AST error handler, don't evict client on completion race (3145)
5529 - zero nameidata in detach_mnt in 2.6 (3118)
5530 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5531 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5532 - initialize RPC timeout timer earlier for 2.6 (3219)
5533 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5534 - print nal/nid for unknown nid (3258)
5535 - additional checks for oscc recovery before doing precreate (3284)
5536 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5537 - don't crash in mdc_close for bad permissions on open (3285)
5538 - zero i_rdev for non-device files (3147)
5539 - clear page->private before handing to FS, better assertion (3119)
5540 - tune the read pipeline (3236)
5541 - fix incorrect decref of invalidated dentry (2350)
5542 - provide read-ahead stats and refine rpc in flight stats (3328)
5543 - don't hold journal transaction open across create RPC (3313)
5544 - update atime on MDS at close time (3265)
5545 - close LDAP connection when recovering to avoid server load (3315)
5546 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5547 - don't leak open file on MDS after open resend (3325)
5548 - serialize filter_precreate and filter_destroy_precreated (3329)
5549 - loop device shouldn't call sync_dev() for nul device (3092)
5550 - clear page cache after eviction (2766)
5551 - resynchronize MDS->OST in background (2824)
5552 - refuse to mount the same filesystem twice on same mountpoint (3394)
5553 - allow llmount to create routes for mounting behind routers (3320)
5554 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5555 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5556 - fix rare problem with rename on htree directories (3417)
5558 - allow default OST striping configuration per directory (1414)
5559 - fix compilation for qswnal for 2.6 kernels (3125)
5560 - increase maximum number of MDS request buffers for large systems
5561 - change liblustreapi to be useful for external progs like lfsck (3098)
5562 - increase local configuration timeout for slow disks (3353)
5563 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5565 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5568 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5569 - don't overwrite extent policy data in reply if lock was blocked (2901)
5570 - drop filter export grants atomically with removal from device (2663)
5571 - del obd_self_export from work_list in class_disconnect_exports (2908)
5572 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5573 - swab reply message in mdc_close, other PPC fixes (2464)
5574 - fix destroying of named logs (2325)
5575 - overwrite old logs when running lconf --write_conf (2264)
5576 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5577 - fix race in target_handle_connect (2898)
5578 - mds_reint_create() should take same inode create lock (2926)
5579 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5580 - don't close files for self_export to avoid uninitialized obd (2936)
5581 - allow MDS with the same name as client node (2939)
5582 - hold dentry reference for closed log files for unlink (2325)
5583 - reserve space for all logs during transactions (2059)
5584 - don't evict page beyond end of stripe extent (2925)
5585 - don't oops on a deleted current working directory (2399)
5586 - handle hard links to targets without a parent properly (2517)
5587 - don't dereference NULL lock when racing during eviction (2867)
5588 - don't grow lock extents when lots of conflicting locks (2919)
5590 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5593 - account for cache space usage on clients to avoid data loss (974)
5594 - lfsck support in lustre kernel code (2349)
5595 - reduce journal credits needed for BRW writes (2370)
5596 - orphan handling to avoid losing space on client/server crashes
5597 - ptlrpcd can be blocked, stopping ALL progress (2477)
5598 - use lock value blocks to assist in proper KMS, faster stat (1021)
5599 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5600 - recovery for initial connections (2355)
5601 - fixes for mds_cleanup_orphans (1934)
5602 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5603 - block all file creations until orphan recovery completes (1901)
5604 - client remove rq_connection from request struct (2423)
5605 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5606 - recovery timer race (2670)
5607 - mdc_close recovey bug (2532)
5608 - ptlrpc cleanup bug (2710)
5609 - mds timeout on local locks (2588)
5610 - namespace lock held during RPCs (2431)
5611 - handle interrupted sync write properly (2503)
5612 - don't try to handle a message that hasn't been replied to (2699)
5613 - client assert failure during cleanup after abort recovery (2701)
5614 - leak mdc device after failed mount (2712)
5615 - ptlrpc_check_set allows timedout requests to complete (2714)
5616 - wait for inflight reqs when ptlrpcd finishes (2710)
5617 - make sure unregistered services are removed from the srv_list
5618 - reset bulk XID's when resending them (caught by 1138 test)
5619 - unregister_bulk after timeout
5620 - fix lconf error (2694)
5621 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5622 - readahead locks pages, leaves pending causing memory pressure (2673)
5623 - increase OST request buffers to 4096 on large machines (2729)
5624 - fix up permission of existing directories in simple_mkdir (2661)
5625 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5626 - don't assign transno to errored transactions (2742)
5627 - don't delete objects on OST if given a bogus objid from MDS (2751)
5628 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5629 - if rq_no_resend, then timeout request after recovery (2432)
5630 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5631 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5632 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5633 - MDS recovery shouldn't delete objects that it has given out (2730)
5634 - if enqueue arrives after completion, don't clobber LVB (2819)
5635 - don't unlock pages twice when trigger_group_io returns error (2814)
5636 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5637 - don't write pages to disk if there was an error (1450)
5638 - don't ping imports that have recovery disabled (2676)
5639 - take buffered bytes into account when balancing socknal conn (2817)
5640 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5641 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5642 - remove little-endian swabbing of llog records (1987)
5643 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5644 - retry reposting request buffers if they fail (1191)
5645 - grow extent at grant time to avoid granting a revoked lock (2809)
5646 - lock revoke doesn't evict page if covered by a second lock (2765)
5647 - disable VM readahead to avoid reading outside lock extents (2805)
5649 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5650 - updated kernel patches for hp-2.4.20 kernel (2681)
5652 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5655 - fix truncated write corruption (2366)
5656 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5658 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5659 - ensure len(uuid) < 37 in lmc (1171)
5660 - fix ia64 OOPS in llog_test (2255)
5661 - zero end of page at obdfilter for partial page writes (2648)
5662 - don't leave stale dentries around after renames (bug 2428)
5663 - fix timeouts when evicting a client with a single lock held (2642)
5664 - set deadline for the initial HELLO message to drain (2634)
5665 - print out dotted-quad IP addresses in the socknal (2302)
5667 - additional debugging for MDS client eviction problem (2443)
5668 - fix mkfsoptions support for osts (2603, 2604)
5670 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5673 - add series for the vanilla 2.6.0 kernel
5674 - add series for the vanilla 2.4.24 kernel
5675 - add series for a cray x86/64 UL kernel drop
5676 - fix xattr patches for the vanilla 2.4.19 series
5678 - generate true UUIDs in lmc (1171)
5679 - have portals stack dumping break in UML (2466)
5680 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5681 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5682 - wake callback waiting threads on client eviction (2460)
5683 - Add --ptldebug and --subsystem to lmc (1719)
5684 - update assertion to allow safe interrupt allocation
5685 - set rq_no_resend for cancel requests (2432)
5686 - recalculate ptlrpcd timeout after resend (2494)
5687 - call vfs_rmdir when removing pending directories (2368)
5688 - fix renaming a file to itself (2429)
5689 - lmc creates a default one-stripe lov (2454)
5690 - expand procfs space to handle large clusters (2326)
5691 - increase UML stack to avoid overflow
5692 - update lconf's list of debug and subsystem masks
5693 - fix lfs find --obd (2510)
5694 - /proc tunable for disabling filter read caching (2591)
5695 - stop rpm packages from altering slapd.conf (2301)
5696 - disable nagle in the socknal under 0conf (2578)
5697 - choose mds inode size based on stripe count (2572)
5698 - fix kernel-source rpm problems (2516)
5700 - add --disable-doc to avoid pdf generation (2421)
5701 - update documentation, tests, type-os, comments
5702 - avoid format warnings on ia64
5703 - remove the TOE NAL
5704 - tiny code cleanups by removing unused fields
5706 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5709 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5710 - remove the most blatant lies from BUILDING file (2371)
5711 - change default debug level to reasonable production setting
5712 - reduce client side cache size to reduce cache flush time
5713 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5714 - make TCP zerocopy and pinger support enabled by default (2476)
5715 - sync writes completed after process exits caused crashes (2319)
5716 - maintain correct mount count on the MDS (2356)
5717 - backout 1557, because 2316 wasn't really fixed
5718 - better file I/O statistics gathering in /proc
5719 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5720 - another kernel patch to fix zero-copy TCP function export
5721 - don't take duplicate lock when processing re-sent getattr (2420)
5722 - lctl uses obd_self_export instead of creating new conn (2353)
5723 - MDS/OST recovery case which requires object creation asserted (2425)
5724 - move lfs from /usr/sbin to /usr/bin in packages
5725 - fix race between mds_client_add and mds_client_free (2417)
5726 - use kmalloc instead of slabs in portals (2430)
5727 - don't create duplicate records when a failover MDS is present (2442)
5728 - remove unnecessary mount age check (2332)
5729 - don't remove directory inodes from locks prematurely (2451)
5730 - don't break if MDS service name is the same as hostname (2103)
5731 - fix races in client write RPC generation when cache full (2482)
5733 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5736 - remove now-unused request->rq_obd (278)
5737 - if an allocation fails, print out how much memory we've used (1933)
5738 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5739 - add the "configurable stack size" patch to most series files (1256)
5740 - ability to write large log records, for 100+ OST configs (2306)
5741 - fix NULL deref when filter_prep fails (2314)
5742 - fix operator precedence error in filter_sync
5743 - dynamic allocation of socknal TX descriptors (2315)
5744 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5745 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5746 - ensure that CWARN messages in Portals always get to the syslog
5747 - __init/__exit are not for prototype decls (ldlm_init/exit)
5748 - x86-64 compile warning fixes
5749 - fix gateway LMC keyword conflict (2318)
5750 - fix MDS lock inversions in getattr/reint paths (1844)
5751 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5752 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5753 - DLM race condition prevented some lock evictions (2328)
5754 - ENOMEM detection and retry on socknal sends (2230)
5755 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5756 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5757 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5758 - better error messages when a client is rejected during recovery (1505)
5759 - avoid cancelling locks which were never granted, after failure (2330)
5760 - fix i_sem/journal inversion in mds_client_add (2333)
5761 - fix truncate/getattr lock cycle deadlock (2334)
5762 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5763 - fix two rare exit paths which could leak an l_lock() ref (2321)
5764 - fencepost error in MDS/OST orphan recovery (2226)
5765 - make log record alignment 8 bytes (1988)
5766 - lstripe now fails when requested offset > ost_count (2237)
5767 - ensure that all kernel series have a complete list.h (1607)
5768 - fix crashes in special-file operations (2316)
5769 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5771 - allow configurable automake binary, for testing new versions
5772 - small update to the lfs documentation
5774 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5776 * fix negative export reference count in fsfilt_sync (2312)
5778 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5779 * release candidate 0.9.1
5781 - orphans are moved into the PENDING directory for possible recovery
5782 - replayed opens now open by fid for orphan/rename safety (1042)
5783 - last close of an orphan inode generates a transno (683)
5784 - chdir() and mount() now pin the directory entry (1020)
5785 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5786 - discard very old requests without processing them (1502)
5787 - remove some common, well-understood CERRORs (1505)
5788 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5789 - clear "grant" flags in OST replies until OST grant code lands (1644)
5790 - fix read performance by not clobbering i_blksize on client (1598)
5791 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5792 - make LRU size a /proc tunable, clears locks when reduced (707)
5793 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5794 - server locks take references on exports now (1558)
5795 - build fixes for 2.4.20-rh trees (1663)
5796 - return an error from lov_create if all OSCs are inactive (1751)
5797 - fix import levels when a reconnect happens without a timeout (1597)
5798 - exit early from mds_open if we get a lookup error (1749)
5799 - partial page read at EOF wouldn't wait for disk before sending (1642)
5800 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5801 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5802 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5803 - write/truncate lock inversion (1639)
5804 - Don't auto-load obdclass, portals modules during cleanup (1495)
5805 - fix timestamps from jumping to "now" (1763)
5806 - extra journal assertions (1648)
5807 - add an extra multiunlink test (1771)
5808 - fix read_record/write_record API (1776)
5809 - fix leak of offset_extent, possible incorrect i_size later (1772)
5810 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5811 - Debugging for the kqswnal_get_idle_tx problems (1820)
5812 - Allow recovery to be attempted multiple times (1536)
5813 - Write out MDS last_rcvd file after it is first created (1600)
5814 - Fix tx_descriptor leak in failed transmit situations (1827)
5815 - ext3 journaling fixes for assertion failure after IO error (1871)
5816 - class_export_put() on freed export after completion AST error (1896)
5817 - Fix revalidate looping in VFS (1322)
5818 - Don't access a freed export during MDS_REINT timeout (1521)
5819 - Add open-unlink recovery support on the MDS (1673,1764)
5820 - Return an error if no MDS data was read from last_rcvd (1946)
5821 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5822 - Don't setup a disk device that doesn't match exported UUID (317)
5823 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5824 - avoid committing NULL handle in force close
5825 - local.sh is now a one-stripe LOV configuration
5826 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5827 - fix ext3 htree duplicate directory entry corruption (1516)
5828 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5829 - update to new LOV EA format (2097)
5830 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5832 2003-06-15 Phil Schwan <phil@clusterfs.com>
5835 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5836 - per-import/export recovery handling (958, 931, 959)
5837 - multiple last-rcvd slots, for serving multiple FSes (949)
5838 - connections are again shared between multiple imp/exports (963, 964)
5839 - "umount -f" would hang if any requests needed to be sent (393, 978)
5840 - avoid pinning large req buffer by copying for queued messages (989)
5841 - add "uuid" to "lctl device" command to help upcalls (991)
5842 - "open" RPCs with transnos would confuse recovery counters (1037)
5843 - do proper endian conversion of all wire messages (288, 340, 891)
5844 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5845 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5846 - fix ptlrpc_connection leak in target_handle_connect (1174)
5847 - fix import refcounting bug in OST and MDS cleanup (1134)
5848 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5849 - fix very unlikely obd_types race condition (501)
5850 - remove osc_open hack for echo_client (1187)
5851 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5852 - a failure in read_inode2 leads to deadlock (1139)
5853 - cancel ack-locks as soon as transaction is committed (1072)
5854 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5855 - make sure to commitrw after any preprw to avoid deadlock (1162)
5856 - failing to execute a file in a lustre FS would lock inode (1203)
5857 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5858 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5859 - obd_uuid2tgt would walk off the end of the list (1255)
5860 - on IA64 the osc would give portals incorrect bulk size (1258)
5861 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5862 - fix lock inversion caused by new llite matching code (1282)
5863 - limit the number of dirty pages on a client to 10MB (1286)
5864 - timed out locks were not being corrected cancelled (1289)
5865 - fix O_DIRECT above 4GB on IA-32 (1292)
5866 * major user-visible changes
5867 - fail out/fail over policy now controlled by the upcall (993)
5869 - add OBD_PING to check server availability and failure (954)
5870 - lustre messages are now sent in sending host order (288, 340, 891)
5871 - add eadatalen to MDS getattr reply (340)
5872 - OST read replies may contain second buffer, with per-page status (593)
5874 2003-03-11 Phil Schwan <phil@clusterfs.com>
5877 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5878 - failed open()s could cause deadlock; fixed (867, 869)
5879 - stop cancelling OST locks when files are closed (481)
5880 - overlapping XID spaces caused network corruption (851, 853)
5881 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5882 - setattr_raw added, to do single-RPC, server-side setattrs
5883 - lmc/lconf syntax change for OST UUIDs
5884 - fix crashy race condition between ptlrpc_free_req and osc_close
5885 - don't use request in mdc_enqueue if we hit a timeout (889)
5886 - don't set the inode i_size for regular files from the MDS (896)
5887 - handle out of order completion AST (842)
5888 - don't LBUG if a lock request times out after receiving AST (913)
5889 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5890 - if a bad lock AST arrives, send an error instead of dropping entirely
5891 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5892 - fix leak in bulk IO when only partially completed (899, 900, 926)
5893 - fix O_DIRECT for ia64 (55)
5894 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5895 - C-z after timeout could hang a process forever; fixed (977)
5897 - client-side I/O cache (678, 924, 929, 941, 970)
5899 - READPAGE and SETATTRs which don't take server-side locks get
5902 2003-02-11 Phil Schwan <phil@clusterfs.com>
5905 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5906 - Fix fsfilt_extN_readpage() to read a full page of directory
5907 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5908 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5909 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5910 - Most intent operations were removed, in favour of a new RPC mode
5911 that does a single RPC to the server and bypasses most of the VFS
5912 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5913 - Aggressively cancel local locks on DLM servers
5914 - mds_reint_unlink sends EA to the client if it's the last nlink.
5915 client uses that EA to unlink OST objects.
5916 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5917 - recursive symlinks were fixed (439)
5918 - fixed NULL deref in DEBUG_REQ
5919 - filter_update_lastobjid no longer calls sync, which annoyed extN
5920 - fixed multi-client small-writes to a single file problem (445)
5921 - fixed mtime updates during file writes (607)
5922 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5923 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5924 - fixed Linux OST size reporting problem (444, 656)
5925 - OST now updates object mtime with writes or setattr (607, 619)
5926 - client verifies file size before zeroing page past EOF (445)
5927 - OST now writes last allocated objid to disk with allocation (108)
5928 - LOV on echo now works (409)
5930 - mds_reint_unlink sends a new buffer, with the EA included. this
5931 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5932 set if a regular file was being unlinked, and it was the last link
5933 - use PtlGet from the target for bulk writes (315)
5934 - OST now updates object mtime with writes or setattr (607, 619)
5935 - LDLM now has a grant-time callback to revalidate locked items, if
5937 - Many MDS operations were reorganized to combat race conditions
5939 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5940 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5942 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5945 - Fully reactivate OST imports after reconnection (512, others)
5946 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5947 - More graceful error handling for truncating on dead OST (515)
5948 - Don't error out unless we're actually accessing dead stripes (474)
5949 - Fix garbage sizes when stripes are missing (410)
5950 - LRU counters were broken, causing constant lock purge (433, 432)
5951 - garbage on read from stripes with failed OSTs (441)
5952 - mark OSCs as active before reconnecting during recovery (438)
5953 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5954 - lfind did not preserve OST order in output (443)
5955 - symlinks cause hung clients, incorrect data (439)
5956 - stop dereferencing request after dropping refcount (457)
5957 - don't LASSERT(spin_is_locked) on non-SMP (455)
5958 - fixes for many rename() bugs
5959 - fstat didn't correctly synchronize attributes (399)
5960 - server must handle lock cancellation during blocking AST prep (487)
5961 - bulk descriptors were free()d too soon (511)
5962 - fix paths in lconf, which would load incorrect modules (451, 507)
5963 - fix confusing lconf 'host not found' error message (386)
5964 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5965 - fix race condition in mdc_blocking_ast() for inode access (526)
5966 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5967 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5968 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5969 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5970 - fix lstripe in several ways (kernel side) (527)
5971 - fix request leak in ldlm_cli_enqueue (262)
5972 - incorrect OSC was marked inactive after OST failure
5973 - call mds_fs_cleanup before unmounting filesystem (524)
5974 - fix races between taking ns_lock and ldlm_lock_change_resource
5975 - fix races updating LOV export open file list
5976 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5977 - fix recovery NULL deref in ldlm_cli_cancel_unused
5978 - fix some DLM races by using new hash table for lock handles (419)
5979 - permit the client to specify desired inodes, at replay
5980 - duplicate requests when we queue them for replay reintegration
5981 - fix last_rcvd offset calculation
5982 - sync after each recovered transaction, so we always make progress
5983 - never, not always, ERESTART requests without transnos
5984 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5986 - skip replay if the MDS says that the client is already connected
5987 - don't check for a recovery-enabled export to match lctl's UUID
5988 - don't INC_USE_COUNT for phantom exports
5989 - don't crash when cleaning up phantom exports (567)
5990 - don't double-finish or set replay data for errored mdc_open requests
5991 - abort requests when they time out, so we don't get old replies
5992 - send/receive replies for AST messages again
5993 - if the client says that it doesn't have the lock, cancel it on the
5995 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5996 mark it as destroyed, it will all work out when decref is called
5997 - fix module use counts (22, 581)
5999 - ASTs now expect a reply (server cancels lock on error reply)
6001 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
6004 - fix many simultaneous client startup (392)
6005 - fix dentry->d_it clobbering
6006 - credentials weren't being shipped for readdir/getattr operations
6007 - remove invalid assertions triggered during some concurrent MD
6009 - proper Lustre versions added (336, 389)
6010 - fix memory leak for create error case (398)
6011 - fix LOV locking bug that would get cli/srv out of sync
6012 - fix echo client over LOV (409)
6013 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
6014 - fix double-O_EXCL intent crash (424)
6015 - avoid sending multiple lock CANCELs (352)
6017 - MDS can do multi-client recovery (modulo bugs in new code)
6019 - many updates, edits, cleanups
6021 2002-11-18 Phil Schwan <phil@clusterfs.com>
6024 - fix null d_it dereference (346)
6025 - fix full OST/dbench hang (333)
6026 - fix permission problem with file removal (286)
6027 - fix removal of OSCs from LOV when they fail
6028 - fix NULL deref during bulk timeout (214)
6029 - fix problems related to multiple filesystems on one MDS (241)
6030 - fixed serious subtle metadata locking bugs
6031 - free locks on clients when inodes are removed due to memory
6033 - fix inode pointer in lock data (285)
6034 - partial support for multiple MDS on a single host (241)
6035 - data locks weren't cancelled at clear_inode time (290, 311)
6036 - intent locks could lead to unbounded lock growth (205)
6037 - added a maximum lock count, an LRU list, and a flusher
6038 - fix multiple rename (365)
6039 - properly abstracted the echo client
6040 - OSC locked 1 byte too many; fixed
6041 - rewrote brw callback code:
6042 - fixed recovery bugs related to LOVs (306)
6043 - fixed too-many-pages-in-one-write crash (191)
6044 - fixed (again) crash in sync_io_timeout (214)
6045 - probably fixed callback-related race (385)
6047 - Add capability to MDS protocol
6048 - LDLM cancellations and callbacks on different portals
6050 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
6053 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
6054 - defer open object destruction to close time (601981, 138)
6055 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
6056 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
6057 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
6058 - don't call obd_setattr in ll_file_release for destroyed objects
6059 * protocol change to lustre_msg: move |version| and add |flags|
6060 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
6061 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
6062 * added replay of create, unlink, link and rename operations during
6063 MDS failover; recovery should be much more robust now
6064 * remove failed OSCs from LOVs (only lov_create uses this so far)
6065 * the lustre-HOWTO was brought (more) up to date (582544)
6067 2002-10-23 Phil Schwan <phil@clusterfs.com>
6070 - in-use dentries weren't being reused properly (617851)
6071 - prevent multiple LDLM setup (599178)
6072 - fix LOV size calculations for truncate (617853)
6073 - fix client handling of MDS intent errors (POSIX)
6074 - fix permission bug in lovstripe.c test (624321)
6075 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
6076 - truncate past end of file could corrupt data
6077 - proper cleanup after timeouts, crashes, etc (592524, 550815)
6078 - a race in recovery could return ETIMEDOUT to apps (623947)
6079 - building outside the source directory was fixed
6080 * the lustre-HOWTO was brought (more) up to date (582544)
6081 * major progress was made on recovery functionality
6083 2002-10-10 Phil Schwan <phil@clusterfs.com>
6086 - recovery deadlock fix
6087 - rm -rf causes LBUG fix (617817)
6088 - file open by multiple tasks fix (618962)
6089 - directory permissions bugs (602707 and 620007)
6090 - journal_stop fixed with locking (611313)
6091 - O_APPEND failures resolved (618273, perhaps 614459)
6092 - lconf PATH fix (619770)
6093 - IA64 build fix (621450)
6094 - RPC buffer sizes scale with amount of memory
6096 2002-10-01 Phil Schwan <phil@clusterfs.com>
6099 - locks would be cancelled without throwing away data pages,
6100 resulting in inconsistent data (605627)
6101 - inode attributes were not always being refreshed (605627, 612449)
6102 - lconf now continues to cleanup after lctl reports an error
6103 - MDS now enforces user permissions (602707)
6104 - lprocfs cleanup fixed, but not yet enabled (614157)
6105 - fixed infinite server hang, should a client not respond to an AST
6106 - avoid going into recovery if user calls readlink() with a buffer
6107 that's too small (613941)
6108 - AST RPCs no longer require replies (614867) -- this may be changed
6109 - don't crash server if client sends an IOV that's too big (611336)
6110 - fixed lock conversion deadlock (611892)
6111 - fixed the following of symlinks (614622)
6112 * recovery: the server can remove locks from a client that dies, other
6113 clients can make progress
6114 * more extN patch fixes
6115 * compile-time configurable ptlrpc buffer allocations
6117 - collaborative read cache document
6118 - Lustre Lite Performance CDR document-in-progress
6120 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6123 - fix typo in patch-2.4.18
6125 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6128 - clear ptlrpc request each time in handle_incoming_request()
6129 - unlink of files now destroys the object on the OST
6131 2002-09-19 Peter Braam <braam@clusterfs.com>
6133 * add hard link support
6134 * change obdfile creation method
6135 * kernel patch changed
6137 2002-09-19 Peter Braam <braam@clusterfs.com>
6140 - stack overflow bug in extN fixed
6142 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
6144 * documentation updates
6145 - add man pages for config tools
6146 - update tests/README to describe testing with new config tools
6147 - finish metadata API descriptions
6148 * bug fixes and cleanups
6149 - statfs workaround for 16TB limit
6150 - LOV stripe allocation improved, can stripe on subset of OSTs
6151 - LOV file size/IO offset was wrong for files > 4GB in size
6152 - object EA data was being dropped, caused files to be unreadable
6153 - memory overflow with non-LOV OST caused memory corruption
6154 - fixed regression tests to work with new config tools, obdfilter
6155 - fixed bug when directory size became larger than 1 block
6156 - fixed bug (for single client case) when PWD was deleted
6157 - invalidate local directory pages when doing intent-based ops
6158 - avoid LDLM oops when lock callback contained bad data
6160 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
6162 * documentation updates
6163 * bug fixes and cleanups
6164 - configuration tools
6167 - 64-bit compile warnings
6168 - 64-bit internal statfs data
6170 * test_brw on persistent OST devices
6172 * lprocfs (disabled)
6174 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
6176 * documentation updates
6177 * bug fixes and cleanups
6178 * configuration tools
6180 2002-08-30 Peter J. Braam <braam@clusterfs.com>
6183 * many small fixes to 0_5_4
6184 * io/network handling
6185 * thinkos in MDS operations
6187 2002-08-24 Peter J. Braam <braam@clusterfs.com>
6190 * crucial basic fixes to 0.5.3
6191 * IOR, Iozone work over Elan
6194 2002-08-07 Phil Schwan <phil@clusterfs.com>
6195 * version 0_5_3, our first alpha
6196 * we use the new Portals iovs
6197 * documentation updates
6198 * bug fixes and cleanups
6199 * small changes in the DLM wire protocol
6201 2002-07-25 Peter J. Braam <braam@clusterfs.com>
6202 * version 0_5_1 with some initial stability,
6203 * locking on MD and file I/O.
6204 * documentation updates
6205 * several bug fixes since 0.5.0
6206 * small changes in wire protocol
6208 2002-07-18 Phil Schwan <phil@clusterfs.com>
6210 * delivered as Lustre Light Alpha
6211 * fixed a crash after handling invalid MDS requests
6212 * fixed directory pages for architectures with non-4k pages sizes
6214 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
6215 * release version v0_4_4
6216 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
6218 * Fixes a number of interruption problems with OST operations.
6219 * Update documentation for portals header changes
6220 * Move all wire protocol structs/defines to lustre_idl.h
6221 * Fixes symlink length bug.
6222 * Add tcpdump to repository.
6224 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
6225 * release version v0_4_3
6226 * Fixes statfs for inodes on extN.
6227 * Fixes bug in runtests which would delete /etc/hosts.
6228 * Use 64-bit object IDs wherever possible (not into VFS though)
6229 Remove ost_get_info, which is unused by lustre, and out of date.
6231 2002-07-03 Peter Braam <braam@clusterfs.com>
6232 * release version v0_4_2 Fixes a lookup error (type not passed)
6233 * move forward to head of Portals
6234 * move forward to latest Lustre kernel
6236 2002-06-25 Peter Braam <braam@clusterfs.com>
6237 * release version v0_4_1. Hopefully stable on single node use.