1 2010-07-15 Oracle, 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
15 Severity : enhancement
17 Description: Distribute lfs_migrate script with lustre
18 Details : The lfs_migrate mentioned in the manual is now included in the
19 lustre tree in lustre/scripts and will distributed in lustre RPMs.
22 Severity : enhancement
24 Description: Failover nid list added to import proc file.
26 Severity : enhancement
28 Description: "som_preview" mount option required to use SOM
29 Details : The clients will need to use som_preview mount option
33 Description: Lustre does not do 1MB IOs to HW RAID
34 Details : Bump MAX_PHYS/HW_SEGMENTS and SG_ALL to 256 in the RHEL5 kernel.
36 Severity : enhancement
38 Description: lfs quota output is not very convenient for awk/sed-parsing
39 Details : Some positions in lfs quota output table could be empty or
40 non-empty which made it hard to parse it with scripts, now
41 a dash is put instead of space where there is not supposed
46 Description: Update to OFED 1.5.1
47 Details : OFED 1.5.1 will be used in environments that don't use
52 Description: Remove lnet. prefix from lctl params display
53 Details : Also renamed lnet.memused to lnet_memused to
58 Description: Use conf_param -d to remove permanant settings.
62 Description: Add list_param -R
63 Details : lctl list_param now supports recursive mode.
66 Bugzilla : 21829, 22312
67 Description: Fix llobdstat.
68 Details : 1. Now updating the write_bytes counter
69 2. Port script fixes from 1.8.
73 Description: decrease the usage of memory on clients.
74 Details : 1. On clients, recycle dentries and inodes unused.
75 2. Delete the code related to ll_deathrow(att 6215 in bug 1443). It
80 Description: llobdstat fix and enhancement
81 Details : add a counter to set a limit to how many samples will be returned
82 fix a wildcard in the path to limit to obdfilter stats only
86 Description: lfs check only allowed for root
87 Details : Code cleanups and added a regresion test of non-root use of
92 Description: MDS Oopses in ptlrpc_dispatch_difficult_reply()/__list_add()
93 Details : Fix hr->hr_index overload.
97 Description: Fix FIEMAP functionality.
101 Description: FC11 patchless client support added
105 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and
106 OEL5.4 kernel to 2.6.18-164.11.1.0.1.el5.
110 Description: New acc-sm script and various test-framework enhancments.
114 Description: Previously landed patch was reverted.
118 Description: Update SLES10 kernel to 2.6.16.60-0.42.8 (patchless)
122 Description: Add ldiskfs maxdirsize mount option.
126 Description: Fix mkfs.lustre support for 16TB LUNs.
130 Description: Print mmp_check_interval
131 Details : print mmp_check_interval and make it possible to abort
132 mount operation in case it takes too long.
136 Description: Add cascading_rw to lustre mpi tests.
140 Description: Store SOM data in EA.
141 Details : Also several SOM related cleanups.
145 Description: Add refcount to md_open_data
149 Description: Use refcount on lov_request_set to prevent memory corruption.
153 Description: Statahead: use parent dir's i_mutex for synchronization.
157 Description: loadgen improvements.
161 Description: Enable peer health detection in ksocklnd and ko2iblnd.
165 Description: Control DCACHE_LUSTRE_INVALID flag with MDS_INODE_LOOKCUP lock
169 Description: Fix race on obd_devcie:md_stats
173 Description: Fix SLES 11 client builds.
174 Details : Resolved uninitalized pointer gcc warnings.
178 Description: Allow non-root access for "lfs check".
179 Details : Added a check in obd_class_ioctl() for OBD_IOC_PING_TARGET.
183 Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
185 Severity : enhancement
187 Description: Update kernel to RHEL5.4 2.6.18-164.6.1.el5 and
188 OEL5 2.6.18-164.6.1.0.1.el5(Both in-kernel OFED enabled).
192 Description: Sanity test_27p fails.
193 Details : OBD_FAIL_OST_ENOSPC has been moved to filter_statfs().
194 Sanity tests 27 were cleaned up and improved.
198 Description: Unify req format on client/servers
199 Details : Use new req_capsule API [almost] everywhere instead of old PTLRPC
200 buffers and swabbers approach..
205 Description: drop join file stuff.
207 Severity : enhancement
209 Description: Add 2.6.27 and SLES11 (patchless client) support
213 Description: Fix in ptlrpc_expire_one_request() to print the signed time
216 Severity : enhancement
218 Description: Build kernels (RHEL5, OEL5 and SLES10/11) using the vendor's own
221 Severity : enhancement
223 Description: Remove set_info(KEY_UNLINKED) from MDS/OSC
225 Severity : enhancement
227 Description: correctly handle big reply message.
228 Details : send LNet event if reply is bigger then buffer and adjust this buffer
233 Description: Drop unnecessary __GFP_NOMEMALLOC flag from filter_get_page()
235 Severity : enhancement
237 Description: Add support for OEL5.
241 Description: Conf-sanity.sh 50g test - deactivated OST should not cause a panic.
245 Description: client could not reconnect to OST because of an active request.
246 Details : The client now retries to reconnect to the same server, if a
247 connect request failed with EBUSY or -EAGAIN. Test case 35b
248 conf-sanity.sh has been created.
252 Description: Don't allow make backward step on assignin osc next id.
253 Details : race between allocation next id and ll_sync thread can be cause
254 of set wrong osc next id and can be kill valid ost objects.
258 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
262 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
266 Description: Error handling in osc_statfs_interpret() has been improved.
267 Details : Check in osc_statfs_interpret() for EBADR.
271 Description: Do not send statfs() requests to OSTs disabled by administrator.
272 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
274 Severity : enhancement
276 Description: Added sanity-quota test to for lfs(1) to work with numeric uid/gid.
278 Severity : enhancement
280 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
282 Severity : enhancement
284 Description: Enhancement for lfs(1) command to use numeric uid/gid.
288 Description: Improved cleanup procedure for conf-sanity test_46a.
292 Description: Do not start bulk transfer, if client tried to reconnect.
293 Details : Do not start bulk transfer, if client tried to reconnect,
294 but failed due to active rpcs. Check for exp_abort_active_req
295 has been implemented.
299 Description: Prevent inconsistences between linux and lustre mount structures.
300 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
301 Make the sleep interruptible.
303 Severity : enhancement
305 Description: provide server to client comms path
306 Details : server to client communications path via new LDLM_SET_INFO rpc
310 Description: truncate starts GFP_FS allocation under transaction causing deadlock
311 Details : ldiskfs_truncate calls grab_cache_page which may start page
312 allocation under an open transaction. This may lead to
313 calling prune_icache with consequent lustre reentrance.
317 Description: send correctly lsm on open replay
318 Details : MDS is trust to LSM size on replay open, but client can set wrong size
323 Description: enable client interface failover
324 Details : When a child reconnects from another NID, properly update export
325 nid hash position and ldlm reverse import.
327 Severity : enhancement
329 Description: Communicate OST degraded/readonly state via statfs to MDS
330 Details : Flags in the statfs returned from OSTs indicate whether the
331 OST is in a degraded RAID state, or if the filesystem has
332 turned read-only after a filesystem error is detected.
337 Description: don't panic if EPROTO was hit when reading symlink
338 Details : correctly handling request reference in error cases.
340 Severity : enhancement
342 Description: Add LustreNetLink, a kernel-userspace communcation path.
344 Severity : enhancement
346 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
351 Description: access to llog context before init.
352 Details : move handling CATALOGS file at osc layer and forbid access to llog
356 Frequency : rare, only when client is writing to OST on local node
358 Description: Avoid deadlock for local client writes
359 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
360 memory freeing context. This allows OST threads to set the
361 PF_MEMALLOC flag on task structures in order to allocate memory
362 from reserved pools and complete IO.
363 Use GFP_HIGHUSER for OST allocations for non-local client writes,
364 so that the OST threads generate memory pressure and allow
365 inactive pages to be reclaimed.
367 Severity : enhancement
369 Description: Update kernel to RHEL5.3 2.6.18-128.1.14.el5.
374 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
375 Details : move ldlm namespace creation in setup phase to avoid grab
376 _lprocfs_lock with cli_sem held.
380 Description: Temporarily disable grant shrink.
381 Details : Disable the feature for debugging.
384 Frequency : only when formatting test filesystems
386 Description: Unable to run several mkfs.lustre on loop devices at the same time
387 Details : mkfs.lustre returns error 256 on the concurrent loop devices
388 formatting. The solution is to proper handle the error.
390 Severity : enhancement
392 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
394 Severity : enhancement
396 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
401 Description: resolve race between obd_disconnect and class_disconnect_exports
402 Details : if obd_disconnect will be called to already disconnected export he
403 forget release one reference and osc module can't unloaded.
405 Severity : enhancement
407 Description: Allow tuning service thread via /proc
408 Details : For each service a new
409 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
410 created that can be used to set min/max thread counts, and get the
411 current number of running threads.
416 Descriptoin: don't return error if have partially created objects for file.
417 Details : lov_update_create_set uses set->set_success as index for created
418 objects, so if some requests will be failed, they will have hole
419 at end of array and qos_shrink_lsm for allocate correct lsm.
421 Severity : enhancement
423 Description: Update OFED support to 1.4.1
428 Description: Reduce small size read RPC
429 Details : Set read-ahead limit for every file and only do read-ahead when
430 available read-ahead pages are bigger than 1M to avoid small size
433 Severity : enhancement
435 Description: move AT tunable parameters for more consistent usage
436 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
438 Severity : enhancement
440 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
441 Details : allow skip disconnected ost for send statfs request and hide error
447 Description: fix racy locking of mballoc block bitmaps causing BUG
448 Details : The locking of the mballoc buddy bitmap and the in-memory
449 block bitmap was using two different spin locks in some
450 cases. This made it possible to incorrectly access the
451 mballoc bitmap while another process was modifying it,
452 causing a sanity assertion to fail. While no on-disk corruption
453 was reported, there was some risk of this happening.
456 Frequency : rare, on llog test 6
458 Descriptoin: don't allow connect to already connected import
459 Details : allowing connect to already connected import is hide connecting problem.
464 Descriptoin: allow kill process which wait statahead result
465 Details : for some reasons 'ls' can stick in waiting result from statahead,
466 in this case need way for kill this process.
468 Severity : enhancement
470 Description: Add state history info file, enhance import info file
471 Details : Track import connection state changes in a new osc/mdc proc file;
472 add overview-type data to the osc/mdc import proc file.
474 Severity : enhancement
476 Description: MDS create should not wait for statfs RPC while holding DLM lock.
478 Severity : enhancement
480 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
485 Descriptoin: OST grant too much space to client even there are not enough space.
486 Details : Client will shrink its grant cache to OST if there are no write
487 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
488 this grant cache if there are already not enough avaible space
489 (left_space < total_clients * 32M).
492 Frequency : start MDS on uncleanly shutdowned MDS device
494 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
495 Details : stay in waiting mds<>ost recovery finished produce random bugs
496 due race between two ll_sync thread for one lov target. send
497 ACTIVATE event only if connect realy finished and import have
501 Frequency : rare, connect and disconnect target at same time
503 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
504 Details : don't call obd_disconnect under lov_lock. this long time
505 operation and can block ptlrpcd which answer to connect request.
510 Descriptoin: don't lose wakeup for imp_recovery_waitq
511 Details : recover_import_no_retry or invalidate_import and import_close can
512 both sleep on imp_recovery_waitq, but we was send only one wakeup
516 Frequency : always with long access acl
518 Descriptoin: mds can't pack reply with long acl.
519 Details : mds don't control size of acl but they limited by reint/getattr
522 Severity : enhancement
524 Description: Update to SLES10 kernel-2.6.16.60-0.33.
526 Severity : enhancement
528 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
531 Frequency : start MDS on uncleanly shutdowned MDS device
533 Descriptoin: aborting recovery hang on MDS
534 Details : don't throttle destroy RPCs for the MDT.
537 Frequency : on remount
539 Description: external journal device not working after the remount
540 Details : clear dev_rdonly flag for external journal devices in
546 Description: shutdown vs evict race
547 Details : client_disconnect_export vs connect request race.
548 if client will evicted at this time - we start invalidate
549 thread without referece to import and import can be freed
553 Frequency : rare, need acl's on inode.
555 Description: client can't handle ost additional correctly
556 Details : if ost was added after client connected to mds client can have
557 hit lnet_try_match_md ... to big messages to wide striped files.
558 in this case need teach client to handle config events about add
559 lov target and update client max ea size at that event.
561 Severity : enhancement
563 Description: Changelogs
564 Details : Changelogs are a lightweight mechanism to track filesystem
565 metadata and namespace changes. The changelog is recorded
566 permanently on the MDTs, and is periodically "consumed" / purged
567 when records are no longer needed.
569 Severity : enhancement
571 Description: compact fld format with extents
572 Details : Store range of seq rather than every seq in FLD. Seq
573 controller update FLD rather than clients. In Case of CMD, mdt0
574 has FLD, all other metadata server act as non persistent proxy
575 for FLD queries and cache fld entries in fld cache.
580 Description: don't skip ost target if they assigned to file
581 Details : Drop slow OSCs if we can, but not for requested start idx.
582 This means "if OSC is slow and it is not the requested
583 start OST, then it can be skipped, otherwise skip it only
584 if it is inactive/recovering/out-of-space.
588 Description: more cleanup in mds_lov
589 Details : not send LOV EA under replay, we can't know about they size at this
590 time. Don't allow client connect to mds before any ost connected,
591 for avoid problems with LOV EA size and returning EIO to client.
593 Severity : enhancement
595 Description: Interoperability at server side (Disk interoperability)
597 Severity : enhancement
599 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
601 Severity : enhancement
603 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
605 Severity : enhancement
607 Description: New client IO stack (CLIO).
609 Severity : enhancement
611 Description: Commit on sharing. Eliminate inter-client dependencies between
612 uncommitted transactions by doing transaction commits.
613 Thereby clients may recovery independently.
616 Frequency : Create a symlink file with a very long name
618 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
619 Details : If there is no extra space in the request for early cancels,
620 ldlm_req_handles_avail() returns 0 instead of a negative value.
622 Severity : enhancement
624 Description: Add /proc entry for import status
625 Details : The mdc, osc, and mgc import directories now have
626 an import directory that contains useful import data for debugging
629 Severity : enhancement
631 Description: Re-disable certain /proc logging
632 Details : Enable and disable client's offset_stats, extents_stats and
633 extents_stats_per_process stats logging on the fly.
636 Frequency : Only on FC kernels 2.6.22+
638 Description: oops in statahead
639 Details : Do not drop reference count for the dentry from VFS when lookup,
640 VFS will do that by itself.
642 Severity : enhancement
644 Description: Generic /proc file permissions
645 Details : Set /Proc file permissions in a more generic way to enable non-
646 root users operate on some /proc files.
650 Description: Hitting mdc_commit_close() ASSERTION
651 Details : Properly handle request reference release in
652 ll_release_openhandle().
656 Description: quota recovery deadlock during mds failover
657 Details : This patch includes att18982, att18236, att18237 in bz14840.
659 1. fix osts hang when mds does failover with quotaon
660 2. prevent watchdog storm when osts threads wait for the
665 Frequency : only patchless client
666 Description: add workaround for race between add/remove dentry from hash
668 Severity : enhancement
670 Description: Allow OST glimpses to return PW locks
674 Description: LBUG when llog conf file is full
675 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
679 Description: Prevent import from entering FULL state when server in recovery
683 Description: service mount cannot take device name with ":"
684 Details : Only when device name contains ":/" will mount treat it as
690 Description: replace ptlrpcd with the statahead thread to interpret the async
691 statahead RPC callback
695 Frequency : on recovery
696 Description: I/O failures after umount during fail back
697 Details : if client reconnected to restarted server we need join to recovery
698 instead of find server handler is changed and process self
699 eviction with cancel all locks.
701 Severity : enhancement
703 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
705 Severity : enhancement
707 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
709 Severity : enhancement
711 Description: Upcall on Lustre log has been dumped
712 Details : Allow for a user mode script to be called once a Lustre log has
713 been dumped. It passes the filename of the dumped log to the
714 script, the location of the script can be specified via
715 /proc/sys/lnet/debug_log_upcall.
720 Description: avoid idr_remove called for id which is not allocated.
721 Details : Move assigment s_dev for clustered nfs to end of initialization,
722 to avoid problem with error handling.
727 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
728 Details : When connection is reused this not moved from CONN_UNUSED_HASH
729 into CONN_USED_HASH and this prodice warning when put connection
730 again in unused hash.
733 Severity : enhancement
735 Description: File striping can now be set to use an arbitrary pool of OSTs.
737 Severity : enhancement
739 Description: Export bytes_read/bytes_write count on OSC/OST.
743 Description: Early reply size mismatch, MGC loses connection
744 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
745 the connect flags are properly negotiated.
750 Description: quotas are not honored with O_DIRECT
751 Details : all writes with the flag O_DIRECT will use grants which leads to
752 this problem. Now using OBD_BRW_SYNC to guard this.
756 Description: add quota statistics
757 Details : 1. sort out quota proc entries and proc code.
758 2. add quota statistics
760 Severity : enhancement
762 Description: enable quota support for HEAD.
766 Description: Properly propagate oinfo flags from lov to osc for statfs
767 Details : restore missing copy oi_flags to lov requests.
769 Severity : enhancement
771 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
772 : l_getgroups(8), lst(8), routerstat(8)
773 Details : included man pages for llobdstat(8), llstat(8),
774 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
776 Severity : enhancement
778 Description: configure's --enable-quota should check the
779 : kernel .config for CONFIG_QUOTA
780 Details : configure is terminated if --enable-quota is passed but
781 : no quota support is in kernel
785 Description: Remove portals compatibility
786 Details : Remove portals compatibility, not interoperable with releases
791 Description: Resolve device initialization race
792 Details : Prevent proc handler from accessing devices added to the
793 obd_devs array but yet be intialized.
795 Severity : enhancement
797 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
799 Severity : enhancement
801 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
806 Description: Using wrong pointer in osc_brw_prep_request
807 Details : Access to array[-1] can produce panic if kernel compiled with
808 CONFIG_PAGE_ALLOC enabled
812 Description: Client runs out of low memory
813 Details : Consider only lowmem when counting initial number of llap pages
817 Description: *optional* service tags registration
818 Details : if the "service tags" package is installed on a Lustre node
819 When the filesystem is mounted, a local-node service tag will
820 be created. See http://inventory.sun.com/ for more information
821 about the Service Tags asset management system.
825 Description: Kernel BUG tries to release flock
826 Details : Lustre does not destroy flock lock before last reference goes
827 away. So always drop flock locks when client is evicted and
828 perform unlock regardless of successfulness of speaking to MDS.
832 Description: add refcount protection for osc callbacks, avoid panic on shutdown
836 Description: sanity test 65a fails if stripecount of -1 is set
837 Details : handle -1 striping on filesystem in ll_dirstripe_verify
842 Description: ASSERTION(CheckWriteback(page,cmd)) failed
843 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
846 Severity : enhancement
848 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
852 Description: do not process already freed flock
853 Details : flock can possibly be freed by another thread before it reaches
854 to ldlm_flock_completion_ast.
858 Description: LBUG during stress test
859 Details : Need properly lock accesses the flock deadlock detection list.
863 Description: oops in page fault handler
864 Details : kernel page fault handler can return two special 'pages' in error
865 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
869 Description: timeout with invalidate import.
870 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
871 should be handled by ptlrpcd. This produce long age waiting and
872 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
874 Severity : enhancement
876 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
880 Description: enable MGS and MDT services start separately
881 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
882 a co-located MGS without starting the MGS, which is a complement
883 to 'nosvc' mount option.
887 Frequency : after recovery
888 Description: precreate to many object's after del orphan.
889 Details : del orphan st in oscc last_id == next_id and this triger growing
890 count of precreated objects. Set flag LOW to skip increase count
891 of precreated objects.
895 Frequency : rare, on clear nid stats
896 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
897 Details : when clean nid stats sometimes try destroy live entry,
898 and this produce panic in free.
902 Description: Stack overflow during MDS log replay
903 ease stack pressure by using a thread dealing llog_process.
907 Description: wait until IO finished before start new when do lock cancel.
908 Details : VM protocol want old IO finished before start new, in this case
909 need wait until PG_writeback is cleared until check dirty flag and
910 call writepages in lock cancel callback.
912 Severity : enhancement
914 Description: using special macro for print time and cleanup in includes.
918 Description: mds_mfd_close() ASSERTION(rc == 0)
919 Details : In mds_mfd_close(), we need protect inode's writecount change
920 within its orphan write semaphore to prevent possible races.
924 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
928 Frequency : rare, on shutdown ost
929 Description: don't hit live lock with umount ost.
930 Details : shrink_dcache_parent can be in long loop with destroy dentries,
931 use shrink_dcache_sb instead.
935 Description: don't panic with use echo client
936 Details : echo client pass NULL as client nid pointer and this produce null
941 Description: fix build on ppc32
942 Details : compile code with -m64 flag produce wrong object file for ppc32.
946 Description: add message levels for liblustreapi
950 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
951 Details : Move the 'good_osts' check before the 'total_bavail' check. This
952 will result in an -EAGAIN and in the exit call path we call
953 alloc_rr() which will with increasing aggressiveness attempt to
954 aquire precreated objects on the minimum number of required OSCs.
958 Description: Use old size assignment to avoid deadlock
959 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
960 the scheduling while holding a spinlock. We do not need locking
961 for size in ll_update_inode() because size is only updated from
962 the MDS for directories or files without objects, so there is no
963 other place to do the update, and concurrent access to such inodes
964 are protected by the inode lock.
968 Description: resolve "_IOWR redefined" build error on SLES10
972 Description: dump the memory debugging after all modules are unloaded to
973 suppress false negative in conf_sanity test 39
975 Severity : enhancement
977 Description: build kernel-ib packages for OFED 1.3 in our release cycle
982 Description: fix SLES kernel versioning
983 Details : the kernel version for our SLES 10 kernel did not include a "-"
984 before the "smp" at the end. while this was not a problem in
985 general, it did mean that software trying to use the kernel
986 version to try to detect a vendor specific kernel would fail.
987 this was most evident by the OFED build scripts.
991 Description: Don't update lov_desc members until making sure they are valid
992 Details : When updating lov_desc members via proc fs, need fix their
993 validities before doing the real update.
997 Description: don't put request into delay list while invalidate in flight.
998 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
999 import in flight. this produce timeout for invalidate and sometimes
1000 can cause stale data.
1004 Frequency : on ppc only
1005 Description: not convert ost objects for directory because it's not exist.
1006 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
1010 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
1012 Severity : enhancement
1014 Description: Update to support 2.6.22.14 vanilla kernel.
1018 Frequency : rare, on recovery
1019 Description: read procfs can produce deadlock in some situation
1020 Details : Holding lprocfs lock which send rpc can produce block for destroy
1021 obd objects and this also block reconnect with -EALREADY. This
1022 isn't fix all lprocfs bugs - but make it rare.
1024 Severity : enhancement
1026 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
1029 Frequency : frequent on X2 node
1031 Description: mdc_set_open_replay_data LBUG
1032 Details : Set replay data for requests that are eligible for replay.
1036 Description: lustre_mgs: operation 101 on unconnected MGS
1037 Details : When MGC is disconnected from MGS long enough, MGS will evict the
1038 MGC, and late on MGC cannot successfully connect to MGS and a lot
1039 of the error messages complaining that MGS is not connected.
1043 Frequency : on network error
1044 Description: panic with double free request if network error
1045 Details : mdc_finish_enqueue is finish request if any network error ocuring,
1046 but it's true only for synchronus enqueue, for async enqueue
1047 (via ptlrpcd) this incorrect and ptlrpcd want finish request
1050 Severity : enhancement
1052 Description: client-side metadata stat-ahead during readdir(directory readahead)
1053 Details : perform client-side metadata stat-ahead when the client detects
1054 readdir and sequential stat of dir entries therein
1057 Frequency : on start mds
1059 Description: Implement get_info(last_id) in obdfilter.
1062 Frequency : occasional
1064 Description: Correctly check stale fid, not start epoch if ost not support SOM
1065 Details : open with flag O_CREATE need set old fid in op_fid3 because
1066 op_fid2 was overwritten with new generated fid, but mds can answer
1067 with one of these two fids and both is not stale. Setattr
1068 incorrectly started an epoch and assume will be called
1069 done_writing, but without SOM done_writing ever being called.
1072 Frequency : rare, depends on device drivers and load
1074 Description: MDS or OSS nodes crash due to stack overflow
1075 Details : Code changes in 1.8.0 increased the stack usage of some functions.
1076 In some cases, in conjunction with device drivers that use a lot
1077 of stack the MDS (or possibly OSS) service threads could overflow
1078 the stack. One change which was identified to consume additional
1079 stack has been reworked to avoid the extra stack usage.
1082 Frequency : occasional
1084 Description: Do not fail import if osc_interpret_create gets -EAGAIN
1085 Details : If osc_interpret_create got -EAGAIN it immediately exits and
1086 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
1087 oscc_has_objects and see OSC has no objests and call
1088 oscc_internal_create to resend create request.
1090 Severity : enhancement
1092 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
1094 Severity : enhancement
1096 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
1099 Frequency : very rare
1101 Description: Fix replay if there is an un-replied request and open
1102 Details : In some cases, older replay request will revert the
1103 mcd->mcd_last_xid on MDS which is used to record the client's
1104 latest sent request.
1106 Severity : enhancement
1108 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
1110 Severity : enhancement
1112 Description: Add rhel5 support to HEAD.
1114 Serverity : enhancement
1116 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
1121 Description: Don't allow skipping OSTs if index has been specified.
1122 Details : Don't allow skipping OSTs if index has been specified, make
1123 locking in internal create lots better.
1127 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
1128 Details : access to bitfield in structure is always rounded to long
1129 and this produce problem with not atomic change any bit.
1133 Description: Lustre make rpms failed.
1134 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
1135 builds Lustre rpms from tarball.
1138 Frequency : rare on shutdown ost
1140 Description: If llog cancel was not send before clean_exports phase, this can
1141 produce deadlock in llog code.
1142 Details : If llog thread has last reference to obd and call class_import_put
1143 this produce deadlock because llog_cleanup_commit_master wait when
1144 last llog_commit_thread exited, but this never success because was
1145 called from llog_commit_thread.
1149 Description: allow userland application know is lost one of stripes.
1150 Details : fill lvb_blocks with error code on ost and return it to
1151 application if error flag found.
1155 Description: NULL lov_tgts causing MDS oops
1156 Details : more safe checks for NULL lov_tgts for avoid oops.
1158 Severity : enhancement
1160 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
1164 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
1165 Details : If an OST is down the MDS will hang indefinitely in
1166 obd_statfs_rqset() waiting for the statfs data. While for
1167 MDS QOS usage of statfs, it should not stuck in waiting.
1169 Severity : enhancement
1171 Description: remote_acl support
1172 Details : Support ACL-based permission check for remote user.
1173 Support setfacl/getfacl for remote user with the utils
1174 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
1175 the system "{s,g}etfacl" utils.
1177 Severity : enhancement
1179 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
1181 Severity : enhancement
1183 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
1187 Description: make mgs_setparam() handle fsname containing dash
1188 Details : fsname containing a dash does not work with lctl conf_param
1190 Severity : enhancement
1192 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
1194 Severity : enhancement
1196 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
1200 Frequency : rare, at shutdown
1201 Description: access already free / zero obd_namespace.
1202 Details : if client_disconnect_export was called without force flag set,
1203 and exist connect request in flight, this can produce access to
1204 NULL pointer (or already free pointer) when connect_interpret
1205 store ocd flags in obd_namespace.
1209 Frequency : only at startup
1210 Description: not alloc memory with spinlock held.
1211 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
1212 if any spinlock held.
1214 Severity : enhancement
1216 Description: make lustre randomly fail allocating memory
1217 Details : Make lustre randomly failed allocating memory for testing purpose.
1219 Severity : enhancement
1221 Description: lost problems with lov objid file
1222 Details : Fixes some scability and access to not inited memory problems
1223 in work with lov objdid file.
1228 Description: lfs find does not continue on file error
1229 Details : Continue other files processing when a file/dir is absent.
1233 Description: Inconsistent usage of lustre_pack_reply()
1234 Details : Standardize the usage of lustre_pack_reply() such that it
1235 always generate a CERROR on failure.
1238 Frequency : occasional
1240 Description: MDS hang or stay in waiting lock
1241 Details : If client receive lock with CBPENDING flag ldlm need send lock
1242 cancel as separate rpc, to avoid situation when cancel request
1243 can't processed due all i/o threads stay in wait lock.
1247 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
1251 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
1255 Description: Update to SLES9 kernel-2.6.5-7.287.3.
1259 Description: Update to RHEL4 latest kernel.
1261 Severity : enhancement
1263 Description: Build SLES10 patchless client fails
1264 Details : The configure was broken by run ./configure with
1265 --with-linux-obj=.... argument for patchless client. When the
1266 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
1267 can't search header adequately. Use absolute path such as
1268 -I($LINUX)/include instead.
1272 Description: interrupt oig_wait produce painc on resend.
1273 Details : brw_redo_request can be used for resend requests from ptlrpcd and
1274 private set, and this produce situation when rq_ptlrpcd_data not
1275 copyed to new allocated request and triggered LBUG on assert
1276 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
1277 ptlrpcd set if request is changed and can be safety changed to use
1282 Description: LASSERT_{REQ,REP}SWAB macros are buggy
1283 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
1284 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
1285 these macros and replace them with their difinitions instead.
1289 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
1290 Details : Update which_patch & target file for SLES10 latest kernel.
1293 Description: Remove client patches from SLES 10 kernel.
1294 Details : This causes SLES 10 clients to behave as patchless clients
1295 even on a Lustre-patched (server) kernel.
1297 Severity : enhancement
1299 Description: self-adjustable client's lru lists
1300 Details : use adaptive algorithm for managing client cached locks lru
1301 lists according to current server load, other client's work
1302 pattern, memory activities, etc. Both, server and client
1303 side namespaces provide number of proc tunables for controlling
1306 Severity : enhancement
1308 Description: light-weight GSS support
1309 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
1310 and reduce performance overhead.
1312 Severity : enhancement
1314 Description: Linux keyring support
1315 Details : Support using service of Linux keyring for Lustre GSS internal
1316 context refresh/cache mechanism.
1320 Description: Fix errors in lfs documentation
1321 Details : Fixes man pages
1325 Description: don't use GFP_* in generic Lustre code.
1326 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
1330 Description: obdclass is limited by single OBD_ALLOC(idarray)
1331 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
1335 Description: warnings with build patchless client with vanila 2.6.19 and up
1336 Details : change old ctl_table style and replace ctl_table/ctl_table_header
1337 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1341 Frequency : Only for SLES
1342 Description: sanity_quota fail test_1
1343 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1344 file, which makes TSTID[2] inunique.
1348 Frequency : Only for SLES9
1349 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1350 Details : Update target/ChangeLog/which_patch .
1354 Frequency : Only for test_10 in sanity-quota.sh
1355 Description: a bug in quota test code
1356 Details : checking a test flag in a wrong place
1360 Frequency : Only for test_13 in sanity-quota.sh
1361 Description: a bug in quota test script
1362 Details : shouldn't check overfull allocated quota
1366 Description: jbd statistics
1367 Details : Port older jbd statistics patch for sles10
1369 Severity : enhancement
1371 Description: Add printing inode info into message about error in writepage.
1376 Description: reply_lock_interpret crash due to race with it and lock cancel.
1377 Details : Do not replay locks that are being cancelled. Do not reference
1378 locks by their address during replay, just by their handle.
1382 Frequency : When flocks are used.
1383 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1384 Details : Flock locks might destroy just granted lock if it could be merged
1385 with another existing flock, this is done in completion handler,
1386 so teach ldlm_cli_enquque_fini that this is a valid case for
1392 Description: Oops in read and write path when failing to allocate lock.
1393 Details : Check if lock allocation failed and return error back.
1397 Description: lstripe command fails for valid OST index
1398 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1399 instead of lov->desc.ld_active_tgt_count.
1403 Description: sanity.sh failed test 103
1404 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1409 Description: df doesn't work properly if diskfs blocksize != 4K
1410 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1414 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1415 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1417 Severity : enhancement
1419 Description: Adaptive timeouts
1420 Details : RPC timeouts adapt to changing server load and network
1421 conditions to reduce resend attempts and improve recovery time.
1425 Description: llapi_file_create() does not allow some changes
1426 Details : add llapi_file_open() that allows specifying the mode and
1427 open flags, and also returns an open file handle.
1431 Description: merge and cleanup kernel patches.
1432 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1436 Description: Add journal checksum support.(Kernel part)
1437 Details : The journal checksum feature adds two new flags i.e
1438 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1439 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1440 indicates that the commit block contains the checksum for
1441 the blocks described by the descriptor blocks. Now commit
1442 record can be sent to disk without waiting for descriptor
1443 blocks to be written to disk. This behavior is controlled
1444 using JBD2_FEATURE_ASYNC_COMMIT flag.
1448 Description: OSS needs mutliple precreate threads
1449 Details : Add ability to start more than one create thread per OSS.
1453 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1454 Details : execute lfs setstripe on client
1458 Description: mds_obd_create error creating tmp object
1459 Details : When the user sets quota on root, llog will be affected and can't
1460 create files and write files.
1463 Frequency : Always on ia64 patchless client, and possibly others.
1465 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1466 Details : This allows the patchless client to be loaded on architectures
1467 without this export.
1471 Description: RedHat Update kernel for RHEL5
1472 Details : Add the kernel config file for RHEL5.
1476 Description: RedHat Update kernel for RHEL5
1477 Details : Modify the kernel config file more closer RHEL5.
1481 Description: Build failure against Centos5 (RHEL5)
1482 Details : Define PAGE_SIZE when it isn't present.
1486 Description: test_7 on sanity-quota.sh can't be used on separate machines
1487 Details : add facet to handle it
1491 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1492 Details : with NFS, the anon dentry's parent was set to itself in
1493 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1494 corresponding dentry other than use rec->ur_name.
1496 Severity : enhancement
1498 Description: lfs setstripe enhancement
1499 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1503 Description: enable data checksumming by default
1504 Details : enable checksum by default, allow --disable-checksum
1505 configure option and "-o nochecksum" mount option. Checksums
1506 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1510 Description: lustre support for RHEL5
1511 Details : Add support for RHEL5.
1515 Description: Client eviction due to failover config
1516 Details : after a connection loss, the lustre client should attempt to
1517 reconnect to the last active server first before trying the
1518 other potential connections.
1522 Description: when mds and osts use different quota unit(32bit and 64bit),
1523 quota will be released repeatly.
1524 Details : void sending multiple quota reqs to mds, which will keep the
1525 status between the reqs.
1529 Description: osts not allocated evenly to files
1530 Details : change the condition to increase offset_idx
1533 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1535 Description: Data corruption for OSTs that are formatted larger than 2TB
1537 Details : When generating the bio request for lustre file writes the
1538 sector number would overflow a temporary variable before being
1539 used for the IO. The data reads correctly from Lustre (which
1540 will overflow in a similar manner) but other file data or
1541 filesystem metadata may be corrupted in some cases.
1545 Description: Tune the kernel for good SCSI performance.
1546 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1547 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1552 Description: rewrite ext2-derived code in obdclass/uuid.c
1553 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1554 from scratch preserving functionality.
1558 Description: "lfs find -obd UUID" prints directories
1559 Details : "lfs find -obd UUID" will return all directory names instead
1560 of just file names. It is incorrect because the directories
1561 do not reside on the OSTs.
1565 Description: use i_size_read and i_size_write in 2.6 port
1566 Details : replace inode->i_size access with i_size_read/write()
1569 Frequency : when removing large files
1571 Description: scheduling issue during removal of large Lustre files
1572 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1573 It causes scheduling issues when removing large files (17TB in the
1577 Frequency : only with liblustre clients on XT3
1579 Description: evictions taking too long
1580 Details : allow llrd to evict clients directly on OSTs
1583 Frequency : only on ppc
1585 Description: /proc/fs/lustre/devices broken on ppc
1586 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1587 We should make sure the type of 'index' is loff_t and then cast
1588 explicitly as needed below. Do not assign an explicitly cast
1592 Frequency : only for rhel5
1594 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1595 Details : Modify the target file & which_kernel.
1597 Severity : enhancement
1599 Description: omit set fsid for export NFS
1600 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1601 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1605 Frequency : Always, for kernels after 2.6.16
1606 Description: Fix warning idr_remove called for id=.. which is not allocated.
1607 Details : Last kernels save old s_dev before kill super and not allow
1608 to restore from callback - restore it before call kill_anon_super.
1612 Description: buffer overruns could theoretically occur
1613 Details : llapi_semantic_traverse() modifies the "path" argument by
1614 appending values to the end of the origin string, and a
1615 overrun may occur. Adding buffer overrun check in liblustreapi.
1619 Description: Fix error on 'ls .' at the top of the Lustre mount.
1620 Details : Don't revalidate dentry if it is a root dentry.
1624 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1625 Details : Modify vm-tunables-rhel4.patch.
1629 Description: Kernel config for 2.6.18-vanilla.
1630 Details : Modify targets/2.6-vanilla.target.in.
1631 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1632 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1633 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1634 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1638 Description: improve handling recoverable errors
1639 Details : if request processig with error which can be recoverable on server
1640 request should be resend, otherwise page released from cache and
1645 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1646 Details : Modify target file & which_patch.
1647 A flaw was found in the IA32 system call emulation provided
1648 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1649 value could be stored in the %RAX register, which could trigger an
1650 out-of-bounds system call table access. An untrusted local user
1651 could exploit this flaw to run code in the kernel
1652 (ie a root privilege escalation). (CVE-2007-4573).
1656 Description: O_DIRECT bypasses client statistics.
1657 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1658 read_ahead_stats were updated. copied the stats section from
1659 osc_send_oap_rpc() into async_internal().
1663 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1667 Description: change order of libsysio includes
1668 Details : '#include sysio.h' should always come before '#include xtio.h'
1672 Description: handle "serious error: objid * already exists" more gracefully
1673 Details : If LAST_ID value on disk is smaller than the objects existing in
1674 the O/0/d* directories, it indicates disk corruption and causes an
1675 LBUG(). If the object is 0-length, then we should use the existing
1676 object. This will help to avoid a full fsck in most cases.
1678 Severity : enhancement
1680 Description: adapt the lustre_config script to support the upgrade case
1681 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1682 server targets to 1.6 in parallel.
1686 Description: To avoid grant space > avaible space when the disk is almost
1687 full. Without this patch you might see the error "grant XXXX >
1688 available" or some LBUG about grant, when the disk is almost
1690 Details : In filter_check_grant, for non_grant cache write, we should
1691 check the left space by if (*left > ungranted + bytes), instead
1692 of (*left > ungranted), because only we are sure the left space
1693 is enough for another "bytes", then the ungrant space should be
1694 increase. In client, we should update cl_avail_grant only there
1695 is OBD_MD_FLGRANT in the reply.
1699 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1700 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1701 described in CVE-2007-4573. This problem could allow untrusted
1702 local users to gain root access.
1705 Frequency : when using O_DIRECT and quotas
1707 Description: Incorrect file ownership on O_DIRECT output files
1708 Details : block usage reported by 'lfs quota' does not take into account
1709 files that have been written with O_DIRECT.
1714 Description: touch file failed when fs is not full
1715 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1716 otherwise we can get ENOSP while fs is not full.
1720 Description: parallel lock callbacks
1721 Details : Instead of sending blocking and completion callbacks as separated
1722 requests, adding them to a set and sending in parallel.
1725 Frequency : only for Cray XT3
1726 Bugzilla : 12829/13455
1727 Description: Changing primary group doesn't change the group lustre assigns to
1729 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1730 overridden with the primary group provided by the group upcall,
1731 whereas the supplied fsgid can be trusted if it is in the list of
1732 supplementary groups returned by the group upcall.
1734 Severity : enhancement
1736 Description: Allow masking D_WARNING, D_ERROR messages from console
1737 Details : Console messages can now be disabled via lnet.printk.
1741 Description: User code with malformed file open parameter crashes client node
1742 Details : Before packing join_file req, all the related reference should be
1743 checked carefully in case some malformed flags cause fake
1744 join_file req on client.
1748 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1749 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1750 mode is equal to the granted mode. However, it should also check
1751 that LDLM_FL_CP_REQD is not set.
1755 Description: Heavy nfs access might result in deadlocks
1756 Details : After ELC code landed, it is now improper to enqueue any mds
1757 locks under och_sem, because enqueue might want to decide to
1758 cancel open locks for same inode we are holding och_sem for.
1762 Description: Client eviction while running blogbench
1763 Details : A lot of unlink operations with concurrent I/O can lead to a
1764 deadlock causing evictions. To address the problem, the number of
1765 oustanding OST_DESTROY requests is now throttled to
1766 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1767 callbacks are processed in priority.
1771 Description: enable ACLs on MDS by default
1772 Details : ACLs must be enabled on MDS by default.
1775 Frequency : PPC/PPC64 only
1777 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1778 Details : fix duplicated definitions between asm-ppc64/types.h and
1779 lustre_types.h on PPC.
1782 Frequency : PPC/PPC64 only
1784 Description: asm-ppc/segment.h does not exist
1785 Details : fix compile issue on PPC.
1789 Description: better handle error messages in extents code
1792 Frequency : RHEL4 only
1794 Description: mkfs is very slow on IA64/RHEL4
1795 Details : A performance regression has been discovered in the MPT Fusion
1796 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1797 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1798 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1800 Severity : enhancement
1802 Description: SNMP support enhancement
1803 Details : Adding total number of sampled request for an MDS node in snmp
1806 Severity : enhancement
1808 Description: Optimize ldlm waiting list processing for PR extent locks
1809 Details : When processing waiting list for read extent lock and meeting read
1810 lock that is same or wider to it that is not contended, skip
1811 processing rest of the list and immediatelly return current
1812 status of conflictness, since we are guaranteed there are no
1813 conflicting locks in the rest of the list.
1817 Description: Time out and refuse to reconnect
1818 Details : When the failover node is the primary node, it is possible
1819 to have two identical connections in imp_conn_list. We must
1820 compare not conn's pointers but NIDs, otherwise we can defeat
1821 connection throttling.
1825 Description: port llog fixes from b1_6 into HEAD
1826 Details : Port llog reference couting and some llog cleanups from b1_6
1827 (bug 10800) into HEAD, for protect from panic and access to already
1828 free llog structures.
1832 Description: Detect stride IO mode in read-ahead
1833 Details : When a client does stride read, read-ahead should detect that and
1834 read-ahead pages according to the detected stride pattern.
1838 Description: data checksumming impacts single node performance
1839 Details : add support for several checksum algorithm. Currently, only CRC32
1840 and Adler-32 are supported. The checksum type can be changed on
1841 the fly via /proc/fs/lustre/osc/*/checksum_type.
1845 Description: use adler32 for page checksums
1846 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1851 Description: build for x2 fails
1852 Details : fix compile issue on Cray systems.
1856 Description: Properly match for duplicate locks
1857 Details : Due to different lock order from skiplists code, we need to
1858 traverse entire list for now
1861 Frequency : only on PPC/SLES10
1863 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1864 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1865 be defined. Add a hack in mkfs_lustre.c to work around this compile
1870 Description: LASSERT on MDS when client holding flock lock dies
1871 Details : ldlm pool logic depends on number of granted locks equal to
1872 number of released locks which is not true for flock locks, so
1873 just exclude such locks from consideration.
1877 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1878 Details : Use fsfilt_sync() for both the whole filesystem sync and
1879 individual file sync to eliminate dangerous inode locking
1880 with I_LOCK that can lead to a deadlock.
1884 Description: performance in 1.6.3
1885 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1886 sure that 1MB requests are not fragmented by the block layer.
1888 Severity : enhancement
1890 Description: organize the server-side client stats on per-nid basis
1891 Details : Change the structure of stats under obdfilter and mds to
1899 The "uuid"s file would list the uuids of _active_ exports.
1900 And the clear entry is to clear all stats and stale nids.
1902 Severity : enhancement
1904 Description: eliminate client locks in face of contention
1905 Details : file contention detection and lockless i/o implementation
1906 for contended files.
1910 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1913 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1915 Description: MDS may lose file striping (and hence file data) in some cases
1916 Details : If there are additional extended attributes stored on the MDS,
1917 in particular ACLs, SELinux, or user attributes (if user_xattr
1918 is specified for the client mount options) then there is a risk
1919 of attribute loss. Additionally, the Lustre file striping
1920 needs to be larger than default (e.g. striped over all OSTs),
1921 and an additional attribute must be stored initially in the
1922 inode and then increase in size enough to be moved to the
1923 external attribute block (e.g. ACL growing in size) for file
1928 Description: skiplist implementation simplification
1929 Details : skiplists are used to group compatible locks on granted list
1930 that was implemented as tracking first and last lock of each
1931 lock group the patch changes that to using doubly linked lists
1935 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1936 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1937 because LASSERT on both the data supplied by a client, and the
1938 data on disk is dangerous and incorrect.
1940 Severity : enhancement
1942 Description: Slow truncate/writes to huge files at high offsets.
1943 Details : Directly associate cached pages to lock that protect those pages,
1944 this allows us to quickly find what pages to write and remove
1945 once lock callback is received.
1949 Description: more ldlm soft lockups
1950 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1951 starve other threads from the resource lock for a long time in
1952 case of long waiting queue, so change the debug level from
1953 D_OTHER to the less frequently used D_INFO.
1955 Severity : enhancement
1957 Description: add -gid, -group, -uid, -user options to lfs find
1961 Description: Hung threads in invalidate_inode_pages2_range
1962 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1963 once one is completed. As a result, some RPCs are stuck in the
1964 queue and are never sent.
1968 Description: filter threads hungs on waiting journal commit
1969 Details : Cleanup filter group llog code, then only filter group llog will
1970 be only created in the MDS/OST syncing process.
1974 Description: Procfs and llog threads access destoryed import sometimes.
1975 Details : Sync the import destoryed process with procfs and llog threads by
1976 the import refcount and semaphore.
1978 Severity : enhancement
1980 Description: openlock cache of b1_6 port to HEAD
1985 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1986 Details : If insertion of an extent fails, then discard the inode
1987 preallocation and free data blocks else it can lead to duplicate
1992 Description: don't always update ctime in ext3_xattr_set_handle()
1993 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1994 In some cases the ctime should not be updated, for example for
1995 2.0->1.8 compatibility it is necessary to delete an xattr and it
1996 should not update the ctime.
2000 Bugzilla : 15713/16362
2001 Description: Assertion in iopen_connect_dentry in 1.6.3
2002 Details : looking up an inode via iopen with the wrong generation number can
2003 populate the dcache with a disconneced dentry while the inode
2004 number is in the process of being reallocated. This causes an
2005 assertion failure in iopen since the inode's dentry list contains
2006 both a connected and disconnected dentry.
2010 Description: assertion failure in ldlm_handle2lock()
2011 Details : fix a race between class_handle_unhash() and class_handle2object()
2012 introduced in lustre 1.6.5 by bug 13622.
2017 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
2018 Details : remove the limit of 256 scsi disks in the sd_iostat patch
2023 Description: oops in sd_iostats_seq_show()
2024 Details : unloading/reloading the scsi low level driver triggers a kernel
2025 bug when trying to access the sd iostat file.
2030 Description: Kernel panics during QLogic driver reload
2031 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
2032 patch, causing memory corruption.
2037 Description: journal_dev option does not work in b1_6
2038 Details : pass mount option during pre-mount.
2040 Severity : enhancement
2042 Description: Add a FIEMAP(FIle Extent MAP) ioctl
2043 Details : FIEMAP ioctl will allow an application to efficiently fetch the
2044 extent information of a file. It can be used to map logical blocks
2045 in a file to physical blocks in the block device.
2049 Description: LDLM soft lockups - improvement
2050 Details : It is be possible to send the lock handle along with each read
2051 or write request because the client is already doing a lock match
2052 itself so there isn't any reason the OST should have to re-do that
2056 Frequency : only on Cray X2
2058 Description: X2 build failures
2059 Details : fix build failures on Cray X2.
2063 Description: xid & resent requests
2064 Details : Initialize RPC XID from clock at startup (randomly if clock is
2067 Severity : enhancement
2069 Description: Add lustre_start utility to start or stop multiple Lustre servers
2074 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
2075 Details : In case of memory pressure, list_del() can be called twice on
2076 req->rq_history_list, causing a kernel oops.
2080 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
2081 Details : fix stack overflow in the distributed lock manager by defering
2082 export eviction after a failed AST to the elt thread instead of
2083 handling it in the dlm interpret routine.
2087 Description: Convert some comments to new format.
2088 Details : Update documenting comments to match doxygen conventions.
2092 Description: Grammar fixes.
2093 Details : A couple of trivial spelling fixes.
2097 Description: OSD_COUNTERS-mandatory
2098 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
2099 to simplify the code.
2103 Description: simplify cmm_device freeing logic.
2104 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
2108 Description: Add lockdep support to dt_object_operations locking interface.
2109 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
2110 parameter indicating lock ordering. Update mdd code to use new
2115 Description: Introduce failloc constants for lockless IO tests.
2116 Details : Add two new failloc constants to test lockless IO. Only one of
2117 them in implemented---another is checked in yet to be landed
2122 Description: Add lockdep support for inode mutex.
2123 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
2124 in the situations where Lustre has to lock more than one inode
2129 Description: Add optional invariants checking support.
2130 Details : Add new LINVRNT() macro, optional on new --enable-invariants
2131 configure switch. This macro is to be used for consistency and
2132 sanity checks that are too expensive to be left in `production'
2137 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
2138 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
2143 Description: Add lockdep support for ldlm_lock and ldlm_resource.
2144 Details : Use spin_lock_nested() in (the only) situation where more than
2145 one ldlm_lock is locked simultaneously. Also, fix possible
2146 dead-lock in ldlm_lock_change_resource() by enforcing particular
2151 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
2152 Details : Instead of specifying each ldlm_lock call-back through separate
2153 parameter, wrap them into struct ldlm_callback_suite.
2157 Description: Kill join_lru obd method and its callers.
2158 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
2159 in memory---a requirement implemented through ->o_join_lru() obd
2160 method in HEAD. Remove this method and its users.
2164 Description: Add asynchronous ldlm ENQUEUE completion handler.
2165 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
2166 case that stock ldlm_completion_ast() cannot handle as it waits
2167 until lock is granted. Introduce new ldlm_completion_ast_async()
2168 for this. Also comment ldlm_completion_ast().
2172 Description: ldlm_error <-> errno conversion.
2173 Details : Add functions to map (rather arbitrary) between LDLM error codes
2174 and standard errno values. CLIO needs this to prevent LDLM specific
2175 constants from escaping ldlm and osc.
2179 Description: Kill unused ldlm_handle2lock_ns() function.
2180 Details : Kill unused ldlm_handle2lock_ns() function.
2184 Description: Add lu_ref support to ldlm_lock
2185 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
2186 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
2187 and ldlm_resource. LDLM interface has to be changed, because code that
2188 releases a reference on a lock, has to "know" what reference this is.
2189 In the most frequent case
2191 lock = ldlm_handle2lock(handle);
2193 LDLM_LOCK_PUT(lock);
2195 no changes are required. When any other reference (received _not_ from
2196 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
2197 instead of LDLM_LOCK_PUT().
2199 Arguably, changes are pervasive, and interface requires some discipline
2200 for proper use. On the other hand, it was very instrumental in finding
2201 a few leaked lock references.
2205 Description: Add ldlm_lock_addref_try().
2206 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
2207 attempts to addref a lock that might be being canceled
2212 Description: Add ldlm_weigh_callback().
2213 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
2214 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
2215 hard-coded `number of pages' logic.
2219 Description: Add lockdep annotations to llog code.
2220 Details : Use appropriately tagged _nested() locking calls in the places
2221 where llog takes more than one ->lgh_lock lock.
2225 Description: Add loi_kms_set().
2226 Details : Wrap kms updates into a helper function.
2230 Description: Constify instances of struct lsm_operations.
2231 Details : Constify instances of struct lsm_operations.
2235 Description: lu_conf support.
2236 Details : On a server, a file system object is uniquely identified
2237 by a fid, which is sufficient to locate and load all object
2238 state (inode). On a client, on the other hand, more data are
2239 necessary instantiate an object. Change lu_object_find() and
2240 friends to take additional `lu_conf' argument describing object.
2241 Typically this includes layout information.
2245 Description: lu_context fixes.
2246 Details : Introduce new lu_context functions that are needed on the client
2247 side, where some system threads (ptlrpcd) are shared by multiple
2248 modules, and so cannot be stopped during module shutdown.
2252 Description: Add start and stop methods to lu_device_type_operations.
2253 Details : Introduce two new methods in lu_device_type_operations, that are
2254 invoked when first instance of a given type is created and last one
2255 is destroyed respectively. This is need by CLIO.
2259 Description: Add lu_ref support to struct lu_device.
2260 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
2261 track leaked references.
2265 Description: Introduce lu_kmem_descr.
2266 Details : lu_kmem_descr and its companion interface allow to create
2267 and destroy a number of kmem caches at once.
2271 Description: Fix lu_object finalization race.
2272 Details : Fix a race between lu_object_find() finding an object and its
2273 concurrent finalization. This race is (most likely) not possible
2274 on the server, but might happen on the client.
2278 Description: Introduce lu_ref interface.
2279 Details : lu_ref is a debugging module allowing to track references to
2280 a given object. It is quite cpu expensive, and has to be
2281 explicitly enabled with --enable-lu_ref. See usage description
2286 Description: Factor lu_site procfs stats into a separate function.
2287 Details : Separate lu_site stats printing code into a separate function
2288 that can be reused on a client.
2292 Description: Constify instances of struct {lu,dt,md}_device_operations.
2293 Details : Constify instances of struct {lu,dt,md}_device_operations.
2297 Description: Introduce struct md_site and move meta-data specific parts of
2298 struct lu_site here.
2299 Details : Move md-specific fields out of struct lu_site into special struct
2300 md_site, so that lu_site can be used on a client.
2304 Description: Kill mdd_lov_destroy().
2305 Details : Remove unused mdd code.
2309 Description: Add st_block checking to multistat.c.
2310 Details : Add st_block checking to multistat.c.
2314 Description: Add lu_ref support to struct obd_device.
2315 Details : Add lu_ref tracking to obd_device.
2319 Description: Kill obd_set_fail_loc().
2320 Details : Remove unused code.
2324 Description: Add special type for ptlrpc_request interpret functions.
2325 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
2326 there. Actual usage will be in CLIO.
2330 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
2331 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
2332 doesn't work with lockdep.
2336 Description: Add rwv.c test program.
2337 Details : New testing program exercising readv(2) and writev(2) (Qian).
2341 Description: Add sendfile.c test program.
2342 Details : New testing program exercising sendfile(2) (Jay).
2346 Description: Ratelimit a message that can be very frequent.
2347 Details : Ratelimit a memory allocation failure message that can
2352 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2353 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2357 Description: Liblustre build fixes.
2358 Details : Liblustre build fixes.
2362 Description: libcfs: add cfs_{need,cond}_resched() interface.
2363 Details : libcfs: add cfs_{need,cond}_resched() definition and
2364 implementations for Linux, NT, and liblustre.
2366 Severity : enhancement
2368 Description: More exported tunables for mballoc
2369 Details : Add support for tunable preallocation window and new tunables for
2370 large/small requests
2374 Description: Detect corruption of block bitmap and checking for preallocations
2375 Details : Checks validity of on-disk block bitmap. Also it does better
2376 checking of number of applied preallocations. When corruption is
2377 found, it turns filesystem readonly to prevent further corruptions.
2381 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2382 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2383 window to 0. If it does hit cache miss, and read-pattern is still
2384 stride-io mode, does not reset the stride window, but also does not
2385 increase the stride window length in this case.
2389 Frequency : only for big-endian servers
2390 Description: Check if system is big-endian while mounting fs with extents feature
2391 Details : Mounting a filesystem with extents feature will fail on big-endian
2392 systems since ext3-based ldiskfs is not supported on big-endian
2393 systems. This can be over-riden with "bigendian_extents" mount option.
2395 Severity : enhancement
2397 Description: The root squash functionality
2398 Details : A security feature, which is to prevent users from being able
2399 to mount lustre on their desktop, run as root, and delete
2400 all of the files in the filesystem. The goal is accomplished by
2401 remapping user id (UID) and group id (GID) of the root user to
2402 a UID and GID specified by the system administartor via Lustre
2403 configuration management server (MGS). The functionality also
2404 allows to specify sets of clients for which the remapping does
2409 Description: Excessive recovery window
2410 Details : With AT enabled, the recovery window can be excessively long (6000+
2411 seconds). To address this problem, we no longer use
2412 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2413 timeout no longer depends on the service time, it is set to
2414 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2415 time via pb_service_time.
2419 Description: Watchdog triggered on MDS failover
2420 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2421 the OSTs know that the MDS "UUID" can be reused for the same export
2422 from a different NID, so we do not need to wait for the export to be
2426 Frequency : rare, only if using MMP with Linux RAID
2428 Description: MMP doesn't work with Linux RAID
2429 Details : While using HA for Lustre servers with Linux RAID, it is possible
2430 that MMP will not detect multiple mounts. To make this work we
2431 need to unplug the device queue in RAID when the MMP block is being
2432 written. Also while reading the MMP block, we should read it from
2433 disk and not the cached one.
2435 Severity : enhancement
2437 Description: open file using fid
2438 Details : A file can be opened using just its fid, like
2439 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2442 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2445 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2446 Details : add proper locking for creation of jbd_4k slab cache
2450 Description: MMP check in ext3_remount() fails without displaying any error
2451 Details : When multiple mount protection fails during remount, proper error
2454 Severity : enhancement
2456 Description: Allow stripe size to be up to 4G-64k
2457 Details : Fix math logic to allow large stripe sizes.
2461 Description: add check for >8TB ldiskfs filesystems
2462 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2463 Don't allow >8TB ldiskfs filesystems to be mounted without
2464 force_over_8tb mount option
2469 Descriptoin: OSS complains "too many blocks" and crashes
2470 Details : In filter_commitrw_write():
2471 1. lquota_chkquota();
2472 2. obdo_from_inode();
2473 3. lquota_pending_commit();
2474 "2." has changed uid and gid in this case, then caused
2475 lquota_chkquota() and lquota_pending_commit() to use
2476 different uid and gid. That is the root of the bug.
2480 Descriptoin: getstripe doesn't report stripe count = -1 sanely
2481 Details : Detect and print out the 'stripe count = -1' case properly.
2485 Description: Out or order replies might be lost on replay
2486 Details : In ptlrpc_retain_replayable_request if we cannot find retained
2487 request with tid smaller then one currently being added, add it
2488 to the start, not end of the list.
2490 --------------------------------------------------------------------------------
2492 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2494 * Support for kernels:
2495 2.6.5-7.283 (SLES 9),
2496 2.6.9-55.EL (RHEL 4),
2497 2.6.16.46-0.14 (SLES 10),
2498 2.6.18.8 vanilla (kernel.org)
2499 * Client support for unpatched kernels:
2500 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2501 2.6.16 - 2.6.22 vanilla (kernel.org)
2502 * Due to recently discovered recovery problems, we do not recommend
2503 using patchless RHEL 4 clients with this or any earlier release.
2504 * Recommended e2fsprogs version: 1.39.cfs8
2505 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2506 * Starting with this release, the ldiskfs backing filesystem required
2507 by Lustre is now in its own package, lustre-ldiskfs. This package
2508 should be installed. It is versioned separately from Lustre and
2509 may be released separately in future.
2513 Description: block reactivating mgc import until all deactivates complete
2514 Details : Fix race when failing back MDT/MGS to itself (testing)
2517 Frequency : at statup only
2519 Description: mds_lov_synchronize race leads to various problems
2520 Details : simultaneous MDT->OST connections at startup can cause the
2521 sync to abort, leaving the OSC in a bad state.
2523 Severity : enhancement
2525 Description: add optional extra BUILD_VERSION info
2526 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2527 to override the lustre version.
2530 Frequency : 2.6.18 servers only
2532 Description: ll_kern_mount() doesn't release the module reference
2533 Details : The ldiskfs module reference count never drops down to 0
2534 because ll_kern_mount() doesn't release the module reference.
2539 Description: server LBUG when using old ost_num_threads parameter
2540 Details : Accept the old ost_num_threads parameter but warn that it
2541 is deprecated, and fix an off-by-one error that caused an LBUG.
2546 Description: Transient SCSI error results in persistent IO issue
2547 Details : iobuf->dr_error is not reinitialized to 0 between two
2551 Frequency : sometimes when underlying device returns I/O errors
2553 Description: OSTs not going read-only during write failures
2554 Details : OSTs are not remounted read-only when the journal commit threads
2555 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2556 instead of the ext3 wrappers.
2560 Description: poor connect scaling with increasing client count
2561 Details : Don't run filter_grant_sanity_check for more than 100 exports
2562 to improve scaling for large numbers of clients.
2565 Frequency : SLES10 only
2567 Description: sanity-quota.sh quotacheck failed: rc = -22
2568 Details : Quotas cannot be enabled on SLES10.
2571 Frequency : liblustre clients only
2573 Description: getdirentries does not give error when run on compute nodes
2574 Details : getdirentries does not fail when the size specified as an argument
2575 is too small to contain at least one entry
2577 Severity : enhancement
2579 Description: Add LNET router traceability for debug purposes
2580 Details : If a checksum failure occurs with a router as part of the
2581 IO path, the NID of the last router that forwarded the bulk data
2582 is printed so it can be identified.
2587 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2588 Details : Due to a race condition, liblustre clients were occasionally
2589 evicted incorrectly.
2591 Severity : enhancement
2593 Description: lfs setstripe use optional parameters instead of postional
2596 Severity : enhancement
2598 Description: Nanosecond timestamp support for ldiskfs
2599 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2600 resolution timestamps. There is not yet support for this at
2601 the Lustre filesystem level.
2604 Frequency : during server recovery
2606 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2607 Details : request with rq_no_resend flag not awake l_wait_event if they get
2611 Frequency : nfs export on patchless client
2613 Description: connectathon hang when test nfs export over patchless client
2614 Details : Disconnected dentry cannot be found with lookup, so we do not need
2615 to unhash it or make it invalid
2618 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2619 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2620 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2625 Description: open req refcounting wrong on reconnect
2626 Details : If reconnect happened between getting open reply from server and
2627 call to mdc_set_replay_data in ll_file_open, we will schedule
2628 replay for unreferenced request that we are about to free.
2629 Subsequent close will crash in variety of ways.
2630 Check that request is still eligible for replay in
2631 mdc_set_replay_data().
2636 Description: disable writes to filesystem when reading health_check file
2637 Details : the default for reading the health_check proc file has changed
2638 to NOT do a journal transaction and write to disk, because this
2639 can cause reads of the /proc file to hang and block HA state
2640 checking on a healthy but otherwise heavily loaded system. It
2641 is possible to return to the previous behaviour during configure
2642 with --enable-health-write.
2644 Severity : enhancement
2646 Description: 64-bit inode version
2647 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2648 to the inode. This will be required for version-based recovery.
2653 Description: MDS fails to start if a duplicate client export is detected
2654 Details : in some rare cases it was possible for a client to connect to
2655 an MDS multiple times. Upon recovery the MDS would detect this
2656 and fail during startup. Handle this more gracefully.
2658 Severity : enhancement
2660 Description: Add -o localflock option to simulate old noflock
2662 Details : This will achieve local-only flock/fcntl locks
2668 Description: log_commit_thread vs filter_destroy race leads to crash
2669 Details : Take import reference before releasing llog record semaphore
2674 Description: Wrong request locking in request set processing
2675 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2676 imp_delayed_list, in this place should be used imp_lock.
2679 Frequency : when reconnection
2681 Description: Grant Leak when osc reconnect to OST
2682 Details : When osc reconnect ost, OST(filter) should check whether it
2683 should grant more space to client by comparing fed_grant and
2684 cl_avail_grant, and return the granted space to client instead
2685 of "new granted" space, because client will call osc_init_grant
2686 to update the client grant space info.
2689 Frequency : when client reconnect to OST
2691 Description: Grant Leak when osc do resend and replay bulk write
2692 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2693 bulk write request, because the grant info while be sync between
2694 OSC and OST when reconnect, and we should ignore the grant info
2695 these of resend/replay write req.
2700 Description: Grant space more than avaiable left space sometimes.
2701 Details : When then OST is about to be full, if two bulk writing from
2702 different clients came to OST. Accord the avaliable space of the
2703 OST, the first req should be permitted, and the second one
2704 should be denied by ENOSPC. But if the seconde arrived before
2705 the first one is commited. The OST might wrongly permit second
2706 writing, which will cause grant space > avaiable space.
2709 Frequency : when client is evicted
2711 Description: Grant might be wrongly erased when osc is evicted by OST
2712 Details : when the import is evicted by server, it will fork another
2713 thread ptlrpc_invalidate_import_thread to invalidate the
2714 import, where the grant will be set to 0. While the original
2715 thread will update the grant it got when connecting. So if
2716 the former happened latter, the grant will be wrongly errased
2717 because of this race.
2722 Description: Checking Stale with correct fid
2723 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2724 to check whether it is stale, because sometimes, we want the
2725 enqueue happened anyway, and op_data.fid2 will not be initialized.
2727 Severity : enhancement
2729 Description: update patchless client
2730 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2733 Frequency : only with 2.4 kernel
2735 Description: random memory corruption
2736 Details : size of struct ll_inode_info is to big for union inode.u and this
2737 can be cause of random memory corruption.
2742 Description: Memory leak in recovery
2743 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2744 It should also check obd_fail before fsfilt_start, otherwise if
2745 fsfilt_start return -EROFS,(failover mds during mds recovery).
2746 then the req will return with repmsg->transno = 0 and rc = EROFS.
2747 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2748 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2749 be freed no matter whether fsfilt_commit success or not.
2752 Frequency : only with huge count clients
2754 Description: Prevents from taking the superblock lock in llap_from_page for
2756 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2757 for llap_from_page call in ll_removepage prevents from taking the
2758 superblock lock for a soon died page.
2763 Description: Not check open intent error before release open handle
2764 Details : in some rare cases, the open intent error is not checked before
2765 release open handle, which may cause
2766 ASSERTION(open_req->rq_transno != 0), because it tries to release
2767 the failed open handle.
2772 Description: Set cat log bitmap only after create log success.
2773 Details : in some rare cases, the cat log bitmap is set too early. and it
2774 should be set only after create log success.
2778 Description: Accessing a block bevice can re-enable I/O when Lustre is
2779 tearing down a device.
2780 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2785 Description: service threads may hog cpus when there are a lot of requests
2786 Details : Insert cond_resched to give other threads a chance to use some CPU
2791 Description: the cat log was not initialized in recovery
2792 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2793 unlink log on mds will not be initialized until mds post
2794 recovery. And also in mds post recovery, the unlink log will
2795 initialization will be done asynchronausly, so there will be race
2796 between add unlink log and unlink log initialization.
2800 Description: brw_stats were being printed incorrectly
2801 Details : brw_stats were being printed as log2 but all of them were not
2802 recorded as log2. Also remove some code duplication arising from
2803 filter_tally_{read,write}.
2807 Frequency : rare, only in recovery.
2808 Description: ASSERTION(req->rq_type != LI_POISON) failed
2809 Details : imp_lock should be held while iterating over imp_sending_list for
2810 prevent destroy request after get timeout in ptlrpc_queue_wait.
2814 Description: replay-single.sh test 52 fails
2815 Details : A lock's skiplist need to be cleanup when it being unlinked
2816 from its resource list.
2820 Description: Short directio read returns full requested size rather than
2822 Details : Direct I/O operations should return actual amount of bytes
2823 transferred rather than requested size.
2825 Severity : enhancement
2827 Description: metadata RPC reduction (e.g. for rm performance)
2828 Details : decrease the amount of synchronous RPC between clients and servers
2829 by canceling conflicing lock before the operation on the client
2830 and packing thier handles into the main operation RPC to server.
2832 Severity : enhancement
2834 Description: Async OSC create to avoid the blocking unnecessarily.
2835 Details : If a OST has no remain object, system will block on the creating
2836 when need to create a new object on this OST. Now, ways use
2837 pre-created objects when available, instead of blocking on an
2838 empty osc while others are not empty. If we must block, we block
2839 for the shortest possible period of time.
2843 Description: improve handling recoverable errors
2844 Details : if request processig with error which can be recoverable on server
2845 request should be resend, otherwise page released from cache and
2848 Severity : enhancement
2850 Description: refine locking for avoid write wrong info into lov_objid file
2851 Details : fix possible races with add new target and write/update data in
2854 --------------------------------------------------------------------------------
2856 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2861 Frequency : on some architectures
2863 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2864 Details : Uninitialized flags sometimes cause configuration commands to
2868 Frequency : patchless clients only
2870 Description: missing __iget() symbol export
2871 Details : The __iget() symbol export is missing. To avoid the need for
2872 this on patchless clients the deathrow inode reaper is turned
2873 off, and we depend on the VM to clean up old inodes. This
2874 dependency was during via the fix for bug 12181.
2876 --------------------------------------------------------------------------------
2878 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2880 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2881 INTEROPERATE with older versions automatically. In many cases a
2882 special upgrade step is needed. Please read the
2883 user documentation before upgrading any part of a live system.
2884 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2885 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2886 * WARNING: Lustre configuration and startup changes are required with
2887 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2891 Severity : enhancement
2893 Description: MountConf
2894 Details : Lustre configuration is now managed via mkfs and mount
2895 commands instead of lmc and lconf. New obd types (MGS, MGC)
2896 are added for dynamic configuration management. See
2897 https://mail.clusterfs.com/wikis/lustre/MountConf for
2900 Severity : enhancement
2902 Description: dynamic OST addition
2903 Details : OSTs can now be added to a live filesystem
2905 Severity : enhancement
2907 Description: startup order invariance
2908 Details : MDTs and OSTs can be started in any order. Clients only
2909 require the MDT to complete startup.
2911 Severity : enhancement
2913 Description: parallel, asynchronous orphan cleanup
2914 Details : orphan cleanup is now performed in separate threads for each
2915 OST, allowing parallel non-blocking operation.
2917 Severity : enhancement
2919 Description: optimized stripe assignment
2920 Details : stripe assignments are now made based on ost space available,
2921 ost previous usage, and OSS previous usage, in order to try
2922 to optimize storage space and networking resources.
2924 Severity : enhancement
2926 Description: Permanently set tunables
2927 Details : All writable /proc/fs/lustre tunables can now be permanently
2928 set on a per-server basis, at mkfs time or on a live system.
2930 Severity : enhancement
2932 Description: Lustre message v2
2933 Details : Add lustre message format v2.
2935 Severity : enhancement
2937 Description: client OST exclusion list
2938 Details : Clients can be started with a list of OSTs that should be
2939 declared "inactive" for known non-responsive OSTs.
2943 Description: SPEC SFS validation failure on NFS v2 over lustre.
2944 Details : Changes the blocksize for regular files to be 2x RPC size,
2945 and not depend on stripe size.
2947 Severity : enhancement
2949 Description: Multiple MD RPCs in flight.
2950 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2951 To avoid overly-overloading MDS, introduce a limit on number of
2952 MDS RPCs in flight for a single client and add /proc controls
2953 to adjust this limit.
2955 Severity : enhancement
2957 Description: client read/write statistics
2958 Details : Add client read/write call usage stats for performance
2959 analysis of user processes.
2960 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2961 file access. extents_stats shows chunk size distribution.
2962 extents_stats_per_process show chunk size distribution per
2965 Severity : enhancement
2967 Description: mds statistics
2968 Details : Add detailed mds operations statistics in
2969 /proc/fs/lustre/mds/*/stats.
2973 Description: Failure of copying files with lustre special EAs.
2974 Details : Client side always return success for setxattr call for lustre
2975 special xattr (currently only "trusted.lov").
2980 Description: Refcount LNET uuids
2981 Details : The global LNET uuid list grew linearly with every startup;
2982 refcount repeated list entries instead of always adding to
2985 Severity : enhancement
2987 Description: Dynamic service threads
2988 Details : Within a small range, start extra service threads
2989 automatically when the request queue builds up.
2992 Frequency : mixed-endian client/server environments
2994 Description: mixed-endian crashes
2995 Details : The new msg_v2 system had some failures in mixed-endian
2998 Severity : enhancement
3000 Description: Easy OST removal
3001 Details : OSTs can be permanently deactivated with e.g. 'lctl
3002 conf_param lustre-OST0001.osc.active=0'
3004 Severity : enhancement
3006 Description: MGS proc entries
3007 Details : Added basic proc entries for the MGS showing what filesystems
3010 Severity : enhancement
3012 Description: provide MGS failover
3013 Details : Added config lock reacquisition after MGS server failover.
3015 Severity : enhancement
3017 Description: add Linux 2.4 support
3018 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
3022 Description: a large application tries to do I/O to the same resource and dies
3023 in the middle of it.
3024 Details : Check the req->rq_arrival time after the call to
3025 ost_brw_lock_get(), but before we do anything about
3026 processing it & sending the BULK transfer request. This
3027 should help move old stale pending locks off the queue as
3028 quickly as obd_timeout.
3031 Frequency : when an incorrect nid is specified during startup
3033 Description: ptlrpc connect to non-existant node causes kernel crash
3034 Details : LNET can't be re-entered from an event callback, which
3035 happened when we expire a message after the export has been
3036 cleaned up. Instead, hand the zombie cleanup off to another
3039 Severity : enhancement
3041 Description: plain/inodebits lock performance improvement
3042 Details : Grouping plain/inodebits in granted list by their request modes
3043 and bits policy, thus improving the performance of search through
3047 Frequency : only if OST filesystem is corrupted
3049 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3050 Details : for a short time RPCs with bulk IO are in the replay list,
3051 but replay of bulk IOs is unimplemented. If the OST filesystem
3052 is corrupted due to disk cache incoherency and then replay is
3053 started it is possible to trip an assertion. Avoid putting
3054 committed RPCs into the replay list at all to avoid this issue.
3057 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3059 Description: System hang on startup
3060 Details : This bug allowed the liblustre (e.g. catamount) client to
3061 return to the app before handling all startup RPCs. This
3062 could leave the node unresponsive to lustre network traffic
3063 and manifested as a server ptllnd timeout.
3065 Severity : enhancement
3067 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3068 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
3069 boolean to control whether to print peer debug info when a
3070 client's RPC times out.
3073 Frequency : only for kernels with patches from Lustre below 1.4.3
3075 Description: Remove old rdonly API
3076 Details : Remove old rdonly API which unused from at least lustre 1.4.3
3079 Frequency : only for devices with external journals
3081 Description: Set external device read-only also
3082 Details : During a commanded failover stop, we set the disk device
3083 read-only while the server shuts down. We now also set any
3084 external journal device read-only at the same time.
3087 Frequency : when upgrading from 1.4 while trying to change parameters
3089 Description: The wrong (new) MDC name was used when setting parameters for
3090 upgraded MDT's. Also allows changing of OSC (and MDC)
3091 parameters if --writeconf is specified at tunefs upgrade time.
3094 Frequency : when setting specific ost indicies
3096 Description: QOS code breaks on skipped indicies
3097 Details : Add checks for missing OST indicies in the QOS code, so OSTs
3098 created with --index need not be sequential.
3100 Severity : enhancement
3102 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3103 Details : The uninit_groups feature works in conjunction with the kernel
3104 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3105 up the pass1 processing of e2fsck. This is a read-only feature
3106 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3107 mount filesystems that have had this feature enabled.
3109 Severity : enhancement
3111 Description: Improve multi-block allocation algorithm to avoid fragmentation
3112 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3113 allocation locality and avoid filesystem fragmentation.
3115 ------------------------------------------------------------------------------
3117 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
3119 * Support for kernels:
3120 2.4.21-47.0.1.EL (RHEL 3)
3121 2.6.5-7.283 (SLES 9)
3122 2.6.9-42.0.10.EL (RHEL 4)
3123 2.6.12.6 vanilla (kernel.org)
3124 2.6.16.27-0.9 (SLES 10)
3125 * Recommended e2fsprogs version: 1.39.cfs5
3127 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3131 Frequency : occasional, depends on client load and configuration
3132 Bugzilla : 12181, 12203
3133 Description: data loss for recently-modified files
3135 Details : In some cases it is possible that recently written or created
3136 files may not be written to disk in a timely manner (this should
3137 normally be within 30s unless client IO load is very high).
3138 The problem appears as zero-length files or files that are a
3139 multiple of 1MB in size after a client crash or client eviction
3140 that are missing data at the end of the file.
3142 This problem is more likely to be hit on clients where files are
3143 repeatedly created and unlinked in the same directory, clients
3144 have a large amount of RAM, have many CPUs, the filesystem has
3145 many OSTs, the clients are rebooted frequently, and/or the files
3146 are not accessed by other nodes after being written.
3148 The presence of the problem can be detected by looking at
3149 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
3150 smaller than the second (nr_unused) then dirty files will not
3151 be flushed automatically to disk. "sync; sleep 10" should be
3152 run several times on the node before unmounting it to update
3153 Lustre (this is also safe to run on nodes without this problem).
3155 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
3156 that can cause this same problem, so customers using that kernel
3157 also need to update the kernel in addition to Lustre. In order
3158 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
3160 It is normal that files written just before a client crash (less
3161 than 30s) may not yet have been flushed to disk, even for local
3165 Frequency : frequent on thin XT3 nodes
3167 Description: UUID collision on thin XT3 Linux nodes
3168 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
3169 randomly, since we relied on an insufficiently-seeded PRNG.
3174 Description: OSS hangs after "All ost request buffers busy"
3175 Details : A deadlock between quota and journal operations caused OSS
3176 hangs after printing "All ost request buffers busy."
3179 Frequency : always on liblustre builds
3181 Description: Cleanup compiler warnings on liblustre
3184 Frequency : always on liblustre builds on XT3
3186 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
3190 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
3191 > 32000 subdirectories
3192 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
3193 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
3194 subdirectory count crosses 32000. This will aid e2fsck to
3195 correctly handle more than 32000 subdirectories.
3198 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3200 Description: System hang on startup
3201 Details : This bug allowed the liblustre (e.g. catamount) client to
3202 return to the app before handling all startup RPCs. This
3203 could leave the node unresponsive to lustre network traffic
3204 and manifested as a server ptllnd timeout.
3206 Severity : enhancement
3208 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3209 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
3210 boolean to control whether to print peer debug info when a
3211 client's RPC times out.
3216 Description: make O_SYNC working on 2.6 kernels
3217 Details : 2.6 kernels use different method for mark pages for write,
3218 so need add a code to lustre for O_SYNC work.
3223 Description: Failure to close file and release space on NFS
3224 Details : Put inode details into lock acquired in ll_intent_file_open.
3225 Use mdc_intent_lock in ll_intent_open to properly
3226 detect all kind of errors unhandled by mdc_enqueue.
3231 Description: proc file read during shutdown sometimes raced obd removal,
3233 Details : Add lock to prevent obd access after proc file removal.
3236 Frequency : Only for files larger than 4GB on 32-bit clients.
3238 Description: improperly doing page alignment of locks
3239 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
3240 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
3245 Description: RPCs being resent when they shouldn't be
3246 Details : Some RPCs that should not be resent are being resent. This
3247 can cause inconsistencies in the RPC state machine. Do not
3248 resend such requests.
3251 Frequency : rare, only with NFS export
3253 Description: Crash on NFS re-export node
3254 Details : under very unusual load conditions an assertion is hit in
3255 ll_intent_file_open()
3258 Frequency : only if OST filesystem is corrupted
3260 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3261 Details : for a short time RPCs with bulk IO are in the replay list,
3262 but replay of bulk IOs is unimplemented. If the OST filesystem
3263 is corrupted due to disk cache incoherency and then replay is
3264 started it is possible to trip an assertion. Avoid putting
3265 committed RPCs into the replay list at all to avoid this issue.
3270 Description: large O_DIRECT requests fail under memory pressure/fragmentation
3271 Details : Large single O_DIRECT read and write calls can fail to allocate
3272 a sufficiently large buffer to process the request. In case of
3273 allocation failure the allocation is retried with a smaller
3274 buffer and broken into smaller requests.
3276 Severity : enhancement
3278 Description: Add -o localflock option to simulate old noflock behaviour.
3279 Details : This will achieve local-only flock/fcntl locks coherentness.
3284 Description: versioning check is incomplete
3285 Details : Checking the version difference of client vs. server, report
3286 error if the gap is too big.
3291 Description: add support PG_writeback bit
3292 Details : add support for PG_writeback bit for Lustre, for more carefull
3293 work with page cache in 2.6 kernel. This also fix some deadlocks
3294 and remove hack for work O_SYNC with 2.6 kernel.
3296 Severity : enhancement
3298 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3299 Details : The uninit_groups feature works in conjunction with the kernel
3300 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3301 up the pass1 processing of e2fsck. This is a read-only feature
3302 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3303 mount filesystems that have had this feature enabled.
3305 Severity : enhancement
3307 Description: Improve multi-block allocation algorithm to avoid fragmentation
3308 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3309 allocation locality and avoid filesystem fragmentation.
3311 ------------------------------------------------------------------------------
3313 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
3315 * Support for kernels:
3316 2.6.9-42.0.3.EL (RHEL 4)
3317 2.6.5-7.276 (SLES 9)
3318 2.4.21-47.0.1.EL (RHEL 3)
3319 2.6.12.6 vanilla (kernel.org)
3320 2.6.16.21-0.8 (SLES10)
3321 * Recommended e2fsprogs version: 1.39.cfs2-0
3323 * The backwards-compatible /proc/sys/portals symlink has been removed
3324 in this release. Before upgrading, please ensure that you change
3325 any configuration scripts or /etc/sysctl.conf files that access
3326 /proc/sys/portals/* or sysctl portals.* to use the corresponding
3327 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
3328 in advance of the upgrade on any system running Lustre 1.4.6 or
3329 newer, since /proc/sys/lnet was added in that version.
3330 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3334 Frequency : only when quota is used
3336 Description: avoid scanning export list for quota master
3337 Details : Change the algorithms to avoid scanning export list in order
3338 to improve the efficiency.
3341 Frequency : MDS failover only, very rarely
3343 Description: "went back in time" messages on mds failover
3344 Details : The greatest transno may be lost when the current operation
3345 finishes with an error (transno==0) and the client's last_rcvd
3346 record is over-written. Save the greatest transno in the
3347 mds_last_transno for this case.
3350 Frequency : always for specific kernels and striping counts
3352 Description: client may get "Matching packet too big" without ACL support
3353 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
3354 when trying to access files in certain configurations. The
3355 clients should in fact be denied when mounting because they do
3356 not understand ACLs.
3359 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3361 Description: many clients connecting with IO in progress causes connect timeouts
3362 Details : Avoid synchronous journal commits to avoid delays caused by many
3363 clients connecting/disconnecting when bulk IO is in progress.
3364 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3365 OST_IO_PORTAL to avoid delays behind potentially many pending
3369 Frequency : occasionally with multiple writers to a single file
3371 Description: shared writes to file may result in wrong size reported by stat()
3372 Details : Allow growing of kms when extent lock is cancelled
3375 Frequency : always with random mmap IO to multi-striped file
3377 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3378 Details : Only if the hole is at the end of OST object so that kms is too
3379 small. Fix is to increase kms accordingly in ll_nopage.
3382 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3384 Description: writes to a missing object would leak memory on the OST
3385 Details : If there is an inconsistency between the MDS and OST filesystems,
3386 such that the MDS references an object that doesn't exist, writes
3387 to that object will leak memory due to incorrect cleanup in the
3388 error handling path, eventually running out of memory on the OST.
3393 Description: Creating too long symlink causes lustre errors
3394 Details : Check symlink and name lengths before sending requests to MDS.
3397 Frequency : only if flock is enabled (not on by default)
3399 Description: posix locks not released on fd closure on 2.6.9+
3400 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3401 kernels, this caused such locks not to be released on fd close and
3402 then assertions on fs unmount about still used locks.
3405 Frequency : MDS failover only, very rarely
3407 Description: clients may get ASSERTION(granted_lock != NULL)
3408 Details : When request was taking a long time, and a client was resending
3409 a getattr by name lock request. The were multiple lock requests
3410 with the same client lock handle and
3411 mds_getattr_name->fixup_handle_for_resent_request found one of the
3412 lock handles but later failed with ASSERTION(granted_lock != NULL).
3417 Description: handle->h_buffer_credits > 0, assertion failure
3418 Details : h_buffer_credits is zero after truncate, causing assertion
3419 failure. This patch extends the transaction or creates a new
3423 Frequency : NFS re-export or patchless client
3424 Bugzilla : 11179, 10796
3425 Description: Crash on NFS re-export node (__d_move)
3426 Details : We do not want to hash the dentry if we don't have a lock.
3427 But if this dentry is later used in d_move, we'd hit uninitialised
3428 list head d_hash, so we just do this to init d_hash field but
3429 leave dentry unhashed.
3432 Frequency : NFS re-export or patchless client
3434 Description: NFS exports has problem with symbolic link
3435 Details : lustre client didn't properly install dentry when re-exported
3436 to NFS or running patchless client.
3439 Frequency : NFS re-export or patchless client
3441 Description: Various nfs/patchless fixes.
3442 Details : fixes reuse disconected alias for lookup process - this fixes
3443 warning "find_exported_dentry: npd != pd",
3444 fix permission error with open files at nfs.
3445 fix apply umask when do revalidate.
3448 Frequency : occasional
3450 Description: Crash on NFS re-export node
3451 Details : calling clear_page() on the wrong pointer triggered oops in
3452 generic_mapping_read().
3455 Frequency : rarely, using O_DIRECT IO
3457 Description: unaligned directio crashes client with LASSERT
3458 Details : check for unaligned buffers before trying any requests.
3461 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3463 Description: stale data returned from RAID cache
3464 Details : If only a small amount of IO is done to the RAID device before
3465 reading it again it is possible to get stale data from the RAID
3466 cache instead of reading it from disk.
3469 Frequency : always for sles10 kernel
3471 Description: sles10 support
3472 Details : ll_follow_link: compile fixes and using of nd_set_link
3473 under newer kernels.
3476 Frequency : depends on arch, kernel and compiler version, always on sles10
3479 Description: recursive or deep enough symlinks cause stack overflow
3480 Details : getting rid of large stack-allocated variable in
3484 Frequency : depends on hardware
3486 Description: lustre write performance loss in the SLES10 kernel
3487 Details : the performance loss is caused by using of write barriers in the
3488 ext3 code. The SLES10 kernel turns barrier support on by
3489 default. The fix is to undo that change for ldiskfs.
3491 ------------------------------------------------------------------------------
3493 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3495 * Support for kernels:
3496 2.6.9-42.0.3EL (RHEL 4)
3497 2.6.5-7.276 (SLES 9)
3498 2.4.21-47.0.1.EL (RHEL 3)
3499 2.6.12.6 vanilla (kernel.org)
3503 Frequency : quota enabled and large files being deleted
3505 Description: releasing more than 4GB of quota at once hangs OST
3506 Details : If a user deletes more than 4GB of files on a single OST it
3507 will cause the OST to spin in an infinite loop. Release
3508 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3513 Description: statfs data retrieved from /proc may be stale or zero
3514 Details : When reading per-device statfs data from /proc, in the
3515 {kbytes,files}_{total,free,avail} files, it may appear
3516 as zero or be out of date.
3519 Frequency : systems with MD RAID1 external journal devices
3521 Description: lconf's call to blkid is confused by RAID1 journal devices
3522 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3523 of returning all block devices that match the journal UUID.
3526 Frequency : always, for aggregate stripe size over 4GB
3528 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3529 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3530 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3533 Frequency : always if "lfs find" used on a local file/directory
3535 Description: "lfs find" segfaults if used on a local file/directory
3536 Details : The case where a directory component was not specified wasn't
3537 handled correctly. Handle this properly.
3540 Frequency : always on ppc64
3542 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3543 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3544 perform endianness conversion.
3547 Frequency : rarely (truncate to non-zero file size after write under load)
3548 Bugzilla : 10730, 10687
3549 Description: Files padded with zeros to next 4K multiple
3550 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3551 it is possible that files that are truncated to a non-zero size
3552 immediately after being written are filled with zero bytes beyond
3553 the truncated size. No file data is lost.
3555 Severity : enhancement
3557 Description: Allow recovery/failover for liblustre clients.
3558 Details : liblustre clients were unaware of failover configurations until
3561 Severity : enhancement
3563 Description: user file locks should fail when not mounting with flock option
3564 Details : Set up an error-returning stub in ll_file_operations.lock field
3565 to prevent incorrect behaviour when client is mounted without
3566 flock option. Also, set up properly f_op->flock field for
3570 Frequency : always on ia64
3572 Description: "lfs df" loops on printing out MDS statfs information
3573 Details : The obd_ioctl_data was not initialized and in some systems
3574 this caused a failure during the ioctl that did not return
3575 an error. Initialize the struct and return an error on failure.
3578 Frequency : SLES 9 only
3580 Description: Error of copying files with lustre special EAs as root
3581 Details : Client side always return success for setxattr call for lustre
3582 special xattr (currently only "trusted.lov").
3585 Frequency : rarely on clusters with both ia64+i386 clients
3587 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3588 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3589 clients are doing concurrent writes to the same file it is
3590 possible that the ia64 clients may LASSERT because the OST
3591 extent locks are not PAGE_SIZE aligned. Ensure that grown
3592 locks are always aligned on the request boundary.
3595 Frequency : specific use, occasional
3597 Description: Overwriting in use executable truncates on-disk binary image
3598 Details : If one node attempts to overwrite an executable in use by
3599 another node, we now correctly return ETXTBSY instead of
3600 truncating the file.
3602 Severity : enhancement
3604 Description: Async OSC create to avoid the blocking unnecessarily.
3605 Details : If a OST has no remain object, system will block on the creating
3606 when need to create a new object on this OST. Now, ways use
3607 pre-created objects when available, instead of blocking on an
3608 empty osc while others are not empty. If we must block, we block
3609 for the shortest possible period of time.
3614 Description: chmod on Lustre root is propagated to other clients
3615 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3616 invalid by the follow_mount time.
3619 Frequency : liblustre clients only
3621 Description: Race in 'instant cancel' lock handling could lead to such locks
3622 never to be granted in case of SMP MDS
3623 Details : Do not destroy not yet granted but cbpending locks in
3627 Frequency : replay/resend of open
3629 Description: non null lock assetion failure in mds_intent_policy
3630 Details : Trying to replay/resend lockless open requests resulted in
3631 mds_open() returning 0 with no lock. Now it sets a flag if
3632 a lock is going to be returned.
3634 Severity : enhancement
3636 Description: Checksum enhancements
3637 Details : New checksum enhancements allow for resending RPCs that failed
3640 Severity : enhancement
3642 Description: Tunables on number of dirty pages in cacche
3643 Details : Allow to set limit on number of dirty pages cached.
3648 Description: client crash on unmount - lock still has references
3649 Details : In some error handling cases it was possible to leak a lock
3650 reference on a client while accessing a file. This was not
3651 harmful to the client during operation, but would cause the
3652 client to crash when the filesystem is unmounted.
3655 Frequency : specific case, rare
3657 Description: ETXTBSY on mds though file not in use by client
3658 Details : ETXTBSY is no longer incorrectly returned when attempting to
3659 chmod or chown a directory that the user previously tried to
3660 execute or a currently-executing binary.
3663 Frequency : extremely rare except on liblustre-based clients
3665 Description: Lustre space not freed when files are deleted
3666 Details : Clean up open-unlinked files after client eviction. Previously
3667 the unlink was skipped and the files remained as orphans.
3672 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3673 Details : In some cases it was possible to send a blocking callback to
3674 a client doing a glimpse, even though that client didn't get
3675 a lock granted. When the glimpse lock is cancelled on the OST
3676 the freed lock is left on the waiting list and corrupted the list.
3679 Frequency : all core dumps
3681 Description: Broke core dumps to lustre
3682 Details : Negative dentry may be unhashed if parent does not have UPDATE
3683 lock, but some callers, e.g. do_coredump, expect dentry to be
3684 hashed after successful create, hash it in ll_create_it.
3686 ------------------------------------------------------------------------------
3688 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3690 * Support for kernels:
3691 2.6.9-42.0.2.EL (RHEL 4)
3692 2.6.5-7.276 (SLES 9)
3693 2.4.21-40.EL (RHEL 3)
3694 2.6.12.6 vanilla (kernel.org)
3698 Frequency : always on RHEL 3
3700 Description: Number of open files grows over time
3701 Details : The number of open files grows over time, whether or not
3702 Lustre is started. This was due to a filp leak introduced
3703 by one of our kernel patches.
3705 ------------------------------------------------------------------------------
3707 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3709 * Support for kernels:
3710 2.6.9-42.EL (RHEL 4)
3711 2.6.5-7.276 (SLES 9)
3712 2.4.21-40.EL (RHEL 3)
3713 2.6.12.6 vanilla (kernel.org)
3718 Bugzilla : 5719, 9635, 9792, 9684
3719 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3720 Details : If a server is under heavy load and cannot reply to new
3721 connection requests before the client resends the (re)connect,
3722 the connection handling code can behave badly if two service
3723 threads are concurrently handing separate (re)connections from
3724 the same client. Add better locking to the connection handling
3725 code, and ensure that only a single connection will be processed
3726 for a given client UUID, even if the lock is dropped.
3728 Severity : enhancement
3730 Description: add TCP zero-copy support to kernel
3731 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3732 sends if the hardware supports scatter-gather and checksumming.
3733 This allows socklnd to do client-write and server-read more
3734 efficiently and reduce CPU utilization from skbuf copying.
3737 Frequency : only if NFS exporting from client
3739 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3740 Details : If setting attributes on a file created under NFS that had
3741 never been opened it would be possible to oops the client
3742 if the file had no objects.
3745 Frequency : always for liblustre
3747 Description: liblustre client does MDS+OSTs setattr RPC for each write
3748 Details : When doing a write from a liblustre client, the client
3749 incorrectly issued an RPC to the MDS and each OST the file was
3750 striped over in order to update the timestamps. When writing
3751 with small chunks and many clients this could overwhelm the MDS
3752 with RPCs. In all cases it would slow down the write because
3753 these RPCs are unnecessary.
3755 Severity : enhancement
3757 Description: allow number of MDS service threads to be changed at module load
3758 Details : It is now possible to change the number of MDS service threads
3759 running. Adding "options mds mds_num_threads={N}" to the MDS's
3760 /etc/modprobe.conf will set the number of threads for the next
3761 time Lustre is restarted (assuming the "mds" module is also
3762 reloaded at that time). The default number of threads will
3763 stay the same, 32 for most systems.
3768 Description: OST crash if filesystem is unformatted or corrupt
3769 Details : If an OST is started on a device that has never been formatted
3770 or if the filesystem is corrupt and cannot even mount then the
3771 error handling cleanup routines would dereference a NULL pointer.
3776 Description: NULL pointer deref in llap_from_page.
3777 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3778 incorrect) mapping if the page was truncated/reclaimed while it was
3779 searched for. Check for this condition and skip such pages when
3780 doing readahead. Introduce extra check to llap_from_page() to
3781 verify page->mapping->host is non-NULL (so page is not anonymous).
3784 Frequency : Sometimes when using sys_sendfile
3786 Description: "page not covered by a lock" warnings from ll_readpage
3787 Details : sendfile called ll_readpage without right page locks present.
3788 Now we introduced ll_file_sendfile that does necessary locking
3789 around call to generic_file_sendfile() much like we do in
3793 Frequency : with certain MDS communication failures at client mount time
3795 Description: NULL pointer deref after failed client mount
3796 Details : a client connection request may delayed by the network layer
3797 and not be sent until after the PTLRPC layer has timed out the
3798 request. If the client fails the mount immediately it will try
3799 to clean up before the network times out the request. Add a
3800 reference from the request import to the obd device and delay
3801 the cleanup until the network drops the request.
3804 Frequency : occasionally during client (re)connect
3806 Description: assertion failure during client (re)connect
3807 Details : processing a client connection request may be delayed by the
3808 client or server longer than the client connect timeout. This
3809 causes the client to resend the connection request. If the
3810 original connection request is replied in this interval, the
3811 client may trip an assertion failure in ptlrpc_connect_interpret()
3812 which thought it would be the only running connect process.
3815 Frequency : only with obd_echo servers and clients that are rebooted
3817 Description: kernel BUG accessing uninitialized data structure
3818 Details : When running an obd_echo server it did not start the ping_evictor
3819 thread, and when a client was evicted an uninitialized data
3820 structure was accessed. Start the ping_evictor in the RPC
3821 service startup instead of the OBD startup.
3823 Severity : enhancement
3824 Bugzilla : 10193 (patchless)
3825 Description: Remove dependency on various unexported kernel interfaces.
3826 Details : No longer need reparent_to_init, exit_mm, exit_files,
3827 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3830 Frequency : rare (only users of deprecated and unsupported LDAP config)
3832 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3833 Details : LDAP apparently contains 'lustreName' attributes instead of
3834 'name'. A simple remapping of the name is sufficient.
3837 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3839 Description: waiting_locks_callback trips kernel BUG if client is evicted
3840 Details : Running with the dump_on_timeout debug flag turned on makes
3841 it possible that the waiting_locks_callback() can try to dump
3842 the Lustre kernel debug logs from an interrupt handler. Defer
3843 this log dumping to the expired_lock_main() thread.
3845 Severity : enhancement
3847 Description: Support NFS exporting on 2.6 kernels.
3848 Details : Implement non-rawops metadata methods for NFS server to use without
3849 changing NFS server code.
3852 Frequency : very rare (synthetic metadata workload only)
3854 Description: two racing renames might cause an MDS thread to deadlock
3855 Details : Running the "racer" program may cause one MDS thread to rename
3856 a file from being the source of a rename to being the target of
3857 a rename at exactly the same time that another thread is doing
3858 so, and the second thread has already enqueued these locks after
3859 doing a lookup of the target and is trying to relock them in
3860 order. Ensure that we don't try to re-lock the same resource.
3863 Frequency : only very large systems with liblustre clients
3865 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3866 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3867 all OSTs in parallel. This allows the eviction of stale liblustre
3868 clients to proceed much faster than if they were done in series,
3869 and also offers similar improvements for other set_info RPCs.
3874 Description: excessive CPU usage during initial read phase on client
3875 Details : During the initial read phase on a client, it would agressively
3876 retry readahead on the file, consuming too much CPU and impacting
3877 performance (since 1.4.5.8). Improve the readahead algorithm
3878 to avoid this, and also improve some other common cases (read
3879 of small files in particular, where "small" is files smaller than
3880 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3885 Description: MDS crash when receiving packet with unknown intent.
3886 Details : Do not LBUG in unknown intent case, just return -EFAULT
3888 Severity : enhancement
3889 Bugzilla : 9293, 9385
3890 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3891 Details : Do not serialize getattr (non-intent version) and statfs.
3894 Frequency : occasional, when OST network is overloaded/intermittent
3896 Description: client evicted by OST after bulk IO timeout
3897 Details : If a client sends a bulk IO request (read or write) the OST
3898 may evict the client if it is unresposive to its data GET/PUT
3899 request. This is incorrect if the network is overloaded (takes
3900 too long to transfer the RPC data) or dropped the OST GET/PUT
3901 request. There is no need to evict the client at all, since
3902 the pinger and/or lock callbacks will handle this, and the
3903 client can restart the bulk request.
3906 Frequency : Always when mmapping file with no objects
3908 Description: client crashes when mmapping file with no objects
3909 Details : Check that we actually have objects in a file before doing any
3910 operations on objects in ll_vm_open, ll_vm_close and
3916 Description: Request leak when working with deleted CWD
3917 Details : Introduce advanced request refcount tracking for requests
3918 referenced from lustre intent.
3920 Severity : Enhancement
3922 Description: Cache open file handles on client.
3923 Details : MDS now will return special lock along with openhandle, if
3924 requested and client is allowed to hold openhandle, even if unused,
3925 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3926 closing files for every read/write openration.
3928 Severity : Enhancement
3930 Description: Cache open negative dentries on client when possible.
3931 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3932 negative dentries on lock revocation.
3937 Description: Remounting a client read-only wasn't possible with a zconf mount
3938 Details : It wasn't possible to remount a client read-only with llmount.
3940 Severity : enhancement
3941 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3942 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3943 MPICH is included to add Lustre-specific ADIO interfaces.
3944 This is based closely on the UFS ADIO layer and only differs
3945 in file creation, in order to allow the OST striping to be set.
3946 This is user-contributed code and not supported by CFS.
3951 Description: extended inode attributes (immutable, append-only) work improperly
3952 when 2.4 and 2.6 kernels are used on client/server or vice versa
3953 Details : Introduce kernel-independent values for these flags.
3955 Severity : enhancement
3958 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3959 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3960 granularity. Now, Fractional MB granularity is be supported,
3961 this is very useful for low memory system.
3963 Severity : enhancement
3965 Description: Getattr by fid
3966 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3967 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3968 have locks and avoids getattr rpc requests for it on every stat.
3971 Frequency : Always, for filesystems larger than 2TB
3973 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3974 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3975 filesystems larger than 2TB to be mounted with mballoc enabled.
3978 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3980 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3981 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3982 filesystems larger than 2TB is mounted then the kernel will
3983 silently corrupt the start of the filesystem. CONFIG_LBD is
3984 enabled for all CFS-supported kernels, but the possibility of
3985 this happening with a modified kernel config exists.
3987 Severity : enhancement
3989 Description: add client O_DIRECT support for 2.6 kernels
3990 Details : It is now possible to do O_DIRECT reads and writes to files
3991 in the Lustre client mountpoint on 2.6 kernel clients.
3993 Severity : enhancement
3995 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3996 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3997 parallel, not waiting for response from every OST before sending
3998 a rpc to the next OST.
4003 Description: setattr vs write race when updating file timestamps
4004 Details : Client processes that update a file timestamp into the past
4005 right after writing to the file (e.g. tar) it is possible that
4006 the updated file modification time can be reset to the current
4007 time due to a race between processing the setattr and write RPC.
4009 Severity : enhancement
4011 Description: Bring 'lfs find' closer in line with regular Linux find.
4012 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
4013 -print0 options and obtains all the needed info through the lustre
4016 Severity : enhancement
4018 Description: support up to 1024 configured devices on one node
4019 Details : change obd_dev array from statically allocated to dynamically
4020 allocated structs as they are first used to reduce memory usage
4025 Description: Flush dirty partially truncated pages during truncate
4026 Details : Immediatelly flush partially truncated pages in filter_setattr,
4027 this way we completely avoid having any pages in page cache on OST
4028 and can retire ugly workarounds during writes to flush such pages.
4033 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
4034 Details : protect inode from truncation within vfs_unlink() context
4035 just take a reference before calling vfs_unlink() and release it
4036 when parent's i_sem is free.
4041 Description: last_id value checked outside lock on OST caused LASSERT failure
4042 Details : If there were multiple MDS->OST object precreate requests in
4043 flight, it was possible that the OST's last object id was checked
4044 outside a lock and incorrectly tripped an assertion. Move checks
4045 inside locks, and discard old precreate requests.
4048 Frequency : always, if extents are used on OSTs
4050 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
4051 Details : OSTs using the extents format would not zero the high 16 bits of
4052 the index physical block number. This is not a problem for any
4053 OST filesystems smaller than 16TB, and no kernels support ext3
4054 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
4055 new/modified files) and can be fixed for existing filesystems
4056 with e2fsprogs-1.39-cfs1.
4061 Description: import connection selection may be incorrect if timer wraps
4062 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
4063 import connections to be ignored if the 32-bit jiffies counter
4064 wraps. Use a 64-bit jiffies counter.
4067 Frequency : very large clusters immediately after boot
4069 Description: LNET request buffers exhausted under heavy short-term load
4070 Details : If a large number of client requests are generated on a service
4071 that has previously never seen so many requests it is possible
4072 that the request buffer growth cannot keep up with the spike in
4073 demand. Instead of dropping incoming requests, they are held in
4074 the LND until the RPC service can accept more requests.
4077 Frequency : Sometimes during replay
4079 Description: Assertion failure in ll_local_open after replay.
4080 Details : If replay happened on an open request reply before we were able
4081 to set replay handler, reply will become not swabbed tripping the
4082 assertion in ll_local_open. Now we set the handler right after
4083 recognising of open request
4086 Frequency : very rare
4088 Description: kernel reports "badness in vsnprintf"
4089 Details : Reading from the "recovery_status" /proc file in small chunks
4090 may cause a negative length in lprocfs_obd_rd_recovery_status()
4091 call to vsnprintf() (which is otherwise harmless). Exit early
4092 if there is no more space in the output buffer.
4094 Severity : enhancement
4096 Description: clear OBD RPC statistics by writing to them
4097 Details : It is now possible to clear the OBD RPC statistics by writing
4098 to the "stats" file.
4103 Description: Client mtime is not the same on different clients after utimes
4104 Details : In some cases, the client was using the utimes() syscall on
4105 a file cached on another node. The clients now validate the
4106 ctime from the MDS + OSTs to determine which one is right.
4111 Description: Inability to activate failout mode
4112 Details : lconf script incorrectly assumed that in pythong string's numeric
4113 value is used in comparisons.
4116 Frequency : always with multiple stripes per file
4118 Description: Inefficient object allocation for mutli-stripe files
4119 Details : When selecting which OSTs to stripe files over, for files with
4120 a stripe count that divides evenly into the number of OSTs,
4121 the MDS is always picking the same starting OST for each file.
4122 Return the OST selection heuristic to the original design.
4127 Description: mount failures may take full timeout to return an error
4128 Details : Under some heavy load conditions it is possible that a
4129 failed mount can wait for the full obd_timeout interval,
4130 possibly several minutes, before reporting an error.
4131 Instead return an error as soon as the status is known.
4133 Frequency : quota enabled and large files being deleted
4135 Description: releasing more than 4GB of quota at once hangs OST
4136 Details : If a user deletes more than 4GB of files on a single OST it
4137 will cause the OST to spin in an infinite loop. Release
4138 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
4143 Description: statfs data retrieved from /proc may be stale or zero
4144 Details : When reading per-device statfs data from /proc, in the
4145 {kbytes,files}_{total,free,avail} files, it may appear
4146 as zero or be out of date.
4149 Frequency : systems with MD RAID1 external journal devices
4151 Description: lconf's call to blkid is confused by RAID1 journal devices
4152 Details : Use the "blkid -l" flag to locate the MD RAID device instead
4153 of returning all block devices that match the journal UUID.
4156 Frequency : always, for aggregate stripe size over 4GB
4158 Description: assertion fails when trying to use 4GB stripe size
4159 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
4160 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
4163 Frequency : always on ppc64
4165 Description: the first write on an ext3 filesystem with mballoc got stuck
4166 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
4167 perform endianness conversion.
4169 ------------------------------------------------------------------------------
4171 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
4173 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
4174 INTEROPERATE with older versions automatically. Please read the
4175 user documentation before upgrading any part of a live system.
4176 * WARNING: Lustre networking configuration changes are required with
4177 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
4180 * Support for kernels:
4181 2.6.9-22.0.2.EL (RHEL 4)
4182 2.6.5-7.244 (SLES 9)
4183 2.6.12.6 vanilla (kernel.org)
4186 Severity : enhancement
4187 Bugzilla : 7981/8208
4188 Description: Introduced Lustre Networking (LNET)
4189 Details : LNET is new networking infrastructure for Lustre, it includes
4190 a reorganized network configuration mode (see the user
4191 documentation for full details) as well as support for routing
4192 between different network fabrics. Lustre Networking Devices
4193 (LNDs) for the supported network fabrics have also been
4194 created for this new infrastructure.
4196 Severity : enhancement
4197 Description: Introduced Access control lists
4198 Details : clients can set ACLs on files and directories in order to have
4199 more fine-grained permissions than the standard Unix UGO+RWX.
4200 The MDS must be started with the "-o acl" mount option.
4202 Severity : enhancement
4203 Description: Introduced filesystem quotas
4204 Details : Administrators may now establish per-user quotas on the
4207 Severity : enhancement
4209 Description: Configuration change for the XT3
4210 The PTLLND is now used to run Lustre over Portals on the XT3
4211 The configure option(s) --with-cray-portals are no longer used.
4212 Rather --with-portals=<path-to-portals-includes> is used to
4213 enable building on the XT3. In addition to enable XT3 specific
4214 features the option --enable-cray-xt3 must be used.
4219 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
4220 Details : A race between allocating a new llcd and re-getting the llcd_lock
4221 allowed another thread to grab newly-allocated llcd.
4223 Severity : enhancement
4225 Description: 2.6 OST async journal commit and locking fix to improve performance
4226 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
4227 2.6 kernels are now async as they already were in 2.4 kernels so
4228 that they can commit concurrently with the network bulk transfer.
4229 For block-allocated files the filter allocation semaphore is held
4230 to avoid filesystem fragmentation during allocation. BKL lock
4231 removed for 2.6 xattr operations where it is no longer needed.
4236 Description: lconf incorrectly determined whether two IP networks could talk
4237 Details : In some more complicated routing and multiple-network
4238 configurations, lconf will avoid trying to make a network
4239 connection to a disjoint part of the IP space. It was doing the
4240 math incorrectly for one set of cases.
4245 Description: Fix for potential infinite loop processing records in an llog.
4246 Details : If an llog record is corrupted/zeroed, it is possible to loop
4247 forever in llog_process(). Validate the llog record length
4248 and skip the remainder of the block on error.
4251 Frequency : occasional (liblustre only)
4253 Description: liblustre could not open files whose last component is a symlink
4254 Details : sysio_path_walk() would incorrectly pass the open intent to
4255 intermediate path components.
4258 Frequency : rare (liblustre only with non-standard tuning)
4259 Bugzilla : 7201 (7350)
4260 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
4261 Details : llu_lookup_finish_locks() tries to set lock data on a lock
4262 after it has been released, only do this for referenced locks
4264 Severity : enhancement
4266 Description: specifying an (invalid) directory default stripe_size of -1
4267 would reset the directory default striping
4268 Details : stripe_size -1 was used internally to signal directory stripe
4269 removal, now use "all default" to signal dir stripe removal
4270 as a directory striping of "all default" is not useful
4273 Frequency : common for large clusters running liblustre clients
4275 Description: doing an ls when liblustre clients are running is slow
4276 Details : sending a glimpse AST to a liblustre client waits for every AST
4277 to time out, as liblustre clients will not respond. Since they
4278 cannot cache data we refresh the OST lock LVB from disk instead.
4280 Severity : enhancement
4282 Description: doing an ls at the same time as file IO can be slow
4283 Details : enqueue and other "small" requests can be blocked behind many
4284 large IO requests. Create a new OST IO portal for non-IO
4285 requests so they can be processed faster.
4288 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
4290 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
4291 created file with mode 000
4292 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
4293 allow an ftruncate() as a non-root user to a file with mode 000.
4294 Lustre now respects this flag to disable mode checks when
4295 truncating a file owned by the user
4298 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
4301 Description: Revoking locks from clients that went dead or catatonic might take
4303 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
4304 cancellation of such locks instant on servers without waiting for
4305 any reply from clients. Clients drops these locks when cancel
4306 notification from server is received without replying.
4309 Frequency : liblustre-only, when liblustre client dies or becomes busy
4311 Description: Doing ls on Linux clients can take a long time with active
4313 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
4314 granting such locks to it in the first place if possible. Locks
4315 are taken by proxy on the OST during the read or write and
4316 dropped immediately afterward. Add connect flags handling, do
4317 not grant locks to liblustre clients for glimpse ASTs.
4319 Severity : enhancement
4321 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
4322 Details : Existing read-ahead algorithm is tuned for the case of streamlined
4323 sequential reads and behaves badly with applications doing random
4324 reads. Improve it by reading ahead at least read region, and
4325 avoiding excessive large RPC for small reads.
4327 Severity : enhancement
4329 Description: Creating more than 1000 files for a single job may cause a load
4330 imbalance on the OSTs if there are also a large number of OSTs.
4331 Details : qos_prep_create() uses an OST index reseed value that is an
4332 even multiple of the number of available OSTs so that if the
4333 reseed happens in the middle of the object allocation it will
4334 still utilize the OSTs as uniformly as possible.
4339 Description: OST or MDS may oops in ping_evictor_main()
4340 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
4341 but doesn't restart at beginning of obd_exports_timed list
4344 Severity : enhancement
4346 Description: improve by-nid export eviction on the MDS and OST
4347 Details : allow multiple exports with the same NID to be evicted at one
4348 time without re-searching the exports list.
4351 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
4353 Description: MDS may oops in groups_free()
4354 Details : in rare race conditions a newly allocated group_info struct is
4355 freed again, and this can be NULL. The 2.4 compatibility code
4356 for groups_free() checked for a NULL pointer, but 2.6 did not.
4359 Frequency : common for liblustre clients doing little filesystem IO
4360 Bugzilla : 9352, 7313
4361 Description: server may evict liblustre clients accessing contended locks
4362 Details : if a client is granted a lock or receives a completion AST
4363 with a blocking AST already set it would not reply to the AST
4364 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4367 Frequency : lfs setstripe, only systems with more than 160 OSTs
4369 Description: unable to set striping with a starting offset beyond OST 160
4370 Details : llapi_create_file() incorrectly limited the starting stripe
4371 index to the maximum single-file stripe count.
4374 Frequency : LDAP users only
4376 Description: lconf did not handle in-kernel recovery with LDAP properly
4377 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4379 Severity : enhancement
4381 Description: bind OST threads to NUMA nodes to improve performance
4382 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4383 node and do their allocations there to localize memory access
4385 Severity : enhancement
4387 Description: llmount can determine client NID directly from Myrinet (GM)
4388 Details : the client NID code from gmnalnid was moved directly into
4389 llmount, removing the need to use this or specifying the
4390 client NID explicitly when mounting GM clients with zeroconf
4393 Frequency : if client is started with down MDS
4395 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4396 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4397 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4398 ptlrpc_import_delayed_req() because we didn't check if the
4399 request was interrupted, and we also didn't break out of the
4400 event loop if there was no timeout
4405 Description: data loss during non-page-aligned writes to a single file from
4406 both multiple nodes and multiple threads on one node at same time
4407 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4408 inconsistency led to false short-reads, that were cached and later
4409 used by ->prepare_write() to fill in partially written page,
4410 leading to data loss.
4413 Frequency : always, if lconf --abort_recovery used
4415 Description: lconf --abort_recovery fails with 'Operation not supported'
4416 Details : lconf was attempting to abort recovery on the MDT device and not
4419 Severity : enhancement
4421 Description: remove cleanup logs
4422 Details : replace lconf-generated cleanup logs with lustre internal
4423 cleanup routines. Eliminates the need for client-cleanup and
4426 Severity : enhancement
4428 Description: add support for EAs (user and system) on lustre filesystems
4429 Details : it is now possible to store extended attributes in the Lustre
4430 client filesystem, and with the user_xattr mount option it
4431 is possible to allow users to store EAs on their files also
4433 Severity : enhancement
4435 Description: Add possibility (config option) to show minimal available OST free
4437 Details : When compiled with --enable-mindf configure option, statfs(2)
4438 (and so, df) will return least minimal free space available from
4439 all OSTs as amount of free space on FS, instead of summary of
4440 free spaces of all OSTs.
4442 Severity : enhancement
4444 Description: do not expand extent locks acquired on OST-side
4445 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4446 by server: they are not cached anyway.
4449 Frequency : when mmap is used/binaries executed from Lustre
4451 Description: Unmmap pages before throwing them away from read cache.
4452 Details : llap_shrink cache now attempts to unmap pages before discarding
4453 them (if unmapping failed - do not discard). SLES9 kernel has
4454 extra checks that trigger if this unmapping is not done first.
4459 Description: lconf didn't resolve symlinks before checking to see whether a
4460 given mountpoint was already in use
4463 Frequency : when migrating failover services
4464 Bugzilla : 6395, 9514
4465 Description: When migrating a subset of services from a node (e.g. failback
4466 from a failover service node) the remaining services would
4467 time out and evict clients.
4468 Details : lconf --force (implied by --failover) sets the global obd_timeout
4469 to 5 seconds in order to quickly disconnect, but this caused
4470 other RPCs to time out too quickly. Do not change the global
4471 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4473 Severity : enhancement
4474 Frequency : if MDS is started with down OST
4475 Bugzilla : 9439,5706
4476 Description: Allow startup/shutdown of an MDS without depending on the
4477 availability of the OSTs.
4478 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4479 Add appropriate locking and lov-osc refcounts for safe
4480 cleaning. Add osc abort_inflight calls in case the
4481 synchronize never started.
4484 Frequency : occasional (Cray XT3 only)
4486 Description: root not authorized to access files in CRAY_PORTALS environment
4487 Details : The client process capabilities were not honoured on the MDS in
4488 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4489 been accessed by an authorized user then root was able to access
4490 the file on the local client also. The root user capabilities
4491 are now allowed on the MDS, as this environment has secure UID.
4494 Frequency : occasional
4496 Description: ldiskfs "too long searching" message happens too often
4497 Details : A debugging message (otherwise harmless) prints too often on
4498 the OST console. This has been reduced to only happen when
4499 there are fragmentation problems on the filesystem.
4504 Description: Division by zero in statfs when all OSCs are inactive
4505 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4506 lov_statfs divides by value returned by lov_get_stripecnt().
4510 Bugzilla : 9489, 3273
4511 Description: First write from each client to each OST was only 4kB in size,
4512 to initialize client writeback cache, which caused sub-optimal
4513 RPCs and poor layout on disk for the first writen file.
4514 Details : Clients now request an initial cache grant at (re)connect time
4515 and so that they can start streaming writes to the cache right
4516 away and always do full-sized RPCs if there is enough data.
4517 If the OST is rebooted the client also re-establishes its grant
4518 so that client cached writes will be honoured under the grant.
4523 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4524 Details : Now I/O RPCs go to different portal number and (presumably) fast
4525 lock requests (and glimses) and other RPCs get their own service
4526 threads pool that should be able to service those RPCs
4529 Severity : enhancement
4531 Description: Ability to exchange lustre version between client and servers and
4532 issue warnings at client side if client is too old. Also for
4533 liblustre clients there is ability to refuse connection of too old
4535 Details : New 'version' field is added to connect data structure that is
4536 filled with version info. That info is later checked by server and
4540 Frequency : rare, liblustre only.
4541 Bugzilla : 9296, 9581
4542 Description: Two simultaneous writes from liblustre at offset within same page
4543 might proceed at the same time overwriting eachother with stale
4545 Details : I/O lock withing llu_file_prwv was released too early, before data
4546 actually was hitting the wire. Extended lock-holding time until
4547 server acknowledges receiving data.
4550 Frequency : extremely rare. Never observed in practice.
4552 Description: avoid generating lustre_handle cookie of 0.
4553 Details : class_handle_hash() generates handle cookies by incrementing
4554 global counter, and can hit 0 occasionaly (this is unlikely, but
4555 not impossible, because initial value of cookie counter is
4556 selected randonly). Value of 0 is used as a sentinel meaning
4557 "unassigned handle" --- avoid it. Also coalesce two critical
4558 sections in this function into one.
4560 Severity : enhancement
4562 Description: allow liblustre clients to delegate truncate locking to OST
4563 Details : To avoid overhead of locking, liblustre client instructs OST to
4564 take extent lock in ost_punch() on client's behalf. New connection
4565 flag is added to handle backward compatibility.
4567 Severity : enhancement
4568 Bugzilla : 4928, 7341, 9758
4569 Description: allow number of OST service threads to be specified
4570 Details : a module parameter allows the number of OST service threads
4571 to be specified via "options ost ost_num_threads={N}" in the
4572 OSS's /etc/modules.conf or /etc/modprobe.conf.
4576 Bugzilla : 6146, 9635, 9895
4577 Description: servers crash with bad pointer in target_handle_connect()
4578 Details : In rare cases when a client is reconnecting it was possible that
4579 the connection request was the last reference for that export.
4580 We would temporarily drop the export reference and get a new
4581 one, but this may have been the last reference and the export
4582 was just destroyed. Get new reference before dropping old one.
4584 Severity : enhancement
4585 Frequency : if client is started with failover MDS
4587 Description: Allow multiple MDS hostnames in the mount command
4588 Details : Try to read the configuration from all specified MDS
4589 hostnames during a client mount in case the "primary"
4592 Severity : enhancement
4594 Description: Stop sending data to evicted clients as soon as possible.
4595 Details : Check if the client we are about to send or are sending data to
4596 was evicted already. (Check is done every second of waiting,
4597 for which l_wait_event interface was extended to allow checking
4598 of exit condition at specified intervals).
4601 Frequency : rare, normally only when NFS exporting is done from client
4603 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4605 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4608 Frequency : always, if srand() is called before liblustre initialization
4610 Description: Liblustre uses system PRNG disturbing its usage by user application
4611 Details : Introduce internal to lustre fast and high-quality PRNG for
4612 lustre usage and make liblustre and some other places in generic
4613 lustre code to use it.
4615 Severity : enhancement
4616 Bugzilla : 9477, 9557, 9870
4617 Description: Verify that the MDS configuration logs are updated when xml is
4618 Details : Check if the .xml configuration logs are newer than the config
4619 logs stored on the MDS and report an error if this is the case.
4620 Request --write-conf, or allow starting with --old_conf.
4622 Severity : enhancement
4624 Description: Handle symlinks in the path when checking if Lustre is mounted.
4625 Details : Resolve intermediate symlinks when checking if a client has
4626 mounted a filesystem to avoid duplicate client mounts.
4631 Description: lconf can hit an error exception but still return success.
4632 Details : The lconf command catches the Command error exception at the top
4633 level script context and will exit with the associated exit
4634 status, but doesn't ensure that this exit status is non-zero.
4639 Description: failure of ptlrpc thread startup can cause oops
4640 Details : Starting a ptlrpc service thread can fail if there are a large
4641 number of threads or the server memory is very fragmented.
4642 Handle this without oopsing.
4645 Frequency : always, only if liblustre and non-default acceptor port was used
4647 Description: liblustre cannot connect to servers with non-default acceptor port
4648 Details : tcpnal_set_default_params() was not called and was therefore
4649 ignoring the environment varaible TCPNAL_PORT, as well as other
4650 TCPNAL_ environment variables
4655 Description: two objects could be created on the same OST for a single file
4656 Details : If an OST is down, in some cases it was possible to create two
4657 objects on a single OST for a single file. No problems other
4658 than potential performance impact and spurious error messages.
4662 Bugzilla : 5681, 9562
4663 Description: Client may oops in ll_unhash_aliases
4664 Details : Client dcache may become inconsistent in race condition.
4665 In some cases "getcwd" can fail if the current directory is
4671 Description: Inode refcounting problems in NFS export code
4672 Details : link_raw functions used to call d_instantiate without obtaining
4673 extra inode reference first.
4677 Bugzilla : 9942, 9903
4678 Description: Referencing freed requests leading to crash, memleaks with NFS.
4679 Details : We used to require that call to ll_revalidate_it was always
4680 followed by ll_lookup_it. Also with revalidate_special() it is
4681 possible to call ll_revalidate_it() twice for the same dentry
4682 even if first occurence returned success. This fix changes semantic
4683 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4684 reference on a request referred from the intent.
4685 ll_intent_release() then releases such a request.
4688 Frequency : rare, normally benchmark loads only
4690 Description: unlinked inodes were kept in memory on the client
4691 Details : If a client is repeatedly creating and unlinking files it
4692 can accumulate a lot of stale inodes in the inode slab cache.
4693 If there is no other client load running this can cause the
4694 client node to run out of memory. Instead flush old inodes
4695 from client cache that have the same inode number as a new inode.
4698 Frequency : SLES9 2.6.5 kernel and long filenames only
4699 Bugzilla : 9969, 10379
4700 Description: utime reports stale NFS file handle
4701 Details : SLES9 uses out-of-dentry names in some cases, which confused
4702 the lustre dentry revalidation. Change it to always use the
4706 Frequency : rare, unless heavy write-truncate concurrency is continuous
4707 Bugzilla : 4180, 6984, 7171, 9963, 9331
4708 Description: OST becomes very slow and/or deadlocked during object unlink
4709 Details : filter_destroy() was holding onto the parent directory lock
4710 while truncating+unlinking objects. For very large objects this
4711 may block other threads for a long time and slow overall OST
4712 responsiveness. It may also be possible to get a lock ordering
4713 deadlock in this case, or run out of journal credits because of
4714 the combined truncate+unlink. Solution is to do object truncate
4715 first in one transaction without parent lock, and then do the
4716 final unlink in a new transaction with the parent lock. This
4717 reduces the lock hold time dramatically.
4720 Frequency : rare, 2.4 kernels only
4722 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4723 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4724 before unmounting their filesystem in order to avoid deadlock.
4725 The kernel_locked() function in 2.4 kernels only checks whether
4726 the kernel lock is held, not whether it is this process that is
4727 holding it as 2.6 kernels do.
4732 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4733 Details : The client ptlrpc code may be trying to reconnect to a down
4734 server before a previous connection attempt has timed out.
4735 Increase the reconnect interval to be longer than the connection
4736 timeout interval to avoid sending duplicate connections to
4740 Frequency : echo_client brw_test command
4742 Description: fix echo_client to work with OST preallocated code
4743 Details : OST preallocation code (5137) didn't take echo_client IO path
4744 into account: echo_client calls filter methods outside of any
4745 OST thread and, hence, there is no per-thread preallocated
4746 pages and buffers to use. Solution: hijack pga pages for IO. As
4747 a byproduct, this avoids unnecessary data copying.
4751 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4752 Description: Client can oops in mdc_commit_close() after open replay
4753 Details : It was possible for the MDS to return an open request with no
4754 transaction number in mds_finish_transno() if the client was
4755 evicted, but without actually returning an error. Clients
4756 would later try to replay that open and may trip an assertion
4757 Simplify the client close codepath, and always return an error
4758 from the MDS in case the open is not successful.
4761 Frequency : rare, 2.6 OSTs only
4763 Description: OST may deadlock under high load on fragmented files
4764 Details : If there was a heavy load and highly-fragmented OST filesystems
4765 it was possible to have all the OST threads deadlock waiting on
4766 allocation of biovecs, because the biovecs were not released
4767 until the entire RPC IO was completed. Instead, release biovecs
4768 as soon as they are complete to ensure forward IO progress.
4770 Severity : enhancement
4772 Description: Support for specifying external journal device at mount
4773 Details : If an OST or MDS device is formatted with an external journal
4774 device, this device major/minor is stored in the ext3 superblock
4775 and may not be valid for failover. Allow detecting and
4776 specifying the external journal at mount time.
4781 Description: Mounting an MDS with pending unlinked files may cause oops
4782 Details : target_finish_recovery() calls mds_postrecov() which returned
4783 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4784 considers this an error and immediately begins cleaning up the
4785 lov, just after starting the mds_lov process
4787 Severity : enhancement
4789 Description: Implement 'lfs df' to report actual free space on per-OST basis
4790 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4791 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4792 usage of objects; '-h' to report in human readable format.
4794 ------------------------------------------------------------------------------
4796 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4803 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4804 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4805 at startup during orphan unlinking. Instead, skip buddy update
4806 before setup, as it will be regenerated after recovery is complete.
4811 Description: If an OST is inactive, its locks might reference stale inodes.
4812 Details : lov_change_cbdata() must iterate over all namespaces, even if
4813 they are inactive to clear inode references from the lock.
4815 Severity : enhancement
4816 Frequency : occasional, if non-standard max_dirty_mb used
4818 Description: Client will block write RPCs if not enough grant
4819 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4820 then the client will block writes while waiting for another RPC
4821 to complete instead of consuming its dirty limit. With change
4822 we get improved performance when max_dirty_mb is small.
4824 Severity : enhancement
4825 Bugzilla : 3389, 6253
4826 Description: Add support for supplementary groups on the MDS.
4827 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4828 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4829 lookups for user supplementary groups into a cache.
4833 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4834 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4835 for liblustre O_EXCL is enforced by MDS.
4839 Description: readdir never returns NULL in liblustre.
4840 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4841 dentry in struct dirent.
4845 Frequency : liblustre only, depends on application IO pattern
4846 Description: liblustre clients evicted if not contacting servers
4847 Details : Don't put liblustre clients into the ping_evictor list, so
4848 they will not be evicted by the pinger ever.
4850 Severity : enhancement
4852 Description: Add ability to evict clients by NID from MDS.
4853 Details : By echoing "nid:$NID" string into
4854 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4855 $NID would be instantly evicted from this MDS and from all active
4856 OSTs connected to it.
4860 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4861 Details : lookup_it_finish() used to query file size from OSTs that was not
4866 Description: service threads change working directory to that of init
4867 Details : Starting lustre service threads may pin the working directory
4868 of the parent thread, making that filesystem busy. Threads
4869 now change to the working directory of init to avoid this.
4873 Frequency : during shutdown only
4874 Description: shutdown with a failed MDS or OST can cause unmount to hang
4875 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4880 Frequency : 2.6 only
4881 Description: chmod/chown may include an extra supplementary group
4882 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4883 supplementary group and if none is specified this is used.
4886 Bugzilla : 5479 (6816)
4888 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4889 Details : If lookup is in progress on a file that is unlinked we might try
4890 to revalidate the inode and fail in revalidate after lookup is
4891 complete and ll_file_open() enqueues the open again but
4892 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4895 Frequency : always, if lconf --abort_recovery used
4897 Description: lconf --abort_recovery fails with 'Operation not supported'
4898 Details : lconf was attempting to abort recovery on the MDT device and not
4901 ------------------------------------------------------------------------------
4903 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4908 Frequency : rare (only unsupported configurations with a node running as an
4910 Bugzilla : 6514, 5137
4911 Description: Mounting a Lustre file system on a node running as an OST could
4913 Details : OSTs now preallocates memory needed to write out data at
4914 startup, instead of when needed, to avoid having to
4915 allocate memory in possibly low memory situations.
4916 Specifically, if the file system is mounted on on OST,
4917 memory pressure could force it to try to write out data,
4918 which it needed to allocate memory to do. Due to the low
4919 memory, it would be unable to do so and the node would
4920 become unresponsive.
4922 Severity : enhancement
4924 Description: Addition of lconf --service command line option
4925 Details : lconf now accepts a '--service <arg>' option, which is
4926 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4928 Severity : enhancement
4930 Description: Failover mode is now the default for OSTs.
4931 Details : By default, OSTs will now run in failover mode. To return to
4932 the old behaviour, add '--failout' to the lmc line for OSTs.
4934 Severity : enhancement
4936 Description: Health checks are now provided for MDS and OSTs
4937 Details : Additional detailed health check information on MSD and OSTs
4938 is now provided through the procfs health_check value.
4941 Frequency : occasional, depends on IO load
4943 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4944 after numerous create/delete cycles
4945 Details : The ext3 inode allocation policy would not allocate new inodes
4946 very well on the OSTs because there are no new directories
4947 being created. Instead we look for groups with free space if
4948 the parent directories are nearly full.
4953 Description: Network or server problems during mount may cause partially
4954 mounted clients instead of returning an error.
4955 Details : The config llog parsing code may overwrite the error return
4956 code during mount error handling, returning success instead
4962 Description: MDS can fail to allocate large reply buffers
4963 Details : After long uptimes the MDS can fail to allocate large reply
4964 buffers (e.g. zconf client mount config records) due to memory
4965 fragmentation or consumption by the buffer cache. Preallocate
4966 some large reply buffers so that these replies can be sent even
4967 under memory pressure.
4971 Frequency : rare (liblustre)
4972 Description: fsx running with liblustre complained that using truncate() to
4973 extend the file doesn't work. This patch corrects that issue.
4974 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4975 ATTR_SIZE and lsm use in llu_setattr_raw.
4979 Frequency : rare, only 2.6 kernels
4980 Description: Unusual file access patterns on the MDS may result in inode
4981 data being lost in very rare circumstances.
4982 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4983 code in the 2.6 kernel caused the inode and/or EA data not to
4984 be read from disk, causing single-file corruption.
4988 Frequency : rare, only 2.6 filesystems using extents
4989 Description: Heavy concurrent write and delete load may cause data corruption.
4990 Details : It was possible under high-load situations to have an extent
4991 metadata block in the block device cache from a just-unlinked
4992 file overwrite a newly-allocated data block. We now unmap any
4993 metadata buffers that alias just-allocated data blocks.
4997 Frequency : filesystems with default stripe_count larger than 77
4998 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4999 Details : lconf specifies an inode size of 4096 bytes when the default
5000 stripe_count is larger than 77. This conflicts with the default
5001 inode density of 1 per 4096 bytes. Allocate smaller inodes in
5002 this case to avoid pinning too much memory for large EAs.
5004 ------------------------------------------------------------------------------
5006 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
5011 Frequency : rare (extremely heavy IO load with hundreds of clients)
5013 Description: Client is evicted, gets IO error writing to file
5014 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
5015 caused clients to be evicted for AST timeouts. The fixes in
5016 bug 5192 mean we no longer need to have such short AST timeouts
5017 so ldlm_timeout has been increased.
5020 Frequency : occasional during --force or --failover shutdown under load
5021 Bugzilla : 5949, 4834
5022 Description: Server oops/LBUG if stopped with --force or --failover under load
5023 Details : a collection of import/export refcount and cleanup ordering
5024 issues fixed for safer force cleanup
5027 Frequency : only filesystems larger than 120 OSTs
5028 Bugzilla : 5990, 6223
5029 Description: lfs getstripe would oops on a very large filesystem
5030 Details : lov_getconfig used kfree on vmalloc'd memory
5033 Frequency : only filesystems exporting via NFS to Solaris 10 clients
5034 Bugzilla : 6242, 6243
5035 Description: reading from files that had been truncated to a non-zero size
5036 but never opened returned no data
5037 Details : ll_file_read() reads zeros from no-object files to EOF
5042 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
5043 Details : There's little harm in aborting MDS/OSS recovery and letting it
5044 try again, so I removed the LASSERT and return an error instead.
5046 Severity : enhancement
5048 Description: New debugging infrastructure for tracking down data corruption
5049 Details : The I/O checksum code was replaced to: (a) control it at runtime,
5050 (b) cover more of the client-side code path, and (c) try to narrow
5051 down where problems occurred
5055 Bugzilla : 3819, 4364, 4397, 6313
5056 Description: Racing close and eviction MDS could cause assertion in mds_close
5057 Details : It was possible to get multiple mfd references during close and
5058 client eviction, leading to one thread referencing a freed mfd.
5060 Severity: : enhancement
5061 Bugzilla : 3262, 6359
5062 Description: Attempts to reconnect to servers are now more aggressive.
5063 Details : This builds on the enhanced upcall-less recovery that was added
5064 in 1.4.2. When trying to reconnect to servers, clients will
5065 now try each server in the failover group every 10 seconds. By
5066 default, clients would previously try one server every 25 seconds.
5071 Description: After recovery, certain operations trigger a failed
5072 assertion on a client.
5073 Details : Failing over an mds, using lconf -d --failover, while a
5074 client was doing a readdir() call would cause the client to
5075 LBUG after recovery completed and the readdir() was resent.
5077 Severity : enhancement
5079 Description: Default groups are now added by lconf
5080 Details : You can now run lconf --group <servicename> without having to
5081 manually add groups with lmc.
5084 Frequency : occasional
5086 Description: Nodes with an elan id of 0 trigger a failed assertion
5089 Frequency : always when accessing e.g. tty/console device nodes
5091 Description: tty and some other devices nodes cannot be used on lustre
5092 Details : file's private_data field is used by device data and lustre
5093 values in there got lost. New field was added to struct file to
5094 store fs-specific private data.
5097 Frequency : when exporting Lustre via NFS
5099 Description: NFSD failed occasionally when looking up a path component
5100 Details : NFSD is looking up ".." which was broken in ext3 directories
5101 that had grown large enough to become hashed.
5104 Frequency : Clusters with multiple interfaces not on the same subnet
5106 Description: Nodes will repeatedly try to reconnect to an interface which it
5107 cannot reach and report an error to the log.
5108 Details : Extra peer list entries will be created by lconf with some peers
5109 unreachable. lconf now validates the peer before adding it.
5112 Frequency : Only if a default stripe is set on the filesystem root.
5114 Description: Setting a default stripe on the filesystem root prevented the
5115 filesystem from being remounted.
5116 Details : The client was sending extra request flags in the root getattr
5117 request and did not allocate a reply buffer for the dir EA.
5120 Frequency : occasional, higher if lots of files are accessed by one client
5121 Bugzilla : 6159, 6097
5122 Description: Client trips assertion regarding lsm mismatch/magic
5123 Details : While revalidating inodes the VFS looks up inodes with ifind()
5124 and in rare cases can find an inode that is being freed.
5125 The ll_test_inode() code will free the lsm during ifind()
5126 when it finds an existing inode and then the VFS later attaches
5127 this free lsm to a new inode.
5131 Bugzilla : 6422, 7030
5132 Description: MDS deadlock between mkdir and client eviction
5133 Details : Creating a new file via mkdir or mknod (starting a transaction
5134 and getting the ns lock) can deadlock with client eviction
5135 (gets ns lock and trying to finish a synchronous transaction).
5138 Frequency : occasional
5139 Description: While starting a server, the fsfilt_ext3 module could not be
5141 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
5142 kernels. Previously, lconf would still use the ext3 name
5143 when trying to load modules. Now, it will correctly use
5144 ext3 on 2.4 and ldiskfs on 2.6.
5146 Severity : enhancement
5147 Description: The default stripe count has been changed to 1
5148 Details : The interpretation of the default stripe count (0, to lfs
5149 or lmc) has been changed to mean striping across a single
5150 OST, rather than all available. For general usage we have
5151 found a stripe count of 1 or 2 works best.
5153 Severity : enhancement
5154 Description: Add support for compiling against Cray portals.
5155 Details : Conditional compiling for some areas that are different
5159 Frequency : occasional
5160 Bugzilla : 6409, 6834
5161 Description: Creating files with an explicit stripe count may lead to
5162 a failed assertion on the MDS
5163 Details : If some OSTs are full or unavailable, creating files may
5164 trigger a failed assertion on the MDS. Now, Lustre will
5165 try to use other servers or return an error to the
5169 Frequency : occasional
5171 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
5172 caused lock timeout during readahead (since 1.4.2).
5173 Details : Processes doing readahead might match a lock that hasn't been
5174 granted yet if there are overlapping and conflicting lock
5175 requests. The readahead process waits on ungranted lock
5176 (original lock is CBPENDING), while OST waits for that process
5177 to cancel CBPENDING read lock and eventually evicts client.
5179 Severity : enhancement
5181 Description: Initial enabling of flock support for clients
5182 Details : Implements fcntl advisory locking and file status functions.
5183 This feature is provided as an optional mount flag (default
5184 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
5185 locking are implemented yet, and those that are are not guaranteed
5186 to be completely correct in production environments.
5187 mount -t lustre -o [flock|noflock] ...
5190 Frequency : occasional
5192 Description: OSTs running 2.4 kernels but with extents enabled might trip an
5193 assertion in the ext3 JBD (journaling) layer.
5194 Details : The b_committed_data struct is protected by the big kernel lock
5195 in 2.4 kernels, serializing journal_commit_transaction() and
5196 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
5197 access to this struct. In 2.6 kernels there is finer grained
5198 locking to improve SMP performance of the JBD layer.
5202 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
5207 Description: merge and cleanup kernel patches.
5210 -----------------------------------------------------------------------------
5212 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
5214 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
5215 versions of Lustre. Please update all servers and clients to
5216 version 1.4.2 or later at the same time. You must also run
5217 "lconf --write-conf {config}.xml" on the MDS while it is stopped
5218 to update the configuration logs.
5220 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
5221 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
5222 - swab llog records of type '0' so we get proper header size/idx (5861)
5223 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
5224 - fix rename of one directory over another leaking an inode (5953)
5225 - avoid SetPageDirty on 2.6 (5981)
5226 - don't re-add just-being-destroyed locks to the waiting list (5653)
5227 - when creating new directories, inherit the parent's custom
5228 striping settings if present parent (3048)
5229 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
5230 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
5231 - don't hold client locks on temporary worklist from l_lru (5666)
5232 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
5233 - automatically evict dead clients (5921)
5234 - Update file size properly in create+truncate+fstat case (6196)
5235 - Do not unhash mountpoint dentries, do not allow removal of
5237 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
5238 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
5239 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
5240 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
5241 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
5242 - Correctly handle reads of files with no objects (6243)
5243 - lctl recover will also mark a device active if deactivate used (5933)
5245 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
5246 - allow --write-conf on an MDS with different nettype than client (5619)
5247 - don't write config llogs to MDS for mounts not from that MDS (5617)
5248 - lconf should create multiple TCP connections from a client (5201)
5249 - init scripts are now turned off by default; run chkconfig --on
5250 lustre and chkconfig --on lustrefs to use them
5251 - upcalls are no longer needed for clients to recover to failover
5253 - add --abort-recovery option to lconf to abort recovery on device
5255 - add support for an arbitrary number of OSTs (3026)
5256 - Quota support protocol changes.
5257 - forward compatibility changes to wire structs (6007)
5258 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
5259 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
5260 - improved "lustre status" script
5261 - initialize blocksize for non-regular files (6062)
5262 - added --disable-server and --disable-client configure options (5782)
5263 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
5264 - Vanilla 2.4.29 support
5265 - increase maximum number of obd devices to 520 (6242)
5266 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
5267 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
5268 SLES 9 2.6.5 kernels
5269 - limit stripes per file to 160 (the maximum EA size) (6093)
5271 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5274 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
5275 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
5276 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
5277 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
5278 - small ext3 extents cleanups and fixes (5733)
5279 - improved mballoc code, several small races and bugs fixed (5733, 5638)
5280 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
5281 - avoid needless client->OST connect, fix handle mismatch (5317)
5282 - fix DLM error path that led to out-of-sync client, long delays (5779)
5283 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
5284 - fix several locking issues related to i_size (5492,5624,5654,5672)
5285 - don't move pending lock onto export if it is already evicted (5683)
5286 - fix kernel oops when creating .foo in unlinked directory (5548)
5287 - fix deadlock in obdfilter statistics vs. object create (5811)
5288 - use time_{before,after} to avoid timer jiffies wrap (5882)
5289 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
5290 - Do not leak request if server was not able to process it (5154)
5291 - If mds_open unable to find parent dir, make that negative lookup(5154)
5292 - don't create new directories with extent-mapping (5909, 5936)
5294 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
5295 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
5296 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
5297 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
5298 - support for RHEL4 kernel on i686 (5773)
5299 - provide error messages when incompatible logs are encountered (5898)
5301 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
5302 * version 1.4.0.10 (1.4.1 release candidate 1)
5304 - don't keep a lock reference when lock is not granted (4238)
5305 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5306 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
5307 - drop import inflight refcount on signal_completed_replay error (5255)
5308 - unlock page after async write error during send (3677)
5309 - handle missing objects in filter_preprw_read properly (5265)
5310 - no transno return for symlink open, don't save no-trasno open (3440)
5311 - don't try to complete elan receive that already failed (4012)
5312 - free RPC server reply state on error (5406)
5313 - clean up thread from ptlrpc_start_thread() on error (5160)
5314 - readahead could read extra page into cache that wasn't ejected (5388)
5315 - prevent races in class_attach/setup/cleanup/detach (5260)
5316 - don't dereference de->d_inode after l_dput of de (5458)
5317 - use "int" for stripe value returned from lock_to_stripe (5544)
5318 - mballoc allocation and error-checking fixes in 2.6 (5504)
5319 - block device patches to fix I/O request sizes in 2.6 (5482)
5320 - look up hostnames for IB nals (5602)
5321 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
5322 - don't start multiple acceptors for the same port (5277)
5323 - fix incorrect LASSERT in mds_getattr_name (5635)
5324 - export a proc file for general "ping" checking (5628)
5325 - fix "lfs check" to not block when the MDS is down (5628)
5327 - service request history (4965)
5328 - put {ll,lov,osc}_async_page structs in a single slab (4699)
5329 - create an "evict_client" /proc entry on OSTs, like the MDS has
5330 - fix mount usage message, return errors per mount(8) (5168)
5331 - change grep [] to grep "[]" in tests so they work in more UMLs
5332 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
5333 - remove ancient LOV_MAGIC_V0 EA support (5047)
5334 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
5335 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
5336 - allow readahead allocations to fail when low on memory (5383)
5337 - mmap locking landed again, after considerable improvement (2828)
5338 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
5340 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
5343 - send OST transaction number in read/write reply to free req (4966)
5344 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5345 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5346 - the watchdog thread now runs as interruptible (5246)
5347 - flock/lockf fixes (but it's still disabled, pending 5135)
5348 - don't use EXT3 constants in llite code (5094)
5349 - memory shortage at startup could cause assertion (5176)
5351 - reorganization of lov code
5352 - single portals codebase
5354 - add extents/mballoc support (5025)
5355 - direct I/O reads in the obdfilter (4048)
5356 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5358 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5361 - send OST transaction number in read/write reply to free req (4966)
5362 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5363 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5364 - the watchdog thread now runs as interruptible (5246)
5365 - handle missing objects in filter_preprw_read properly (5265)
5366 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5367 - drop import inflight refcount on signal_completed_replay error (5255)
5368 - unlock page after async write error during send (3677)
5369 - return original error code on reconstructed replies (3761)
5370 - no transno return for symlink open, don't save no-trasno open (3440)
5372 - add pid to ldlm debugging output (4922)
5373 - bump the watchdog timeouts -- we can't handle 30sec yet
5374 - extra debugging for orphan dentry/inode bug (5259)
5376 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5379 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5380 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5381 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5382 - fix NULL dereference in /proc/sys/portals/routes (4827)
5383 - allow failed mdc_close() operations to be interrupted (4561)
5384 - stop precreate on OST before MDS would time out on it (4778)
5385 - don't send partial-page writes before EOF from client (4410)
5386 - discard client grant for sub-page writes on large-page clients (4520)
5387 - don't free dentries not owned by NFS code, check generation (4806)
5388 - fix lsm leak if mds_create_objects() fails (4801)
5389 - limit debug_daemon file size, always print CERROR messages (4789)
5390 - use transno after validating reply (3892)
5391 - process timed out requests if import state changes (3754)
5392 - update mtime on OST during writes, return in glimpse (4829)
5393 - add mkfsoptions to LDAP (4679)
5394 - use ->max_readahead method instead of zapping global ra (5039)
5395 - don't interrupt __l_wait_event() during strace
5397 - add software watchdogs to catch hung threads quickly (4941)
5398 - make lustrefs init script start after nfs is mounted
5399 - fix CWARN/ERROR duplication (4930)
5400 - return async write errors to application if possible (2248)
5401 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5402 - print NAL number in %x format (4645)
5403 - update barely-supported suse-2.4.21-171 series (4842)
5404 - support for sles 9 %post scripts
5405 - support for building 2.6 kernel-source packages
5406 - support for sles km_* packages
5408 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5411 - ignore -ENOENT errors in osc_destroy (3639)
5412 - notify osc create thread that OSC is being cleaned up (4600)
5413 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5414 - reconstruct ost_handle() like mds_handle() (4657)
5415 - create a new thread to do import eviction to avoid deadlock (3969)
5416 - let lconf resolve symlinked-to devices (4629)
5417 - don't unlink "objects" from directory with default EA (4554)
5418 - hold socknal file ref over connect in case target is down (4394)
5419 - allow more than 32000 subdirectories in a single directory (3244)
5420 - fix blocks count for O_DIRECT writes (3751)
5421 - OST returns ENOSPC from object create when no space left (4539)
5422 - don't send truncate RPC if file size isn't changing (4410)
5423 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5424 - bind to privileged port in socknal and tcpnal (3689)
5426 - rate limit CERROR/CWARN console message to avoid overload (4519)
5427 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5428 - basic mmap support (3918)
5429 - kernel patch series update from b1_4 (4711)
5431 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5434 - avoid crash during MDS cleanup with OST shut down (2775)
5435 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5436 - don't use bad inodes on the MDS (3744)
5437 - dynamic object preallocation to improve recovery speed (4236)
5438 - don't hold spinlock over lock dumping or change debug flags (4401)
5439 - don't zero obd_dev when it is force cleaned (3651)
5440 - print grants to console if they go negative (4431)
5441 - "lctl deactivate" will stop automatic recovery attempts (3406)
5442 - look for existing locks in ldlm_handle_enqueue() (3764)
5443 - don't resolve lock handle twice in recovery avoiding race (4401)
5444 - revalidate should check working dir is a directory (4134)
5446 - don't always mark "slow" obdfilter messages as errors (4418)
5448 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5451 - don't close LustreDB during write_conf until it is done (3860)
5452 - fix typo in lconf for_each_profile (3821)
5453 - allow dumping logs from multiple threads at one time (3820)
5454 - don't allow multiple threads in OSC recovery (3812)
5455 - fix debug_size parameters (3864)
5456 - fix mds_postrecov to initialize import for llog ctxt (3121)
5457 - replace config semaphore with spinlock (3306)
5458 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5459 - don't allow enqueue to complete on a destroyed export (3822)
5460 - down write_lock before checking llog header bitmap (3825)
5461 - recover from lock replay timeout (3764)
5462 - up llog sem before sending rpc (3652)
5463 - reduce ns lock hold times when setting kms (3267)
5464 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5465 - fix NULL deref and obd_dev leak on setup error (3312)
5466 - replace some LBUG about llog ops with error handling (3841)
5467 - don't match INVALID dentries from d_lookup and spin (3784)
5468 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5469 - fix invalid assertion in ptlrpc_set_wait (3880)
5471 - add libwrap support for the TCP acceptor (3996)
5472 - add /proc/sys/portals/routes for non-root route listing (3994)
5473 - allow setting MDS UUID in .xml (2580)
5474 - print the stack of a process that LBUGs (4228)
5476 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5479 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5480 - make sure to unset replay flag from failed open requests (3440)
5481 - if default stripe count is 0, use OST count for inode size (3636)
5482 - update parent mtime/ctime on client for create/unlink (2611)
5483 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5484 - free recovery state on server during a forced cleanup (3571)
5485 - unregister_reply for resent reqs (3063)
5486 - loop back devices mounting and status check on 2.6 (3563)
5487 - fix resource-creation race that can provoke i_size == 0 (3513)
5488 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5489 - more debugging for page-accounting assertion (3746)
5490 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5491 - avoid deadlock after precreation failure (3758)
5492 - fix race and lock order deadlock in orphan handling (3450, 3750)
5493 - add validity checks when grabbing inodes from l_ast_data (3599)
5495 - add /proc/.../recovery_status to obdfilter (3428)
5496 - lightweight CDEBUG infrastructure, debug daemon (3668)
5497 - change default OSC RPC parameters to be better on small clusters
5498 - turn off OST read cache for files smaller than 32MB
5499 - install man pages and include them in rpms (3100)
5500 - add new init script for (un)mounting lustre filesystems (2593)
5501 - run chkconfig in %post for init scripts (3701)
5502 - drop scimac NAL (unmaintained)
5504 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5507 - clean kiobufs before and after use (3485)
5508 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5509 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5510 - change init script to fail more clearly if not run as root (1528)
5511 - allow clients to reconnect during replay (1742)
5512 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5513 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5514 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5515 - fix cond_resched (3554)
5516 - only evict unfinished clients after recovery (3515)
5517 - allow bulk resend, prevent data loss (3570)
5518 - dynamic ptlrpc request buffer allocation (2102)
5519 - don't allow unlinking open directory if it isn't empty (2904)
5520 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5521 - remove extraneous obd dereference causing LASSERT failure (3334)
5522 - don't use get_cycles() when creating temp. files on the mds (3156)
5523 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5524 - handle EEXIST for set-stripe, set proper directory name (3336)
5526 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5527 - fix ksocknal_fmb_callback() error messages (2918)
5529 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5532 - don't copy lvb into (possibly NULL) reply on error (2983)
5533 - don't deref dentry after dput, don't free lvb on error (2922)
5534 - use the kms to determine writeback rpc length (2947)
5535 - increment oti_logcookies when osc is inactive (2948)
5536 - update client's i_blocks count via lvb messages (2543)
5537 - handle intent open/close of special files properly (1557)
5538 - mount MDS with errors=remount-ro, like obdfilter (2009)
5539 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5540 - don't use cancelling-locks' kms values (2947)
5541 - use highest lock extent for kms, not last one (2925)
5542 - don't dereference ERR_PTR() dentry in error handling path (3107)
5543 - fix thread race in portals_debug_dumplog() (3122)
5544 - create lprocfs device entries at setup instead of at attach (1519)
5545 - common AST error handler, don't evict client on completion race (3145)
5546 - zero nameidata in detach_mnt in 2.6 (3118)
5547 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5548 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5549 - initialize RPC timeout timer earlier for 2.6 (3219)
5550 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5551 - print nal/nid for unknown nid (3258)
5552 - additional checks for oscc recovery before doing precreate (3284)
5553 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5554 - don't crash in mdc_close for bad permissions on open (3285)
5555 - zero i_rdev for non-device files (3147)
5556 - clear page->private before handing to FS, better assertion (3119)
5557 - tune the read pipeline (3236)
5558 - fix incorrect decref of invalidated dentry (2350)
5559 - provide read-ahead stats and refine rpc in flight stats (3328)
5560 - don't hold journal transaction open across create RPC (3313)
5561 - update atime on MDS at close time (3265)
5562 - close LDAP connection when recovering to avoid server load (3315)
5563 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5564 - don't leak open file on MDS after open resend (3325)
5565 - serialize filter_precreate and filter_destroy_precreated (3329)
5566 - loop device shouldn't call sync_dev() for nul device (3092)
5567 - clear page cache after eviction (2766)
5568 - resynchronize MDS->OST in background (2824)
5569 - refuse to mount the same filesystem twice on same mountpoint (3394)
5570 - allow llmount to create routes for mounting behind routers (3320)
5571 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5572 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5573 - fix rare problem with rename on htree directories (3417)
5575 - allow default OST striping configuration per directory (1414)
5576 - fix compilation for qswnal for 2.6 kernels (3125)
5577 - increase maximum number of MDS request buffers for large systems
5578 - change liblustreapi to be useful for external progs like lfsck (3098)
5579 - increase local configuration timeout for slow disks (3353)
5580 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5582 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5585 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5586 - don't overwrite extent policy data in reply if lock was blocked (2901)
5587 - drop filter export grants atomically with removal from device (2663)
5588 - del obd_self_export from work_list in class_disconnect_exports (2908)
5589 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5590 - swab reply message in mdc_close, other PPC fixes (2464)
5591 - fix destroying of named logs (2325)
5592 - overwrite old logs when running lconf --write_conf (2264)
5593 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5594 - fix race in target_handle_connect (2898)
5595 - mds_reint_create() should take same inode create lock (2926)
5596 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5597 - don't close files for self_export to avoid uninitialized obd (2936)
5598 - allow MDS with the same name as client node (2939)
5599 - hold dentry reference for closed log files for unlink (2325)
5600 - reserve space for all logs during transactions (2059)
5601 - don't evict page beyond end of stripe extent (2925)
5602 - don't oops on a deleted current working directory (2399)
5603 - handle hard links to targets without a parent properly (2517)
5604 - don't dereference NULL lock when racing during eviction (2867)
5605 - don't grow lock extents when lots of conflicting locks (2919)
5607 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5610 - account for cache space usage on clients to avoid data loss (974)
5611 - lfsck support in lustre kernel code (2349)
5612 - reduce journal credits needed for BRW writes (2370)
5613 - orphan handling to avoid losing space on client/server crashes
5614 - ptlrpcd can be blocked, stopping ALL progress (2477)
5615 - use lock value blocks to assist in proper KMS, faster stat (1021)
5616 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5617 - recovery for initial connections (2355)
5618 - fixes for mds_cleanup_orphans (1934)
5619 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5620 - block all file creations until orphan recovery completes (1901)
5621 - client remove rq_connection from request struct (2423)
5622 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5623 - recovery timer race (2670)
5624 - mdc_close recovey bug (2532)
5625 - ptlrpc cleanup bug (2710)
5626 - mds timeout on local locks (2588)
5627 - namespace lock held during RPCs (2431)
5628 - handle interrupted sync write properly (2503)
5629 - don't try to handle a message that hasn't been replied to (2699)
5630 - client assert failure during cleanup after abort recovery (2701)
5631 - leak mdc device after failed mount (2712)
5632 - ptlrpc_check_set allows timedout requests to complete (2714)
5633 - wait for inflight reqs when ptlrpcd finishes (2710)
5634 - make sure unregistered services are removed from the srv_list
5635 - reset bulk XID's when resending them (caught by 1138 test)
5636 - unregister_bulk after timeout
5637 - fix lconf error (2694)
5638 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5639 - readahead locks pages, leaves pending causing memory pressure (2673)
5640 - increase OST request buffers to 4096 on large machines (2729)
5641 - fix up permission of existing directories in simple_mkdir (2661)
5642 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5643 - don't assign transno to errored transactions (2742)
5644 - don't delete objects on OST if given a bogus objid from MDS (2751)
5645 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5646 - if rq_no_resend, then timeout request after recovery (2432)
5647 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5648 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5649 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5650 - MDS recovery shouldn't delete objects that it has given out (2730)
5651 - if enqueue arrives after completion, don't clobber LVB (2819)
5652 - don't unlock pages twice when trigger_group_io returns error (2814)
5653 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5654 - don't write pages to disk if there was an error (1450)
5655 - don't ping imports that have recovery disabled (2676)
5656 - take buffered bytes into account when balancing socknal conn (2817)
5657 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5658 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5659 - remove little-endian swabbing of llog records (1987)
5660 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5661 - retry reposting request buffers if they fail (1191)
5662 - grow extent at grant time to avoid granting a revoked lock (2809)
5663 - lock revoke doesn't evict page if covered by a second lock (2765)
5664 - disable VM readahead to avoid reading outside lock extents (2805)
5666 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5667 - updated kernel patches for hp-2.4.20 kernel (2681)
5669 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5672 - fix truncated write corruption (2366)
5673 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5675 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5676 - ensure len(uuid) < 37 in lmc (1171)
5677 - fix ia64 OOPS in llog_test (2255)
5678 - zero end of page at obdfilter for partial page writes (2648)
5679 - don't leave stale dentries around after renames (bug 2428)
5680 - fix timeouts when evicting a client with a single lock held (2642)
5681 - set deadline for the initial HELLO message to drain (2634)
5682 - print out dotted-quad IP addresses in the socknal (2302)
5684 - additional debugging for MDS client eviction problem (2443)
5685 - fix mkfsoptions support for osts (2603, 2604)
5687 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5690 - add series for the vanilla 2.6.0 kernel
5691 - add series for the vanilla 2.4.24 kernel
5692 - add series for a cray x86/64 UL kernel drop
5693 - fix xattr patches for the vanilla 2.4.19 series
5695 - generate true UUIDs in lmc (1171)
5696 - have portals stack dumping break in UML (2466)
5697 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5698 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5699 - wake callback waiting threads on client eviction (2460)
5700 - Add --ptldebug and --subsystem to lmc (1719)
5701 - update assertion to allow safe interrupt allocation
5702 - set rq_no_resend for cancel requests (2432)
5703 - recalculate ptlrpcd timeout after resend (2494)
5704 - call vfs_rmdir when removing pending directories (2368)
5705 - fix renaming a file to itself (2429)
5706 - lmc creates a default one-stripe lov (2454)
5707 - expand procfs space to handle large clusters (2326)
5708 - increase UML stack to avoid overflow
5709 - update lconf's list of debug and subsystem masks
5710 - fix lfs find --obd (2510)
5711 - /proc tunable for disabling filter read caching (2591)
5712 - stop rpm packages from altering slapd.conf (2301)
5713 - disable nagle in the socknal under 0conf (2578)
5714 - choose mds inode size based on stripe count (2572)
5715 - fix kernel-source rpm problems (2516)
5717 - add --disable-doc to avoid pdf generation (2421)
5718 - update documentation, tests, type-os, comments
5719 - avoid format warnings on ia64
5720 - remove the TOE NAL
5721 - tiny code cleanups by removing unused fields
5723 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5726 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5727 - remove the most blatant lies from BUILDING file (2371)
5728 - change default debug level to reasonable production setting
5729 - reduce client side cache size to reduce cache flush time
5730 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5731 - make TCP zerocopy and pinger support enabled by default (2476)
5732 - sync writes completed after process exits caused crashes (2319)
5733 - maintain correct mount count on the MDS (2356)
5734 - backout 1557, because 2316 wasn't really fixed
5735 - better file I/O statistics gathering in /proc
5736 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5737 - another kernel patch to fix zero-copy TCP function export
5738 - don't take duplicate lock when processing re-sent getattr (2420)
5739 - lctl uses obd_self_export instead of creating new conn (2353)
5740 - MDS/OST recovery case which requires object creation asserted (2425)
5741 - move lfs from /usr/sbin to /usr/bin in packages
5742 - fix race between mds_client_add and mds_client_free (2417)
5743 - use kmalloc instead of slabs in portals (2430)
5744 - don't create duplicate records when a failover MDS is present (2442)
5745 - remove unnecessary mount age check (2332)
5746 - don't remove directory inodes from locks prematurely (2451)
5747 - don't break if MDS service name is the same as hostname (2103)
5748 - fix races in client write RPC generation when cache full (2482)
5750 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5753 - remove now-unused request->rq_obd (278)
5754 - if an allocation fails, print out how much memory we've used (1933)
5755 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5756 - add the "configurable stack size" patch to most series files (1256)
5757 - ability to write large log records, for 100+ OST configs (2306)
5758 - fix NULL deref when filter_prep fails (2314)
5759 - fix operator precedence error in filter_sync
5760 - dynamic allocation of socknal TX descriptors (2315)
5761 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5762 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5763 - ensure that CWARN messages in Portals always get to the syslog
5764 - __init/__exit are not for prototype decls (ldlm_init/exit)
5765 - x86-64 compile warning fixes
5766 - fix gateway LMC keyword conflict (2318)
5767 - fix MDS lock inversions in getattr/reint paths (1844)
5768 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5769 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5770 - DLM race condition prevented some lock evictions (2328)
5771 - ENOMEM detection and retry on socknal sends (2230)
5772 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5773 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5774 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5775 - better error messages when a client is rejected during recovery (1505)
5776 - avoid cancelling locks which were never granted, after failure (2330)
5777 - fix i_sem/journal inversion in mds_client_add (2333)
5778 - fix truncate/getattr lock cycle deadlock (2334)
5779 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5780 - fix two rare exit paths which could leak an l_lock() ref (2321)
5781 - fencepost error in MDS/OST orphan recovery (2226)
5782 - make log record alignment 8 bytes (1988)
5783 - lstripe now fails when requested offset > ost_count (2237)
5784 - ensure that all kernel series have a complete list.h (1607)
5785 - fix crashes in special-file operations (2316)
5786 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5788 - allow configurable automake binary, for testing new versions
5789 - small update to the lfs documentation
5791 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5793 * fix negative export reference count in fsfilt_sync (2312)
5795 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5796 * release candidate 0.9.1
5798 - orphans are moved into the PENDING directory for possible recovery
5799 - replayed opens now open by fid for orphan/rename safety (1042)
5800 - last close of an orphan inode generates a transno (683)
5801 - chdir() and mount() now pin the directory entry (1020)
5802 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5803 - discard very old requests without processing them (1502)
5804 - remove some common, well-understood CERRORs (1505)
5805 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5806 - clear "grant" flags in OST replies until OST grant code lands (1644)
5807 - fix read performance by not clobbering i_blksize on client (1598)
5808 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5809 - make LRU size a /proc tunable, clears locks when reduced (707)
5810 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5811 - server locks take references on exports now (1558)
5812 - build fixes for 2.4.20-rh trees (1663)
5813 - return an error from lov_create if all OSCs are inactive (1751)
5814 - fix import levels when a reconnect happens without a timeout (1597)
5815 - exit early from mds_open if we get a lookup error (1749)
5816 - partial page read at EOF wouldn't wait for disk before sending (1642)
5817 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5818 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5819 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5820 - write/truncate lock inversion (1639)
5821 - Don't auto-load obdclass, portals modules during cleanup (1495)
5822 - fix timestamps from jumping to "now" (1763)
5823 - extra journal assertions (1648)
5824 - add an extra multiunlink test (1771)
5825 - fix read_record/write_record API (1776)
5826 - fix leak of offset_extent, possible incorrect i_size later (1772)
5827 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5828 - Debugging for the kqswnal_get_idle_tx problems (1820)
5829 - Allow recovery to be attempted multiple times (1536)
5830 - Write out MDS last_rcvd file after it is first created (1600)
5831 - Fix tx_descriptor leak in failed transmit situations (1827)
5832 - ext3 journaling fixes for assertion failure after IO error (1871)
5833 - class_export_put() on freed export after completion AST error (1896)
5834 - Fix revalidate looping in VFS (1322)
5835 - Don't access a freed export during MDS_REINT timeout (1521)
5836 - Add open-unlink recovery support on the MDS (1673,1764)
5837 - Return an error if no MDS data was read from last_rcvd (1946)
5838 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5839 - Don't setup a disk device that doesn't match exported UUID (317)
5840 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5841 - avoid committing NULL handle in force close
5842 - local.sh is now a one-stripe LOV configuration
5843 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5844 - fix ext3 htree duplicate directory entry corruption (1516)
5845 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5846 - update to new LOV EA format (2097)
5847 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5849 2003-06-15 Phil Schwan <phil@clusterfs.com>
5852 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5853 - per-import/export recovery handling (958, 931, 959)
5854 - multiple last-rcvd slots, for serving multiple FSes (949)
5855 - connections are again shared between multiple imp/exports (963, 964)
5856 - "umount -f" would hang if any requests needed to be sent (393, 978)
5857 - avoid pinning large req buffer by copying for queued messages (989)
5858 - add "uuid" to "lctl device" command to help upcalls (991)
5859 - "open" RPCs with transnos would confuse recovery counters (1037)
5860 - do proper endian conversion of all wire messages (288, 340, 891)
5861 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5862 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5863 - fix ptlrpc_connection leak in target_handle_connect (1174)
5864 - fix import refcounting bug in OST and MDS cleanup (1134)
5865 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5866 - fix very unlikely obd_types race condition (501)
5867 - remove osc_open hack for echo_client (1187)
5868 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5869 - a failure in read_inode2 leads to deadlock (1139)
5870 - cancel ack-locks as soon as transaction is committed (1072)
5871 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5872 - make sure to commitrw after any preprw to avoid deadlock (1162)
5873 - failing to execute a file in a lustre FS would lock inode (1203)
5874 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5875 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5876 - obd_uuid2tgt would walk off the end of the list (1255)
5877 - on IA64 the osc would give portals incorrect bulk size (1258)
5878 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5879 - fix lock inversion caused by new llite matching code (1282)
5880 - limit the number of dirty pages on a client to 10MB (1286)
5881 - timed out locks were not being corrected cancelled (1289)
5882 - fix O_DIRECT above 4GB on IA-32 (1292)
5883 * major user-visible changes
5884 - fail out/fail over policy now controlled by the upcall (993)
5886 - add OBD_PING to check server availability and failure (954)
5887 - lustre messages are now sent in sending host order (288, 340, 891)
5888 - add eadatalen to MDS getattr reply (340)
5889 - OST read replies may contain second buffer, with per-page status (593)
5891 2003-03-11 Phil Schwan <phil@clusterfs.com>
5894 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5895 - failed open()s could cause deadlock; fixed (867, 869)
5896 - stop cancelling OST locks when files are closed (481)
5897 - overlapping XID spaces caused network corruption (851, 853)
5898 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5899 - setattr_raw added, to do single-RPC, server-side setattrs
5900 - lmc/lconf syntax change for OST UUIDs
5901 - fix crashy race condition between ptlrpc_free_req and osc_close
5902 - don't use request in mdc_enqueue if we hit a timeout (889)
5903 - don't set the inode i_size for regular files from the MDS (896)
5904 - handle out of order completion AST (842)
5905 - don't LBUG if a lock request times out after receiving AST (913)
5906 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5907 - if a bad lock AST arrives, send an error instead of dropping entirely
5908 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5909 - fix leak in bulk IO when only partially completed (899, 900, 926)
5910 - fix O_DIRECT for ia64 (55)
5911 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5912 - C-z after timeout could hang a process forever; fixed (977)
5914 - client-side I/O cache (678, 924, 929, 941, 970)
5916 - READPAGE and SETATTRs which don't take server-side locks get
5919 2003-02-11 Phil Schwan <phil@clusterfs.com>
5922 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5923 - Fix fsfilt_extN_readpage() to read a full page of directory
5924 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5925 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5926 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5927 - Most intent operations were removed, in favour of a new RPC mode
5928 that does a single RPC to the server and bypasses most of the VFS
5929 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5930 - Aggressively cancel local locks on DLM servers
5931 - mds_reint_unlink sends EA to the client if it's the last nlink.
5932 client uses that EA to unlink OST objects.
5933 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5934 - recursive symlinks were fixed (439)
5935 - fixed NULL deref in DEBUG_REQ
5936 - filter_update_lastobjid no longer calls sync, which annoyed extN
5937 - fixed multi-client small-writes to a single file problem (445)
5938 - fixed mtime updates during file writes (607)
5939 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5940 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5941 - fixed Linux OST size reporting problem (444, 656)
5942 - OST now updates object mtime with writes or setattr (607, 619)
5943 - client verifies file size before zeroing page past EOF (445)
5944 - OST now writes last allocated objid to disk with allocation (108)
5945 - LOV on echo now works (409)
5947 - mds_reint_unlink sends a new buffer, with the EA included. this
5948 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5949 set if a regular file was being unlinked, and it was the last link
5950 - use PtlGet from the target for bulk writes (315)
5951 - OST now updates object mtime with writes or setattr (607, 619)
5952 - LDLM now has a grant-time callback to revalidate locked items, if
5954 - Many MDS operations were reorganized to combat race conditions
5956 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5957 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5959 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5962 - Fully reactivate OST imports after reconnection (512, others)
5963 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5964 - More graceful error handling for truncating on dead OST (515)
5965 - Don't error out unless we're actually accessing dead stripes (474)
5966 - Fix garbage sizes when stripes are missing (410)
5967 - LRU counters were broken, causing constant lock purge (433, 432)
5968 - garbage on read from stripes with failed OSTs (441)
5969 - mark OSCs as active before reconnecting during recovery (438)
5970 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5971 - lfind did not preserve OST order in output (443)
5972 - symlinks cause hung clients, incorrect data (439)
5973 - stop dereferencing request after dropping refcount (457)
5974 - don't LASSERT(spin_is_locked) on non-SMP (455)
5975 - fixes for many rename() bugs
5976 - fstat didn't correctly synchronize attributes (399)
5977 - server must handle lock cancellation during blocking AST prep (487)
5978 - bulk descriptors were free()d too soon (511)
5979 - fix paths in lconf, which would load incorrect modules (451, 507)
5980 - fix confusing lconf 'host not found' error message (386)
5981 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5982 - fix race condition in mdc_blocking_ast() for inode access (526)
5983 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5984 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5985 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5986 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5987 - fix lstripe in several ways (kernel side) (527)
5988 - fix request leak in ldlm_cli_enqueue (262)
5989 - incorrect OSC was marked inactive after OST failure
5990 - call mds_fs_cleanup before unmounting filesystem (524)
5991 - fix races between taking ns_lock and ldlm_lock_change_resource
5992 - fix races updating LOV export open file list
5993 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5994 - fix recovery NULL deref in ldlm_cli_cancel_unused
5995 - fix some DLM races by using new hash table for lock handles (419)
5996 - permit the client to specify desired inodes, at replay
5997 - duplicate requests when we queue them for replay reintegration
5998 - fix last_rcvd offset calculation
5999 - sync after each recovered transaction, so we always make progress
6000 - never, not always, ERESTART requests without transnos
6001 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
6003 - skip replay if the MDS says that the client is already connected
6004 - don't check for a recovery-enabled export to match lctl's UUID
6005 - don't INC_USE_COUNT for phantom exports
6006 - don't crash when cleaning up phantom exports (567)
6007 - don't double-finish or set replay data for errored mdc_open requests
6008 - abort requests when they time out, so we don't get old replies
6009 - send/receive replies for AST messages again
6010 - if the client says that it doesn't have the lock, cancel it on the
6012 - if we timeout during I/O, don't try to cancel an in-use lock; instead
6013 mark it as destroyed, it will all work out when decref is called
6014 - fix module use counts (22, 581)
6016 - ASTs now expect a reply (server cancels lock on error reply)
6018 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
6021 - fix many simultaneous client startup (392)
6022 - fix dentry->d_it clobbering
6023 - credentials weren't being shipped for readdir/getattr operations
6024 - remove invalid assertions triggered during some concurrent MD
6026 - proper Lustre versions added (336, 389)
6027 - fix memory leak for create error case (398)
6028 - fix LOV locking bug that would get cli/srv out of sync
6029 - fix echo client over LOV (409)
6030 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
6031 - fix double-O_EXCL intent crash (424)
6032 - avoid sending multiple lock CANCELs (352)
6034 - MDS can do multi-client recovery (modulo bugs in new code)
6036 - many updates, edits, cleanups
6038 2002-11-18 Phil Schwan <phil@clusterfs.com>
6041 - fix null d_it dereference (346)
6042 - fix full OST/dbench hang (333)
6043 - fix permission problem with file removal (286)
6044 - fix removal of OSCs from LOV when they fail
6045 - fix NULL deref during bulk timeout (214)
6046 - fix problems related to multiple filesystems on one MDS (241)
6047 - fixed serious subtle metadata locking bugs
6048 - free locks on clients when inodes are removed due to memory
6050 - fix inode pointer in lock data (285)
6051 - partial support for multiple MDS on a single host (241)
6052 - data locks weren't cancelled at clear_inode time (290, 311)
6053 - intent locks could lead to unbounded lock growth (205)
6054 - added a maximum lock count, an LRU list, and a flusher
6055 - fix multiple rename (365)
6056 - properly abstracted the echo client
6057 - OSC locked 1 byte too many; fixed
6058 - rewrote brw callback code:
6059 - fixed recovery bugs related to LOVs (306)
6060 - fixed too-many-pages-in-one-write crash (191)
6061 - fixed (again) crash in sync_io_timeout (214)
6062 - probably fixed callback-related race (385)
6064 - Add capability to MDS protocol
6065 - LDLM cancellations and callbacks on different portals
6067 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
6070 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
6071 - defer open object destruction to close time (601981, 138)
6072 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
6073 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
6074 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
6075 - don't call obd_setattr in ll_file_release for destroyed objects
6076 * protocol change to lustre_msg: move |version| and add |flags|
6077 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
6078 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
6079 * added replay of create, unlink, link and rename operations during
6080 MDS failover; recovery should be much more robust now
6081 * remove failed OSCs from LOVs (only lov_create uses this so far)
6082 * the lustre-HOWTO was brought (more) up to date (582544)
6084 2002-10-23 Phil Schwan <phil@clusterfs.com>
6087 - in-use dentries weren't being reused properly (617851)
6088 - prevent multiple LDLM setup (599178)
6089 - fix LOV size calculations for truncate (617853)
6090 - fix client handling of MDS intent errors (POSIX)
6091 - fix permission bug in lovstripe.c test (624321)
6092 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
6093 - truncate past end of file could corrupt data
6094 - proper cleanup after timeouts, crashes, etc (592524, 550815)
6095 - a race in recovery could return ETIMEDOUT to apps (623947)
6096 - building outside the source directory was fixed
6097 * the lustre-HOWTO was brought (more) up to date (582544)
6098 * major progress was made on recovery functionality
6100 2002-10-10 Phil Schwan <phil@clusterfs.com>
6103 - recovery deadlock fix
6104 - rm -rf causes LBUG fix (617817)
6105 - file open by multiple tasks fix (618962)
6106 - directory permissions bugs (602707 and 620007)
6107 - journal_stop fixed with locking (611313)
6108 - O_APPEND failures resolved (618273, perhaps 614459)
6109 - lconf PATH fix (619770)
6110 - IA64 build fix (621450)
6111 - RPC buffer sizes scale with amount of memory
6113 2002-10-01 Phil Schwan <phil@clusterfs.com>
6116 - locks would be cancelled without throwing away data pages,
6117 resulting in inconsistent data (605627)
6118 - inode attributes were not always being refreshed (605627, 612449)
6119 - lconf now continues to cleanup after lctl reports an error
6120 - MDS now enforces user permissions (602707)
6121 - lprocfs cleanup fixed, but not yet enabled (614157)
6122 - fixed infinite server hang, should a client not respond to an AST
6123 - avoid going into recovery if user calls readlink() with a buffer
6124 that's too small (613941)
6125 - AST RPCs no longer require replies (614867) -- this may be changed
6126 - don't crash server if client sends an IOV that's too big (611336)
6127 - fixed lock conversion deadlock (611892)
6128 - fixed the following of symlinks (614622)
6129 * recovery: the server can remove locks from a client that dies, other
6130 clients can make progress
6131 * more extN patch fixes
6132 * compile-time configurable ptlrpc buffer allocations
6134 - collaborative read cache document
6135 - Lustre Lite Performance CDR document-in-progress
6137 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6140 - fix typo in patch-2.4.18
6142 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6145 - clear ptlrpc request each time in handle_incoming_request()
6146 - unlink of files now destroys the object on the OST
6148 2002-09-19 Peter Braam <braam@clusterfs.com>
6150 * add hard link support
6151 * change obdfile creation method
6152 * kernel patch changed
6154 2002-09-19 Peter Braam <braam@clusterfs.com>
6157 - stack overflow bug in extN fixed
6159 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
6161 * documentation updates
6162 - add man pages for config tools
6163 - update tests/README to describe testing with new config tools
6164 - finish metadata API descriptions
6165 * bug fixes and cleanups
6166 - statfs workaround for 16TB limit
6167 - LOV stripe allocation improved, can stripe on subset of OSTs
6168 - LOV file size/IO offset was wrong for files > 4GB in size
6169 - object EA data was being dropped, caused files to be unreadable
6170 - memory overflow with non-LOV OST caused memory corruption
6171 - fixed regression tests to work with new config tools, obdfilter
6172 - fixed bug when directory size became larger than 1 block
6173 - fixed bug (for single client case) when PWD was deleted
6174 - invalidate local directory pages when doing intent-based ops
6175 - avoid LDLM oops when lock callback contained bad data
6177 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
6179 * documentation updates
6180 * bug fixes and cleanups
6181 - configuration tools
6184 - 64-bit compile warnings
6185 - 64-bit internal statfs data
6187 * test_brw on persistent OST devices
6189 * lprocfs (disabled)
6191 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
6193 * documentation updates
6194 * bug fixes and cleanups
6195 * configuration tools
6197 2002-08-30 Peter J. Braam <braam@clusterfs.com>
6200 * many small fixes to 0_5_4
6201 * io/network handling
6202 * thinkos in MDS operations
6204 2002-08-24 Peter J. Braam <braam@clusterfs.com>
6207 * crucial basic fixes to 0.5.3
6208 * IOR, Iozone work over Elan
6211 2002-08-07 Phil Schwan <phil@clusterfs.com>
6212 * version 0_5_3, our first alpha
6213 * we use the new Portals iovs
6214 * documentation updates
6215 * bug fixes and cleanups
6216 * small changes in the DLM wire protocol
6218 2002-07-25 Peter J. Braam <braam@clusterfs.com>
6219 * version 0_5_1 with some initial stability,
6220 * locking on MD and file I/O.
6221 * documentation updates
6222 * several bug fixes since 0.5.0
6223 * small changes in wire protocol
6225 2002-07-18 Phil Schwan <phil@clusterfs.com>
6227 * delivered as Lustre Light Alpha
6228 * fixed a crash after handling invalid MDS requests
6229 * fixed directory pages for architectures with non-4k pages sizes
6231 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
6232 * release version v0_4_4
6233 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
6235 * Fixes a number of interruption problems with OST operations.
6236 * Update documentation for portals header changes
6237 * Move all wire protocol structs/defines to lustre_idl.h
6238 * Fixes symlink length bug.
6239 * Add tcpdump to repository.
6241 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
6242 * release version v0_4_3
6243 * Fixes statfs for inodes on extN.
6244 * Fixes bug in runtests which would delete /etc/hosts.
6245 * Use 64-bit object IDs wherever possible (not into VFS though)
6246 Remove ost_get_info, which is unused by lustre, and out of date.
6248 2002-07-03 Peter Braam <braam@clusterfs.com>
6249 * release version v0_4_2 Fixes a lookup error (type not passed)
6250 * move forward to head of Portals
6251 * move forward to latest Lustre kernel
6253 2002-06-25 Peter Braam <braam@clusterfs.com>
6254 * release version v0_4_1. Hopefully stable on single node use.