1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.39.3 (SLES 10),
5 2.6.18-164.6.1.el5 (RHEL 5),
6 2.6.18-164.6.1.0.1.el5 (OEL 5),
8 2.6.27 vanilla (kernel.org).
9 * Client support for unpatched kernels:
10 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
11 2.6.16 - 2.6.27 vanilla (kernel.org)
12 * Recommended e2fsprogs version: 1.41.6.sun1
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
14 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
15 removed cwd "./" (refer to Bugzilla 14399).
19 Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
21 Severity : enhancement
23 Description: Update kernel to RHEL5.4 2.6.18-164.6.1.el5 and
24 OEL5 2.6.18-164.6.1.0.1.el5(Both in-kernel OFED enabled).
28 Description: Sanity test_27p fails.
29 Details : OBD_FAIL_OST_ENOSPC has been moved to filter_statfs().
30 Sanity tests 27 were cleaned up and improved.
34 Description: Unify req format on client/servers
35 Details : Use new req_capsule API [almost] everywhere instead of old PTLRPC
36 buffers and swabbers approach..
41 Description: drop join file stuff.
43 Severity : enhancement
45 Description: Add 2.6.27 and SLES11 (patchless client) support
47 Severity : enhancement
49 Description: Adjust locks' extents on their first enqueue, so that at the time
50 they get granted, there is no need for another pass through the
51 queues since they are already shaped into the proper forms.
55 Description: Fix in ptlrpc_expire_one_request() to print the signed time
58 Severity : enhancement
60 Description: Build kernels (RHEL5, OEL5 and SLES10/11) using the vendor's own
63 Severity : enhancement
65 Description: Remove set_info(KEY_UNLINKED) from MDS/OSC
67 Severity : enhancement
69 Description: correctly handle big reply message.
70 Details : send LNet event if reply is bigger then buffer and adjust this buffer
75 Description: Drop unnecessary __GFP_NOMEMALLOC flag from filter_get_page()
77 Severity : enhancement
79 Description: Add support for OEL5.
83 Description: Conf-sanity.sh 50g test - deactivated OST should not cause a panic.
87 Description: client could not reconnect to OST because of an active request.
88 Details : The client now retries to reconnect to the same server, if a
89 connect request failed with EBUSY or -EAGAIN. Test case 35b
90 conf-sanity.sh has been created.
94 Description: Don't allow make backward step on assignin osc next id.
95 Details : race between allocation next id and ll_sync thread can be cause
96 of set wrong osc next id and can be kill valid ost objects.
100 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
104 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
108 Description: Error handling in osc_statfs_interpret() has been improved.
109 Details : Check in osc_statfs_interpret() for EBADR.
113 Description: Do not send statfs() requests to OSTs disabled by administrator.
114 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
116 Severity : enhancement
118 Description: Added sanity-quota test to for lfs(1) to work with numeric uid/gid.
120 Severity : enhancement
122 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
124 Severity : enhancement
126 Description: Enhancement for lfs(1) command to use numeric uid/gid.
130 Description: Improved cleanup procedure for conf-sanity test_46a.
134 Description: Do not start bulk transfer, if client tried to reconnect.
135 Details : Do not start bulk transfer, if client tried to reconnect,
136 but failed due to active rpcs. Check for exp_abort_active_req
137 has been implemented.
141 Description: Prevent inconsistences between linux and lustre mount structures.
142 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
143 Make the sleep interruptible.
145 Severity : enhancement
147 Description: provide server to client comms path
148 Details : server to client communications path via new LDLM_SET_INFO rpc
152 Description: truncate starts GFP_FS allocation under transaction causing deadlock
153 Details : ldiskfs_truncate calls grab_cache_page which may start page
154 allocation under an open transaction. This may lead to
155 calling prune_icache with consequent lustre reentrance.
159 Description: send correctly lsm on open replay
160 Details : MDS is trust to LSM size on replay open, but client can set wrong size
165 Description: enable client interface failover
166 Details : When a child reconnects from another NID, properly update export
167 nid hash position and ldlm reverse import.
169 Severity : enhancement
171 Description: Communicate OST degraded/readonly state via statfs to MDS
172 Details : Flags in the statfs returned from OSTs indicate whether the
173 OST is in a degraded RAID state, or if the filesystem has
174 turned read-only after a filesystem error is detected.
179 Description: don't panic if EPROTO was hit when reading symlink
180 Details : correctly handling request reference in error cases.
182 Severity : enhancement
184 Description: Add LustreNetLink, a kernel-userspace communcation path.
186 Severity : enhancement
188 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
193 Description: access to llog context before init.
194 Details : move handling CATALOGS file at osc layer and forbid access to llog
198 Frequency : rare, only when client is writing to OST on local node
200 Description: Avoid deadlock for local client writes
201 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
202 memory freeing context. This allows OST threads to set the
203 PF_MEMALLOC flag on task structures in order to allocate memory
204 from reserved pools and complete IO.
205 Use GFP_HIGHUSER for OST allocations for non-local client writes,
206 so that the OST threads generate memory pressure and allow
207 inactive pages to be reclaimed.
209 Severity : enhancement
211 Description: Update kernel to RHEL5.3 2.6.18-128.1.14.el5.
216 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
217 Details : move ldlm namespace creation in setup phase to avoid grab
218 _lprocfs_lock with cli_sem held.
222 Description: Temporarily disable grant shrink.
223 Details : Disable the feature for debugging.
226 Frequency : only when formatting test filesystems
228 Description: Unable to run several mkfs.lustre on loop devices at the same time
229 Details : mkfs.lustre returns error 256 on the concurrent loop devices
230 formatting. The solution is to proper handle the error.
232 Severity : enhancement
234 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
236 Severity : enhancement
238 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
243 Description: resolve race between obd_disconnect and class_disconnect_exports
244 Details : if obd_disconnect will be called to already disconnected export he
245 forget release one reference and osc module can't unloaded.
247 Severity : enhancement
249 Description: Allow tuning service thread via /proc
250 Details : For each service a new
251 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
252 created that can be used to set min/max thread counts, and get the
253 current number of running threads.
258 Descriptoin: don't return error if have partially created objects for file.
259 Details : lov_update_create_set uses set->set_success as index for created
260 objects, so if some requests will be failed, they will have hole
261 at end of array and qos_shrink_lsm for allocate correct lsm.
263 Severity : enhancement
265 Description: Update OFED support to 1.4.1
270 Description: Reduce small size read RPC
271 Details : Set read-ahead limit for every file and only do read-ahead when
272 available read-ahead pages are bigger than 1M to avoid small size
275 Severity : enhancement
277 Description: move AT tunable parameters for more consistent usage
278 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
280 Severity : enhancement
282 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
283 Details : allow skip disconnected ost for send statfs request and hide error
289 Description: fix racy locking of mballoc block bitmaps causing BUG
290 Details : The locking of the mballoc buddy bitmap and the in-memory
291 block bitmap was using two different spin locks in some
292 cases. This made it possible to incorrectly access the
293 mballoc bitmap while another process was modifying it,
294 causing a sanity assertion to fail. While no on-disk corruption
295 was reported, there was some risk of this happening.
298 Frequency : rare, on llog test 6
300 Descriptoin: don't allow connect to already connected import
301 Details : allowing connect to already connected import is hide connecting problem.
306 Descriptoin: allow kill process which wait statahead result
307 Details : for some reasons 'ls' can stick in waiting result from statahead,
308 in this case need way for kill this process.
310 Severity : enhancement
312 Description: Add state history info file, enhance import info file
313 Details : Track import connection state changes in a new osc/mdc proc file;
314 add overview-type data to the osc/mdc import proc file.
316 Severity : enhancement
318 Description: MDS create should not wait for statfs RPC while holding DLM lock.
320 Severity : enhancement
322 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
327 Descriptoin: OST grant too much space to client even there are not enough space.
328 Details : Client will shrink its grant cache to OST if there are no write
329 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
330 this grant cache if there are already not enough avaible space
331 (left_space < total_clients * 32M).
334 Frequency : start MDS on uncleanly shutdowned MDS device
336 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
337 Details : stay in waiting mds<>ost recovery finished produce random bugs
338 due race between two ll_sync thread for one lov target. send
339 ACTIVATE event only if connect realy finished and import have
343 Frequency : rare, connect and disconnect target at same time
345 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
346 Details : don't call obd_disconnect under lov_lock. this long time
347 operation and can block ptlrpcd which answer to connect request.
352 Descriptoin: don't lose wakeup for imp_recovery_waitq
353 Details : recover_import_no_retry or invalidate_import and import_close can
354 both sleep on imp_recovery_waitq, but we was send only one wakeup
358 Frequency : always with long access acl
360 Descriptoin: mds can't pack reply with long acl.
361 Details : mds don't control size of acl but they limited by reint/getattr
364 Severity : enhancement
366 Description: Update to SLES10 kernel-2.6.16.60-0.33.
368 Severity : enhancement
370 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
373 Frequency : start MDS on uncleanly shutdowned MDS device
375 Descriptoin: aborting recovery hang on MDS
376 Details : don't throttle destroy RPCs for the MDT.
379 Frequency : on remount
381 Description: external journal device not working after the remount
382 Details : clear dev_rdonly flag for external journal devices in
388 Description: shutdown vs evict race
389 Details : client_disconnect_export vs connect request race.
390 if client will evicted at this time - we start invalidate
391 thread without referece to import and import can be freed
395 Frequency : rare, need acl's on inode.
397 Description: client can't handle ost additional correctly
398 Details : if ost was added after client connected to mds client can have
399 hit lnet_try_match_md ... to big messages to wide striped files.
400 in this case need teach client to handle config events about add
401 lov target and update client max ea size at that event.
403 Severity : enhancement
405 Description: Changelogs
406 Details : Changelogs are a lightweight mechanism to track filesystem
407 metadata and namespace changes. The changelog is recorded
408 permanently on the MDTs, and is periodically "consumed" / purged
409 when records are no longer needed.
411 Severity : enhancement
413 Description: compact fld format with extents
414 Details : Store range of seq rather than every seq in FLD. Seq
415 controller update FLD rather than clients. In Case of CMD, mdt0
416 has FLD, all other metadata server act as non persistent proxy
417 for FLD queries and cache fld entries in fld cache.
422 Description: don't skip ost target if they assigned to file
423 Details : Drop slow OSCs if we can, but not for requested start idx.
424 This means "if OSC is slow and it is not the requested
425 start OST, then it can be skipped, otherwise skip it only
426 if it is inactive/recovering/out-of-space.
430 Description: more cleanup in mds_lov
431 Details : not send LOV EA under replay, we can't know about they size at this
432 time. Don't allow client connect to mds before any ost connected,
433 for avoid problems with LOV EA size and returning EIO to client.
435 Severity : enhancement
437 Description: Interoperability at server side (Disk interoperability)
439 Severity : enhancement
441 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
443 Severity : enhancement
445 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
447 Severity : enhancement
449 Description: New client IO stack (CLIO).
451 Severity : enhancement
453 Description: Commit on sharing. Eliminate inter-client dependencies between
454 uncommitted transactions by doing transaction commits.
455 Thereby clients may recovery independently.
458 Frequency : Create a symlink file with a very long name
460 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
461 Details : If there is no extra space in the request for early cancels,
462 ldlm_req_handles_avail() returns 0 instead of a negative value.
464 Severity : enhancement
466 Description: Add /proc entry for import status
467 Details : The mdc, osc, and mgc import directories now have
468 an import directory that contains useful import data for debugging
471 Severity : enhancement
473 Description: Re-disable certain /proc logging
474 Details : Enable and disable client's offset_stats, extents_stats and
475 extents_stats_per_process stats logging on the fly.
478 Frequency : Only on FC kernels 2.6.22+
480 Description: oops in statahead
481 Details : Do not drop reference count for the dentry from VFS when lookup,
482 VFS will do that by itself.
484 Severity : enhancement
486 Description: Generic /proc file permissions
487 Details : Set /Proc file permissions in a more generic way to enable non-
488 root users operate on some /proc files.
492 Description: Hitting mdc_commit_close() ASSERTION
493 Details : Properly handle request reference release in
494 ll_release_openhandle().
498 Description: quota recovery deadlock during mds failover
499 Details : This patch includes att18982, att18236, att18237 in bz14840.
501 1. fix osts hang when mds does failover with quotaon
502 2. prevent watchdog storm when osts threads wait for the
507 Frequency : only patchless client
508 Description: add workaround for race between add/remove dentry from hash
510 Severity : enhancement
512 Description: Allow OST glimpses to return PW locks
516 Description: LBUG when llog conf file is full
517 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
521 Description: Prevent import from entering FULL state when server in recovery
525 Description: service mount cannot take device name with ":"
526 Details : Only when device name contains ":/" will mount treat it as
532 Description: replace ptlrpcd with the statahead thread to interpret the async
533 statahead RPC callback
537 Frequency : on recovery
538 Description: I/O failures after umount during fail back
539 Details : if client reconnected to restarted server we need join to recovery
540 instead of find server handler is changed and process self
541 eviction with cancel all locks.
543 Severity : enhancement
545 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
547 Severity : enhancement
549 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
551 Severity : enhancement
553 Description: Upcall on Lustre log has been dumped
554 Details : Allow for a user mode script to be called once a Lustre log has
555 been dumped. It passes the filename of the dumped log to the
556 script, the location of the script can be specified via
557 /proc/sys/lnet/debug_log_upcall.
562 Description: avoid idr_remove called for id which is not allocated.
563 Details : Move assigment s_dev for clustered nfs to end of initialization,
564 to avoid problem with error handling.
569 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
570 Details : When connection is reused this not moved from CONN_UNUSED_HASH
571 into CONN_USED_HASH and this prodice warning when put connection
572 again in unused hash.
575 Severity : enhancement
577 Description: File striping can now be set to use an arbitrary pool of OSTs.
579 Severity : enhancement
581 Description: Export bytes_read/bytes_write count on OSC/OST.
585 Description: Early reply size mismatch, MGC loses connection
586 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
587 the connect flags are properly negotiated.
592 Description: quotas are not honored with O_DIRECT
593 Details : all writes with the flag O_DIRECT will use grants which leads to
594 this problem. Now using OBD_BRW_SYNC to guard this.
598 Description: add quota statistics
599 Details : 1. sort out quota proc entries and proc code.
600 2. add quota statistics
602 Severity : enhancement
604 Description: enable quota support for HEAD.
608 Description: Properly propagate oinfo flags from lov to osc for statfs
609 Details : restore missing copy oi_flags to lov requests.
611 Severity : enhancement
613 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
614 : l_getgroups(8), lst(8), routerstat(8)
615 Details : included man pages for llobdstat(8), llstat(8),
616 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
618 Severity : enhancement
620 Description: configure's --enable-quota should check the
621 : kernel .config for CONFIG_QUOTA
622 Details : configure is terminated if --enable-quota is passed but
623 : no quota support is in kernel
627 Description: Remove portals compatibility
628 Details : Remove portals compatibility, not interoperable with releases
633 Description: Resolve device initialization race
634 Details : Prevent proc handler from accessing devices added to the
635 obd_devs array but yet be intialized.
637 Severity : enhancement
639 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
641 Severity : enhancement
643 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
648 Description: Using wrong pointer in osc_brw_prep_request
649 Details : Access to array[-1] can produce panic if kernel compiled with
650 CONFIG_PAGE_ALLOC enabled
654 Description: Client runs out of low memory
655 Details : Consider only lowmem when counting initial number of llap pages
659 Description: *optional* service tags registration
660 Details : if the "service tags" package is installed on a Lustre node
661 When the filesystem is mounted, a local-node service tag will
662 be created. See http://inventory.sun.com/ for more information
663 about the Service Tags asset management system.
667 Description: Kernel BUG tries to release flock
668 Details : Lustre does not destroy flock lock before last reference goes
669 away. So always drop flock locks when client is evicted and
670 perform unlock regardless of successfulness of speaking to MDS.
674 Description: add refcount protection for osc callbacks, avoid panic on shutdown
678 Description: sanity test 65a fails if stripecount of -1 is set
679 Details : handle -1 striping on filesystem in ll_dirstripe_verify
684 Description: ASSERTION(CheckWriteback(page,cmd)) failed
685 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
688 Severity : enhancement
690 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
694 Description: do not process already freed flock
695 Details : flock can possibly be freed by another thread before it reaches
696 to ldlm_flock_completion_ast.
700 Description: LBUG during stress test
701 Details : Need properly lock accesses the flock deadlock detection list.
705 Description: oops in page fault handler
706 Details : kernel page fault handler can return two special 'pages' in error
707 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
711 Description: timeout with invalidate import.
712 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
713 should be handled by ptlrpcd. This produce long age waiting and
714 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
716 Severity : enhancement
718 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
722 Description: enable MGS and MDT services start separately
723 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
724 a co-located MGS without starting the MGS, which is a complement
725 to 'nosvc' mount option.
729 Frequency : after recovery
730 Description: precreate to many object's after del orphan.
731 Details : del orphan st in oscc last_id == next_id and this triger growing
732 count of precreated objects. Set flag LOW to skip increase count
733 of precreated objects.
737 Frequency : rare, on clear nid stats
738 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
739 Details : when clean nid stats sometimes try destroy live entry,
740 and this produce panic in free.
744 Description: Stack overflow during MDS log replay
745 ease stack pressure by using a thread dealing llog_process.
749 Description: wait until IO finished before start new when do lock cancel.
750 Details : VM protocol want old IO finished before start new, in this case
751 need wait until PG_writeback is cleared until check dirty flag and
752 call writepages in lock cancel callback.
754 Severity : enhancement
756 Description: using special macro for print time and cleanup in includes.
760 Description: mds_mfd_close() ASSERTION(rc == 0)
761 Details : In mds_mfd_close(), we need protect inode's writecount change
762 within its orphan write semaphore to prevent possible races.
766 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
770 Frequency : rare, on shutdown ost
771 Description: don't hit live lock with umount ost.
772 Details : shrink_dcache_parent can be in long loop with destroy dentries,
773 use shrink_dcache_sb instead.
777 Description: don't panic with use echo client
778 Details : echo client pass NULL as client nid pointer and this produce null
783 Description: fix build on ppc32
784 Details : compile code with -m64 flag produce wrong object file for ppc32.
788 Description: add message levels for liblustreapi
792 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
793 Details : Move the 'good_osts' check before the 'total_bavail' check. This
794 will result in an -EAGAIN and in the exit call path we call
795 alloc_rr() which will with increasing aggressiveness attempt to
796 aquire precreated objects on the minimum number of required OSCs.
800 Description: Use old size assignment to avoid deadlock
801 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
802 the scheduling while holding a spinlock. We do not need locking
803 for size in ll_update_inode() because size is only updated from
804 the MDS for directories or files without objects, so there is no
805 other place to do the update, and concurrent access to such inodes
806 are protected by the inode lock.
810 Description: resolve "_IOWR redefined" build error on SLES10
814 Description: dump the memory debugging after all modules are unloaded to
815 suppress false negative in conf_sanity test 39
817 Severity : enhancement
819 Description: build kernel-ib packages for OFED 1.3 in our release cycle
824 Description: fix SLES kernel versioning
825 Details : the kernel version for our SLES 10 kernel did not include a "-"
826 before the "smp" at the end. while this was not a problem in
827 general, it did mean that software trying to use the kernel
828 version to try to detect a vendor specific kernel would fail.
829 this was most evident by the OFED build scripts.
833 Description: Don't update lov_desc members until making sure they are valid
834 Details : When updating lov_desc members via proc fs, need fix their
835 validities before doing the real update.
839 Description: don't put request into delay list while invalidate in flight.
840 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
841 import in flight. this produce timeout for invalidate and sometimes
842 can cause stale data.
846 Frequency : on ppc only
847 Description: not convert ost objects for directory because it's not exist.
848 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
852 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
854 Severity : enhancement
856 Description: Update to support 2.6.22.14 vanilla kernel.
860 Frequency : rare, on recovery
861 Description: read procfs can produce deadlock in some situation
862 Details : Holding lprocfs lock which send rpc can produce block for destroy
863 obd objects and this also block reconnect with -EALREADY. This
864 isn't fix all lprocfs bugs - but make it rare.
866 Severity : enhancement
868 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
871 Frequency : frequent on X2 node
873 Description: mdc_set_open_replay_data LBUG
874 Details : Set replay data for requests that are eligible for replay.
878 Description: lustre_mgs: operation 101 on unconnected MGS
879 Details : When MGC is disconnected from MGS long enough, MGS will evict the
880 MGC, and late on MGC cannot successfully connect to MGS and a lot
881 of the error messages complaining that MGS is not connected.
885 Frequency : on network error
886 Description: panic with double free request if network error
887 Details : mdc_finish_enqueue is finish request if any network error ocuring,
888 but it's true only for synchronus enqueue, for async enqueue
889 (via ptlrpcd) this incorrect and ptlrpcd want finish request
892 Severity : enhancement
894 Description: client-side metadata stat-ahead during readdir(directory readahead)
895 Details : perform client-side metadata stat-ahead when the client detects
896 readdir and sequential stat of dir entries therein
899 Frequency : on start mds
901 Description: Implement get_info(last_id) in obdfilter.
904 Frequency : occasional
906 Description: Correctly check stale fid, not start epoch if ost not support SOM
907 Details : open with flag O_CREATE need set old fid in op_fid3 because
908 op_fid2 was overwritten with new generated fid, but mds can answer
909 with one of these two fids and both is not stale. Setattr
910 incorrectly started an epoch and assume will be called
911 done_writing, but without SOM done_writing ever being called.
914 Frequency : rare, depends on device drivers and load
916 Description: MDS or OSS nodes crash due to stack overflow
917 Details : Code changes in 1.8.0 increased the stack usage of some functions.
918 In some cases, in conjunction with device drivers that use a lot
919 of stack the MDS (or possibly OSS) service threads could overflow
920 the stack. One change which was identified to consume additional
921 stack has been reworked to avoid the extra stack usage.
924 Frequency : occasional
926 Description: Do not fail import if osc_interpret_create gets -EAGAIN
927 Details : If osc_interpret_create got -EAGAIN it immediately exits and
928 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
929 oscc_has_objects and see OSC has no objests and call
930 oscc_internal_create to resend create request.
932 Severity : enhancement
934 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
936 Severity : enhancement
938 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
941 Frequency : very rare
943 Description: Fix replay if there is an un-replied request and open
944 Details : In some cases, older replay request will revert the
945 mcd->mcd_last_xid on MDS which is used to record the client's
948 Severity : enhancement
950 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
952 Severity : enhancement
954 Description: Add rhel5 support to HEAD.
956 Serverity : enhancement
958 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
963 Description: Don't allow skipping OSTs if index has been specified.
964 Details : Don't allow skipping OSTs if index has been specified, make
965 locking in internal create lots better.
969 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
970 Details : access to bitfield in structure is always rounded to long
971 and this produce problem with not atomic change any bit.
975 Description: Lustre make rpms failed.
976 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
977 builds Lustre rpms from tarball.
980 Frequency : rare on shutdown ost
982 Description: If llog cancel was not send before clean_exports phase, this can
983 produce deadlock in llog code.
984 Details : If llog thread has last reference to obd and call class_import_put
985 this produce deadlock because llog_cleanup_commit_master wait when
986 last llog_commit_thread exited, but this never success because was
987 called from llog_commit_thread.
991 Description: allow userland application know is lost one of stripes.
992 Details : fill lvb_blocks with error code on ost and return it to
993 application if error flag found.
997 Description: NULL lov_tgts causing MDS oops
998 Details : more safe checks for NULL lov_tgts for avoid oops.
1000 Severity : enhancement
1002 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
1006 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
1007 Details : If an OST is down the MDS will hang indefinitely in
1008 obd_statfs_rqset() waiting for the statfs data. While for
1009 MDS QOS usage of statfs, it should not stuck in waiting.
1011 Severity : enhancement
1013 Description: remote_acl support
1014 Details : Support ACL-based permission check for remote user.
1015 Support setfacl/getfacl for remote user with the utils
1016 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
1017 the system "{s,g}etfacl" utils.
1019 Severity : enhancement
1021 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
1023 Severity : enhancement
1025 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
1029 Description: make mgs_setparam() handle fsname containing dash
1030 Details : fsname containing a dash does not work with lctl conf_param
1032 Severity : enhancement
1034 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
1036 Severity : enhancement
1038 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
1042 Frequency : rare, at shutdown
1043 Description: access already free / zero obd_namespace.
1044 Details : if client_disconnect_export was called without force flag set,
1045 and exist connect request in flight, this can produce access to
1046 NULL pointer (or already free pointer) when connect_interpret
1047 store ocd flags in obd_namespace.
1051 Frequency : only at startup
1052 Description: not alloc memory with spinlock held.
1053 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
1054 if any spinlock held.
1056 Severity : enhancement
1058 Description: make lustre randomly fail allocating memory
1059 Details : Make lustre randomly failed allocating memory for testing purpose.
1061 Severity : enhancement
1063 Description: lost problems with lov objid file
1064 Details : Fixes some scability and access to not inited memory problems
1065 in work with lov objdid file.
1070 Description: lfs find does not continue on file error
1071 Details : Continue other files processing when a file/dir is absent.
1075 Description: Inconsistent usage of lustre_pack_reply()
1076 Details : Standardize the usage of lustre_pack_reply() such that it
1077 always generate a CERROR on failure.
1080 Frequency : occasional
1082 Description: MDS hang or stay in waiting lock
1083 Details : If client receive lock with CBPENDING flag ldlm need send lock
1084 cancel as separate rpc, to avoid situation when cancel request
1085 can't processed due all i/o threads stay in wait lock.
1089 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
1093 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
1097 Description: Update to SLES9 kernel-2.6.5-7.287.3.
1101 Description: Update to RHEL4 latest kernel.
1103 Severity : enhancement
1105 Description: Build SLES10 patchless client fails
1106 Details : The configure was broken by run ./configure with
1107 --with-linux-obj=.... argument for patchless client. When the
1108 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
1109 can't search header adequately. Use absolute path such as
1110 -I($LINUX)/include instead.
1114 Description: interrupt oig_wait produce painc on resend.
1115 Details : brw_redo_request can be used for resend requests from ptlrpcd and
1116 private set, and this produce situation when rq_ptlrpcd_data not
1117 copyed to new allocated request and triggered LBUG on assert
1118 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
1119 ptlrpcd set if request is changed and can be safety changed to use
1124 Description: LASSERT_{REQ,REP}SWAB macros are buggy
1125 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
1126 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
1127 these macros and replace them with their difinitions instead.
1131 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
1132 Details : Update which_patch & target file for SLES10 latest kernel.
1135 Description: Remove client patches from SLES 10 kernel.
1136 Details : This causes SLES 10 clients to behave as patchless clients
1137 even on a Lustre-patched (server) kernel.
1139 Severity : enhancement
1141 Description: self-adjustable client's lru lists
1142 Details : use adaptive algorithm for managing client cached locks lru
1143 lists according to current server load, other client's work
1144 pattern, memory activities, etc. Both, server and client
1145 side namespaces provide number of proc tunables for controlling
1148 Severity : enhancement
1150 Description: light-weight GSS support
1151 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
1152 and reduce performance overhead.
1154 Severity : enhancement
1156 Description: Linux keyring support
1157 Details : Support using service of Linux keyring for Lustre GSS internal
1158 context refresh/cache mechanism.
1162 Description: Fix errors in lfs documentation
1163 Details : Fixes man pages
1167 Description: don't use GFP_* in generic Lustre code.
1168 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
1172 Description: obdclass is limited by single OBD_ALLOC(idarray)
1173 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
1177 Description: warnings with build patchless client with vanila 2.6.19 and up
1178 Details : change old ctl_table style and replace ctl_table/ctl_table_header
1179 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1183 Frequency : Only for SLES
1184 Description: sanity_quota fail test_1
1185 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1186 file, which makes TSTID[2] inunique.
1190 Frequency : Only for SLES9
1191 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1192 Details : Update target/ChangeLog/which_patch .
1196 Frequency : Only for test_10 in sanity-quota.sh
1197 Description: a bug in quota test code
1198 Details : checking a test flag in a wrong place
1202 Frequency : Only for test_13 in sanity-quota.sh
1203 Description: a bug in quota test script
1204 Details : shouldn't check overfull allocated quota
1208 Description: jbd statistics
1209 Details : Port older jbd statistics patch for sles10
1211 Severity : enhancement
1213 Description: Add printing inode info into message about error in writepage.
1218 Description: reply_lock_interpret crash due to race with it and lock cancel.
1219 Details : Do not replay locks that are being cancelled. Do not reference
1220 locks by their address during replay, just by their handle.
1224 Frequency : When flocks are used.
1225 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1226 Details : Flock locks might destroy just granted lock if it could be merged
1227 with another existing flock, this is done in completion handler,
1228 so teach ldlm_cli_enquque_fini that this is a valid case for
1234 Description: Oops in read and write path when failing to allocate lock.
1235 Details : Check if lock allocation failed and return error back.
1239 Description: lstripe command fails for valid OST index
1240 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1241 instead of lov->desc.ld_active_tgt_count.
1245 Description: sanity.sh failed test 103
1246 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1251 Description: df doesn't work properly if diskfs blocksize != 4K
1252 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1256 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1257 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1259 Severity : enhancement
1261 Description: Adaptive timeouts
1262 Details : RPC timeouts adapt to changing server load and network
1263 conditions to reduce resend attempts and improve recovery time.
1267 Description: llapi_file_create() does not allow some changes
1268 Details : add llapi_file_open() that allows specifying the mode and
1269 open flags, and also returns an open file handle.
1273 Description: merge and cleanup kernel patches.
1274 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1278 Description: Add journal checksum support.(Kernel part)
1279 Details : The journal checksum feature adds two new flags i.e
1280 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1281 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1282 indicates that the commit block contains the checksum for
1283 the blocks described by the descriptor blocks. Now commit
1284 record can be sent to disk without waiting for descriptor
1285 blocks to be written to disk. This behavior is controlled
1286 using JBD2_FEATURE_ASYNC_COMMIT flag.
1290 Description: OSS needs mutliple precreate threads
1291 Details : Add ability to start more than one create thread per OSS.
1295 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1296 Details : execute lfs setstripe on client
1300 Description: mds_obd_create error creating tmp object
1301 Details : When the user sets quota on root, llog will be affected and can't
1302 create files and write files.
1305 Frequency : Always on ia64 patchless client, and possibly others.
1307 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1308 Details : This allows the patchless client to be loaded on architectures
1309 without this export.
1313 Description: RedHat Update kernel for RHEL5
1314 Details : Add the kernel config file for RHEL5.
1318 Description: RedHat Update kernel for RHEL5
1319 Details : Modify the kernel config file more closer RHEL5.
1323 Description: Build failure against Centos5 (RHEL5)
1324 Details : Define PAGE_SIZE when it isn't present.
1328 Description: test_7 on sanity-quota.sh can't be used on separate machines
1329 Details : add facet to handle it
1333 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1334 Details : with NFS, the anon dentry's parent was set to itself in
1335 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1336 corresponding dentry other than use rec->ur_name.
1338 Severity : enhancement
1340 Description: lfs setstripe enhancement
1341 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1345 Description: enable data checksumming by default
1346 Details : enable checksum by default, allow --disable-checksum
1347 configure option and "-o nochecksum" mount option. Checksums
1348 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1352 Description: lustre support for RHEL5
1353 Details : Add support for RHEL5.
1357 Description: Client eviction due to failover config
1358 Details : after a connection loss, the lustre client should attempt to
1359 reconnect to the last active server first before trying the
1360 other potential connections.
1364 Description: when mds and osts use different quota unit(32bit and 64bit),
1365 quota will be released repeatly.
1366 Details : void sending multiple quota reqs to mds, which will keep the
1367 status between the reqs.
1371 Description: osts not allocated evenly to files
1372 Details : change the condition to increase offset_idx
1375 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1377 Description: Data corruption for OSTs that are formatted larger than 2TB
1379 Details : When generating the bio request for lustre file writes the
1380 sector number would overflow a temporary variable before being
1381 used for the IO. The data reads correctly from Lustre (which
1382 will overflow in a similar manner) but other file data or
1383 filesystem metadata may be corrupted in some cases.
1387 Description: Tune the kernel for good SCSI performance.
1388 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1389 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1394 Description: rewrite ext2-derived code in obdclass/uuid.c
1395 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1396 from scratch preserving functionality.
1400 Description: "lfs find -obd UUID" prints directories
1401 Details : "lfs find -obd UUID" will return all directory names instead
1402 of just file names. It is incorrect because the directories
1403 do not reside on the OSTs.
1407 Description: use i_size_read and i_size_write in 2.6 port
1408 Details : replace inode->i_size access with i_size_read/write()
1411 Frequency : when removing large files
1413 Description: scheduling issue during removal of large Lustre files
1414 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1415 It causes scheduling issues when removing large files (17TB in the
1419 Frequency : only with liblustre clients on XT3
1421 Description: evictions taking too long
1422 Details : allow llrd to evict clients directly on OSTs
1425 Frequency : only on ppc
1427 Description: /proc/fs/lustre/devices broken on ppc
1428 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1429 We should make sure the type of 'index' is loff_t and then cast
1430 explicitly as needed below. Do not assign an explicitly cast
1434 Frequency : only for rhel5
1436 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1437 Details : Modify the target file & which_kernel.
1439 Severity : enhancement
1441 Description: omit set fsid for export NFS
1442 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1443 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1447 Frequency : Always, for kernels after 2.6.16
1448 Description: Fix warning idr_remove called for id=.. which is not allocated.
1449 Details : Last kernels save old s_dev before kill super and not allow
1450 to restore from callback - restore it before call kill_anon_super.
1454 Description: buffer overruns could theoretically occur
1455 Details : llapi_semantic_traverse() modifies the "path" argument by
1456 appending values to the end of the origin string, and a
1457 overrun may occur. Adding buffer overrun check in liblustreapi.
1461 Description: Fix error on 'ls .' at the top of the Lustre mount.
1462 Details : Don't revalidate dentry if it is a root dentry.
1466 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1467 Details : Modify vm-tunables-rhel4.patch.
1471 Description: Kernel config for 2.6.18-vanilla.
1472 Details : Modify targets/2.6-vanilla.target.in.
1473 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1474 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1475 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1476 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1480 Description: improve handling recoverable errors
1481 Details : if request processig with error which can be recoverable on server
1482 request should be resend, otherwise page released from cache and
1487 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1488 Details : Modify target file & which_patch.
1489 A flaw was found in the IA32 system call emulation provided
1490 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1491 value could be stored in the %RAX register, which could trigger an
1492 out-of-bounds system call table access. An untrusted local user
1493 could exploit this flaw to run code in the kernel
1494 (ie a root privilege escalation). (CVE-2007-4573).
1498 Description: O_DIRECT bypasses client statistics.
1499 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1500 read_ahead_stats were updated. copied the stats section from
1501 osc_send_oap_rpc() into async_internal().
1505 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1509 Description: change order of libsysio includes
1510 Details : '#include sysio.h' should always come before '#include xtio.h'
1514 Description: handle "serious error: objid * already exists" more gracefully
1515 Details : If LAST_ID value on disk is smaller than the objects existing in
1516 the O/0/d* directories, it indicates disk corruption and causes an
1517 LBUG(). If the object is 0-length, then we should use the existing
1518 object. This will help to avoid a full fsck in most cases.
1520 Severity : enhancement
1522 Description: adapt the lustre_config script to support the upgrade case
1523 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1524 server targets to 1.6 in parallel.
1528 Description: To avoid grant space > avaible space when the disk is almost
1529 full. Without this patch you might see the error "grant XXXX >
1530 available" or some LBUG about grant, when the disk is almost
1532 Details : In filter_check_grant, for non_grant cache write, we should
1533 check the left space by if (*left > ungranted + bytes), instead
1534 of (*left > ungranted), because only we are sure the left space
1535 is enough for another "bytes", then the ungrant space should be
1536 increase. In client, we should update cl_avail_grant only there
1537 is OBD_MD_FLGRANT in the reply.
1541 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1542 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1543 described in CVE-2007-4573. This problem could allow untrusted
1544 local users to gain root access.
1547 Frequency : when using O_DIRECT and quotas
1549 Description: Incorrect file ownership on O_DIRECT output files
1550 Details : block usage reported by 'lfs quota' does not take into account
1551 files that have been written with O_DIRECT.
1556 Description: touch file failed when fs is not full
1557 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1558 otherwise we can get ENOSP while fs is not full.
1562 Description: parallel lock callbacks
1563 Details : Instead of sending blocking and completion callbacks as separated
1564 requests, adding them to a set and sending in parallel.
1567 Frequency : only for Cray XT3
1568 Bugzilla : 12829/13455
1569 Description: Changing primary group doesn't change the group lustre assigns to
1571 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1572 overridden with the primary group provided by the group upcall,
1573 whereas the supplied fsgid can be trusted if it is in the list of
1574 supplementary groups returned by the group upcall.
1576 Severity : enhancement
1578 Description: Allow masking D_WARNING, D_ERROR messages from console
1579 Details : Console messages can now be disabled via lnet.printk.
1583 Description: User code with malformed file open parameter crashes client node
1584 Details : Before packing join_file req, all the related reference should be
1585 checked carefully in case some malformed flags cause fake
1586 join_file req on client.
1590 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1591 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1592 mode is equal to the granted mode. However, it should also check
1593 that LDLM_FL_CP_REQD is not set.
1597 Description: Heavy nfs access might result in deadlocks
1598 Details : After ELC code landed, it is now improper to enqueue any mds
1599 locks under och_sem, because enqueue might want to decide to
1600 cancel open locks for same inode we are holding och_sem for.
1604 Description: Client eviction while running blogbench
1605 Details : A lot of unlink operations with concurrent I/O can lead to a
1606 deadlock causing evictions. To address the problem, the number of
1607 oustanding OST_DESTROY requests is now throttled to
1608 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1609 callbacks are processed in priority.
1613 Description: enable ACLs on MDS by default
1614 Details : ACLs must be enabled on MDS by default.
1617 Frequency : PPC/PPC64 only
1619 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1620 Details : fix duplicated definitions between asm-ppc64/types.h and
1621 lustre_types.h on PPC.
1624 Frequency : PPC/PPC64 only
1626 Description: asm-ppc/segment.h does not exist
1627 Details : fix compile issue on PPC.
1631 Description: better handle error messages in extents code
1634 Frequency : RHEL4 only
1636 Description: mkfs is very slow on IA64/RHEL4
1637 Details : A performance regression has been discovered in the MPT Fusion
1638 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1639 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1640 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1642 Severity : enhancement
1644 Description: SNMP support enhancement
1645 Details : Adding total number of sampled request for an MDS node in snmp
1648 Severity : enhancement
1650 Description: Optimize ldlm waiting list processing for PR extent locks
1651 Details : When processing waiting list for read extent lock and meeting read
1652 lock that is same or wider to it that is not contended, skip
1653 processing rest of the list and immediatelly return current
1654 status of conflictness, since we are guaranteed there are no
1655 conflicting locks in the rest of the list.
1659 Description: Time out and refuse to reconnect
1660 Details : When the failover node is the primary node, it is possible
1661 to have two identical connections in imp_conn_list. We must
1662 compare not conn's pointers but NIDs, otherwise we can defeat
1663 connection throttling.
1667 Description: port llog fixes from b1_6 into HEAD
1668 Details : Port llog reference couting and some llog cleanups from b1_6
1669 (bug 10800) into HEAD, for protect from panic and access to already
1670 free llog structures.
1674 Description: Detect stride IO mode in read-ahead
1675 Details : When a client does stride read, read-ahead should detect that and
1676 read-ahead pages according to the detected stride pattern.
1680 Description: data checksumming impacts single node performance
1681 Details : add support for several checksum algorithm. Currently, only CRC32
1682 and Adler-32 are supported. The checksum type can be changed on
1683 the fly via /proc/fs/lustre/osc/*/checksum_type.
1687 Description: use adler32 for page checksums
1688 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1693 Description: build for x2 fails
1694 Details : fix compile issue on Cray systems.
1698 Description: Properly match for duplicate locks
1699 Details : Due to different lock order from skiplists code, we need to
1700 traverse entire list for now
1703 Frequency : only on PPC/SLES10
1705 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1706 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1707 be defined. Add a hack in mkfs_lustre.c to work around this compile
1712 Description: LASSERT on MDS when client holding flock lock dies
1713 Details : ldlm pool logic depends on number of granted locks equal to
1714 number of released locks which is not true for flock locks, so
1715 just exclude such locks from consideration.
1719 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1720 Details : Use fsfilt_sync() for both the whole filesystem sync and
1721 individual file sync to eliminate dangerous inode locking
1722 with I_LOCK that can lead to a deadlock.
1726 Description: performance in 1.6.3
1727 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1728 sure that 1MB requests are not fragmented by the block layer.
1730 Severity : enhancement
1732 Description: organize the server-side client stats on per-nid basis
1733 Details : Change the structure of stats under obdfilter and mds to
1741 The "uuid"s file would list the uuids of _active_ exports.
1742 And the clear entry is to clear all stats and stale nids.
1744 Severity : enhancement
1746 Description: eliminate client locks in face of contention
1747 Details : file contention detection and lockless i/o implementation
1748 for contended files.
1752 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1755 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1757 Description: MDS may lose file striping (and hence file data) in some cases
1758 Details : If there are additional extended attributes stored on the MDS,
1759 in particular ACLs, SELinux, or user attributes (if user_xattr
1760 is specified for the client mount options) then there is a risk
1761 of attribute loss. Additionally, the Lustre file striping
1762 needs to be larger than default (e.g. striped over all OSTs),
1763 and an additional attribute must be stored initially in the
1764 inode and then increase in size enough to be moved to the
1765 external attribute block (e.g. ACL growing in size) for file
1770 Description: skiplist implementation simplification
1771 Details : skiplists are used to group compatible locks on granted list
1772 that was implemented as tracking first and last lock of each
1773 lock group the patch changes that to using doubly linked lists
1777 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1778 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1779 because LASSERT on both the data supplied by a client, and the
1780 data on disk is dangerous and incorrect.
1782 Severity : enhancement
1784 Description: Slow truncate/writes to huge files at high offsets.
1785 Details : Directly associate cached pages to lock that protect those pages,
1786 this allows us to quickly find what pages to write and remove
1787 once lock callback is received.
1791 Description: more ldlm soft lockups
1792 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1793 starve other threads from the resource lock for a long time in
1794 case of long waiting queue, so change the debug level from
1795 D_OTHER to the less frequently used D_INFO.
1797 Severity : enhancement
1799 Description: add -gid, -group, -uid, -user options to lfs find
1803 Description: Hung threads in invalidate_inode_pages2_range
1804 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1805 once one is completed. As a result, some RPCs are stuck in the
1806 queue and are never sent.
1810 Description: filter threads hungs on waiting journal commit
1811 Details : Cleanup filter group llog code, then only filter group llog will
1812 be only created in the MDS/OST syncing process.
1816 Description: Procfs and llog threads access destoryed import sometimes.
1817 Details : Sync the import destoryed process with procfs and llog threads by
1818 the import refcount and semaphore.
1820 Severity : enhancement
1822 Description: openlock cache of b1_6 port to HEAD
1827 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1828 Details : If insertion of an extent fails, then discard the inode
1829 preallocation and free data blocks else it can lead to duplicate
1834 Description: don't always update ctime in ext3_xattr_set_handle()
1835 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1836 In some cases the ctime should not be updated, for example for
1837 2.0->1.8 compatibility it is necessary to delete an xattr and it
1838 should not update the ctime.
1842 Bugzilla : 15713/16362
1843 Description: Assertion in iopen_connect_dentry in 1.6.3
1844 Details : looking up an inode via iopen with the wrong generation number can
1845 populate the dcache with a disconneced dentry while the inode
1846 number is in the process of being reallocated. This causes an
1847 assertion failure in iopen since the inode's dentry list contains
1848 both a connected and disconnected dentry.
1852 Description: assertion failure in ldlm_handle2lock()
1853 Details : fix a race between class_handle_unhash() and class_handle2object()
1854 introduced in lustre 1.6.5 by bug 13622.
1859 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1860 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1865 Description: oops in sd_iostats_seq_show()
1866 Details : unloading/reloading the scsi low level driver triggers a kernel
1867 bug when trying to access the sd iostat file.
1872 Description: Kernel panics during QLogic driver reload
1873 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1874 patch, causing memory corruption.
1879 Description: journal_dev option does not work in b1_6
1880 Details : pass mount option during pre-mount.
1882 Severity : enhancement
1884 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1885 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1886 extent information of a file. It can be used to map logical blocks
1887 in a file to physical blocks in the block device.
1891 Description: LDLM soft lockups - improvement
1892 Details : It is be possible to send the lock handle along with each read
1893 or write request because the client is already doing a lock match
1894 itself so there isn't any reason the OST should have to re-do that
1898 Frequency : only on Cray X2
1900 Description: X2 build failures
1901 Details : fix build failures on Cray X2.
1905 Description: xid & resent requests
1906 Details : Initialize RPC XID from clock at startup (randomly if clock is
1909 Severity : enhancement
1911 Description: Add lustre_start utility to start or stop multiple Lustre servers
1916 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1917 Details : In case of memory pressure, list_del() can be called twice on
1918 req->rq_history_list, causing a kernel oops.
1922 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1923 Details : fix stack overflow in the distributed lock manager by defering
1924 export eviction after a failed AST to the elt thread instead of
1925 handling it in the dlm interpret routine.
1929 Description: Convert some comments to new format.
1930 Details : Update documenting comments to match doxygen conventions.
1934 Description: Grammar fixes.
1935 Details : A couple of trivial spelling fixes.
1939 Description: OSD_COUNTERS-mandatory
1940 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1941 to simplify the code.
1945 Description: simplify cmm_device freeing logic.
1946 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1950 Description: Add lockdep support to dt_object_operations locking interface.
1951 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
1952 parameter indicating lock ordering. Update mdd code to use new
1957 Description: Introduce failloc constants for lockless IO tests.
1958 Details : Add two new failloc constants to test lockless IO. Only one of
1959 them in implemented---another is checked in yet to be landed
1964 Description: Add lockdep support for inode mutex.
1965 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1966 in the situations where Lustre has to lock more than one inode
1971 Description: Add optional invariants checking support.
1972 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1973 configure switch. This macro is to be used for consistency and
1974 sanity checks that are too expensive to be left in `production'
1979 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1980 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1985 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1986 Details : Use spin_lock_nested() in (the only) situation where more than
1987 one ldlm_lock is locked simultaneously. Also, fix possible
1988 dead-lock in ldlm_lock_change_resource() by enforcing particular
1993 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1994 Details : Instead of specifying each ldlm_lock call-back through separate
1995 parameter, wrap them into struct ldlm_callback_suite.
1999 Description: Kill join_lru obd method and its callers.
2000 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
2001 in memory---a requirement implemented through ->o_join_lru() obd
2002 method in HEAD. Remove this method and its users.
2006 Description: Add asynchronous ldlm ENQUEUE completion handler.
2007 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
2008 case that stock ldlm_completion_ast() cannot handle as it waits
2009 until lock is granted. Introduce new ldlm_completion_ast_async()
2010 for this. Also comment ldlm_completion_ast().
2014 Description: ldlm_error <-> errno conversion.
2015 Details : Add functions to map (rather arbitrary) between LDLM error codes
2016 and standard errno values. CLIO needs this to prevent LDLM specific
2017 constants from escaping ldlm and osc.
2021 Description: Kill unused ldlm_handle2lock_ns() function.
2022 Details : Kill unused ldlm_handle2lock_ns() function.
2026 Description: Add lu_ref support to ldlm_lock
2027 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
2028 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
2029 and ldlm_resource. LDLM interface has to be changed, because code that
2030 releases a reference on a lock, has to "know" what reference this is.
2031 In the most frequent case
2033 lock = ldlm_handle2lock(handle);
2035 LDLM_LOCK_PUT(lock);
2037 no changes are required. When any other reference (received _not_ from
2038 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
2039 instead of LDLM_LOCK_PUT().
2041 Arguably, changes are pervasive, and interface requires some discipline
2042 for proper use. On the other hand, it was very instrumental in finding
2043 a few leaked lock references.
2047 Description: Add ldlm_lock_addref_try().
2048 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
2049 attempts to addref a lock that might be being canceled
2054 Description: Add ldlm_weigh_callback().
2055 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
2056 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
2057 hard-coded `number of pages' logic.
2061 Description: Add lockdep annotations to llog code.
2062 Details : Use appropriately tagged _nested() locking calls in the places
2063 where llog takes more than one ->lgh_lock lock.
2067 Description: Add loi_kms_set().
2068 Details : Wrap kms updates into a helper function.
2072 Description: Constify instances of struct lsm_operations.
2073 Details : Constify instances of struct lsm_operations.
2077 Description: lu_conf support.
2078 Details : On a server, a file system object is uniquely identified
2079 by a fid, which is sufficient to locate and load all object
2080 state (inode). On a client, on the other hand, more data are
2081 necessary instantiate an object. Change lu_object_find() and
2082 friends to take additional `lu_conf' argument describing object.
2083 Typically this includes layout information.
2087 Description: lu_context fixes.
2088 Details : Introduce new lu_context functions that are needed on the client
2089 side, where some system threads (ptlrpcd) are shared by multiple
2090 modules, and so cannot be stopped during module shutdown.
2094 Description: Add start and stop methods to lu_device_type_operations.
2095 Details : Introduce two new methods in lu_device_type_operations, that are
2096 invoked when first instance of a given type is created and last one
2097 is destroyed respectively. This is need by CLIO.
2101 Description: Add lu_ref support to struct lu_device.
2102 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
2103 track leaked references.
2107 Description: Introduce lu_kmem_descr.
2108 Details : lu_kmem_descr and its companion interface allow to create
2109 and destroy a number of kmem caches at once.
2113 Description: Fix lu_object finalization race.
2114 Details : Fix a race between lu_object_find() finding an object and its
2115 concurrent finalization. This race is (most likely) not possible
2116 on the server, but might happen on the client.
2120 Description: Introduce lu_ref interface.
2121 Details : lu_ref is a debugging module allowing to track references to
2122 a given object. It is quite cpu expensive, and has to be
2123 explicitly enabled with --enable-lu_ref. See usage description
2128 Description: Factor lu_site procfs stats into a separate function.
2129 Details : Separate lu_site stats printing code into a separate function
2130 that can be reused on a client.
2134 Description: Constify instances of struct {lu,dt,md}_device_operations.
2135 Details : Constify instances of struct {lu,dt,md}_device_operations.
2139 Description: Introduce struct md_site and move meta-data specific parts of
2140 struct lu_site here.
2141 Details : Move md-specific fields out of struct lu_site into special struct
2142 md_site, so that lu_site can be used on a client.
2146 Description: Kill mdd_lov_destroy().
2147 Details : Remove unused mdd code.
2151 Description: Add st_block checking to multistat.c.
2152 Details : Add st_block checking to multistat.c.
2156 Description: Add lu_ref support to struct obd_device.
2157 Details : Add lu_ref tracking to obd_device.
2161 Description: Kill obd_set_fail_loc().
2162 Details : Remove unused code.
2166 Description: Add special type for ptlrpc_request interpret functions.
2167 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
2168 there. Actual usage will be in CLIO.
2172 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
2173 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
2174 doesn't work with lockdep.
2178 Description: Add rwv.c test program.
2179 Details : New testing program exercising readv(2) and writev(2) (Qian).
2183 Description: Add sendfile.c test program.
2184 Details : New testing program exercising sendfile(2) (Jay).
2188 Description: Ratelimit a message that can be very frequent.
2189 Details : Ratelimit a memory allocation failure message that can
2194 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2195 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2199 Description: Liblustre build fixes.
2200 Details : Liblustre build fixes.
2204 Description: libcfs: add cfs_{need,cond}_resched() interface.
2205 Details : libcfs: add cfs_{need,cond}_resched() definition and
2206 implementations for Linux, NT, and liblustre.
2208 Severity : enhancement
2210 Description: More exported tunables for mballoc
2211 Details : Add support for tunable preallocation window and new tunables for
2212 large/small requests
2216 Description: Detect corruption of block bitmap and checking for preallocations
2217 Details : Checks validity of on-disk block bitmap. Also it does better
2218 checking of number of applied preallocations. When corruption is
2219 found, it turns filesystem readonly to prevent further corruptions.
2223 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2224 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2225 window to 0. If it does hit cache miss, and read-pattern is still
2226 stride-io mode, does not reset the stride window, but also does not
2227 increase the stride window length in this case.
2231 Frequency : only for big-endian servers
2232 Description: Check if system is big-endian while mounting fs with extents feature
2233 Details : Mounting a filesystem with extents feature will fail on big-endian
2234 systems since ext3-based ldiskfs is not supported on big-endian
2235 systems. This can be over-riden with "bigendian_extents" mount option.
2237 Severity : enhancement
2239 Description: The root squash functionality
2240 Details : A security feature, which is to prevent users from being able
2241 to mount lustre on their desktop, run as root, and delete
2242 all of the files in the filesystem. The goal is accomplished by
2243 remapping user id (UID) and group id (GID) of the root user to
2244 a UID and GID specified by the system administartor via Lustre
2245 configuration management server (MGS). The functionality also
2246 allows to specify sets of clients for which the remapping does
2251 Description: Excessive recovery window
2252 Details : With AT enabled, the recovery window can be excessively long (6000+
2253 seconds). To address this problem, we no longer use
2254 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2255 timeout no longer depends on the service time, it is set to
2256 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2257 time via pb_service_time.
2261 Description: Watchdog triggered on MDS failover
2262 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2263 the OSTs know that the MDS "UUID" can be reused for the same export
2264 from a different NID, so we do not need to wait for the export to be
2268 Frequency : rare, only if using MMP with Linux RAID
2270 Description: MMP doesn't work with Linux RAID
2271 Details : While using HA for Lustre servers with Linux RAID, it is possible
2272 that MMP will not detect multiple mounts. To make this work we
2273 need to unplug the device queue in RAID when the MMP block is being
2274 written. Also while reading the MMP block, we should read it from
2275 disk and not the cached one.
2277 Severity : enhancement
2279 Description: open file using fid
2280 Details : A file can be opened using just its fid, like
2281 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2284 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2287 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2288 Details : add proper locking for creation of jbd_4k slab cache
2292 Description: MMP check in ext3_remount() fails without displaying any error
2293 Details : When multiple mount protection fails during remount, proper error
2296 Severity : enhancement
2298 Description: Allow stripe size to be up to 4G-64k
2299 Details : Fix math logic to allow large stripe sizes.
2303 Description: add check for >8TB ldiskfs filesystems
2304 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2305 Don't allow >8TB ldiskfs filesystems to be mounted without
2306 force_over_8tb mount option
2311 Descriptoin: OSS complains "too many blocks" and crashes
2312 Details : In filter_commitrw_write():
2313 1. lquota_chkquota();
2314 2. obdo_from_inode();
2315 3. lquota_pending_commit();
2316 "2." has changed uid and gid in this case, then caused
2317 lquota_chkquota() and lquota_pending_commit() to use
2318 different uid and gid. That is the root of the bug.
2322 Descriptoin: getstripe doesn't report stripe count = -1 sanely
2323 Details : Detect and print out the 'stripe count = -1' case properly.
2327 Description: Out or order replies might be lost on replay
2328 Details : In ptlrpc_retain_replayable_request if we cannot find retained
2329 request with tid smaller then one currently being added, add it
2330 to the start, not end of the list.
2332 --------------------------------------------------------------------------------
2334 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2336 * Support for kernels:
2337 2.6.5-7.283 (SLES 9),
2338 2.6.9-55.EL (RHEL 4),
2339 2.6.16.46-0.14 (SLES 10),
2340 2.6.18.8 vanilla (kernel.org)
2341 * Client support for unpatched kernels:
2342 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2343 2.6.16 - 2.6.22 vanilla (kernel.org)
2344 * Due to recently discovered recovery problems, we do not recommend
2345 using patchless RHEL 4 clients with this or any earlier release.
2346 * Recommended e2fsprogs version: 1.39.cfs8
2347 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2348 * Starting with this release, the ldiskfs backing filesystem required
2349 by Lustre is now in its own package, lustre-ldiskfs. This package
2350 should be installed. It is versioned separately from Lustre and
2351 may be released separately in future.
2355 Description: block reactivating mgc import until all deactivates complete
2356 Details : Fix race when failing back MDT/MGS to itself (testing)
2359 Frequency : at statup only
2361 Description: mds_lov_synchronize race leads to various problems
2362 Details : simultaneous MDT->OST connections at startup can cause the
2363 sync to abort, leaving the OSC in a bad state.
2365 Severity : enhancement
2367 Description: add optional extra BUILD_VERSION info
2368 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2369 to override the lustre version.
2372 Frequency : 2.6.18 servers only
2374 Description: ll_kern_mount() doesn't release the module reference
2375 Details : The ldiskfs module reference count never drops down to 0
2376 because ll_kern_mount() doesn't release the module reference.
2381 Description: server LBUG when using old ost_num_threads parameter
2382 Details : Accept the old ost_num_threads parameter but warn that it
2383 is deprecated, and fix an off-by-one error that caused an LBUG.
2388 Description: Transient SCSI error results in persistent IO issue
2389 Details : iobuf->dr_error is not reinitialized to 0 between two
2393 Frequency : sometimes when underlying device returns I/O errors
2395 Description: OSTs not going read-only during write failures
2396 Details : OSTs are not remounted read-only when the journal commit threads
2397 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2398 instead of the ext3 wrappers.
2402 Description: poor connect scaling with increasing client count
2403 Details : Don't run filter_grant_sanity_check for more than 100 exports
2404 to improve scaling for large numbers of clients.
2407 Frequency : SLES10 only
2409 Description: sanity-quota.sh quotacheck failed: rc = -22
2410 Details : Quotas cannot be enabled on SLES10.
2413 Frequency : liblustre clients only
2415 Description: getdirentries does not give error when run on compute nodes
2416 Details : getdirentries does not fail when the size specified as an argument
2417 is too small to contain at least one entry
2419 Severity : enhancement
2421 Description: Add LNET router traceability for debug purposes
2422 Details : If a checksum failure occurs with a router as part of the
2423 IO path, the NID of the last router that forwarded the bulk data
2424 is printed so it can be identified.
2429 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2430 Details : Due to a race condition, liblustre clients were occasionally
2431 evicted incorrectly.
2433 Severity : enhancement
2435 Description: lfs setstripe use optional parameters instead of postional
2438 Severity : enhancement
2440 Description: Nanosecond timestamp support for ldiskfs
2441 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2442 resolution timestamps. There is not yet support for this at
2443 the Lustre filesystem level.
2446 Frequency : during server recovery
2448 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2449 Details : request with rq_no_resend flag not awake l_wait_event if they get
2453 Frequency : nfs export on patchless client
2455 Description: connectathon hang when test nfs export over patchless client
2456 Details : Disconnected dentry cannot be found with lookup, so we do not need
2457 to unhash it or make it invalid
2460 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2461 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2462 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2467 Description: open req refcounting wrong on reconnect
2468 Details : If reconnect happened between getting open reply from server and
2469 call to mdc_set_replay_data in ll_file_open, we will schedule
2470 replay for unreferenced request that we are about to free.
2471 Subsequent close will crash in variety of ways.
2472 Check that request is still eligible for replay in
2473 mdc_set_replay_data().
2478 Description: disable writes to filesystem when reading health_check file
2479 Details : the default for reading the health_check proc file has changed
2480 to NOT do a journal transaction and write to disk, because this
2481 can cause reads of the /proc file to hang and block HA state
2482 checking on a healthy but otherwise heavily loaded system. It
2483 is possible to return to the previous behaviour during configure
2484 with --enable-health-write.
2486 Severity : enhancement
2488 Description: 64-bit inode version
2489 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2490 to the inode. This will be required for version-based recovery.
2495 Description: MDS fails to start if a duplicate client export is detected
2496 Details : in some rare cases it was possible for a client to connect to
2497 an MDS multiple times. Upon recovery the MDS would detect this
2498 and fail during startup. Handle this more gracefully.
2500 Severity : enhancement
2502 Description: Add -o localflock option to simulate old noflock
2504 Details : This will achieve local-only flock/fcntl locks
2510 Description: log_commit_thread vs filter_destroy race leads to crash
2511 Details : Take import reference before releasing llog record semaphore
2516 Description: Wrong request locking in request set processing
2517 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2518 imp_delayed_list, in this place should be used imp_lock.
2521 Frequency : when reconnection
2523 Description: Grant Leak when osc reconnect to OST
2524 Details : When osc reconnect ost, OST(filter) should check whether it
2525 should grant more space to client by comparing fed_grant and
2526 cl_avail_grant, and return the granted space to client instead
2527 of "new granted" space, because client will call osc_init_grant
2528 to update the client grant space info.
2531 Frequency : when client reconnect to OST
2533 Description: Grant Leak when osc do resend and replay bulk write
2534 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2535 bulk write request, because the grant info while be sync between
2536 OSC and OST when reconnect, and we should ignore the grant info
2537 these of resend/replay write req.
2542 Description: Grant space more than avaiable left space sometimes.
2543 Details : When then OST is about to be full, if two bulk writing from
2544 different clients came to OST. Accord the avaliable space of the
2545 OST, the first req should be permitted, and the second one
2546 should be denied by ENOSPC. But if the seconde arrived before
2547 the first one is commited. The OST might wrongly permit second
2548 writing, which will cause grant space > avaiable space.
2551 Frequency : when client is evicted
2553 Description: Grant might be wrongly erased when osc is evicted by OST
2554 Details : when the import is evicted by server, it will fork another
2555 thread ptlrpc_invalidate_import_thread to invalidate the
2556 import, where the grant will be set to 0. While the original
2557 thread will update the grant it got when connecting. So if
2558 the former happened latter, the grant will be wrongly errased
2559 because of this race.
2564 Description: Checking Stale with correct fid
2565 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2566 to check whether it is stale, because sometimes, we want the
2567 enqueue happened anyway, and op_data.fid2 will not be initialized.
2569 Severity : enhancement
2571 Description: update patchless client
2572 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2575 Frequency : only with 2.4 kernel
2577 Description: random memory corruption
2578 Details : size of struct ll_inode_info is to big for union inode.u and this
2579 can be cause of random memory corruption.
2584 Description: Memory leak in recovery
2585 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2586 It should also check obd_fail before fsfilt_start, otherwise if
2587 fsfilt_start return -EROFS,(failover mds during mds recovery).
2588 then the req will return with repmsg->transno = 0 and rc = EROFS.
2589 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2590 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2591 be freed no matter whether fsfilt_commit success or not.
2594 Frequency : only with huge count clients
2596 Description: Prevents from taking the superblock lock in llap_from_page for
2598 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2599 for llap_from_page call in ll_removepage prevents from taking the
2600 superblock lock for a soon died page.
2605 Description: Not check open intent error before release open handle
2606 Details : in some rare cases, the open intent error is not checked before
2607 release open handle, which may cause
2608 ASSERTION(open_req->rq_transno != 0), because it tries to release
2609 the failed open handle.
2614 Description: Set cat log bitmap only after create log success.
2615 Details : in some rare cases, the cat log bitmap is set too early. and it
2616 should be set only after create log success.
2620 Description: Accessing a block bevice can re-enable I/O when Lustre is
2621 tearing down a device.
2622 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2627 Description: service threads may hog cpus when there are a lot of requests
2628 Details : Insert cond_resched to give other threads a chance to use some CPU
2633 Description: the cat log was not initialized in recovery
2634 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2635 unlink log on mds will not be initialized until mds post
2636 recovery. And also in mds post recovery, the unlink log will
2637 initialization will be done asynchronausly, so there will be race
2638 between add unlink log and unlink log initialization.
2642 Description: brw_stats were being printed incorrectly
2643 Details : brw_stats were being printed as log2 but all of them were not
2644 recorded as log2. Also remove some code duplication arising from
2645 filter_tally_{read,write}.
2649 Frequency : rare, only in recovery.
2650 Description: ASSERTION(req->rq_type != LI_POISON) failed
2651 Details : imp_lock should be held while iterating over imp_sending_list for
2652 prevent destroy request after get timeout in ptlrpc_queue_wait.
2656 Description: replay-single.sh test 52 fails
2657 Details : A lock's skiplist need to be cleanup when it being unlinked
2658 from its resource list.
2662 Description: Short directio read returns full requested size rather than
2664 Details : Direct I/O operations should return actual amount of bytes
2665 transferred rather than requested size.
2667 Severity : enhancement
2669 Description: metadata RPC reduction (e.g. for rm performance)
2670 Details : decrease the amount of synchronous RPC between clients and servers
2671 by canceling conflicing lock before the operation on the client
2672 and packing thier handles into the main operation RPC to server.
2674 Severity : enhancement
2676 Description: Async OSC create to avoid the blocking unnecessarily.
2677 Details : If a OST has no remain object, system will block on the creating
2678 when need to create a new object on this OST. Now, ways use
2679 pre-created objects when available, instead of blocking on an
2680 empty osc while others are not empty. If we must block, we block
2681 for the shortest possible period of time.
2685 Description: improve handling recoverable errors
2686 Details : if request processig with error which can be recoverable on server
2687 request should be resend, otherwise page released from cache and
2690 Severity : enhancement
2692 Description: refine locking for avoid write wrong info into lov_objid file
2693 Details : fix possible races with add new target and write/update data in
2696 --------------------------------------------------------------------------------
2698 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2703 Frequency : on some architectures
2705 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2706 Details : Uninitialized flags sometimes cause configuration commands to
2710 Frequency : patchless clients only
2712 Description: missing __iget() symbol export
2713 Details : The __iget() symbol export is missing. To avoid the need for
2714 this on patchless clients the deathrow inode reaper is turned
2715 off, and we depend on the VM to clean up old inodes. This
2716 dependency was during via the fix for bug 12181.
2718 --------------------------------------------------------------------------------
2720 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2722 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2723 INTEROPERATE with older versions automatically. In many cases a
2724 special upgrade step is needed. Please read the
2725 user documentation before upgrading any part of a live system.
2726 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2727 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2728 * WARNING: Lustre configuration and startup changes are required with
2729 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2733 Severity : enhancement
2735 Description: MountConf
2736 Details : Lustre configuration is now managed via mkfs and mount
2737 commands instead of lmc and lconf. New obd types (MGS, MGC)
2738 are added for dynamic configuration management. See
2739 https://mail.clusterfs.com/wikis/lustre/MountConf for
2742 Severity : enhancement
2744 Description: dynamic OST addition
2745 Details : OSTs can now be added to a live filesystem
2747 Severity : enhancement
2749 Description: startup order invariance
2750 Details : MDTs and OSTs can be started in any order. Clients only
2751 require the MDT to complete startup.
2753 Severity : enhancement
2755 Description: parallel, asynchronous orphan cleanup
2756 Details : orphan cleanup is now performed in separate threads for each
2757 OST, allowing parallel non-blocking operation.
2759 Severity : enhancement
2761 Description: optimized stripe assignment
2762 Details : stripe assignments are now made based on ost space available,
2763 ost previous usage, and OSS previous usage, in order to try
2764 to optimize storage space and networking resources.
2766 Severity : enhancement
2768 Description: Permanently set tunables
2769 Details : All writable /proc/fs/lustre tunables can now be permanently
2770 set on a per-server basis, at mkfs time or on a live system.
2772 Severity : enhancement
2774 Description: Lustre message v2
2775 Details : Add lustre message format v2.
2777 Severity : enhancement
2779 Description: client OST exclusion list
2780 Details : Clients can be started with a list of OSTs that should be
2781 declared "inactive" for known non-responsive OSTs.
2785 Description: SPEC SFS validation failure on NFS v2 over lustre.
2786 Details : Changes the blocksize for regular files to be 2x RPC size,
2787 and not depend on stripe size.
2789 Severity : enhancement
2791 Description: Multiple MD RPCs in flight.
2792 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2793 To avoid overly-overloading MDS, introduce a limit on number of
2794 MDS RPCs in flight for a single client and add /proc controls
2795 to adjust this limit.
2797 Severity : enhancement
2799 Description: client read/write statistics
2800 Details : Add client read/write call usage stats for performance
2801 analysis of user processes.
2802 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2803 file access. extents_stats shows chunk size distribution.
2804 extents_stats_per_process show chunk size distribution per
2807 Severity : enhancement
2809 Description: mds statistics
2810 Details : Add detailed mds operations statistics in
2811 /proc/fs/lustre/mds/*/stats.
2815 Description: Failure of copying files with lustre special EAs.
2816 Details : Client side always return success for setxattr call for lustre
2817 special xattr (currently only "trusted.lov").
2822 Description: Refcount LNET uuids
2823 Details : The global LNET uuid list grew linearly with every startup;
2824 refcount repeated list entries instead of always adding to
2827 Severity : enhancement
2829 Description: Dynamic service threads
2830 Details : Within a small range, start extra service threads
2831 automatically when the request queue builds up.
2834 Frequency : mixed-endian client/server environments
2836 Description: mixed-endian crashes
2837 Details : The new msg_v2 system had some failures in mixed-endian
2840 Severity : enhancement
2842 Description: Easy OST removal
2843 Details : OSTs can be permanently deactivated with e.g. 'lctl
2844 conf_param lustre-OST0001.osc.active=0'
2846 Severity : enhancement
2848 Description: MGS proc entries
2849 Details : Added basic proc entries for the MGS showing what filesystems
2852 Severity : enhancement
2854 Description: provide MGS failover
2855 Details : Added config lock reacquisition after MGS server failover.
2857 Severity : enhancement
2859 Description: add Linux 2.4 support
2860 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2864 Description: a large application tries to do I/O to the same resource and dies
2865 in the middle of it.
2866 Details : Check the req->rq_arrival time after the call to
2867 ost_brw_lock_get(), but before we do anything about
2868 processing it & sending the BULK transfer request. This
2869 should help move old stale pending locks off the queue as
2870 quickly as obd_timeout.
2873 Frequency : when an incorrect nid is specified during startup
2875 Description: ptlrpc connect to non-existant node causes kernel crash
2876 Details : LNET can't be re-entered from an event callback, which
2877 happened when we expire a message after the export has been
2878 cleaned up. Instead, hand the zombie cleanup off to another
2881 Severity : enhancement
2883 Description: plain/inodebits lock performance improvement
2884 Details : Grouping plain/inodebits in granted list by their request modes
2885 and bits policy, thus improving the performance of search through
2889 Frequency : only if OST filesystem is corrupted
2891 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2892 Details : for a short time RPCs with bulk IO are in the replay list,
2893 but replay of bulk IOs is unimplemented. If the OST filesystem
2894 is corrupted due to disk cache incoherency and then replay is
2895 started it is possible to trip an assertion. Avoid putting
2896 committed RPCs into the replay list at all to avoid this issue.
2899 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2901 Description: System hang on startup
2902 Details : This bug allowed the liblustre (e.g. catamount) client to
2903 return to the app before handling all startup RPCs. This
2904 could leave the node unresponsive to lustre network traffic
2905 and manifested as a server ptllnd timeout.
2907 Severity : enhancement
2909 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2910 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2911 boolean to control whether to print peer debug info when a
2912 client's RPC times out.
2915 Frequency : only for kernels with patches from Lustre below 1.4.3
2917 Description: Remove old rdonly API
2918 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2921 Frequency : only for devices with external journals
2923 Description: Set external device read-only also
2924 Details : During a commanded failover stop, we set the disk device
2925 read-only while the server shuts down. We now also set any
2926 external journal device read-only at the same time.
2929 Frequency : when upgrading from 1.4 while trying to change parameters
2931 Description: The wrong (new) MDC name was used when setting parameters for
2932 upgraded MDT's. Also allows changing of OSC (and MDC)
2933 parameters if --writeconf is specified at tunefs upgrade time.
2936 Frequency : when setting specific ost indicies
2938 Description: QOS code breaks on skipped indicies
2939 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2940 created with --index need not be sequential.
2942 Severity : enhancement
2944 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2945 Details : The uninit_groups feature works in conjunction with the kernel
2946 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2947 up the pass1 processing of e2fsck. This is a read-only feature
2948 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2949 mount filesystems that have had this feature enabled.
2951 Severity : enhancement
2953 Description: Improve multi-block allocation algorithm to avoid fragmentation
2954 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2955 allocation locality and avoid filesystem fragmentation.
2957 ------------------------------------------------------------------------------
2959 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2961 * Support for kernels:
2962 2.4.21-47.0.1.EL (RHEL 3)
2963 2.6.5-7.283 (SLES 9)
2964 2.6.9-42.0.10.EL (RHEL 4)
2965 2.6.12.6 vanilla (kernel.org)
2966 2.6.16.27-0.9 (SLES 10)
2967 * Recommended e2fsprogs version: 1.39.cfs5
2969 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2973 Frequency : occasional, depends on client load and configuration
2974 Bugzilla : 12181, 12203
2975 Description: data loss for recently-modified files
2977 Details : In some cases it is possible that recently written or created
2978 files may not be written to disk in a timely manner (this should
2979 normally be within 30s unless client IO load is very high).
2980 The problem appears as zero-length files or files that are a
2981 multiple of 1MB in size after a client crash or client eviction
2982 that are missing data at the end of the file.
2984 This problem is more likely to be hit on clients where files are
2985 repeatedly created and unlinked in the same directory, clients
2986 have a large amount of RAM, have many CPUs, the filesystem has
2987 many OSTs, the clients are rebooted frequently, and/or the files
2988 are not accessed by other nodes after being written.
2990 The presence of the problem can be detected by looking at
2991 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2992 smaller than the second (nr_unused) then dirty files will not
2993 be flushed automatically to disk. "sync; sleep 10" should be
2994 run several times on the node before unmounting it to update
2995 Lustre (this is also safe to run on nodes without this problem).
2997 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2998 that can cause this same problem, so customers using that kernel
2999 also need to update the kernel in addition to Lustre. In order
3000 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
3002 It is normal that files written just before a client crash (less
3003 than 30s) may not yet have been flushed to disk, even for local
3007 Frequency : frequent on thin XT3 nodes
3009 Description: UUID collision on thin XT3 Linux nodes
3010 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
3011 randomly, since we relied on an insufficiently-seeded PRNG.
3016 Description: OSS hangs after "All ost request buffers busy"
3017 Details : A deadlock between quota and journal operations caused OSS
3018 hangs after printing "All ost request buffers busy."
3021 Frequency : always on liblustre builds
3023 Description: Cleanup compiler warnings on liblustre
3026 Frequency : always on liblustre builds on XT3
3028 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
3032 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
3033 > 32000 subdirectories
3034 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
3035 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
3036 subdirectory count crosses 32000. This will aid e2fsck to
3037 correctly handle more than 32000 subdirectories.
3040 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3042 Description: System hang on startup
3043 Details : This bug allowed the liblustre (e.g. catamount) client to
3044 return to the app before handling all startup RPCs. This
3045 could leave the node unresponsive to lustre network traffic
3046 and manifested as a server ptllnd timeout.
3048 Severity : enhancement
3050 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3051 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
3052 boolean to control whether to print peer debug info when a
3053 client's RPC times out.
3058 Description: make O_SYNC working on 2.6 kernels
3059 Details : 2.6 kernels use different method for mark pages for write,
3060 so need add a code to lustre for O_SYNC work.
3065 Description: Failure to close file and release space on NFS
3066 Details : Put inode details into lock acquired in ll_intent_file_open.
3067 Use mdc_intent_lock in ll_intent_open to properly
3068 detect all kind of errors unhandled by mdc_enqueue.
3073 Description: proc file read during shutdown sometimes raced obd removal,
3075 Details : Add lock to prevent obd access after proc file removal.
3078 Frequency : Only for files larger than 4GB on 32-bit clients.
3080 Description: improperly doing page alignment of locks
3081 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
3082 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
3087 Description: RPCs being resent when they shouldn't be
3088 Details : Some RPCs that should not be resent are being resent. This
3089 can cause inconsistencies in the RPC state machine. Do not
3090 resend such requests.
3093 Frequency : rare, only with NFS export
3095 Description: Crash on NFS re-export node
3096 Details : under very unusual load conditions an assertion is hit in
3097 ll_intent_file_open()
3100 Frequency : only if OST filesystem is corrupted
3102 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3103 Details : for a short time RPCs with bulk IO are in the replay list,
3104 but replay of bulk IOs is unimplemented. If the OST filesystem
3105 is corrupted due to disk cache incoherency and then replay is
3106 started it is possible to trip an assertion. Avoid putting
3107 committed RPCs into the replay list at all to avoid this issue.
3112 Description: large O_DIRECT requests fail under memory pressure/fragmentation
3113 Details : Large single O_DIRECT read and write calls can fail to allocate
3114 a sufficiently large buffer to process the request. In case of
3115 allocation failure the allocation is retried with a smaller
3116 buffer and broken into smaller requests.
3118 Severity : enhancement
3120 Description: Add -o localflock option to simulate old noflock behaviour.
3121 Details : This will achieve local-only flock/fcntl locks coherentness.
3126 Description: versioning check is incomplete
3127 Details : Checking the version difference of client vs. server, report
3128 error if the gap is too big.
3133 Description: add support PG_writeback bit
3134 Details : add support for PG_writeback bit for Lustre, for more carefull
3135 work with page cache in 2.6 kernel. This also fix some deadlocks
3136 and remove hack for work O_SYNC with 2.6 kernel.
3138 Severity : enhancement
3140 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3141 Details : The uninit_groups feature works in conjunction with the kernel
3142 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3143 up the pass1 processing of e2fsck. This is a read-only feature
3144 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3145 mount filesystems that have had this feature enabled.
3147 Severity : enhancement
3149 Description: Improve multi-block allocation algorithm to avoid fragmentation
3150 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3151 allocation locality and avoid filesystem fragmentation.
3153 ------------------------------------------------------------------------------
3155 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
3157 * Support for kernels:
3158 2.6.9-42.0.3.EL (RHEL 4)
3159 2.6.5-7.276 (SLES 9)
3160 2.4.21-47.0.1.EL (RHEL 3)
3161 2.6.12.6 vanilla (kernel.org)
3162 2.6.16.21-0.8 (SLES10)
3163 * Recommended e2fsprogs version: 1.39.cfs2-0
3165 * The backwards-compatible /proc/sys/portals symlink has been removed
3166 in this release. Before upgrading, please ensure that you change
3167 any configuration scripts or /etc/sysctl.conf files that access
3168 /proc/sys/portals/* or sysctl portals.* to use the corresponding
3169 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
3170 in advance of the upgrade on any system running Lustre 1.4.6 or
3171 newer, since /proc/sys/lnet was added in that version.
3172 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3176 Frequency : only when quota is used
3178 Description: avoid scanning export list for quota master
3179 Details : Change the algorithms to avoid scanning export list in order
3180 to improve the efficiency.
3183 Frequency : MDS failover only, very rarely
3185 Description: "went back in time" messages on mds failover
3186 Details : The greatest transno may be lost when the current operation
3187 finishes with an error (transno==0) and the client's last_rcvd
3188 record is over-written. Save the greatest transno in the
3189 mds_last_transno for this case.
3192 Frequency : always for specific kernels and striping counts
3194 Description: client may get "Matching packet too big" without ACL support
3195 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
3196 when trying to access files in certain configurations. The
3197 clients should in fact be denied when mounting because they do
3198 not understand ACLs.
3201 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3203 Description: many clients connecting with IO in progress causes connect timeouts
3204 Details : Avoid synchronous journal commits to avoid delays caused by many
3205 clients connecting/disconnecting when bulk IO is in progress.
3206 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3207 OST_IO_PORTAL to avoid delays behind potentially many pending
3211 Frequency : occasionally with multiple writers to a single file
3213 Description: shared writes to file may result in wrong size reported by stat()
3214 Details : Allow growing of kms when extent lock is cancelled
3217 Frequency : always with random mmap IO to multi-striped file
3219 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3220 Details : Only if the hole is at the end of OST object so that kms is too
3221 small. Fix is to increase kms accordingly in ll_nopage.
3224 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3226 Description: writes to a missing object would leak memory on the OST
3227 Details : If there is an inconsistency between the MDS and OST filesystems,
3228 such that the MDS references an object that doesn't exist, writes
3229 to that object will leak memory due to incorrect cleanup in the
3230 error handling path, eventually running out of memory on the OST.
3235 Description: Creating too long symlink causes lustre errors
3236 Details : Check symlink and name lengths before sending requests to MDS.
3239 Frequency : only if flock is enabled (not on by default)
3241 Description: posix locks not released on fd closure on 2.6.9+
3242 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3243 kernels, this caused such locks not to be released on fd close and
3244 then assertions on fs unmount about still used locks.
3247 Frequency : MDS failover only, very rarely
3249 Description: clients may get ASSERTION(granted_lock != NULL)
3250 Details : When request was taking a long time, and a client was resending
3251 a getattr by name lock request. The were multiple lock requests
3252 with the same client lock handle and
3253 mds_getattr_name->fixup_handle_for_resent_request found one of the
3254 lock handles but later failed with ASSERTION(granted_lock != NULL).
3259 Description: handle->h_buffer_credits > 0, assertion failure
3260 Details : h_buffer_credits is zero after truncate, causing assertion
3261 failure. This patch extends the transaction or creates a new
3265 Frequency : NFS re-export or patchless client
3266 Bugzilla : 11179, 10796
3267 Description: Crash on NFS re-export node (__d_move)
3268 Details : We do not want to hash the dentry if we don't have a lock.
3269 But if this dentry is later used in d_move, we'd hit uninitialised
3270 list head d_hash, so we just do this to init d_hash field but
3271 leave dentry unhashed.
3274 Frequency : NFS re-export or patchless client
3276 Description: NFS exports has problem with symbolic link
3277 Details : lustre client didn't properly install dentry when re-exported
3278 to NFS or running patchless client.
3281 Frequency : NFS re-export or patchless client
3283 Description: Various nfs/patchless fixes.
3284 Details : fixes reuse disconected alias for lookup process - this fixes
3285 warning "find_exported_dentry: npd != pd",
3286 fix permission error with open files at nfs.
3287 fix apply umask when do revalidate.
3290 Frequency : occasional
3292 Description: Crash on NFS re-export node
3293 Details : calling clear_page() on the wrong pointer triggered oops in
3294 generic_mapping_read().
3297 Frequency : rarely, using O_DIRECT IO
3299 Description: unaligned directio crashes client with LASSERT
3300 Details : check for unaligned buffers before trying any requests.
3303 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3305 Description: stale data returned from RAID cache
3306 Details : If only a small amount of IO is done to the RAID device before
3307 reading it again it is possible to get stale data from the RAID
3308 cache instead of reading it from disk.
3311 Frequency : always for sles10 kernel
3313 Description: sles10 support
3314 Details : ll_follow_link: compile fixes and using of nd_set_link
3315 under newer kernels.
3318 Frequency : depends on arch, kernel and compiler version, always on sles10
3321 Description: recursive or deep enough symlinks cause stack overflow
3322 Details : getting rid of large stack-allocated variable in
3326 Frequency : depends on hardware
3328 Description: lustre write performance loss in the SLES10 kernel
3329 Details : the performance loss is caused by using of write barriers in the
3330 ext3 code. The SLES10 kernel turns barrier support on by
3331 default. The fix is to undo that change for ldiskfs.
3333 ------------------------------------------------------------------------------
3335 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3337 * Support for kernels:
3338 2.6.9-42.0.3EL (RHEL 4)
3339 2.6.5-7.276 (SLES 9)
3340 2.4.21-47.0.1.EL (RHEL 3)
3341 2.6.12.6 vanilla (kernel.org)
3345 Frequency : quota enabled and large files being deleted
3347 Description: releasing more than 4GB of quota at once hangs OST
3348 Details : If a user deletes more than 4GB of files on a single OST it
3349 will cause the OST to spin in an infinite loop. Release
3350 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3355 Description: statfs data retrieved from /proc may be stale or zero
3356 Details : When reading per-device statfs data from /proc, in the
3357 {kbytes,files}_{total,free,avail} files, it may appear
3358 as zero or be out of date.
3361 Frequency : systems with MD RAID1 external journal devices
3363 Description: lconf's call to blkid is confused by RAID1 journal devices
3364 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3365 of returning all block devices that match the journal UUID.
3368 Frequency : always, for aggregate stripe size over 4GB
3370 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3371 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3372 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3375 Frequency : always if "lfs find" used on a local file/directory
3377 Description: "lfs find" segfaults if used on a local file/directory
3378 Details : The case where a directory component was not specified wasn't
3379 handled correctly. Handle this properly.
3382 Frequency : always on ppc64
3384 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3385 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3386 perform endianness conversion.
3389 Frequency : rarely (truncate to non-zero file size after write under load)
3390 Bugzilla : 10730, 10687
3391 Description: Files padded with zeros to next 4K multiple
3392 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3393 it is possible that files that are truncated to a non-zero size
3394 immediately after being written are filled with zero bytes beyond
3395 the truncated size. No file data is lost.
3397 Severity : enhancement
3399 Description: Allow recovery/failover for liblustre clients.
3400 Details : liblustre clients were unaware of failover configurations until
3403 Severity : enhancement
3405 Description: user file locks should fail when not mounting with flock option
3406 Details : Set up an error-returning stub in ll_file_operations.lock field
3407 to prevent incorrect behaviour when client is mounted without
3408 flock option. Also, set up properly f_op->flock field for
3412 Frequency : always on ia64
3414 Description: "lfs df" loops on printing out MDS statfs information
3415 Details : The obd_ioctl_data was not initialized and in some systems
3416 this caused a failure during the ioctl that did not return
3417 an error. Initialize the struct and return an error on failure.
3420 Frequency : SLES 9 only
3422 Description: Error of copying files with lustre special EAs as root
3423 Details : Client side always return success for setxattr call for lustre
3424 special xattr (currently only "trusted.lov").
3427 Frequency : rarely on clusters with both ia64+i386 clients
3429 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3430 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3431 clients are doing concurrent writes to the same file it is
3432 possible that the ia64 clients may LASSERT because the OST
3433 extent locks are not PAGE_SIZE aligned. Ensure that grown
3434 locks are always aligned on the request boundary.
3437 Frequency : specific use, occasional
3439 Description: Overwriting in use executable truncates on-disk binary image
3440 Details : If one node attempts to overwrite an executable in use by
3441 another node, we now correctly return ETXTBSY instead of
3442 truncating the file.
3444 Severity : enhancement
3446 Description: Async OSC create to avoid the blocking unnecessarily.
3447 Details : If a OST has no remain object, system will block on the creating
3448 when need to create a new object on this OST. Now, ways use
3449 pre-created objects when available, instead of blocking on an
3450 empty osc while others are not empty. If we must block, we block
3451 for the shortest possible period of time.
3456 Description: chmod on Lustre root is propagated to other clients
3457 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3458 invalid by the follow_mount time.
3461 Frequency : liblustre clients only
3463 Description: Race in 'instant cancel' lock handling could lead to such locks
3464 never to be granted in case of SMP MDS
3465 Details : Do not destroy not yet granted but cbpending locks in
3469 Frequency : replay/resend of open
3471 Description: non null lock assetion failure in mds_intent_policy
3472 Details : Trying to replay/resend lockless open requests resulted in
3473 mds_open() returning 0 with no lock. Now it sets a flag if
3474 a lock is going to be returned.
3476 Severity : enhancement
3478 Description: Checksum enhancements
3479 Details : New checksum enhancements allow for resending RPCs that failed
3482 Severity : enhancement
3484 Description: Tunables on number of dirty pages in cacche
3485 Details : Allow to set limit on number of dirty pages cached.
3490 Description: client crash on unmount - lock still has references
3491 Details : In some error handling cases it was possible to leak a lock
3492 reference on a client while accessing a file. This was not
3493 harmful to the client during operation, but would cause the
3494 client to crash when the filesystem is unmounted.
3497 Frequency : specific case, rare
3499 Description: ETXTBSY on mds though file not in use by client
3500 Details : ETXTBSY is no longer incorrectly returned when attempting to
3501 chmod or chown a directory that the user previously tried to
3502 execute or a currently-executing binary.
3505 Frequency : extremely rare except on liblustre-based clients
3507 Description: Lustre space not freed when files are deleted
3508 Details : Clean up open-unlinked files after client eviction. Previously
3509 the unlink was skipped and the files remained as orphans.
3514 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3515 Details : In some cases it was possible to send a blocking callback to
3516 a client doing a glimpse, even though that client didn't get
3517 a lock granted. When the glimpse lock is cancelled on the OST
3518 the freed lock is left on the waiting list and corrupted the list.
3521 Frequency : all core dumps
3523 Description: Broke core dumps to lustre
3524 Details : Negative dentry may be unhashed if parent does not have UPDATE
3525 lock, but some callers, e.g. do_coredump, expect dentry to be
3526 hashed after successful create, hash it in ll_create_it.
3528 ------------------------------------------------------------------------------
3530 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3532 * Support for kernels:
3533 2.6.9-42.0.2.EL (RHEL 4)
3534 2.6.5-7.276 (SLES 9)
3535 2.4.21-40.EL (RHEL 3)
3536 2.6.12.6 vanilla (kernel.org)
3540 Frequency : always on RHEL 3
3542 Description: Number of open files grows over time
3543 Details : The number of open files grows over time, whether or not
3544 Lustre is started. This was due to a filp leak introduced
3545 by one of our kernel patches.
3547 ------------------------------------------------------------------------------
3549 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3551 * Support for kernels:
3552 2.6.9-42.EL (RHEL 4)
3553 2.6.5-7.276 (SLES 9)
3554 2.4.21-40.EL (RHEL 3)
3555 2.6.12.6 vanilla (kernel.org)
3560 Bugzilla : 5719, 9635, 9792, 9684
3561 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3562 Details : If a server is under heavy load and cannot reply to new
3563 connection requests before the client resends the (re)connect,
3564 the connection handling code can behave badly if two service
3565 threads are concurrently handing separate (re)connections from
3566 the same client. Add better locking to the connection handling
3567 code, and ensure that only a single connection will be processed
3568 for a given client UUID, even if the lock is dropped.
3570 Severity : enhancement
3572 Description: add TCP zero-copy support to kernel
3573 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3574 sends if the hardware supports scatter-gather and checksumming.
3575 This allows socklnd to do client-write and server-read more
3576 efficiently and reduce CPU utilization from skbuf copying.
3579 Frequency : only if NFS exporting from client
3581 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3582 Details : If setting attributes on a file created under NFS that had
3583 never been opened it would be possible to oops the client
3584 if the file had no objects.
3587 Frequency : always for liblustre
3589 Description: liblustre client does MDS+OSTs setattr RPC for each write
3590 Details : When doing a write from a liblustre client, the client
3591 incorrectly issued an RPC to the MDS and each OST the file was
3592 striped over in order to update the timestamps. When writing
3593 with small chunks and many clients this could overwhelm the MDS
3594 with RPCs. In all cases it would slow down the write because
3595 these RPCs are unnecessary.
3597 Severity : enhancement
3599 Description: allow number of MDS service threads to be changed at module load
3600 Details : It is now possible to change the number of MDS service threads
3601 running. Adding "options mds mds_num_threads={N}" to the MDS's
3602 /etc/modprobe.conf will set the number of threads for the next
3603 time Lustre is restarted (assuming the "mds" module is also
3604 reloaded at that time). The default number of threads will
3605 stay the same, 32 for most systems.
3610 Description: OST crash if filesystem is unformatted or corrupt
3611 Details : If an OST is started on a device that has never been formatted
3612 or if the filesystem is corrupt and cannot even mount then the
3613 error handling cleanup routines would dereference a NULL pointer.
3618 Description: NULL pointer deref in llap_from_page.
3619 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3620 incorrect) mapping if the page was truncated/reclaimed while it was
3621 searched for. Check for this condition and skip such pages when
3622 doing readahead. Introduce extra check to llap_from_page() to
3623 verify page->mapping->host is non-NULL (so page is not anonymous).
3626 Frequency : Sometimes when using sys_sendfile
3628 Description: "page not covered by a lock" warnings from ll_readpage
3629 Details : sendfile called ll_readpage without right page locks present.
3630 Now we introduced ll_file_sendfile that does necessary locking
3631 around call to generic_file_sendfile() much like we do in
3635 Frequency : with certain MDS communication failures at client mount time
3637 Description: NULL pointer deref after failed client mount
3638 Details : a client connection request may delayed by the network layer
3639 and not be sent until after the PTLRPC layer has timed out the
3640 request. If the client fails the mount immediately it will try
3641 to clean up before the network times out the request. Add a
3642 reference from the request import to the obd device and delay
3643 the cleanup until the network drops the request.
3646 Frequency : occasionally during client (re)connect
3648 Description: assertion failure during client (re)connect
3649 Details : processing a client connection request may be delayed by the
3650 client or server longer than the client connect timeout. This
3651 causes the client to resend the connection request. If the
3652 original connection request is replied in this interval, the
3653 client may trip an assertion failure in ptlrpc_connect_interpret()
3654 which thought it would be the only running connect process.
3657 Frequency : only with obd_echo servers and clients that are rebooted
3659 Description: kernel BUG accessing uninitialized data structure
3660 Details : When running an obd_echo server it did not start the ping_evictor
3661 thread, and when a client was evicted an uninitialized data
3662 structure was accessed. Start the ping_evictor in the RPC
3663 service startup instead of the OBD startup.
3665 Severity : enhancement
3666 Bugzilla : 10193 (patchless)
3667 Description: Remove dependency on various unexported kernel interfaces.
3668 Details : No longer need reparent_to_init, exit_mm, exit_files,
3669 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3672 Frequency : rare (only users of deprecated and unsupported LDAP config)
3674 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3675 Details : LDAP apparently contains 'lustreName' attributes instead of
3676 'name'. A simple remapping of the name is sufficient.
3679 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3681 Description: waiting_locks_callback trips kernel BUG if client is evicted
3682 Details : Running with the dump_on_timeout debug flag turned on makes
3683 it possible that the waiting_locks_callback() can try to dump
3684 the Lustre kernel debug logs from an interrupt handler. Defer
3685 this log dumping to the expired_lock_main() thread.
3687 Severity : enhancement
3689 Description: Support NFS exporting on 2.6 kernels.
3690 Details : Implement non-rawops metadata methods for NFS server to use without
3691 changing NFS server code.
3694 Frequency : very rare (synthetic metadata workload only)
3696 Description: two racing renames might cause an MDS thread to deadlock
3697 Details : Running the "racer" program may cause one MDS thread to rename
3698 a file from being the source of a rename to being the target of
3699 a rename at exactly the same time that another thread is doing
3700 so, and the second thread has already enqueued these locks after
3701 doing a lookup of the target and is trying to relock them in
3702 order. Ensure that we don't try to re-lock the same resource.
3705 Frequency : only very large systems with liblustre clients
3707 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3708 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3709 all OSTs in parallel. This allows the eviction of stale liblustre
3710 clients to proceed much faster than if they were done in series,
3711 and also offers similar improvements for other set_info RPCs.
3716 Description: excessive CPU usage during initial read phase on client
3717 Details : During the initial read phase on a client, it would agressively
3718 retry readahead on the file, consuming too much CPU and impacting
3719 performance (since 1.4.5.8). Improve the readahead algorithm
3720 to avoid this, and also improve some other common cases (read
3721 of small files in particular, where "small" is files smaller than
3722 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3727 Description: MDS crash when receiving packet with unknown intent.
3728 Details : Do not LBUG in unknown intent case, just return -EFAULT
3730 Severity : enhancement
3731 Bugzilla : 9293, 9385
3732 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3733 Details : Do not serialize getattr (non-intent version) and statfs.
3736 Frequency : occasional, when OST network is overloaded/intermittent
3738 Description: client evicted by OST after bulk IO timeout
3739 Details : If a client sends a bulk IO request (read or write) the OST
3740 may evict the client if it is unresposive to its data GET/PUT
3741 request. This is incorrect if the network is overloaded (takes
3742 too long to transfer the RPC data) or dropped the OST GET/PUT
3743 request. There is no need to evict the client at all, since
3744 the pinger and/or lock callbacks will handle this, and the
3745 client can restart the bulk request.
3748 Frequency : Always when mmapping file with no objects
3750 Description: client crashes when mmapping file with no objects
3751 Details : Check that we actually have objects in a file before doing any
3752 operations on objects in ll_vm_open, ll_vm_close and
3758 Description: Request leak when working with deleted CWD
3759 Details : Introduce advanced request refcount tracking for requests
3760 referenced from lustre intent.
3762 Severity : Enhancement
3764 Description: Cache open file handles on client.
3765 Details : MDS now will return special lock along with openhandle, if
3766 requested and client is allowed to hold openhandle, even if unused,
3767 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3768 closing files for every read/write openration.
3770 Severity : Enhancement
3772 Description: Cache open negative dentries on client when possible.
3773 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3774 negative dentries on lock revocation.
3779 Description: Remounting a client read-only wasn't possible with a zconf mount
3780 Details : It wasn't possible to remount a client read-only with llmount.
3782 Severity : enhancement
3783 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3784 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3785 MPICH is included to add Lustre-specific ADIO interfaces.
3786 This is based closely on the UFS ADIO layer and only differs
3787 in file creation, in order to allow the OST striping to be set.
3788 This is user-contributed code and not supported by CFS.
3793 Description: extended inode attributes (immutable, append-only) work improperly
3794 when 2.4 and 2.6 kernels are used on client/server or vice versa
3795 Details : Introduce kernel-independent values for these flags.
3797 Severity : enhancement
3800 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3801 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3802 granularity. Now, Fractional MB granularity is be supported,
3803 this is very useful for low memory system.
3805 Severity : enhancement
3807 Description: Getattr by fid
3808 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3809 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3810 have locks and avoids getattr rpc requests for it on every stat.
3813 Frequency : Always, for filesystems larger than 2TB
3815 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3816 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3817 filesystems larger than 2TB to be mounted with mballoc enabled.
3820 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3822 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3823 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3824 filesystems larger than 2TB is mounted then the kernel will
3825 silently corrupt the start of the filesystem. CONFIG_LBD is
3826 enabled for all CFS-supported kernels, but the possibility of
3827 this happening with a modified kernel config exists.
3829 Severity : enhancement
3831 Description: add client O_DIRECT support for 2.6 kernels
3832 Details : It is now possible to do O_DIRECT reads and writes to files
3833 in the Lustre client mountpoint on 2.6 kernel clients.
3835 Severity : enhancement
3837 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3838 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3839 parallel, not waiting for response from every OST before sending
3840 a rpc to the next OST.
3845 Description: setattr vs write race when updating file timestamps
3846 Details : Client processes that update a file timestamp into the past
3847 right after writing to the file (e.g. tar) it is possible that
3848 the updated file modification time can be reset to the current
3849 time due to a race between processing the setattr and write RPC.
3851 Severity : enhancement
3853 Description: Bring 'lfs find' closer in line with regular Linux find.
3854 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3855 -print0 options and obtains all the needed info through the lustre
3858 Severity : enhancement
3860 Description: support up to 1024 configured devices on one node
3861 Details : change obd_dev array from statically allocated to dynamically
3862 allocated structs as they are first used to reduce memory usage
3867 Description: Flush dirty partially truncated pages during truncate
3868 Details : Immediatelly flush partially truncated pages in filter_setattr,
3869 this way we completely avoid having any pages in page cache on OST
3870 and can retire ugly workarounds during writes to flush such pages.
3875 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3876 Details : protect inode from truncation within vfs_unlink() context
3877 just take a reference before calling vfs_unlink() and release it
3878 when parent's i_sem is free.
3883 Description: last_id value checked outside lock on OST caused LASSERT failure
3884 Details : If there were multiple MDS->OST object precreate requests in
3885 flight, it was possible that the OST's last object id was checked
3886 outside a lock and incorrectly tripped an assertion. Move checks
3887 inside locks, and discard old precreate requests.
3890 Frequency : always, if extents are used on OSTs
3892 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3893 Details : OSTs using the extents format would not zero the high 16 bits of
3894 the index physical block number. This is not a problem for any
3895 OST filesystems smaller than 16TB, and no kernels support ext3
3896 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3897 new/modified files) and can be fixed for existing filesystems
3898 with e2fsprogs-1.39-cfs1.
3903 Description: import connection selection may be incorrect if timer wraps
3904 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3905 import connections to be ignored if the 32-bit jiffies counter
3906 wraps. Use a 64-bit jiffies counter.
3909 Frequency : very large clusters immediately after boot
3911 Description: LNET request buffers exhausted under heavy short-term load
3912 Details : If a large number of client requests are generated on a service
3913 that has previously never seen so many requests it is possible
3914 that the request buffer growth cannot keep up with the spike in
3915 demand. Instead of dropping incoming requests, they are held in
3916 the LND until the RPC service can accept more requests.
3919 Frequency : Sometimes during replay
3921 Description: Assertion failure in ll_local_open after replay.
3922 Details : If replay happened on an open request reply before we were able
3923 to set replay handler, reply will become not swabbed tripping the
3924 assertion in ll_local_open. Now we set the handler right after
3925 recognising of open request
3928 Frequency : very rare
3930 Description: kernel reports "badness in vsnprintf"
3931 Details : Reading from the "recovery_status" /proc file in small chunks
3932 may cause a negative length in lprocfs_obd_rd_recovery_status()
3933 call to vsnprintf() (which is otherwise harmless). Exit early
3934 if there is no more space in the output buffer.
3936 Severity : enhancement
3938 Description: clear OBD RPC statistics by writing to them
3939 Details : It is now possible to clear the OBD RPC statistics by writing
3940 to the "stats" file.
3945 Description: Client mtime is not the same on different clients after utimes
3946 Details : In some cases, the client was using the utimes() syscall on
3947 a file cached on another node. The clients now validate the
3948 ctime from the MDS + OSTs to determine which one is right.
3953 Description: Inability to activate failout mode
3954 Details : lconf script incorrectly assumed that in pythong string's numeric
3955 value is used in comparisons.
3958 Frequency : always with multiple stripes per file
3960 Description: Inefficient object allocation for mutli-stripe files
3961 Details : When selecting which OSTs to stripe files over, for files with
3962 a stripe count that divides evenly into the number of OSTs,
3963 the MDS is always picking the same starting OST for each file.
3964 Return the OST selection heuristic to the original design.
3969 Description: mount failures may take full timeout to return an error
3970 Details : Under some heavy load conditions it is possible that a
3971 failed mount can wait for the full obd_timeout interval,
3972 possibly several minutes, before reporting an error.
3973 Instead return an error as soon as the status is known.
3975 Frequency : quota enabled and large files being deleted
3977 Description: releasing more than 4GB of quota at once hangs OST
3978 Details : If a user deletes more than 4GB of files on a single OST it
3979 will cause the OST to spin in an infinite loop. Release
3980 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3985 Description: statfs data retrieved from /proc may be stale or zero
3986 Details : When reading per-device statfs data from /proc, in the
3987 {kbytes,files}_{total,free,avail} files, it may appear
3988 as zero or be out of date.
3991 Frequency : systems with MD RAID1 external journal devices
3993 Description: lconf's call to blkid is confused by RAID1 journal devices
3994 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3995 of returning all block devices that match the journal UUID.
3998 Frequency : always, for aggregate stripe size over 4GB
4000 Description: assertion fails when trying to use 4GB stripe size
4001 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
4002 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
4005 Frequency : always on ppc64
4007 Description: the first write on an ext3 filesystem with mballoc got stuck
4008 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
4009 perform endianness conversion.
4011 ------------------------------------------------------------------------------
4013 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
4015 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
4016 INTEROPERATE with older versions automatically. Please read the
4017 user documentation before upgrading any part of a live system.
4018 * WARNING: Lustre networking configuration changes are required with
4019 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
4022 * Support for kernels:
4023 2.6.9-22.0.2.EL (RHEL 4)
4024 2.6.5-7.244 (SLES 9)
4025 2.6.12.6 vanilla (kernel.org)
4028 Severity : enhancement
4029 Bugzilla : 7981/8208
4030 Description: Introduced Lustre Networking (LNET)
4031 Details : LNET is new networking infrastructure for Lustre, it includes
4032 a reorganized network configuration mode (see the user
4033 documentation for full details) as well as support for routing
4034 between different network fabrics. Lustre Networking Devices
4035 (LNDs) for the supported network fabrics have also been
4036 created for this new infrastructure.
4038 Severity : enhancement
4039 Description: Introduced Access control lists
4040 Details : clients can set ACLs on files and directories in order to have
4041 more fine-grained permissions than the standard Unix UGO+RWX.
4042 The MDS must be started with the "-o acl" mount option.
4044 Severity : enhancement
4045 Description: Introduced filesystem quotas
4046 Details : Administrators may now establish per-user quotas on the
4049 Severity : enhancement
4051 Description: Configuration change for the XT3
4052 The PTLLND is now used to run Lustre over Portals on the XT3
4053 The configure option(s) --with-cray-portals are no longer used.
4054 Rather --with-portals=<path-to-portals-includes> is used to
4055 enable building on the XT3. In addition to enable XT3 specific
4056 features the option --enable-cray-xt3 must be used.
4061 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
4062 Details : A race between allocating a new llcd and re-getting the llcd_lock
4063 allowed another thread to grab newly-allocated llcd.
4065 Severity : enhancement
4067 Description: 2.6 OST async journal commit and locking fix to improve performance
4068 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
4069 2.6 kernels are now async as they already were in 2.4 kernels so
4070 that they can commit concurrently with the network bulk transfer.
4071 For block-allocated files the filter allocation semaphore is held
4072 to avoid filesystem fragmentation during allocation. BKL lock
4073 removed for 2.6 xattr operations where it is no longer needed.
4078 Description: lconf incorrectly determined whether two IP networks could talk
4079 Details : In some more complicated routing and multiple-network
4080 configurations, lconf will avoid trying to make a network
4081 connection to a disjoint part of the IP space. It was doing the
4082 math incorrectly for one set of cases.
4087 Description: Fix for potential infinite loop processing records in an llog.
4088 Details : If an llog record is corrupted/zeroed, it is possible to loop
4089 forever in llog_process(). Validate the llog record length
4090 and skip the remainder of the block on error.
4093 Frequency : occasional (liblustre only)
4095 Description: liblustre could not open files whose last component is a symlink
4096 Details : sysio_path_walk() would incorrectly pass the open intent to
4097 intermediate path components.
4100 Frequency : rare (liblustre only with non-standard tuning)
4101 Bugzilla : 7201 (7350)
4102 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
4103 Details : llu_lookup_finish_locks() tries to set lock data on a lock
4104 after it has been released, only do this for referenced locks
4106 Severity : enhancement
4108 Description: specifying an (invalid) directory default stripe_size of -1
4109 would reset the directory default striping
4110 Details : stripe_size -1 was used internally to signal directory stripe
4111 removal, now use "all default" to signal dir stripe removal
4112 as a directory striping of "all default" is not useful
4115 Frequency : common for large clusters running liblustre clients
4117 Description: doing an ls when liblustre clients are running is slow
4118 Details : sending a glimpse AST to a liblustre client waits for every AST
4119 to time out, as liblustre clients will not respond. Since they
4120 cannot cache data we refresh the OST lock LVB from disk instead.
4122 Severity : enhancement
4124 Description: doing an ls at the same time as file IO can be slow
4125 Details : enqueue and other "small" requests can be blocked behind many
4126 large IO requests. Create a new OST IO portal for non-IO
4127 requests so they can be processed faster.
4130 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
4132 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
4133 created file with mode 000
4134 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
4135 allow an ftruncate() as a non-root user to a file with mode 000.
4136 Lustre now respects this flag to disable mode checks when
4137 truncating a file owned by the user
4140 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
4143 Description: Revoking locks from clients that went dead or catatonic might take
4145 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
4146 cancellation of such locks instant on servers without waiting for
4147 any reply from clients. Clients drops these locks when cancel
4148 notification from server is received without replying.
4151 Frequency : liblustre-only, when liblustre client dies or becomes busy
4153 Description: Doing ls on Linux clients can take a long time with active
4155 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
4156 granting such locks to it in the first place if possible. Locks
4157 are taken by proxy on the OST during the read or write and
4158 dropped immediately afterward. Add connect flags handling, do
4159 not grant locks to liblustre clients for glimpse ASTs.
4161 Severity : enhancement
4163 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
4164 Details : Existing read-ahead algorithm is tuned for the case of streamlined
4165 sequential reads and behaves badly with applications doing random
4166 reads. Improve it by reading ahead at least read region, and
4167 avoiding excessive large RPC for small reads.
4169 Severity : enhancement
4171 Description: Creating more than 1000 files for a single job may cause a load
4172 imbalance on the OSTs if there are also a large number of OSTs.
4173 Details : qos_prep_create() uses an OST index reseed value that is an
4174 even multiple of the number of available OSTs so that if the
4175 reseed happens in the middle of the object allocation it will
4176 still utilize the OSTs as uniformly as possible.
4181 Description: OST or MDS may oops in ping_evictor_main()
4182 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
4183 but doesn't restart at beginning of obd_exports_timed list
4186 Severity : enhancement
4188 Description: improve by-nid export eviction on the MDS and OST
4189 Details : allow multiple exports with the same NID to be evicted at one
4190 time without re-searching the exports list.
4193 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
4195 Description: MDS may oops in groups_free()
4196 Details : in rare race conditions a newly allocated group_info struct is
4197 freed again, and this can be NULL. The 2.4 compatibility code
4198 for groups_free() checked for a NULL pointer, but 2.6 did not.
4201 Frequency : common for liblustre clients doing little filesystem IO
4202 Bugzilla : 9352, 7313
4203 Description: server may evict liblustre clients accessing contended locks
4204 Details : if a client is granted a lock or receives a completion AST
4205 with a blocking AST already set it would not reply to the AST
4206 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4209 Frequency : lfs setstripe, only systems with more than 160 OSTs
4211 Description: unable to set striping with a starting offset beyond OST 160
4212 Details : llapi_create_file() incorrectly limited the starting stripe
4213 index to the maximum single-file stripe count.
4216 Frequency : LDAP users only
4218 Description: lconf did not handle in-kernel recovery with LDAP properly
4219 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4221 Severity : enhancement
4223 Description: bind OST threads to NUMA nodes to improve performance
4224 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4225 node and do their allocations there to localize memory access
4227 Severity : enhancement
4229 Description: llmount can determine client NID directly from Myrinet (GM)
4230 Details : the client NID code from gmnalnid was moved directly into
4231 llmount, removing the need to use this or specifying the
4232 client NID explicitly when mounting GM clients with zeroconf
4235 Frequency : if client is started with down MDS
4237 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4238 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4239 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4240 ptlrpc_import_delayed_req() because we didn't check if the
4241 request was interrupted, and we also didn't break out of the
4242 event loop if there was no timeout
4247 Description: data loss during non-page-aligned writes to a single file from
4248 both multiple nodes and multiple threads on one node at same time
4249 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4250 inconsistency led to false short-reads, that were cached and later
4251 used by ->prepare_write() to fill in partially written page,
4252 leading to data loss.
4255 Frequency : always, if lconf --abort_recovery used
4257 Description: lconf --abort_recovery fails with 'Operation not supported'
4258 Details : lconf was attempting to abort recovery on the MDT device and not
4261 Severity : enhancement
4263 Description: remove cleanup logs
4264 Details : replace lconf-generated cleanup logs with lustre internal
4265 cleanup routines. Eliminates the need for client-cleanup and
4268 Severity : enhancement
4270 Description: add support for EAs (user and system) on lustre filesystems
4271 Details : it is now possible to store extended attributes in the Lustre
4272 client filesystem, and with the user_xattr mount option it
4273 is possible to allow users to store EAs on their files also
4275 Severity : enhancement
4277 Description: Add possibility (config option) to show minimal available OST free
4279 Details : When compiled with --enable-mindf configure option, statfs(2)
4280 (and so, df) will return least minimal free space available from
4281 all OSTs as amount of free space on FS, instead of summary of
4282 free spaces of all OSTs.
4284 Severity : enhancement
4286 Description: do not expand extent locks acquired on OST-side
4287 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4288 by server: they are not cached anyway.
4291 Frequency : when mmap is used/binaries executed from Lustre
4293 Description: Unmmap pages before throwing them away from read cache.
4294 Details : llap_shrink cache now attempts to unmap pages before discarding
4295 them (if unmapping failed - do not discard). SLES9 kernel has
4296 extra checks that trigger if this unmapping is not done first.
4301 Description: lconf didn't resolve symlinks before checking to see whether a
4302 given mountpoint was already in use
4305 Frequency : when migrating failover services
4306 Bugzilla : 6395, 9514
4307 Description: When migrating a subset of services from a node (e.g. failback
4308 from a failover service node) the remaining services would
4309 time out and evict clients.
4310 Details : lconf --force (implied by --failover) sets the global obd_timeout
4311 to 5 seconds in order to quickly disconnect, but this caused
4312 other RPCs to time out too quickly. Do not change the global
4313 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4315 Severity : enhancement
4316 Frequency : if MDS is started with down OST
4317 Bugzilla : 9439,5706
4318 Description: Allow startup/shutdown of an MDS without depending on the
4319 availability of the OSTs.
4320 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4321 Add appropriate locking and lov-osc refcounts for safe
4322 cleaning. Add osc abort_inflight calls in case the
4323 synchronize never started.
4326 Frequency : occasional (Cray XT3 only)
4328 Description: root not authorized to access files in CRAY_PORTALS environment
4329 Details : The client process capabilities were not honoured on the MDS in
4330 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4331 been accessed by an authorized user then root was able to access
4332 the file on the local client also. The root user capabilities
4333 are now allowed on the MDS, as this environment has secure UID.
4336 Frequency : occasional
4338 Description: ldiskfs "too long searching" message happens too often
4339 Details : A debugging message (otherwise harmless) prints too often on
4340 the OST console. This has been reduced to only happen when
4341 there are fragmentation problems on the filesystem.
4346 Description: Division by zero in statfs when all OSCs are inactive
4347 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4348 lov_statfs divides by value returned by lov_get_stripecnt().
4352 Bugzilla : 9489, 3273
4353 Description: First write from each client to each OST was only 4kB in size,
4354 to initialize client writeback cache, which caused sub-optimal
4355 RPCs and poor layout on disk for the first writen file.
4356 Details : Clients now request an initial cache grant at (re)connect time
4357 and so that they can start streaming writes to the cache right
4358 away and always do full-sized RPCs if there is enough data.
4359 If the OST is rebooted the client also re-establishes its grant
4360 so that client cached writes will be honoured under the grant.
4365 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4366 Details : Now I/O RPCs go to different portal number and (presumably) fast
4367 lock requests (and glimses) and other RPCs get their own service
4368 threads pool that should be able to service those RPCs
4371 Severity : enhancement
4373 Description: Ability to exchange lustre version between client and servers and
4374 issue warnings at client side if client is too old. Also for
4375 liblustre clients there is ability to refuse connection of too old
4377 Details : New 'version' field is added to connect data structure that is
4378 filled with version info. That info is later checked by server and
4382 Frequency : rare, liblustre only.
4383 Bugzilla : 9296, 9581
4384 Description: Two simultaneous writes from liblustre at offset within same page
4385 might proceed at the same time overwriting eachother with stale
4387 Details : I/O lock withing llu_file_prwv was released too early, before data
4388 actually was hitting the wire. Extended lock-holding time until
4389 server acknowledges receiving data.
4392 Frequency : extremely rare. Never observed in practice.
4394 Description: avoid generating lustre_handle cookie of 0.
4395 Details : class_handle_hash() generates handle cookies by incrementing
4396 global counter, and can hit 0 occasionaly (this is unlikely, but
4397 not impossible, because initial value of cookie counter is
4398 selected randonly). Value of 0 is used as a sentinel meaning
4399 "unassigned handle" --- avoid it. Also coalesce two critical
4400 sections in this function into one.
4402 Severity : enhancement
4404 Description: allow liblustre clients to delegate truncate locking to OST
4405 Details : To avoid overhead of locking, liblustre client instructs OST to
4406 take extent lock in ost_punch() on client's behalf. New connection
4407 flag is added to handle backward compatibility.
4409 Severity : enhancement
4410 Bugzilla : 4928, 7341, 9758
4411 Description: allow number of OST service threads to be specified
4412 Details : a module parameter allows the number of OST service threads
4413 to be specified via "options ost ost_num_threads={N}" in the
4414 OSS's /etc/modules.conf or /etc/modprobe.conf.
4418 Bugzilla : 6146, 9635, 9895
4419 Description: servers crash with bad pointer in target_handle_connect()
4420 Details : In rare cases when a client is reconnecting it was possible that
4421 the connection request was the last reference for that export.
4422 We would temporarily drop the export reference and get a new
4423 one, but this may have been the last reference and the export
4424 was just destroyed. Get new reference before dropping old one.
4426 Severity : enhancement
4427 Frequency : if client is started with failover MDS
4429 Description: Allow multiple MDS hostnames in the mount command
4430 Details : Try to read the configuration from all specified MDS
4431 hostnames during a client mount in case the "primary"
4434 Severity : enhancement
4436 Description: Stop sending data to evicted clients as soon as possible.
4437 Details : Check if the client we are about to send or are sending data to
4438 was evicted already. (Check is done every second of waiting,
4439 for which l_wait_event interface was extended to allow checking
4440 of exit condition at specified intervals).
4443 Frequency : rare, normally only when NFS exporting is done from client
4445 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4447 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4450 Frequency : always, if srand() is called before liblustre initialization
4452 Description: Liblustre uses system PRNG disturbing its usage by user application
4453 Details : Introduce internal to lustre fast and high-quality PRNG for
4454 lustre usage and make liblustre and some other places in generic
4455 lustre code to use it.
4457 Severity : enhancement
4458 Bugzilla : 9477, 9557, 9870
4459 Description: Verify that the MDS configuration logs are updated when xml is
4460 Details : Check if the .xml configuration logs are newer than the config
4461 logs stored on the MDS and report an error if this is the case.
4462 Request --write-conf, or allow starting with --old_conf.
4464 Severity : enhancement
4466 Description: Handle symlinks in the path when checking if Lustre is mounted.
4467 Details : Resolve intermediate symlinks when checking if a client has
4468 mounted a filesystem to avoid duplicate client mounts.
4473 Description: lconf can hit an error exception but still return success.
4474 Details : The lconf command catches the Command error exception at the top
4475 level script context and will exit with the associated exit
4476 status, but doesn't ensure that this exit status is non-zero.
4481 Description: failure of ptlrpc thread startup can cause oops
4482 Details : Starting a ptlrpc service thread can fail if there are a large
4483 number of threads or the server memory is very fragmented.
4484 Handle this without oopsing.
4487 Frequency : always, only if liblustre and non-default acceptor port was used
4489 Description: liblustre cannot connect to servers with non-default acceptor port
4490 Details : tcpnal_set_default_params() was not called and was therefore
4491 ignoring the environment varaible TCPNAL_PORT, as well as other
4492 TCPNAL_ environment variables
4497 Description: two objects could be created on the same OST for a single file
4498 Details : If an OST is down, in some cases it was possible to create two
4499 objects on a single OST for a single file. No problems other
4500 than potential performance impact and spurious error messages.
4504 Bugzilla : 5681, 9562
4505 Description: Client may oops in ll_unhash_aliases
4506 Details : Client dcache may become inconsistent in race condition.
4507 In some cases "getcwd" can fail if the current directory is
4513 Description: Inode refcounting problems in NFS export code
4514 Details : link_raw functions used to call d_instantiate without obtaining
4515 extra inode reference first.
4519 Bugzilla : 9942, 9903
4520 Description: Referencing freed requests leading to crash, memleaks with NFS.
4521 Details : We used to require that call to ll_revalidate_it was always
4522 followed by ll_lookup_it. Also with revalidate_special() it is
4523 possible to call ll_revalidate_it() twice for the same dentry
4524 even if first occurence returned success. This fix changes semantic
4525 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4526 reference on a request referred from the intent.
4527 ll_intent_release() then releases such a request.
4530 Frequency : rare, normally benchmark loads only
4532 Description: unlinked inodes were kept in memory on the client
4533 Details : If a client is repeatedly creating and unlinking files it
4534 can accumulate a lot of stale inodes in the inode slab cache.
4535 If there is no other client load running this can cause the
4536 client node to run out of memory. Instead flush old inodes
4537 from client cache that have the same inode number as a new inode.
4540 Frequency : SLES9 2.6.5 kernel and long filenames only
4541 Bugzilla : 9969, 10379
4542 Description: utime reports stale NFS file handle
4543 Details : SLES9 uses out-of-dentry names in some cases, which confused
4544 the lustre dentry revalidation. Change it to always use the
4548 Frequency : rare, unless heavy write-truncate concurrency is continuous
4549 Bugzilla : 4180, 6984, 7171, 9963, 9331
4550 Description: OST becomes very slow and/or deadlocked during object unlink
4551 Details : filter_destroy() was holding onto the parent directory lock
4552 while truncating+unlinking objects. For very large objects this
4553 may block other threads for a long time and slow overall OST
4554 responsiveness. It may also be possible to get a lock ordering
4555 deadlock in this case, or run out of journal credits because of
4556 the combined truncate+unlink. Solution is to do object truncate
4557 first in one transaction without parent lock, and then do the
4558 final unlink in a new transaction with the parent lock. This
4559 reduces the lock hold time dramatically.
4562 Frequency : rare, 2.4 kernels only
4564 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4565 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4566 before unmounting their filesystem in order to avoid deadlock.
4567 The kernel_locked() function in 2.4 kernels only checks whether
4568 the kernel lock is held, not whether it is this process that is
4569 holding it as 2.6 kernels do.
4574 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4575 Details : The client ptlrpc code may be trying to reconnect to a down
4576 server before a previous connection attempt has timed out.
4577 Increase the reconnect interval to be longer than the connection
4578 timeout interval to avoid sending duplicate connections to
4582 Frequency : echo_client brw_test command
4584 Description: fix echo_client to work with OST preallocated code
4585 Details : OST preallocation code (5137) didn't take echo_client IO path
4586 into account: echo_client calls filter methods outside of any
4587 OST thread and, hence, there is no per-thread preallocated
4588 pages and buffers to use. Solution: hijack pga pages for IO. As
4589 a byproduct, this avoids unnecessary data copying.
4593 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4594 Description: Client can oops in mdc_commit_close() after open replay
4595 Details : It was possible for the MDS to return an open request with no
4596 transaction number in mds_finish_transno() if the client was
4597 evicted, but without actually returning an error. Clients
4598 would later try to replay that open and may trip an assertion
4599 Simplify the client close codepath, and always return an error
4600 from the MDS in case the open is not successful.
4603 Frequency : rare, 2.6 OSTs only
4605 Description: OST may deadlock under high load on fragmented files
4606 Details : If there was a heavy load and highly-fragmented OST filesystems
4607 it was possible to have all the OST threads deadlock waiting on
4608 allocation of biovecs, because the biovecs were not released
4609 until the entire RPC IO was completed. Instead, release biovecs
4610 as soon as they are complete to ensure forward IO progress.
4612 Severity : enhancement
4614 Description: Support for specifying external journal device at mount
4615 Details : If an OST or MDS device is formatted with an external journal
4616 device, this device major/minor is stored in the ext3 superblock
4617 and may not be valid for failover. Allow detecting and
4618 specifying the external journal at mount time.
4623 Description: Mounting an MDS with pending unlinked files may cause oops
4624 Details : target_finish_recovery() calls mds_postrecov() which returned
4625 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4626 considers this an error and immediately begins cleaning up the
4627 lov, just after starting the mds_lov process
4629 Severity : enhancement
4631 Description: Implement 'lfs df' to report actual free space on per-OST basis
4632 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4633 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4634 usage of objects; '-h' to report in human readable format.
4636 ------------------------------------------------------------------------------
4638 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4645 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4646 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4647 at startup during orphan unlinking. Instead, skip buddy update
4648 before setup, as it will be regenerated after recovery is complete.
4653 Description: If an OST is inactive, its locks might reference stale inodes.
4654 Details : lov_change_cbdata() must iterate over all namespaces, even if
4655 they are inactive to clear inode references from the lock.
4657 Severity : enhancement
4658 Frequency : occasional, if non-standard max_dirty_mb used
4660 Description: Client will block write RPCs if not enough grant
4661 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4662 then the client will block writes while waiting for another RPC
4663 to complete instead of consuming its dirty limit. With change
4664 we get improved performance when max_dirty_mb is small.
4666 Severity : enhancement
4667 Bugzilla : 3389, 6253
4668 Description: Add support for supplementary groups on the MDS.
4669 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4670 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4671 lookups for user supplementary groups into a cache.
4675 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4676 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4677 for liblustre O_EXCL is enforced by MDS.
4681 Description: readdir never returns NULL in liblustre.
4682 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4683 dentry in struct dirent.
4687 Frequency : liblustre only, depends on application IO pattern
4688 Description: liblustre clients evicted if not contacting servers
4689 Details : Don't put liblustre clients into the ping_evictor list, so
4690 they will not be evicted by the pinger ever.
4692 Severity : enhancement
4694 Description: Add ability to evict clients by NID from MDS.
4695 Details : By echoing "nid:$NID" string into
4696 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4697 $NID would be instantly evicted from this MDS and from all active
4698 OSTs connected to it.
4702 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4703 Details : lookup_it_finish() used to query file size from OSTs that was not
4708 Description: service threads change working directory to that of init
4709 Details : Starting lustre service threads may pin the working directory
4710 of the parent thread, making that filesystem busy. Threads
4711 now change to the working directory of init to avoid this.
4715 Frequency : during shutdown only
4716 Description: shutdown with a failed MDS or OST can cause unmount to hang
4717 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4722 Frequency : 2.6 only
4723 Description: chmod/chown may include an extra supplementary group
4724 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4725 supplementary group and if none is specified this is used.
4728 Bugzilla : 5479 (6816)
4730 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4731 Details : If lookup is in progress on a file that is unlinked we might try
4732 to revalidate the inode and fail in revalidate after lookup is
4733 complete and ll_file_open() enqueues the open again but
4734 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4737 Frequency : always, if lconf --abort_recovery used
4739 Description: lconf --abort_recovery fails with 'Operation not supported'
4740 Details : lconf was attempting to abort recovery on the MDT device and not
4743 ------------------------------------------------------------------------------
4745 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4750 Frequency : rare (only unsupported configurations with a node running as an
4752 Bugzilla : 6514, 5137
4753 Description: Mounting a Lustre file system on a node running as an OST could
4755 Details : OSTs now preallocates memory needed to write out data at
4756 startup, instead of when needed, to avoid having to
4757 allocate memory in possibly low memory situations.
4758 Specifically, if the file system is mounted on on OST,
4759 memory pressure could force it to try to write out data,
4760 which it needed to allocate memory to do. Due to the low
4761 memory, it would be unable to do so and the node would
4762 become unresponsive.
4764 Severity : enhancement
4766 Description: Addition of lconf --service command line option
4767 Details : lconf now accepts a '--service <arg>' option, which is
4768 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4770 Severity : enhancement
4772 Description: Failover mode is now the default for OSTs.
4773 Details : By default, OSTs will now run in failover mode. To return to
4774 the old behaviour, add '--failout' to the lmc line for OSTs.
4776 Severity : enhancement
4778 Description: Health checks are now provided for MDS and OSTs
4779 Details : Additional detailed health check information on MSD and OSTs
4780 is now provided through the procfs health_check value.
4783 Frequency : occasional, depends on IO load
4785 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4786 after numerous create/delete cycles
4787 Details : The ext3 inode allocation policy would not allocate new inodes
4788 very well on the OSTs because there are no new directories
4789 being created. Instead we look for groups with free space if
4790 the parent directories are nearly full.
4795 Description: Network or server problems during mount may cause partially
4796 mounted clients instead of returning an error.
4797 Details : The config llog parsing code may overwrite the error return
4798 code during mount error handling, returning success instead
4804 Description: MDS can fail to allocate large reply buffers
4805 Details : After long uptimes the MDS can fail to allocate large reply
4806 buffers (e.g. zconf client mount config records) due to memory
4807 fragmentation or consumption by the buffer cache. Preallocate
4808 some large reply buffers so that these replies can be sent even
4809 under memory pressure.
4813 Frequency : rare (liblustre)
4814 Description: fsx running with liblustre complained that using truncate() to
4815 extend the file doesn't work. This patch corrects that issue.
4816 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4817 ATTR_SIZE and lsm use in llu_setattr_raw.
4821 Frequency : rare, only 2.6 kernels
4822 Description: Unusual file access patterns on the MDS may result in inode
4823 data being lost in very rare circumstances.
4824 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4825 code in the 2.6 kernel caused the inode and/or EA data not to
4826 be read from disk, causing single-file corruption.
4830 Frequency : rare, only 2.6 filesystems using extents
4831 Description: Heavy concurrent write and delete load may cause data corruption.
4832 Details : It was possible under high-load situations to have an extent
4833 metadata block in the block device cache from a just-unlinked
4834 file overwrite a newly-allocated data block. We now unmap any
4835 metadata buffers that alias just-allocated data blocks.
4839 Frequency : filesystems with default stripe_count larger than 77
4840 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4841 Details : lconf specifies an inode size of 4096 bytes when the default
4842 stripe_count is larger than 77. This conflicts with the default
4843 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4844 this case to avoid pinning too much memory for large EAs.
4846 ------------------------------------------------------------------------------
4848 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4853 Frequency : rare (extremely heavy IO load with hundreds of clients)
4855 Description: Client is evicted, gets IO error writing to file
4856 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4857 caused clients to be evicted for AST timeouts. The fixes in
4858 bug 5192 mean we no longer need to have such short AST timeouts
4859 so ldlm_timeout has been increased.
4862 Frequency : occasional during --force or --failover shutdown under load
4863 Bugzilla : 5949, 4834
4864 Description: Server oops/LBUG if stopped with --force or --failover under load
4865 Details : a collection of import/export refcount and cleanup ordering
4866 issues fixed for safer force cleanup
4869 Frequency : only filesystems larger than 120 OSTs
4870 Bugzilla : 5990, 6223
4871 Description: lfs getstripe would oops on a very large filesystem
4872 Details : lov_getconfig used kfree on vmalloc'd memory
4875 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4876 Bugzilla : 6242, 6243
4877 Description: reading from files that had been truncated to a non-zero size
4878 but never opened returned no data
4879 Details : ll_file_read() reads zeros from no-object files to EOF
4884 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4885 Details : There's little harm in aborting MDS/OSS recovery and letting it
4886 try again, so I removed the LASSERT and return an error instead.
4888 Severity : enhancement
4890 Description: New debugging infrastructure for tracking down data corruption
4891 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4892 (b) cover more of the client-side code path, and (c) try to narrow
4893 down where problems occurred
4897 Bugzilla : 3819, 4364, 4397, 6313
4898 Description: Racing close and eviction MDS could cause assertion in mds_close
4899 Details : It was possible to get multiple mfd references during close and
4900 client eviction, leading to one thread referencing a freed mfd.
4902 Severity: : enhancement
4903 Bugzilla : 3262, 6359
4904 Description: Attempts to reconnect to servers are now more aggressive.
4905 Details : This builds on the enhanced upcall-less recovery that was added
4906 in 1.4.2. When trying to reconnect to servers, clients will
4907 now try each server in the failover group every 10 seconds. By
4908 default, clients would previously try one server every 25 seconds.
4913 Description: After recovery, certain operations trigger a failed
4914 assertion on a client.
4915 Details : Failing over an mds, using lconf -d --failover, while a
4916 client was doing a readdir() call would cause the client to
4917 LBUG after recovery completed and the readdir() was resent.
4919 Severity : enhancement
4921 Description: Default groups are now added by lconf
4922 Details : You can now run lconf --group <servicename> without having to
4923 manually add groups with lmc.
4926 Frequency : occasional
4928 Description: Nodes with an elan id of 0 trigger a failed assertion
4931 Frequency : always when accessing e.g. tty/console device nodes
4933 Description: tty and some other devices nodes cannot be used on lustre
4934 Details : file's private_data field is used by device data and lustre
4935 values in there got lost. New field was added to struct file to
4936 store fs-specific private data.
4939 Frequency : when exporting Lustre via NFS
4941 Description: NFSD failed occasionally when looking up a path component
4942 Details : NFSD is looking up ".." which was broken in ext3 directories
4943 that had grown large enough to become hashed.
4946 Frequency : Clusters with multiple interfaces not on the same subnet
4948 Description: Nodes will repeatedly try to reconnect to an interface which it
4949 cannot reach and report an error to the log.
4950 Details : Extra peer list entries will be created by lconf with some peers
4951 unreachable. lconf now validates the peer before adding it.
4954 Frequency : Only if a default stripe is set on the filesystem root.
4956 Description: Setting a default stripe on the filesystem root prevented the
4957 filesystem from being remounted.
4958 Details : The client was sending extra request flags in the root getattr
4959 request and did not allocate a reply buffer for the dir EA.
4962 Frequency : occasional, higher if lots of files are accessed by one client
4963 Bugzilla : 6159, 6097
4964 Description: Client trips assertion regarding lsm mismatch/magic
4965 Details : While revalidating inodes the VFS looks up inodes with ifind()
4966 and in rare cases can find an inode that is being freed.
4967 The ll_test_inode() code will free the lsm during ifind()
4968 when it finds an existing inode and then the VFS later attaches
4969 this free lsm to a new inode.
4973 Bugzilla : 6422, 7030
4974 Description: MDS deadlock between mkdir and client eviction
4975 Details : Creating a new file via mkdir or mknod (starting a transaction
4976 and getting the ns lock) can deadlock with client eviction
4977 (gets ns lock and trying to finish a synchronous transaction).
4980 Frequency : occasional
4981 Description: While starting a server, the fsfilt_ext3 module could not be
4983 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4984 kernels. Previously, lconf would still use the ext3 name
4985 when trying to load modules. Now, it will correctly use
4986 ext3 on 2.4 and ldiskfs on 2.6.
4988 Severity : enhancement
4989 Description: The default stripe count has been changed to 1
4990 Details : The interpretation of the default stripe count (0, to lfs
4991 or lmc) has been changed to mean striping across a single
4992 OST, rather than all available. For general usage we have
4993 found a stripe count of 1 or 2 works best.
4995 Severity : enhancement
4996 Description: Add support for compiling against Cray portals.
4997 Details : Conditional compiling for some areas that are different
5001 Frequency : occasional
5002 Bugzilla : 6409, 6834
5003 Description: Creating files with an explicit stripe count may lead to
5004 a failed assertion on the MDS
5005 Details : If some OSTs are full or unavailable, creating files may
5006 trigger a failed assertion on the MDS. Now, Lustre will
5007 try to use other servers or return an error to the
5011 Frequency : occasional
5013 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
5014 caused lock timeout during readahead (since 1.4.2).
5015 Details : Processes doing readahead might match a lock that hasn't been
5016 granted yet if there are overlapping and conflicting lock
5017 requests. The readahead process waits on ungranted lock
5018 (original lock is CBPENDING), while OST waits for that process
5019 to cancel CBPENDING read lock and eventually evicts client.
5021 Severity : enhancement
5023 Description: Initial enabling of flock support for clients
5024 Details : Implements fcntl advisory locking and file status functions.
5025 This feature is provided as an optional mount flag (default
5026 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
5027 locking are implemented yet, and those that are are not guaranteed
5028 to be completely correct in production environments.
5029 mount -t lustre -o [flock|noflock] ...
5032 Frequency : occasional
5034 Description: OSTs running 2.4 kernels but with extents enabled might trip an
5035 assertion in the ext3 JBD (journaling) layer.
5036 Details : The b_committed_data struct is protected by the big kernel lock
5037 in 2.4 kernels, serializing journal_commit_transaction() and
5038 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
5039 access to this struct. In 2.6 kernels there is finer grained
5040 locking to improve SMP performance of the JBD layer.
5044 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
5049 Description: merge and cleanup kernel patches.
5052 -----------------------------------------------------------------------------
5054 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
5056 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
5057 versions of Lustre. Please update all servers and clients to
5058 version 1.4.2 or later at the same time. You must also run
5059 "lconf --write-conf {config}.xml" on the MDS while it is stopped
5060 to update the configuration logs.
5062 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
5063 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
5064 - swab llog records of type '0' so we get proper header size/idx (5861)
5065 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
5066 - fix rename of one directory over another leaking an inode (5953)
5067 - avoid SetPageDirty on 2.6 (5981)
5068 - don't re-add just-being-destroyed locks to the waiting list (5653)
5069 - when creating new directories, inherit the parent's custom
5070 striping settings if present parent (3048)
5071 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
5072 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
5073 - don't hold client locks on temporary worklist from l_lru (5666)
5074 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
5075 - automatically evict dead clients (5921)
5076 - Update file size properly in create+truncate+fstat case (6196)
5077 - Do not unhash mountpoint dentries, do not allow removal of
5079 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
5080 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
5081 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
5082 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
5083 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
5084 - Correctly handle reads of files with no objects (6243)
5085 - lctl recover will also mark a device active if deactivate used (5933)
5087 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
5088 - allow --write-conf on an MDS with different nettype than client (5619)
5089 - don't write config llogs to MDS for mounts not from that MDS (5617)
5090 - lconf should create multiple TCP connections from a client (5201)
5091 - init scripts are now turned off by default; run chkconfig --on
5092 lustre and chkconfig --on lustrefs to use them
5093 - upcalls are no longer needed for clients to recover to failover
5095 - add --abort-recovery option to lconf to abort recovery on device
5097 - add support for an arbitrary number of OSTs (3026)
5098 - Quota support protocol changes.
5099 - forward compatibility changes to wire structs (6007)
5100 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
5101 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
5102 - improved "lustre status" script
5103 - initialize blocksize for non-regular files (6062)
5104 - added --disable-server and --disable-client configure options (5782)
5105 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
5106 - Vanilla 2.4.29 support
5107 - increase maximum number of obd devices to 520 (6242)
5108 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
5109 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
5110 SLES 9 2.6.5 kernels
5111 - limit stripes per file to 160 (the maximum EA size) (6093)
5113 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5116 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
5117 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
5118 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
5119 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
5120 - small ext3 extents cleanups and fixes (5733)
5121 - improved mballoc code, several small races and bugs fixed (5733, 5638)
5122 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
5123 - avoid needless client->OST connect, fix handle mismatch (5317)
5124 - fix DLM error path that led to out-of-sync client, long delays (5779)
5125 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
5126 - fix several locking issues related to i_size (5492,5624,5654,5672)
5127 - don't move pending lock onto export if it is already evicted (5683)
5128 - fix kernel oops when creating .foo in unlinked directory (5548)
5129 - fix deadlock in obdfilter statistics vs. object create (5811)
5130 - use time_{before,after} to avoid timer jiffies wrap (5882)
5131 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
5132 - Do not leak request if server was not able to process it (5154)
5133 - If mds_open unable to find parent dir, make that negative lookup(5154)
5134 - don't create new directories with extent-mapping (5909, 5936)
5136 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
5137 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
5138 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
5139 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
5140 - support for RHEL4 kernel on i686 (5773)
5141 - provide error messages when incompatible logs are encountered (5898)
5143 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
5144 * version 1.4.0.10 (1.4.1 release candidate 1)
5146 - don't keep a lock reference when lock is not granted (4238)
5147 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5148 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
5149 - drop import inflight refcount on signal_completed_replay error (5255)
5150 - unlock page after async write error during send (3677)
5151 - handle missing objects in filter_preprw_read properly (5265)
5152 - no transno return for symlink open, don't save no-trasno open (3440)
5153 - don't try to complete elan receive that already failed (4012)
5154 - free RPC server reply state on error (5406)
5155 - clean up thread from ptlrpc_start_thread() on error (5160)
5156 - readahead could read extra page into cache that wasn't ejected (5388)
5157 - prevent races in class_attach/setup/cleanup/detach (5260)
5158 - don't dereference de->d_inode after l_dput of de (5458)
5159 - use "int" for stripe value returned from lock_to_stripe (5544)
5160 - mballoc allocation and error-checking fixes in 2.6 (5504)
5161 - block device patches to fix I/O request sizes in 2.6 (5482)
5162 - look up hostnames for IB nals (5602)
5163 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
5164 - don't start multiple acceptors for the same port (5277)
5165 - fix incorrect LASSERT in mds_getattr_name (5635)
5166 - export a proc file for general "ping" checking (5628)
5167 - fix "lfs check" to not block when the MDS is down (5628)
5169 - service request history (4965)
5170 - put {ll,lov,osc}_async_page structs in a single slab (4699)
5171 - create an "evict_client" /proc entry on OSTs, like the MDS has
5172 - fix mount usage message, return errors per mount(8) (5168)
5173 - change grep [] to grep "[]" in tests so they work in more UMLs
5174 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
5175 - remove ancient LOV_MAGIC_V0 EA support (5047)
5176 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
5177 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
5178 - allow readahead allocations to fail when low on memory (5383)
5179 - mmap locking landed again, after considerable improvement (2828)
5180 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
5182 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
5185 - send OST transaction number in read/write reply to free req (4966)
5186 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5187 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5188 - the watchdog thread now runs as interruptible (5246)
5189 - flock/lockf fixes (but it's still disabled, pending 5135)
5190 - don't use EXT3 constants in llite code (5094)
5191 - memory shortage at startup could cause assertion (5176)
5193 - reorganization of lov code
5194 - single portals codebase
5196 - add extents/mballoc support (5025)
5197 - direct I/O reads in the obdfilter (4048)
5198 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5200 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5203 - send OST transaction number in read/write reply to free req (4966)
5204 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5205 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5206 - the watchdog thread now runs as interruptible (5246)
5207 - handle missing objects in filter_preprw_read properly (5265)
5208 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5209 - drop import inflight refcount on signal_completed_replay error (5255)
5210 - unlock page after async write error during send (3677)
5211 - return original error code on reconstructed replies (3761)
5212 - no transno return for symlink open, don't save no-trasno open (3440)
5214 - add pid to ldlm debugging output (4922)
5215 - bump the watchdog timeouts -- we can't handle 30sec yet
5216 - extra debugging for orphan dentry/inode bug (5259)
5218 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5221 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5222 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5223 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5224 - fix NULL dereference in /proc/sys/portals/routes (4827)
5225 - allow failed mdc_close() operations to be interrupted (4561)
5226 - stop precreate on OST before MDS would time out on it (4778)
5227 - don't send partial-page writes before EOF from client (4410)
5228 - discard client grant for sub-page writes on large-page clients (4520)
5229 - don't free dentries not owned by NFS code, check generation (4806)
5230 - fix lsm leak if mds_create_objects() fails (4801)
5231 - limit debug_daemon file size, always print CERROR messages (4789)
5232 - use transno after validating reply (3892)
5233 - process timed out requests if import state changes (3754)
5234 - update mtime on OST during writes, return in glimpse (4829)
5235 - add mkfsoptions to LDAP (4679)
5236 - use ->max_readahead method instead of zapping global ra (5039)
5237 - don't interrupt __l_wait_event() during strace
5239 - add software watchdogs to catch hung threads quickly (4941)
5240 - make lustrefs init script start after nfs is mounted
5241 - fix CWARN/ERROR duplication (4930)
5242 - return async write errors to application if possible (2248)
5243 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5244 - print NAL number in %x format (4645)
5245 - update barely-supported suse-2.4.21-171 series (4842)
5246 - support for sles 9 %post scripts
5247 - support for building 2.6 kernel-source packages
5248 - support for sles km_* packages
5250 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5253 - ignore -ENOENT errors in osc_destroy (3639)
5254 - notify osc create thread that OSC is being cleaned up (4600)
5255 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5256 - reconstruct ost_handle() like mds_handle() (4657)
5257 - create a new thread to do import eviction to avoid deadlock (3969)
5258 - let lconf resolve symlinked-to devices (4629)
5259 - don't unlink "objects" from directory with default EA (4554)
5260 - hold socknal file ref over connect in case target is down (4394)
5261 - allow more than 32000 subdirectories in a single directory (3244)
5262 - fix blocks count for O_DIRECT writes (3751)
5263 - OST returns ENOSPC from object create when no space left (4539)
5264 - don't send truncate RPC if file size isn't changing (4410)
5265 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5266 - bind to privileged port in socknal and tcpnal (3689)
5268 - rate limit CERROR/CWARN console message to avoid overload (4519)
5269 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5270 - basic mmap support (3918)
5271 - kernel patch series update from b1_4 (4711)
5273 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5276 - avoid crash during MDS cleanup with OST shut down (2775)
5277 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5278 - don't use bad inodes on the MDS (3744)
5279 - dynamic object preallocation to improve recovery speed (4236)
5280 - don't hold spinlock over lock dumping or change debug flags (4401)
5281 - don't zero obd_dev when it is force cleaned (3651)
5282 - print grants to console if they go negative (4431)
5283 - "lctl deactivate" will stop automatic recovery attempts (3406)
5284 - look for existing locks in ldlm_handle_enqueue() (3764)
5285 - don't resolve lock handle twice in recovery avoiding race (4401)
5286 - revalidate should check working dir is a directory (4134)
5288 - don't always mark "slow" obdfilter messages as errors (4418)
5290 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5293 - don't close LustreDB during write_conf until it is done (3860)
5294 - fix typo in lconf for_each_profile (3821)
5295 - allow dumping logs from multiple threads at one time (3820)
5296 - don't allow multiple threads in OSC recovery (3812)
5297 - fix debug_size parameters (3864)
5298 - fix mds_postrecov to initialize import for llog ctxt (3121)
5299 - replace config semaphore with spinlock (3306)
5300 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5301 - don't allow enqueue to complete on a destroyed export (3822)
5302 - down write_lock before checking llog header bitmap (3825)
5303 - recover from lock replay timeout (3764)
5304 - up llog sem before sending rpc (3652)
5305 - reduce ns lock hold times when setting kms (3267)
5306 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5307 - fix NULL deref and obd_dev leak on setup error (3312)
5308 - replace some LBUG about llog ops with error handling (3841)
5309 - don't match INVALID dentries from d_lookup and spin (3784)
5310 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5311 - fix invalid assertion in ptlrpc_set_wait (3880)
5313 - add libwrap support for the TCP acceptor (3996)
5314 - add /proc/sys/portals/routes for non-root route listing (3994)
5315 - allow setting MDS UUID in .xml (2580)
5316 - print the stack of a process that LBUGs (4228)
5318 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5321 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5322 - make sure to unset replay flag from failed open requests (3440)
5323 - if default stripe count is 0, use OST count for inode size (3636)
5324 - update parent mtime/ctime on client for create/unlink (2611)
5325 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5326 - free recovery state on server during a forced cleanup (3571)
5327 - unregister_reply for resent reqs (3063)
5328 - loop back devices mounting and status check on 2.6 (3563)
5329 - fix resource-creation race that can provoke i_size == 0 (3513)
5330 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5331 - more debugging for page-accounting assertion (3746)
5332 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5333 - avoid deadlock after precreation failure (3758)
5334 - fix race and lock order deadlock in orphan handling (3450, 3750)
5335 - add validity checks when grabbing inodes from l_ast_data (3599)
5337 - add /proc/.../recovery_status to obdfilter (3428)
5338 - lightweight CDEBUG infrastructure, debug daemon (3668)
5339 - change default OSC RPC parameters to be better on small clusters
5340 - turn off OST read cache for files smaller than 32MB
5341 - install man pages and include them in rpms (3100)
5342 - add new init script for (un)mounting lustre filesystems (2593)
5343 - run chkconfig in %post for init scripts (3701)
5344 - drop scimac NAL (unmaintained)
5346 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5349 - clean kiobufs before and after use (3485)
5350 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5351 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5352 - change init script to fail more clearly if not run as root (1528)
5353 - allow clients to reconnect during replay (1742)
5354 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5355 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5356 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5357 - fix cond_resched (3554)
5358 - only evict unfinished clients after recovery (3515)
5359 - allow bulk resend, prevent data loss (3570)
5360 - dynamic ptlrpc request buffer allocation (2102)
5361 - don't allow unlinking open directory if it isn't empty (2904)
5362 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5363 - remove extraneous obd dereference causing LASSERT failure (3334)
5364 - don't use get_cycles() when creating temp. files on the mds (3156)
5365 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5366 - handle EEXIST for set-stripe, set proper directory name (3336)
5368 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5369 - fix ksocknal_fmb_callback() error messages (2918)
5371 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5374 - don't copy lvb into (possibly NULL) reply on error (2983)
5375 - don't deref dentry after dput, don't free lvb on error (2922)
5376 - use the kms to determine writeback rpc length (2947)
5377 - increment oti_logcookies when osc is inactive (2948)
5378 - update client's i_blocks count via lvb messages (2543)
5379 - handle intent open/close of special files properly (1557)
5380 - mount MDS with errors=remount-ro, like obdfilter (2009)
5381 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5382 - don't use cancelling-locks' kms values (2947)
5383 - use highest lock extent for kms, not last one (2925)
5384 - don't dereference ERR_PTR() dentry in error handling path (3107)
5385 - fix thread race in portals_debug_dumplog() (3122)
5386 - create lprocfs device entries at setup instead of at attach (1519)
5387 - common AST error handler, don't evict client on completion race (3145)
5388 - zero nameidata in detach_mnt in 2.6 (3118)
5389 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5390 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5391 - initialize RPC timeout timer earlier for 2.6 (3219)
5392 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5393 - print nal/nid for unknown nid (3258)
5394 - additional checks for oscc recovery before doing precreate (3284)
5395 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5396 - don't crash in mdc_close for bad permissions on open (3285)
5397 - zero i_rdev for non-device files (3147)
5398 - clear page->private before handing to FS, better assertion (3119)
5399 - tune the read pipeline (3236)
5400 - fix incorrect decref of invalidated dentry (2350)
5401 - provide read-ahead stats and refine rpc in flight stats (3328)
5402 - don't hold journal transaction open across create RPC (3313)
5403 - update atime on MDS at close time (3265)
5404 - close LDAP connection when recovering to avoid server load (3315)
5405 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5406 - don't leak open file on MDS after open resend (3325)
5407 - serialize filter_precreate and filter_destroy_precreated (3329)
5408 - loop device shouldn't call sync_dev() for nul device (3092)
5409 - clear page cache after eviction (2766)
5410 - resynchronize MDS->OST in background (2824)
5411 - refuse to mount the same filesystem twice on same mountpoint (3394)
5412 - allow llmount to create routes for mounting behind routers (3320)
5413 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5414 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5415 - fix rare problem with rename on htree directories (3417)
5417 - allow default OST striping configuration per directory (1414)
5418 - fix compilation for qswnal for 2.6 kernels (3125)
5419 - increase maximum number of MDS request buffers for large systems
5420 - change liblustreapi to be useful for external progs like lfsck (3098)
5421 - increase local configuration timeout for slow disks (3353)
5422 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5424 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5427 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5428 - don't overwrite extent policy data in reply if lock was blocked (2901)
5429 - drop filter export grants atomically with removal from device (2663)
5430 - del obd_self_export from work_list in class_disconnect_exports (2908)
5431 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5432 - swab reply message in mdc_close, other PPC fixes (2464)
5433 - fix destroying of named logs (2325)
5434 - overwrite old logs when running lconf --write_conf (2264)
5435 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5436 - fix race in target_handle_connect (2898)
5437 - mds_reint_create() should take same inode create lock (2926)
5438 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5439 - don't close files for self_export to avoid uninitialized obd (2936)
5440 - allow MDS with the same name as client node (2939)
5441 - hold dentry reference for closed log files for unlink (2325)
5442 - reserve space for all logs during transactions (2059)
5443 - don't evict page beyond end of stripe extent (2925)
5444 - don't oops on a deleted current working directory (2399)
5445 - handle hard links to targets without a parent properly (2517)
5446 - don't dereference NULL lock when racing during eviction (2867)
5447 - don't grow lock extents when lots of conflicting locks (2919)
5449 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5452 - account for cache space usage on clients to avoid data loss (974)
5453 - lfsck support in lustre kernel code (2349)
5454 - reduce journal credits needed for BRW writes (2370)
5455 - orphan handling to avoid losing space on client/server crashes
5456 - ptlrpcd can be blocked, stopping ALL progress (2477)
5457 - use lock value blocks to assist in proper KMS, faster stat (1021)
5458 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5459 - recovery for initial connections (2355)
5460 - fixes for mds_cleanup_orphans (1934)
5461 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5462 - block all file creations until orphan recovery completes (1901)
5463 - client remove rq_connection from request struct (2423)
5464 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5465 - recovery timer race (2670)
5466 - mdc_close recovey bug (2532)
5467 - ptlrpc cleanup bug (2710)
5468 - mds timeout on local locks (2588)
5469 - namespace lock held during RPCs (2431)
5470 - handle interrupted sync write properly (2503)
5471 - don't try to handle a message that hasn't been replied to (2699)
5472 - client assert failure during cleanup after abort recovery (2701)
5473 - leak mdc device after failed mount (2712)
5474 - ptlrpc_check_set allows timedout requests to complete (2714)
5475 - wait for inflight reqs when ptlrpcd finishes (2710)
5476 - make sure unregistered services are removed from the srv_list
5477 - reset bulk XID's when resending them (caught by 1138 test)
5478 - unregister_bulk after timeout
5479 - fix lconf error (2694)
5480 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5481 - readahead locks pages, leaves pending causing memory pressure (2673)
5482 - increase OST request buffers to 4096 on large machines (2729)
5483 - fix up permission of existing directories in simple_mkdir (2661)
5484 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5485 - don't assign transno to errored transactions (2742)
5486 - don't delete objects on OST if given a bogus objid from MDS (2751)
5487 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5488 - if rq_no_resend, then timeout request after recovery (2432)
5489 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5490 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5491 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5492 - MDS recovery shouldn't delete objects that it has given out (2730)
5493 - if enqueue arrives after completion, don't clobber LVB (2819)
5494 - don't unlock pages twice when trigger_group_io returns error (2814)
5495 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5496 - don't write pages to disk if there was an error (1450)
5497 - don't ping imports that have recovery disabled (2676)
5498 - take buffered bytes into account when balancing socknal conn (2817)
5499 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5500 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5501 - remove little-endian swabbing of llog records (1987)
5502 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5503 - retry reposting request buffers if they fail (1191)
5504 - grow extent at grant time to avoid granting a revoked lock (2809)
5505 - lock revoke doesn't evict page if covered by a second lock (2765)
5506 - disable VM readahead to avoid reading outside lock extents (2805)
5508 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5509 - updated kernel patches for hp-2.4.20 kernel (2681)
5511 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5514 - fix truncated write corruption (2366)
5515 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5517 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5518 - ensure len(uuid) < 37 in lmc (1171)
5519 - fix ia64 OOPS in llog_test (2255)
5520 - zero end of page at obdfilter for partial page writes (2648)
5521 - don't leave stale dentries around after renames (bug 2428)
5522 - fix timeouts when evicting a client with a single lock held (2642)
5523 - set deadline for the initial HELLO message to drain (2634)
5524 - print out dotted-quad IP addresses in the socknal (2302)
5526 - additional debugging for MDS client eviction problem (2443)
5527 - fix mkfsoptions support for osts (2603, 2604)
5529 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5532 - add series for the vanilla 2.6.0 kernel
5533 - add series for the vanilla 2.4.24 kernel
5534 - add series for a cray x86/64 UL kernel drop
5535 - fix xattr patches for the vanilla 2.4.19 series
5537 - generate true UUIDs in lmc (1171)
5538 - have portals stack dumping break in UML (2466)
5539 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5540 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5541 - wake callback waiting threads on client eviction (2460)
5542 - Add --ptldebug and --subsystem to lmc (1719)
5543 - update assertion to allow safe interrupt allocation
5544 - set rq_no_resend for cancel requests (2432)
5545 - recalculate ptlrpcd timeout after resend (2494)
5546 - call vfs_rmdir when removing pending directories (2368)
5547 - fix renaming a file to itself (2429)
5548 - lmc creates a default one-stripe lov (2454)
5549 - expand procfs space to handle large clusters (2326)
5550 - increase UML stack to avoid overflow
5551 - update lconf's list of debug and subsystem masks
5552 - fix lfs find --obd (2510)
5553 - /proc tunable for disabling filter read caching (2591)
5554 - stop rpm packages from altering slapd.conf (2301)
5555 - disable nagle in the socknal under 0conf (2578)
5556 - choose mds inode size based on stripe count (2572)
5557 - fix kernel-source rpm problems (2516)
5559 - add --disable-doc to avoid pdf generation (2421)
5560 - update documentation, tests, type-os, comments
5561 - avoid format warnings on ia64
5562 - remove the TOE NAL
5563 - tiny code cleanups by removing unused fields
5565 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5568 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5569 - remove the most blatant lies from BUILDING file (2371)
5570 - change default debug level to reasonable production setting
5571 - reduce client side cache size to reduce cache flush time
5572 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5573 - make TCP zerocopy and pinger support enabled by default (2476)
5574 - sync writes completed after process exits caused crashes (2319)
5575 - maintain correct mount count on the MDS (2356)
5576 - backout 1557, because 2316 wasn't really fixed
5577 - better file I/O statistics gathering in /proc
5578 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5579 - another kernel patch to fix zero-copy TCP function export
5580 - don't take duplicate lock when processing re-sent getattr (2420)
5581 - lctl uses obd_self_export instead of creating new conn (2353)
5582 - MDS/OST recovery case which requires object creation asserted (2425)
5583 - move lfs from /usr/sbin to /usr/bin in packages
5584 - fix race between mds_client_add and mds_client_free (2417)
5585 - use kmalloc instead of slabs in portals (2430)
5586 - don't create duplicate records when a failover MDS is present (2442)
5587 - remove unnecessary mount age check (2332)
5588 - don't remove directory inodes from locks prematurely (2451)
5589 - don't break if MDS service name is the same as hostname (2103)
5590 - fix races in client write RPC generation when cache full (2482)
5592 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5595 - remove now-unused request->rq_obd (278)
5596 - if an allocation fails, print out how much memory we've used (1933)
5597 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5598 - add the "configurable stack size" patch to most series files (1256)
5599 - ability to write large log records, for 100+ OST configs (2306)
5600 - fix NULL deref when filter_prep fails (2314)
5601 - fix operator precedence error in filter_sync
5602 - dynamic allocation of socknal TX descriptors (2315)
5603 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5604 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5605 - ensure that CWARN messages in Portals always get to the syslog
5606 - __init/__exit are not for prototype decls (ldlm_init/exit)
5607 - x86-64 compile warning fixes
5608 - fix gateway LMC keyword conflict (2318)
5609 - fix MDS lock inversions in getattr/reint paths (1844)
5610 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5611 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5612 - DLM race condition prevented some lock evictions (2328)
5613 - ENOMEM detection and retry on socknal sends (2230)
5614 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5615 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5616 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5617 - better error messages when a client is rejected during recovery (1505)
5618 - avoid cancelling locks which were never granted, after failure (2330)
5619 - fix i_sem/journal inversion in mds_client_add (2333)
5620 - fix truncate/getattr lock cycle deadlock (2334)
5621 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5622 - fix two rare exit paths which could leak an l_lock() ref (2321)
5623 - fencepost error in MDS/OST orphan recovery (2226)
5624 - make log record alignment 8 bytes (1988)
5625 - lstripe now fails when requested offset > ost_count (2237)
5626 - ensure that all kernel series have a complete list.h (1607)
5627 - fix crashes in special-file operations (2316)
5628 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5630 - allow configurable automake binary, for testing new versions
5631 - small update to the lfs documentation
5633 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5635 * fix negative export reference count in fsfilt_sync (2312)
5637 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5638 * release candidate 0.9.1
5640 - orphans are moved into the PENDING directory for possible recovery
5641 - replayed opens now open by fid for orphan/rename safety (1042)
5642 - last close of an orphan inode generates a transno (683)
5643 - chdir() and mount() now pin the directory entry (1020)
5644 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5645 - discard very old requests without processing them (1502)
5646 - remove some common, well-understood CERRORs (1505)
5647 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5648 - clear "grant" flags in OST replies until OST grant code lands (1644)
5649 - fix read performance by not clobbering i_blksize on client (1598)
5650 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5651 - make LRU size a /proc tunable, clears locks when reduced (707)
5652 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5653 - server locks take references on exports now (1558)
5654 - build fixes for 2.4.20-rh trees (1663)
5655 - return an error from lov_create if all OSCs are inactive (1751)
5656 - fix import levels when a reconnect happens without a timeout (1597)
5657 - exit early from mds_open if we get a lookup error (1749)
5658 - partial page read at EOF wouldn't wait for disk before sending (1642)
5659 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5660 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5661 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5662 - write/truncate lock inversion (1639)
5663 - Don't auto-load obdclass, portals modules during cleanup (1495)
5664 - fix timestamps from jumping to "now" (1763)
5665 - extra journal assertions (1648)
5666 - add an extra multiunlink test (1771)
5667 - fix read_record/write_record API (1776)
5668 - fix leak of offset_extent, possible incorrect i_size later (1772)
5669 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5670 - Debugging for the kqswnal_get_idle_tx problems (1820)
5671 - Allow recovery to be attempted multiple times (1536)
5672 - Write out MDS last_rcvd file after it is first created (1600)
5673 - Fix tx_descriptor leak in failed transmit situations (1827)
5674 - ext3 journaling fixes for assertion failure after IO error (1871)
5675 - class_export_put() on freed export after completion AST error (1896)
5676 - Fix revalidate looping in VFS (1322)
5677 - Don't access a freed export during MDS_REINT timeout (1521)
5678 - Add open-unlink recovery support on the MDS (1673,1764)
5679 - Return an error if no MDS data was read from last_rcvd (1946)
5680 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5681 - Don't setup a disk device that doesn't match exported UUID (317)
5682 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5683 - avoid committing NULL handle in force close
5684 - local.sh is now a one-stripe LOV configuration
5685 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5686 - fix ext3 htree duplicate directory entry corruption (1516)
5687 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5688 - update to new LOV EA format (2097)
5689 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5691 2003-06-15 Phil Schwan <phil@clusterfs.com>
5694 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5695 - per-import/export recovery handling (958, 931, 959)
5696 - multiple last-rcvd slots, for serving multiple FSes (949)
5697 - connections are again shared between multiple imp/exports (963, 964)
5698 - "umount -f" would hang if any requests needed to be sent (393, 978)
5699 - avoid pinning large req buffer by copying for queued messages (989)
5700 - add "uuid" to "lctl device" command to help upcalls (991)
5701 - "open" RPCs with transnos would confuse recovery counters (1037)
5702 - do proper endian conversion of all wire messages (288, 340, 891)
5703 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5704 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5705 - fix ptlrpc_connection leak in target_handle_connect (1174)
5706 - fix import refcounting bug in OST and MDS cleanup (1134)
5707 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5708 - fix very unlikely obd_types race condition (501)
5709 - remove osc_open hack for echo_client (1187)
5710 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5711 - a failure in read_inode2 leads to deadlock (1139)
5712 - cancel ack-locks as soon as transaction is committed (1072)
5713 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5714 - make sure to commitrw after any preprw to avoid deadlock (1162)
5715 - failing to execute a file in a lustre FS would lock inode (1203)
5716 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5717 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5718 - obd_uuid2tgt would walk off the end of the list (1255)
5719 - on IA64 the osc would give portals incorrect bulk size (1258)
5720 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5721 - fix lock inversion caused by new llite matching code (1282)
5722 - limit the number of dirty pages on a client to 10MB (1286)
5723 - timed out locks were not being corrected cancelled (1289)
5724 - fix O_DIRECT above 4GB on IA-32 (1292)
5725 * major user-visible changes
5726 - fail out/fail over policy now controlled by the upcall (993)
5728 - add OBD_PING to check server availability and failure (954)
5729 - lustre messages are now sent in sending host order (288, 340, 891)
5730 - add eadatalen to MDS getattr reply (340)
5731 - OST read replies may contain second buffer, with per-page status (593)
5733 2003-03-11 Phil Schwan <phil@clusterfs.com>
5736 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5737 - failed open()s could cause deadlock; fixed (867, 869)
5738 - stop cancelling OST locks when files are closed (481)
5739 - overlapping XID spaces caused network corruption (851, 853)
5740 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5741 - setattr_raw added, to do single-RPC, server-side setattrs
5742 - lmc/lconf syntax change for OST UUIDs
5743 - fix crashy race condition between ptlrpc_free_req and osc_close
5744 - don't use request in mdc_enqueue if we hit a timeout (889)
5745 - don't set the inode i_size for regular files from the MDS (896)
5746 - handle out of order completion AST (842)
5747 - don't LBUG if a lock request times out after receiving AST (913)
5748 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5749 - if a bad lock AST arrives, send an error instead of dropping entirely
5750 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5751 - fix leak in bulk IO when only partially completed (899, 900, 926)
5752 - fix O_DIRECT for ia64 (55)
5753 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5754 - C-z after timeout could hang a process forever; fixed (977)
5756 - client-side I/O cache (678, 924, 929, 941, 970)
5758 - READPAGE and SETATTRs which don't take server-side locks get
5761 2003-02-11 Phil Schwan <phil@clusterfs.com>
5764 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5765 - Fix fsfilt_extN_readpage() to read a full page of directory
5766 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5767 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5768 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5769 - Most intent operations were removed, in favour of a new RPC mode
5770 that does a single RPC to the server and bypasses most of the VFS
5771 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5772 - Aggressively cancel local locks on DLM servers
5773 - mds_reint_unlink sends EA to the client if it's the last nlink.
5774 client uses that EA to unlink OST objects.
5775 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5776 - recursive symlinks were fixed (439)
5777 - fixed NULL deref in DEBUG_REQ
5778 - filter_update_lastobjid no longer calls sync, which annoyed extN
5779 - fixed multi-client small-writes to a single file problem (445)
5780 - fixed mtime updates during file writes (607)
5781 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5782 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5783 - fixed Linux OST size reporting problem (444, 656)
5784 - OST now updates object mtime with writes or setattr (607, 619)
5785 - client verifies file size before zeroing page past EOF (445)
5786 - OST now writes last allocated objid to disk with allocation (108)
5787 - LOV on echo now works (409)
5789 - mds_reint_unlink sends a new buffer, with the EA included. this
5790 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5791 set if a regular file was being unlinked, and it was the last link
5792 - use PtlGet from the target for bulk writes (315)
5793 - OST now updates object mtime with writes or setattr (607, 619)
5794 - LDLM now has a grant-time callback to revalidate locked items, if
5796 - Many MDS operations were reorganized to combat race conditions
5798 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5799 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5801 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5804 - Fully reactivate OST imports after reconnection (512, others)
5805 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5806 - More graceful error handling for truncating on dead OST (515)
5807 - Don't error out unless we're actually accessing dead stripes (474)
5808 - Fix garbage sizes when stripes are missing (410)
5809 - LRU counters were broken, causing constant lock purge (433, 432)
5810 - garbage on read from stripes with failed OSTs (441)
5811 - mark OSCs as active before reconnecting during recovery (438)
5812 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5813 - lfind did not preserve OST order in output (443)
5814 - symlinks cause hung clients, incorrect data (439)
5815 - stop dereferencing request after dropping refcount (457)
5816 - don't LASSERT(spin_is_locked) on non-SMP (455)
5817 - fixes for many rename() bugs
5818 - fstat didn't correctly synchronize attributes (399)
5819 - server must handle lock cancellation during blocking AST prep (487)
5820 - bulk descriptors were free()d too soon (511)
5821 - fix paths in lconf, which would load incorrect modules (451, 507)
5822 - fix confusing lconf 'host not found' error message (386)
5823 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5824 - fix race condition in mdc_blocking_ast() for inode access (526)
5825 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5826 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5827 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5828 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5829 - fix lstripe in several ways (kernel side) (527)
5830 - fix request leak in ldlm_cli_enqueue (262)
5831 - incorrect OSC was marked inactive after OST failure
5832 - call mds_fs_cleanup before unmounting filesystem (524)
5833 - fix races between taking ns_lock and ldlm_lock_change_resource
5834 - fix races updating LOV export open file list
5835 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5836 - fix recovery NULL deref in ldlm_cli_cancel_unused
5837 - fix some DLM races by using new hash table for lock handles (419)
5838 - permit the client to specify desired inodes, at replay
5839 - duplicate requests when we queue them for replay reintegration
5840 - fix last_rcvd offset calculation
5841 - sync after each recovered transaction, so we always make progress
5842 - never, not always, ERESTART requests without transnos
5843 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5845 - skip replay if the MDS says that the client is already connected
5846 - don't check for a recovery-enabled export to match lctl's UUID
5847 - don't INC_USE_COUNT for phantom exports
5848 - don't crash when cleaning up phantom exports (567)
5849 - don't double-finish or set replay data for errored mdc_open requests
5850 - abort requests when they time out, so we don't get old replies
5851 - send/receive replies for AST messages again
5852 - if the client says that it doesn't have the lock, cancel it on the
5854 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5855 mark it as destroyed, it will all work out when decref is called
5856 - fix module use counts (22, 581)
5858 - ASTs now expect a reply (server cancels lock on error reply)
5860 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5863 - fix many simultaneous client startup (392)
5864 - fix dentry->d_it clobbering
5865 - credentials weren't being shipped for readdir/getattr operations
5866 - remove invalid assertions triggered during some concurrent MD
5868 - proper Lustre versions added (336, 389)
5869 - fix memory leak for create error case (398)
5870 - fix LOV locking bug that would get cli/srv out of sync
5871 - fix echo client over LOV (409)
5872 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5873 - fix double-O_EXCL intent crash (424)
5874 - avoid sending multiple lock CANCELs (352)
5876 - MDS can do multi-client recovery (modulo bugs in new code)
5878 - many updates, edits, cleanups
5880 2002-11-18 Phil Schwan <phil@clusterfs.com>
5883 - fix null d_it dereference (346)
5884 - fix full OST/dbench hang (333)
5885 - fix permission problem with file removal (286)
5886 - fix removal of OSCs from LOV when they fail
5887 - fix NULL deref during bulk timeout (214)
5888 - fix problems related to multiple filesystems on one MDS (241)
5889 - fixed serious subtle metadata locking bugs
5890 - free locks on clients when inodes are removed due to memory
5892 - fix inode pointer in lock data (285)
5893 - partial support for multiple MDS on a single host (241)
5894 - data locks weren't cancelled at clear_inode time (290, 311)
5895 - intent locks could lead to unbounded lock growth (205)
5896 - added a maximum lock count, an LRU list, and a flusher
5897 - fix multiple rename (365)
5898 - properly abstracted the echo client
5899 - OSC locked 1 byte too many; fixed
5900 - rewrote brw callback code:
5901 - fixed recovery bugs related to LOVs (306)
5902 - fixed too-many-pages-in-one-write crash (191)
5903 - fixed (again) crash in sync_io_timeout (214)
5904 - probably fixed callback-related race (385)
5906 - Add capability to MDS protocol
5907 - LDLM cancellations and callbacks on different portals
5909 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5912 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5913 - defer open object destruction to close time (601981, 138)
5914 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5915 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5916 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5917 - don't call obd_setattr in ll_file_release for destroyed objects
5918 * protocol change to lustre_msg: move |version| and add |flags|
5919 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5920 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5921 * added replay of create, unlink, link and rename operations during
5922 MDS failover; recovery should be much more robust now
5923 * remove failed OSCs from LOVs (only lov_create uses this so far)
5924 * the lustre-HOWTO was brought (more) up to date (582544)
5926 2002-10-23 Phil Schwan <phil@clusterfs.com>
5929 - in-use dentries weren't being reused properly (617851)
5930 - prevent multiple LDLM setup (599178)
5931 - fix LOV size calculations for truncate (617853)
5932 - fix client handling of MDS intent errors (POSIX)
5933 - fix permission bug in lovstripe.c test (624321)
5934 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5935 - truncate past end of file could corrupt data
5936 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5937 - a race in recovery could return ETIMEDOUT to apps (623947)
5938 - building outside the source directory was fixed
5939 * the lustre-HOWTO was brought (more) up to date (582544)
5940 * major progress was made on recovery functionality
5942 2002-10-10 Phil Schwan <phil@clusterfs.com>
5945 - recovery deadlock fix
5946 - rm -rf causes LBUG fix (617817)
5947 - file open by multiple tasks fix (618962)
5948 - directory permissions bugs (602707 and 620007)
5949 - journal_stop fixed with locking (611313)
5950 - O_APPEND failures resolved (618273, perhaps 614459)
5951 - lconf PATH fix (619770)
5952 - IA64 build fix (621450)
5953 - RPC buffer sizes scale with amount of memory
5955 2002-10-01 Phil Schwan <phil@clusterfs.com>
5958 - locks would be cancelled without throwing away data pages,
5959 resulting in inconsistent data (605627)
5960 - inode attributes were not always being refreshed (605627, 612449)
5961 - lconf now continues to cleanup after lctl reports an error
5962 - MDS now enforces user permissions (602707)
5963 - lprocfs cleanup fixed, but not yet enabled (614157)
5964 - fixed infinite server hang, should a client not respond to an AST
5965 - avoid going into recovery if user calls readlink() with a buffer
5966 that's too small (613941)
5967 - AST RPCs no longer require replies (614867) -- this may be changed
5968 - don't crash server if client sends an IOV that's too big (611336)
5969 - fixed lock conversion deadlock (611892)
5970 - fixed the following of symlinks (614622)
5971 * recovery: the server can remove locks from a client that dies, other
5972 clients can make progress
5973 * more extN patch fixes
5974 * compile-time configurable ptlrpc buffer allocations
5976 - collaborative read cache document
5977 - Lustre Lite Performance CDR document-in-progress
5979 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5982 - fix typo in patch-2.4.18
5984 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5987 - clear ptlrpc request each time in handle_incoming_request()
5988 - unlink of files now destroys the object on the OST
5990 2002-09-19 Peter Braam <braam@clusterfs.com>
5992 * add hard link support
5993 * change obdfile creation method
5994 * kernel patch changed
5996 2002-09-19 Peter Braam <braam@clusterfs.com>
5999 - stack overflow bug in extN fixed
6001 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
6003 * documentation updates
6004 - add man pages for config tools
6005 - update tests/README to describe testing with new config tools
6006 - finish metadata API descriptions
6007 * bug fixes and cleanups
6008 - statfs workaround for 16TB limit
6009 - LOV stripe allocation improved, can stripe on subset of OSTs
6010 - LOV file size/IO offset was wrong for files > 4GB in size
6011 - object EA data was being dropped, caused files to be unreadable
6012 - memory overflow with non-LOV OST caused memory corruption
6013 - fixed regression tests to work with new config tools, obdfilter
6014 - fixed bug when directory size became larger than 1 block
6015 - fixed bug (for single client case) when PWD was deleted
6016 - invalidate local directory pages when doing intent-based ops
6017 - avoid LDLM oops when lock callback contained bad data
6019 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
6021 * documentation updates
6022 * bug fixes and cleanups
6023 - configuration tools
6026 - 64-bit compile warnings
6027 - 64-bit internal statfs data
6029 * test_brw on persistent OST devices
6031 * lprocfs (disabled)
6033 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
6035 * documentation updates
6036 * bug fixes and cleanups
6037 * configuration tools
6039 2002-08-30 Peter J. Braam <braam@clusterfs.com>
6042 * many small fixes to 0_5_4
6043 * io/network handling
6044 * thinkos in MDS operations
6046 2002-08-24 Peter J. Braam <braam@clusterfs.com>
6049 * crucial basic fixes to 0.5.3
6050 * IOR, Iozone work over Elan
6053 2002-08-07 Phil Schwan <phil@clusterfs.com>
6054 * version 0_5_3, our first alpha
6055 * we use the new Portals iovs
6056 * documentation updates
6057 * bug fixes and cleanups
6058 * small changes in the DLM wire protocol
6060 2002-07-25 Peter J. Braam <braam@clusterfs.com>
6061 * version 0_5_1 with some initial stability,
6062 * locking on MD and file I/O.
6063 * documentation updates
6064 * several bug fixes since 0.5.0
6065 * small changes in wire protocol
6067 2002-07-18 Phil Schwan <phil@clusterfs.com>
6069 * delivered as Lustre Light Alpha
6070 * fixed a crash after handling invalid MDS requests
6071 * fixed directory pages for architectures with non-4k pages sizes
6073 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
6074 * release version v0_4_4
6075 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
6077 * Fixes a number of interruption problems with OST operations.
6078 * Update documentation for portals header changes
6079 * Move all wire protocol structs/defines to lustre_idl.h
6080 * Fixes symlink length bug.
6081 * Add tcpdump to repository.
6083 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
6084 * release version v0_4_3
6085 * Fixes statfs for inodes on extN.
6086 * Fixes bug in runtests which would delete /etc/hosts.
6087 * Use 64-bit object IDs wherever possible (not into VFS though)
6088 Remove ost_get_info, which is unused by lustre, and out of date.
6090 2002-07-03 Peter Braam <braam@clusterfs.com>
6091 * release version v0_4_2 Fixes a lookup error (type not passed)
6092 * move forward to head of Portals
6093 * move forward to latest Lustre kernel
6095 2002-06-25 Peter Braam <braam@clusterfs.com>
6096 * release version v0_4_1. Hopefully stable on single node use.