1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.37 (SLES 10),
5 2.6.18-128.1.6.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.41.6.sun1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
14 * File join has been disabled in this release, refer to Bugzilla 16929.
19 Descriptoin: lock ordering violation between &cli->cl_sem and _lprocfs_lock
20 Details : move ldlm namespace creation in setup phase to avoid grab
21 _lprocfs_lock with cli_sem held.
25 Description: Temporarily disable grant shrink.
26 Details : Disable the feature for debugging.
30 Description: Unable to run several mkfs.lustre on loop devices at the same
32 Details : mkfs.lustre returns error 256 on the concurrent loop devices
33 formatting. The solution is to proper handle the error.
35 Severity : enhancement
37 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
39 Severity : enhancement
41 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
45 Description: resolve race between obd_disconnect and class_disconnect_exports
46 Details : if obd_disconnect will be called to already disconnected export he
47 forget release one reference and osc module can't unloaded.
49 Severity : enhancement
51 Description: Allow tuning service thread via /proc
52 Details : For each service a new
53 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
54 created that can be used to set min/max thread counts, and get the
55 current number of running threads.
59 Descriptoin: don't return error if have particaly created objects for file.
60 Details : lov_update_create_set uses set->set_success as index for created objects,
61 so if some requests will be failed, they will have hole at end of
62 array and we can use qos_shrink_lsm for allocate correct lsm.
64 Severity : enhancement
66 Description: Update OFED support to 1.4.1
70 Description: Reduce small size read RPC
71 Details : Set read-ahead limite for every file and only do read-ahead when
72 available read-ahead pages are bigger than 1M to avoid small size
75 Severity : enhancement
77 Description: move AT tunable parameters for more consistent usage
78 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
80 Severity : enhancement
82 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
83 Details : allow skip disconnected ost for send statfs request and hide error
89 Description: fix racy locking of mballoc block bitmaps causing BUG
90 Details : The locking of the mballoc buddy bitmap and the in-memory
91 block bitmap was using two different spin locks in some
92 cases. This made it possible to incorrectly access the
93 mballoc bitmap while another process was modifying it,
94 causing a sanity assertion to fail. While no on-disk corruption
95 was reported, there was some risk of this happening.
98 Frequency : rare, on llog test 6
100 Descriptoin: don't allow connect to already connected import
101 Details : allowing connect to already connected import is hide connecting problem.
106 Descriptoin: allow kill process which wait statahead result
107 Details : for some reasons 'ls' can stick in waiting result from statahead,
108 in this case need way for kill this process.
110 Severity : enhancement
112 Description: Add state history info file, enhance import info file
113 Details : Track import connection state changes in a new osc/mdc proc file;
114 add overview-type data to the osc/mdc import proc file.
116 Severity : enhancement
118 Description: MDS create should not wait for statfs RPC while holding DLM lock.
120 Severity : enhancement
122 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
127 Descriptoin: OST grant too much space to client even there are not enough space.
128 Details : Client will shrink its grant cache to OST if there are no write
129 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
130 this grant cache if there are already not enough avaible space
131 (left_space < total_clients * 32M).
134 Frequency : start MDS on uncleanly shutdowned MDS device
136 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
137 Details : stay in waiting mds<>ost recovery finished produce random bugs
138 due race between two ll_sync thread for one lov target. send
139 ACTIVATE event only if connect realy finished and import have
143 Frequency : rare, connect and disconnect target at same time
145 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
146 Details : don't call obd_disconnect under lov_lock. this long time
147 operation and can block ptlrpcd which answer to connect request.
152 Descriptoin: don't lose wakeup for imp_recovery_waitq
153 Details : recover_import_no_retry or invalidate_import and import_close can
154 both sleep on imp_recovery_waitq, but we was send only one wakeup
158 Frequency : always with long access acl
160 Descriptoin: mds can't pack reply with long acl.
161 Details : mds don't control size of acl but they limited by reint/getattr
164 Severity : enhancement
166 Description: Update to SLES10 kernel-2.6.16.60-0.33.
168 Severity : enhancement
170 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
173 Frequency : start MDS on uncleanly shutdowned MDS device
175 Descriptoin: aborting recovery hang on MDS
176 Details : don't throttle destroy RPCs for the MDT.
179 Frequency : on remount
181 Description: external journal device not working after the remount
182 Details : clear dev_rdonly flag for external journal devices in
188 Description: shutdown vs evict race
189 Details : client_disconnect_export vs connect request race.
190 if client will evicted at this time - we start invalidate
191 thread without referece to import and import can be freed
195 Frequency : rare, need acl's on inode.
197 Description: client can't handle ost additional correctly
198 Details : if ost was added after client connected to mds client can have
199 hit lnet_try_match_md ... to big messages to wide striped files.
200 in this case need teach client to handle config events about add
201 lov target and update client max ea size at that event.
203 Severity : enhancement
205 Description: Changelogs
206 Details : Changelogs are a lightweight mechanism to track filesystem
207 metadata and namespace changes. The changelog is recorded
208 permanently on the MDTs, and is periodically "consumed" / purged
209 when records are no longer needed.
211 Severity : enhancement
213 Description: compact fld format with extents
214 Details : Store range of seq rather than every seq in FLD. Seq
215 controller update FLD rather than clients. In Case of CMD, mdt0
216 has FLD, all other metadata server act as non persistent proxy
217 for FLD queries and cache fld entries in fld cache.
222 Description: don't skip ost target if they assigned to file
223 Details : Drop slow OSCs if we can, but not for requested start idx.
224 This means "if OSC is slow and it is not the requested
225 start OST, then it can be skipped, otherwise skip it only
226 if it is inactive/recovering/out-of-space.
230 Description: more cleanup in mds_lov
231 Details : not send LOV EA under replay, we can't know about they size at this
232 time. Don't allow client connect to mds before any ost connected,
233 for avoid problems with LOV EA size and returning EIO to client.
235 Severity : enhancement
237 Description: Interoperability at server side (Disk interoperability)
239 Severity : enhancement
241 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
243 Severity : enhancement
245 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
247 Severity : enhancement
249 Description: New client IO stack (CLIO).
251 Severity : enhancement
253 Description: Commit on sharing. Eliminate inter-client dependencies between
254 uncommitted transactions by doing transaction commits.
255 Thereby clients may recovery independently.
258 Frequency : Create a symlink file with a very long name
260 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
261 Details : If there is no extra space in the request for early cancels,
262 ldlm_req_handles_avail() returns 0 instead of a negative value.
264 Severity : enhancement
266 Description: Add /proc entry for import status
267 Details : The mdc, osc, and mgc import directories now have
268 an import directory that contains useful import data for debugging
271 Severity : enhancement
273 Description: Re-disable certain /proc logging
274 Details : Enable and disable client's offset_stats, extents_stats and
275 extents_stats_per_process stats logging on the fly.
278 Frequency : Only on FC kernels 2.6.22+
280 Description: oops in statahead
281 Details : Do not drop reference count for the dentry from VFS when lookup,
282 VFS will do that by itself.
284 Severity : enhancement
286 Description: Generic /proc file permissions
287 Details : Set /Proc file permissions in a more generic way to enable non-
288 root users operate on some /proc files.
292 Description: Hitting mdc_commit_close() ASSERTION
293 Details : Properly handle request reference release in
294 ll_release_openhandle().
298 Description: quota recovery deadlock during mds failover
299 Details : This patch includes att18982, att18236, att18237 in bz14840.
301 1. fix osts hang when mds does failover with quotaon
302 2. prevent watchdog storm when osts threads wait for the
307 Frequency : only patchless client
308 Description: add workaround for race between add/remove dentry from hash
310 Severity : enhancement
312 Description: Allow OST glimpses to return PW locks
316 Description: LBUG when llog conf file is full
317 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
321 Description: Prevent import from entering FULL state when server in recovery
325 Description: service mount cannot take device name with ":"
326 Details : Only when device name contains ":/" will mount treat it as
332 Description: replace ptlrpcd with the statahead thread to interpret the async
333 statahead RPC callback
337 Frequency : on recovery
338 Description: I/O failures after umount during fail back
339 Details : if client reconnected to restarted server we need join to recovery
340 instead of find server handler is changed and process self
341 eviction with cancel all locks.
343 Severity : enhancement
345 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
347 Severity : enhancement
349 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
351 Severity : enhancement
353 Description: Upcall on Lustre log has been dumped
354 Details : Allow for a user mode script to be called once a Lustre log has
355 been dumped. It passes the filename of the dumped log to the
356 script, the location of the script can be specified via
357 /proc/sys/lnet/debug_log_upcall.
362 Description: avoid idr_remove called for id which is not allocated.
363 Details : Move assigment s_dev for clustered nfs to end of initialization,
364 to avoid problem with error handling.
369 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
370 Details : When connection is reused this not moved from CONN_UNUSED_HASH
371 into CONN_USED_HASH and this prodice warning when put connection
372 again in unused hash.
375 Severity : enhancement
377 Description: File striping can now be set to use an arbitrary pool of OSTs.
379 Severity : enhancement
381 Description: Export bytes_read/bytes_write count on OSC/OST.
385 Description: Early reply size mismatch, MGC loses connection
386 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
387 the connect flags are properly negotiated.
392 Description: quotas are not honored with O_DIRECT
393 Details : all writes with the flag O_DIRECT will use grants which leads to
394 this problem. Now using OBD_BRW_SYNC to guard this.
398 Description: add quota statistics
399 Details : 1. sort out quota proc entries and proc code.
400 2. add quota statistics
402 Severity : enhancement
404 Description: enable quota support for HEAD.
408 Description: Properly propagate oinfo flags from lov to osc for statfs
409 Details : restore missing copy oi_flags to lov requests.
411 Severity : enhancement
413 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
414 : l_getgroups(8), lst(8), routerstat(8)
415 Details : included man pages for llobdstat(8), llstat(8),
416 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
418 Severity : enhancement
420 Description: configure's --enable-quota should check the
421 : kernel .config for CONFIG_QUOTA
422 Details : configure is terminated if --enable-quota is passed but
423 : no quota support is in kernel
427 Description: Remove portals compatibility
428 Details : Remove portals compatibility, not interoperable with releases
433 Description: Resolve device initialization race
434 Details : Prevent proc handler from accessing devices added to the
435 obd_devs array but yet be intialized.
437 Severity : enhancement
439 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
441 Severity : enhancement
443 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
448 Description: Using wrong pointer in osc_brw_prep_request
449 Details : Access to array[-1] can produce panic if kernel compiled with
450 CONFIG_PAGE_ALLOC enabled
454 Description: Client runs out of low memory
455 Details : Consider only lowmem when counting initial number of llap pages
459 Description: *optional* service tags registration
460 Details : if the "service tags" package is installed on a Lustre node
461 When the filesystem is mounted, a local-node service tag will
462 be created. See http://inventory.sun.com/ for more information
463 about the Service Tags asset management system.
467 Description: Kernel BUG tries to release flock
468 Details : Lustre does not destroy flock lock before last reference goes
469 away. So always drop flock locks when client is evicted and
470 perform unlock regardless of successfulness of speaking to MDS.
474 Description: add refcount protection for osc callbacks, avoid panic on shutdown
478 Description: sanity test 65a fails if stripecount of -1 is set
479 Details : handle -1 striping on filesystem in ll_dirstripe_verify
484 Description: ASSERTION(CheckWriteback(page,cmd)) failed
485 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
488 Severity : enhancement
490 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
494 Description: do not process already freed flock
495 Details : flock can possibly be freed by another thread before it reaches
496 to ldlm_flock_completion_ast.
500 Description: LBUG during stress test
501 Details : Need properly lock accesses the flock deadlock detection list.
505 Description: oops in page fault handler
506 Details : kernel page fault handler can return two special 'pages' in error
507 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
511 Description: timeout with invalidate import.
512 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
513 should be handled by ptlrpcd. This produce long age waiting and
514 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
516 Severity : enhancement
518 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
522 Description: enable MGS and MDT services start separately
523 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
524 a co-located MGS without starting the MGS, which is a complement
525 to 'nosvc' mount option.
529 Frequency : after recovery
530 Description: precreate to many object's after del orphan.
531 Details : del orphan st in oscc last_id == next_id and this triger growing
532 count of precreated objects. Set flag LOW to skip increase count
533 of precreated objects.
537 Frequency : rare, on clear nid stats
538 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
539 Details : when clean nid stats sometimes try destroy live entry,
540 and this produce panic in free.
544 Description: Stack overflow during MDS log replay
545 ease stack pressure by using a thread dealing llog_process.
549 Description: wait until IO finished before start new when do lock cancel.
550 Details : VM protocol want old IO finished before start new, in this case
551 need wait until PG_writeback is cleared until check dirty flag and
552 call writepages in lock cancel callback.
554 Severity : enhancement
556 Description: using special macro for print time and cleanup in includes.
560 Description: mds_mfd_close() ASSERTION(rc == 0)
561 Details : In mds_mfd_close(), we need protect inode's writecount change
562 within its orphan write semaphore to prevent possible races.
566 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
570 Frequency : rare, on shutdown ost
571 Description: don't hit live lock with umount ost.
572 Details : shrink_dcache_parent can be in long loop with destroy dentries,
573 use shrink_dcache_sb instead.
577 Description: don't panic with use echo client
578 Details : echo client pass NULL as client nid pointer and this produce null
583 Description: fix build on ppc32
584 Details : compile code with -m64 flag produce wrong object file for ppc32.
588 Description: add message levels for liblustreapi
592 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
593 Details : Move the 'good_osts' check before the 'total_bavail' check. This
594 will result in an -EAGAIN and in the exit call path we call
595 alloc_rr() which will with increasing aggressiveness attempt to
596 aquire precreated objects on the minimum number of required OSCs.
600 Description: Use old size assignment to avoid deadlock
601 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
602 the scheduling while holding a spinlock. We do not need locking
603 for size in ll_update_inode() because size is only updated from
604 the MDS for directories or files without objects, so there is no
605 other place to do the update, and concurrent access to such inodes
606 are protected by the inode lock.
610 Description: resolve "_IOWR redefined" build error on SLES10
614 Description: dump the memory debugging after all modules are unloaded to
615 suppress false negative in conf_sanity test 39
617 Severity : enhancement
619 Description: build kernel-ib packages for OFED 1.3 in our release cycle
624 Description: fix SLES kernel versioning
625 Details : the kernel version for our SLES 10 kernel did not include a "-"
626 before the "smp" at the end. while this was not a problem in
627 general, it did mean that software trying to use the kernel
628 version to try to detect a vendor specific kernel would fail.
629 this was most evident by the OFED build scripts.
633 Description: Don't update lov_desc members until making sure they are valid
634 Details : When updating lov_desc members via proc fs, need fix their
635 validities before doing the real update.
639 Description: don't put request into delay list while invalidate in flight.
640 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
641 import in flight. this produce timeout for invalidate and sometimes
642 can cause stale data.
646 Frequency : on ppc only
647 Description: not convert ost objects for directory because it's not exist.
648 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
652 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
654 Severity : enhancement
656 Description: Update to support 2.6.22.14 vanilla kernel.
660 Frequency : rare, on recovery
661 Description: read procfs can produce deadlock in some situation
662 Details : Holding lprocfs lock which send rpc can produce block for destroy
663 obd objects and this also block reconnect with -EALREADY. This
664 isn't fix all lprocfs bugs - but make it rare.
666 Severity : enhancement
668 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
671 Frequency : frequent on X2 node
673 Description: mdc_set_open_replay_data LBUG
674 Details : Set replay data for requests that are eligible for replay.
678 Description: lustre_mgs: operation 101 on unconnected MGS
679 Details : When MGC is disconnected from MGS long enough, MGS will evict the
680 MGC, and late on MGC cannot successfully connect to MGS and a lot
681 of the error messages complaining that MGS is not connected.
685 Frequency : on network error
686 Description: panic with double free request if network error
687 Details : mdc_finish_enqueue is finish request if any network error ocuring,
688 but it's true only for synchronus enqueue, for async enqueue
689 (via ptlrpcd) this incorrect and ptlrpcd want finish request
692 Severity : enhancement
694 Description: client-side metadata stat-ahead during readdir(directory readahead)
695 Details : perform client-side metadata stat-ahead when the client detects
696 readdir and sequential stat of dir entries therein
699 Frequency : on start mds
701 Description: Implement get_info(last_id) in obdfilter.
704 Frequency : occasional
706 Description: Correctly check stale fid, not start epoch if ost not support SOM
707 Details : open with flag O_CREATE need set old fid in op_fid3 because
708 op_fid2 was overwritten with new generated fid, but mds can answer
709 with one of these two fids and both is not stale. Setattr
710 incorrectly started an epoch and assume will be called
711 done_writing, but without SOM done_writing ever being called.
714 Frequency : rare, depends on device drivers and load
716 Description: MDS or OSS nodes crash due to stack overflow
717 Details : Code changes in 1.8.0 increased the stack usage of some functions.
718 In some cases, in conjunction with device drivers that use a lot
719 of stack the MDS (or possibly OSS) service threads could overflow
720 the stack. One change which was identified to consume additional
721 stack has been reworked to avoid the extra stack usage.
724 Frequency : occasional
726 Description: Do not fail import if osc_interpret_create gets -EAGAIN
727 Details : If osc_interpret_create got -EAGAIN it immediately exits and
728 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
729 oscc_has_objects and see OSC has no objests and call
730 oscc_internal_create to resend create request.
732 Severity : enhancement
734 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
736 Severity : enhancement
738 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
741 Frequency : very rare
743 Description: Fix replay if there is an un-replied request and open
744 Details : In some cases, older replay request will revert the
745 mcd->mcd_last_xid on MDS which is used to record the client's
748 Severity : enhancement
750 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
752 Severity : enhancement
754 Description: Add rhel5 support to HEAD.
756 Serverity : enhancement
758 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
763 Description: Don't allow skipping OSTs if index has been specified.
764 Details : Don't allow skipping OSTs if index has been specified, make
765 locking in internal create lots better.
769 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
770 Details : access to bitfield in structure is always rounded to long
771 and this produce problem with not atomic change any bit.
775 Description: Lustre make rpms failed.
776 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
777 builds Lustre rpms from tarball.
780 Frequency : rare on shutdown ost
782 Description: If llog cancel was not send before clean_exports phase, this can
783 produce deadlock in llog code.
784 Details : If llog thread has last reference to obd and call class_import_put
785 this produce deadlock because llog_cleanup_commit_master wait when
786 last llog_commit_thread exited, but this never success because was
787 called from llog_commit_thread.
791 Description: allow userland application know is lost one of stripes.
792 Details : fill lvb_blocks with error code on ost and return it to
793 application if error flag found.
797 Description: NULL lov_tgts causing MDS oops
798 Details : more safe checks for NULL lov_tgts for avoid oops.
800 Severity : enhancement
802 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
806 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
807 Details : If an OST is down the MDS will hang indefinitely in
808 obd_statfs_rqset() waiting for the statfs data. While for
809 MDS QOS usage of statfs, it should not stuck in waiting.
811 Severity : enhancement
813 Description: remote_acl support
814 Details : Support ACL-based permission check for remote user.
815 Support setfacl/getfacl for remote user with the utils
816 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
817 the system "{s,g}etfacl" utils.
819 Severity : enhancement
821 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
823 Severity : enhancement
825 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
829 Description: make mgs_setparam() handle fsname containing dash
830 Details : fsname containing a dash does not work with lctl conf_param
832 Severity : enhancement
834 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
836 Severity : enhancement
838 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
842 Frequency : rare, at shutdown
843 Description: access already free / zero obd_namespace.
844 Details : if client_disconnect_export was called without force flag set,
845 and exist connect request in flight, this can produce access to
846 NULL pointer (or already free pointer) when connect_interpret
847 store ocd flags in obd_namespace.
851 Frequency : only at startup
852 Description: not alloc memory with spinlock held.
853 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
854 if any spinlock held.
856 Severity : enhancement
858 Description: make lustre randomly fail allocating memory
859 Details : Make lustre randomly failed allocating memory for testing purpose.
861 Severity : enhancement
863 Description: lost problems with lov objid file
864 Details : Fixes some scability and access to not inited memory problems
865 in work with lov objdid file.
870 Description: lfs find does not continue on file error
871 Details : Continue other files processing when a file/dir is absent.
875 Description: Inconsistent usage of lustre_pack_reply()
876 Details : Standardize the usage of lustre_pack_reply() such that it
877 always generate a CERROR on failure.
880 Frequency : occasional
882 Description: MDS hang or stay in waiting lock
883 Details : If client receive lock with CBPENDING flag ldlm need send lock
884 cancel as separate rpc, to avoid situation when cancel request
885 can't processed due all i/o threads stay in wait lock.
889 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
893 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
897 Description: Update to SLES9 kernel-2.6.5-7.287.3.
901 Description: Update to RHEL4 latest kernel.
903 Severity : enhancement
905 Description: Build SLES10 patchless client fails
906 Details : The configure was broken by run ./configure with
907 --with-linux-obj=.... argument for patchless client. When the
908 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
909 can't search header adequately. Use absolute path such as
910 -I($LINUX)/include instead.
914 Description: interrupt oig_wait produce painc on resend.
915 Details : brw_redo_request can be used for resend requests from ptlrpcd and
916 private set, and this produce situation when rq_ptlrpcd_data not
917 copyed to new allocated request and triggered LBUG on assert
918 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
919 ptlrpcd set if request is changed and can be safety changed to use
924 Description: LASSERT_{REQ,REP}SWAB macros are buggy
925 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
926 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
927 these macros and replace them with their difinitions instead.
931 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
932 Details : Update which_patch & target file for SLES10 latest kernel.
935 Description: Remove client patches from SLES 10 kernel.
936 Details : This causes SLES 10 clients to behave as patchless clients
937 even on a Lustre-patched (server) kernel.
939 Severity : enhancement
941 Description: self-adjustable client's lru lists
942 Details : use adaptive algorithm for managing client cached locks lru
943 lists according to current server load, other client's work
944 pattern, memory activities, etc. Both, server and client
945 side namespaces provide number of proc tunables for controlling
948 Severity : enhancement
950 Description: light-weight GSS support
951 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
952 and reduce performance overhead.
954 Severity : enhancement
956 Description: Linux keyring support
957 Details : Support using service of Linux keyring for Lustre GSS internal
958 context refresh/cache mechanism.
962 Description: Fix errors in lfs documentation
963 Details : Fixes man pages
967 Description: don't use GFP_* in generic Lustre code.
968 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
972 Description: obdclass is limited by single OBD_ALLOC(idarray)
973 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
977 Description: warnings with build patchless client with vanila 2.6.19 and up
978 Details : change old ctl_table style and replace ctl_table/ctl_table_header
979 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
983 Frequency : Only for SLES
984 Description: sanity_quota fail test_1
985 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
986 file, which makes TSTID[2] inunique.
990 Frequency : Only for SLES9
991 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
992 Details : Update target/ChangeLog/which_patch .
996 Frequency : Only for test_10 in sanity-quota.sh
997 Description: a bug in quota test code
998 Details : checking a test flag in a wrong place
1002 Frequency : Only for test_13 in sanity-quota.sh
1003 Description: a bug in quota test script
1004 Details : shouldn't check overfull allocated quota
1008 Description: jbd statistics
1009 Details : Port older jbd statistics patch for sles10
1011 Severity : enhancement
1013 Description: Add printing inode info into message about error in writepage.
1018 Description: reply_lock_interpret crash due to race with it and lock cancel.
1019 Details : Do not replay locks that are being cancelled. Do not reference
1020 locks by their address during replay, just by their handle.
1024 Frequency : When flocks are used.
1025 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1026 Details : Flock locks might destroy just granted lock if it could be merged
1027 with another existing flock, this is done in completion handler,
1028 so teach ldlm_cli_enquque_fini that this is a valid case for
1034 Description: Oops in read and write path when failing to allocate lock.
1035 Details : Check if lock allocation failed and return error back.
1039 Description: lstripe command fails for valid OST index
1040 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1041 instead of lov->desc.ld_active_tgt_count.
1045 Description: sanity.sh failed test 103
1046 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1051 Description: df doesn't work properly if diskfs blocksize != 4K
1052 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1056 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1057 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1059 Severity : enhancement
1061 Description: Adaptive timeouts
1062 Details : RPC timeouts adapt to changing server load and network
1063 conditions to reduce resend attempts and improve recovery time.
1067 Description: llapi_file_create() does not allow some changes
1068 Details : add llapi_file_open() that allows specifying the mode and
1069 open flags, and also returns an open file handle.
1073 Description: merge and cleanup kernel patches.
1074 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1078 Description: Add journal checksum support.(Kernel part)
1079 Details : The journal checksum feature adds two new flags i.e
1080 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1081 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1082 indicates that the commit block contains the checksum for
1083 the blocks described by the descriptor blocks. Now commit
1084 record can be sent to disk without waiting for descriptor
1085 blocks to be written to disk. This behavior is controlled
1086 using JBD2_FEATURE_ASYNC_COMMIT flag.
1090 Description: OSS needs mutliple precreate threads
1091 Details : Add ability to start more than one create thread per OSS.
1095 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1096 Details : execute lfs setstripe on client
1100 Description: mds_obd_create error creating tmp object
1101 Details : When the user sets quota on root, llog will be affected and can't
1102 create files and write files.
1105 Frequency : Always on ia64 patchless client, and possibly others.
1107 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1108 Details : This allows the patchless client to be loaded on architectures
1109 without this export.
1113 Description: RedHat Update kernel for RHEL5
1114 Details : Add the kernel config file for RHEL5.
1118 Description: RedHat Update kernel for RHEL5
1119 Details : Modify the kernel config file more closer RHEL5.
1123 Description: Build failure against Centos5 (RHEL5)
1124 Details : Define PAGE_SIZE when it isn't present.
1128 Description: test_7 on sanity-quota.sh can't be used on separate machines
1129 Details : add facet to handle it
1133 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1134 Details : with NFS, the anon dentry's parent was set to itself in
1135 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1136 corresponding dentry other than use rec->ur_name.
1138 Severity : enhancement
1140 Description: lfs setstripe enhancement
1141 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1145 Description: enable data checksumming by default
1146 Details : enable checksum by default, allow --disable-checksum
1147 configure option and "-o nochecksum" mount option. Checksums
1148 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1152 Description: lustre support for RHEL5
1153 Details : Add support for RHEL5.
1157 Description: Client eviction due to failover config
1158 Details : after a connection loss, the lustre client should attempt to
1159 reconnect to the last active server first before trying the
1160 other potential connections.
1164 Description: when mds and osts use different quota unit(32bit and 64bit),
1165 quota will be released repeatly.
1166 Details : void sending multiple quota reqs to mds, which will keep the
1167 status between the reqs.
1171 Description: osts not allocated evenly to files
1172 Details : change the condition to increase offset_idx
1175 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1177 Description: Data corruption for OSTs that are formatted larger than 2TB
1179 Details : When generating the bio request for lustre file writes the
1180 sector number would overflow a temporary variable before being
1181 used for the IO. The data reads correctly from Lustre (which
1182 will overflow in a similar manner) but other file data or
1183 filesystem metadata may be corrupted in some cases.
1187 Description: Tune the kernel for good SCSI performance.
1188 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1189 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1194 Description: rewrite ext2-derived code in obdclass/uuid.c
1195 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1196 from scratch preserving functionality.
1200 Description: "lfs find -obd UUID" prints directories
1201 Details : "lfs find -obd UUID" will return all directory names instead
1202 of just file names. It is incorrect because the directories
1203 do not reside on the OSTs.
1207 Description: use i_size_read and i_size_write in 2.6 port
1208 Details : replace inode->i_size access with i_size_read/write()
1211 Frequency : when removing large files
1213 Description: scheduling issue during removal of large Lustre files
1214 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1215 It causes scheduling issues when removing large files (17TB in the
1219 Frequency : only with liblustre clients on XT3
1221 Description: evictions taking too long
1222 Details : allow llrd to evict clients directly on OSTs
1225 Frequency : only on ppc
1227 Description: /proc/fs/lustre/devices broken on ppc
1228 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1229 We should make sure the type of 'index' is loff_t and then cast
1230 explicitly as needed below. Do not assign an explicitly cast
1234 Frequency : only for rhel5
1236 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1237 Details : Modify the target file & which_kernel.
1239 Severity : enhancement
1241 Description: omit set fsid for export NFS
1242 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1243 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1247 Frequency : Always, for kernels after 2.6.16
1248 Description: Fix warning idr_remove called for id=.. which is not allocated.
1249 Details : Last kernels save old s_dev before kill super and not allow
1250 to restore from callback - restore it before call kill_anon_super.
1254 Description: buffer overruns could theoretically occur
1255 Details : llapi_semantic_traverse() modifies the "path" argument by
1256 appending values to the end of the origin string, and a
1257 overrun may occur. Adding buffer overrun check in liblustreapi.
1261 Description: Fix error on 'ls .' at the top of the Lustre mount.
1262 Details : Don't revalidate dentry if it is a root dentry.
1266 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1267 Details : Modify vm-tunables-rhel4.patch.
1271 Description: Kernel config for 2.6.18-vanilla.
1272 Details : Modify targets/2.6-vanilla.target.in.
1273 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1274 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1275 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1276 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1280 Description: improve handling recoverable errors
1281 Details : if request processig with error which can be recoverable on server
1282 request should be resend, otherwise page released from cache and
1287 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1288 Details : Modify target file & which_patch.
1289 A flaw was found in the IA32 system call emulation provided
1290 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1291 value could be stored in the %RAX register, which could trigger an
1292 out-of-bounds system call table access. An untrusted local user
1293 could exploit this flaw to run code in the kernel
1294 (ie a root privilege escalation). (CVE-2007-4573).
1298 Description: O_DIRECT bypasses client statistics.
1299 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1300 read_ahead_stats were updated. copied the stats section from
1301 osc_send_oap_rpc() into async_internal().
1305 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1309 Description: change order of libsysio includes
1310 Details : '#include sysio.h' should always come before '#include xtio.h'
1314 Description: handle "serious error: objid * already exists" more gracefully
1315 Details : If LAST_ID value on disk is smaller than the objects existing in
1316 the O/0/d* directories, it indicates disk corruption and causes an
1317 LBUG(). If the object is 0-length, then we should use the existing
1318 object. This will help to avoid a full fsck in most cases.
1320 Severity : enhancement
1322 Description: adapt the lustre_config script to support the upgrade case
1323 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1324 server targets to 1.6 in parallel.
1328 Description: To avoid grant space > avaible space when the disk is almost
1329 full. Without this patch you might see the error "grant XXXX >
1330 available" or some LBUG about grant, when the disk is almost
1332 Details : In filter_check_grant, for non_grant cache write, we should
1333 check the left space by if (*left > ungranted + bytes), instead
1334 of (*left > ungranted), because only we are sure the left space
1335 is enough for another "bytes", then the ungrant space should be
1336 increase. In client, we should update cl_avail_grant only there
1337 is OBD_MD_FLGRANT in the reply.
1341 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1342 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1343 described in CVE-2007-4573. This problem could allow untrusted
1344 local users to gain root access.
1347 Frequency : when using O_DIRECT and quotas
1349 Description: Incorrect file ownership on O_DIRECT output files
1350 Details : block usage reported by 'lfs quota' does not take into account
1351 files that have been written with O_DIRECT.
1356 Description: touch file failed when fs is not full
1357 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1358 otherwise we can get ENOSP while fs is not full.
1362 Description: parallel lock callbacks
1363 Details : Instead of sending blocking and completion callbacks as separated
1364 requests, adding them to a set and sending in parallel.
1367 Frequency : only for Cray XT3
1368 Bugzilla : 12829/13455
1369 Description: Changing primary group doesn't change the group lustre assigns to
1371 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1372 overridden with the primary group provided by the group upcall,
1373 whereas the supplied fsgid can be trusted if it is in the list of
1374 supplementary groups returned by the group upcall.
1376 Severity : enhancement
1378 Description: Allow masking D_WARNING, D_ERROR messages from console
1379 Details : Console messages can now be disabled via lnet.printk.
1383 Description: User code with malformed file open parameter crashes client node
1384 Details : Before packing join_file req, all the related reference should be
1385 checked carefully in case some malformed flags cause fake
1386 join_file req on client.
1390 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1391 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1392 mode is equal to the granted mode. However, it should also check
1393 that LDLM_FL_CP_REQD is not set.
1397 Description: Heavy nfs access might result in deadlocks
1398 Details : After ELC code landed, it is now improper to enqueue any mds
1399 locks under och_sem, because enqueue might want to decide to
1400 cancel open locks for same inode we are holding och_sem for.
1404 Description: Client eviction while running blogbench
1405 Details : A lot of unlink operations with concurrent I/O can lead to a
1406 deadlock causing evictions. To address the problem, the number of
1407 oustanding OST_DESTROY requests is now throttled to
1408 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1409 callbacks are processed in priority.
1413 Description: enable ACLs on MDS by default
1414 Details : ACLs must be enabled on MDS by default.
1417 Frequency : PPC/PPC64 only
1419 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1420 Details : fix duplicated definitions between asm-ppc64/types.h and
1421 lustre_types.h on PPC.
1424 Frequency : PPC/PPC64 only
1426 Description: asm-ppc/segment.h does not exist
1427 Details : fix compile issue on PPC.
1431 Description: better handle error messages in extents code
1434 Frequency : RHEL4 only
1436 Description: mkfs is very slow on IA64/RHEL4
1437 Details : A performance regression has been discovered in the MPT Fusion
1438 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1439 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1440 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1442 Severity : enhancement
1444 Description: SNMP support enhancement
1445 Details : Adding total number of sampled request for an MDS node in snmp
1448 Severity : enhancement
1450 Description: Optimize ldlm waiting list processing for PR extent locks
1451 Details : When processing waiting list for read extent lock and meeting read
1452 lock that is same or wider to it that is not contended, skip
1453 processing rest of the list and immediatelly return current
1454 status of conflictness, since we are guaranteed there are no
1455 conflicting locks in the rest of the list.
1459 Description: Time out and refuse to reconnect
1460 Details : When the failover node is the primary node, it is possible
1461 to have two identical connections in imp_conn_list. We must
1462 compare not conn's pointers but NIDs, otherwise we can defeat
1463 connection throttling.
1467 Description: port llog fixes from b1_6 into HEAD
1468 Details : Port llog reference couting and some llog cleanups from b1_6
1469 (bug 10800) into HEAD, for protect from panic and access to already
1470 free llog structures.
1474 Description: Detect stride IO mode in read-ahead
1475 Details : When a client does stride read, read-ahead should detect that and
1476 read-ahead pages according to the detected stride pattern.
1480 Description: data checksumming impacts single node performance
1481 Details : add support for several checksum algorithm. Currently, only CRC32
1482 and Adler-32 are supported. The checksum type can be changed on
1483 the fly via /proc/fs/lustre/osc/*/checksum_type.
1487 Description: use adler32 for page checksums
1488 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1493 Description: build for x2 fails
1494 Details : fix compile issue on Cray systems.
1498 Description: Properly match for duplicate locks
1499 Details : Due to different lock order from skiplists code, we need to
1500 traverse entire list for now
1503 Frequency : only on PPC/SLES10
1505 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1506 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1507 be defined. Add a hack in mkfs_lustre.c to work around this compile
1512 Description: LASSERT on MDS when client holding flock lock dies
1513 Details : ldlm pool logic depends on number of granted locks equal to
1514 number of released locks which is not true for flock locks, so
1515 just exclude such locks from consideration.
1519 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1520 Details : Use fsfilt_sync() for both the whole filesystem sync and
1521 individual file sync to eliminate dangerous inode locking
1522 with I_LOCK that can lead to a deadlock.
1526 Description: performance in 1.6.3
1527 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1528 sure that 1MB requests are not fragmented by the block layer.
1530 Severity : enhancement
1532 Description: organize the server-side client stats on per-nid basis
1533 Details : Change the structure of stats under obdfilter and mds to
1541 The "uuid"s file would list the uuids of _active_ exports.
1542 And the clear entry is to clear all stats and stale nids.
1544 Severity : enhancement
1546 Description: eliminate client locks in face of contention
1547 Details : file contention detection and lockless i/o implementation
1548 for contended files.
1552 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1555 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1557 Description: MDS may lose file striping (and hence file data) in some cases
1558 Details : If there are additional extended attributes stored on the MDS,
1559 in particular ACLs, SELinux, or user attributes (if user_xattr
1560 is specified for the client mount options) then there is a risk
1561 of attribute loss. Additionally, the Lustre file striping
1562 needs to be larger than default (e.g. striped over all OSTs),
1563 and an additional attribute must be stored initially in the
1564 inode and then increase in size enough to be moved to the
1565 external attribute block (e.g. ACL growing in size) for file
1570 Description: skiplist implementation simplification
1571 Details : skiplists are used to group compatible locks on granted list
1572 that was implemented as tracking first and last lock of each
1573 lock group the patch changes that to using doubly linked lists
1577 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1578 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1579 because LASSERT on both the data supplied by a client, and the
1580 data on disk is dangerous and incorrect.
1582 Severity : enhancement
1584 Description: Slow truncate/writes to huge files at high offsets.
1585 Details : Directly associate cached pages to lock that protect those pages,
1586 this allows us to quickly find what pages to write and remove
1587 once lock callback is received.
1591 Description: more ldlm soft lockups
1592 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1593 starve other threads from the resource lock for a long time in
1594 case of long waiting queue, so change the debug level from
1595 D_OTHER to the less frequently used D_INFO.
1597 Severity : enhancement
1599 Description: add -gid, -group, -uid, -user options to lfs find
1603 Description: Hung threads in invalidate_inode_pages2_range
1604 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1605 once one is completed. As a result, some RPCs are stuck in the
1606 queue and are never sent.
1610 Description: filter threads hungs on waiting journal commit
1611 Details : Cleanup filter group llog code, then only filter group llog will
1612 be only created in the MDS/OST syncing process.
1616 Description: Procfs and llog threads access destoryed import sometimes.
1617 Details : Sync the import destoryed process with procfs and llog threads by
1618 the import refcount and semaphore.
1620 Severity : enhancement
1622 Description: openlock cache of b1_6 port to HEAD
1627 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1628 Details : If insertion of an extent fails, then discard the inode
1629 preallocation and free data blocks else it can lead to duplicate
1634 Description: don't always update ctime in ext3_xattr_set_handle()
1635 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1636 In some cases the ctime should not be updated, for example for
1637 2.0->1.8 compatibility it is necessary to delete an xattr and it
1638 should not update the ctime.
1642 Bugzilla : 15713/16362
1643 Description: Assertion in iopen_connect_dentry in 1.6.3
1644 Details : looking up an inode via iopen with the wrong generation number can
1645 populate the dcache with a disconneced dentry while the inode
1646 number is in the process of being reallocated. This causes an
1647 assertion failure in iopen since the inode's dentry list contains
1648 both a connected and disconnected dentry.
1652 Description: assertion failure in ldlm_handle2lock()
1653 Details : fix a race between class_handle_unhash() and class_handle2object()
1654 introduced in lustre 1.6.5 by bug 13622.
1659 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1660 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1665 Description: oops in sd_iostats_seq_show()
1666 Details : unloading/reloading the scsi low level driver triggers a kernel
1667 bug when trying to access the sd iostat file.
1672 Description: Kernel panics during QLogic driver reload
1673 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1674 patch, causing memory corruption.
1679 Description: journal_dev option does not work in b1_6
1680 Details : pass mount option during pre-mount.
1682 Severity : enhancement
1684 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1685 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1686 extent information of a file. It can be used to map logical blocks
1687 in a file to physical blocks in the block device.
1691 Description: LDLM soft lockups - improvement
1692 Details : It is be possible to send the lock handle along with each read
1693 or write request because the client is already doing a lock match
1694 itself so there isn't any reason the OST should have to re-do that
1698 Frequency : only on Cray X2
1700 Description: X2 build failures
1701 Details : fix build failures on Cray X2.
1705 Description: xid & resent requests
1706 Details : Initialize RPC XID from clock at startup (randomly if clock is
1709 Severity : enhancement
1711 Description: Add lustre_start utility to start or stop multiple Lustre servers
1716 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1717 Details : In case of memory pressure, list_del() can be called twice on
1718 req->rq_history_list, causing a kernel oops.
1722 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1723 Details : fix stack overflow in the distributed lock manager by defering
1724 export eviction after a failed AST to the elt thread instead of
1725 handling it in the dlm interpret routine.
1729 Description: Convert some comments to new format.
1730 Details : Update documenting comments to match doxygen conventions.
1734 Description: Grammar fixes.
1735 Details : A couple of trivial spelling fixes.
1739 Description: OSD_COUNTERS-mandatory
1740 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1741 to simplify the code.
1745 Description: simplify cmm_device freeing logic.
1746 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1750 Description: Add lockdep support to dt_object_operations locking interface.
1751 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
1752 parameter indicating lock ordering. Update mdd code to use new
1757 Description: Introduce failloc constants for lockless IO tests.
1758 Details : Add two new failloc constants to test lockless IO. Only one of
1759 them in implemented---another is checked in yet to be landed
1764 Description: Add lockdep support for inode mutex.
1765 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1766 in the situations where Lustre has to lock more than one inode
1771 Description: Add optional invariants checking support.
1772 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1773 configure switch. This macro is to be used for consistency and
1774 sanity checks that are too expensive to be left in `production'
1779 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1780 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1785 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1786 Details : Use spin_lock_nested() in (the only) situation where more than
1787 one ldlm_lock is locked simultaneously. Also, fix possible
1788 dead-lock in ldlm_lock_change_resource() by enforcing particular
1793 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1794 Details : Instead of specifying each ldlm_lock call-back through separate
1795 parameter, wrap them into struct ldlm_callback_suite.
1799 Description: Kill join_lru obd method and its callers.
1800 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
1801 in memory---a requirement implemented through ->o_join_lru() obd
1802 method in HEAD. Remove this method and its users.
1806 Description: Add asynchronous ldlm ENQUEUE completion handler.
1807 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
1808 case that stock ldlm_completion_ast() cannot handle as it waits
1809 until lock is granted. Introduce new ldlm_completion_ast_async()
1810 for this. Also comment ldlm_completion_ast().
1814 Description: ldlm_error <-> errno conversion.
1815 Details : Add functions to map (rather arbitrary) between LDLM error codes
1816 and standard errno values. CLIO needs this to prevent LDLM specific
1817 constants from escaping ldlm and osc.
1821 Description: Kill unused ldlm_handle2lock_ns() function.
1822 Details : Kill unused ldlm_handle2lock_ns() function.
1826 Description: Add lu_ref support to ldlm_lock
1827 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
1828 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
1829 and ldlm_resource. LDLM interface has to be changed, because code that
1830 releases a reference on a lock, has to "know" what reference this is.
1831 In the most frequent case
1833 lock = ldlm_handle2lock(handle);
1835 LDLM_LOCK_PUT(lock);
1837 no changes are required. When any other reference (received _not_ from
1838 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
1839 instead of LDLM_LOCK_PUT().
1841 Arguably, changes are pervasive, and interface requires some discipline
1842 for proper use. On the other hand, it was very instrumental in finding
1843 a few leaked lock references.
1847 Description: Add ldlm_lock_addref_try().
1848 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
1849 attempts to addref a lock that might be being canceled
1854 Description: Add ldlm_weigh_callback().
1855 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
1856 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
1857 hard-coded `number of pages' logic.
1861 Description: Add lockdep annotations to llog code.
1862 Details : Use appropriately tagged _nested() locking calls in the places
1863 where llog takes more than one ->lgh_lock lock.
1867 Description: Add loi_kms_set().
1868 Details : Wrap kms updates into a helper function.
1872 Description: Constify instances of struct lsm_operations.
1873 Details : Constify instances of struct lsm_operations.
1877 Description: lu_conf support.
1878 Details : On a server, a file system object is uniquely identified
1879 by a fid, which is sufficient to locate and load all object
1880 state (inode). On a client, on the other hand, more data are
1881 necessary instantiate an object. Change lu_object_find() and
1882 friends to take additional `lu_conf' argument describing object.
1883 Typically this includes layout information.
1887 Description: lu_context fixes.
1888 Details : Introduce new lu_context functions that are needed on the client
1889 side, where some system threads (ptlrpcd) are shared by multiple
1890 modules, and so cannot be stopped during module shutdown.
1894 Description: Add start and stop methods to lu_device_type_operations.
1895 Details : Introduce two new methods in lu_device_type_operations, that are
1896 invoked when first instance of a given type is created and last one
1897 is destroyed respectively. This is need by CLIO.
1901 Description: Add lu_ref support to struct lu_device.
1902 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
1903 track leaked references.
1907 Description: Introduce lu_kmem_descr.
1908 Details : lu_kmem_descr and its companion interface allow to create
1909 and destroy a number of kmem caches at once.
1913 Description: Fix lu_object finalization race.
1914 Details : Fix a race between lu_object_find() finding an object and its
1915 concurrent finalization. This race is (most likely) not possible
1916 on the server, but might happen on the client.
1920 Description: Introduce lu_ref interface.
1921 Details : lu_ref is a debugging module allowing to track references to
1922 a given object. It is quite cpu expensive, and has to be
1923 explicitly enabled with --enable-lu_ref. See usage description
1928 Description: Factor lu_site procfs stats into a separate function.
1929 Details : Separate lu_site stats printing code into a separate function
1930 that can be reused on a client.
1934 Description: Constify instances of struct {lu,dt,md}_device_operations.
1935 Details : Constify instances of struct {lu,dt,md}_device_operations.
1939 Description: Introduce struct md_site and move meta-data specific parts of
1940 struct lu_site here.
1941 Details : Move md-specific fields out of struct lu_site into special struct
1942 md_site, so that lu_site can be used on a client.
1946 Description: Kill mdd_lov_destroy().
1947 Details : Remove unused mdd code.
1951 Description: Add st_block checking to multistat.c.
1952 Details : Add st_block checking to multistat.c.
1956 Description: Add lu_ref support to struct obd_device.
1957 Details : Add lu_ref tracking to obd_device.
1961 Description: Kill obd_set_fail_loc().
1962 Details : Remove unused code.
1966 Description: Add special type for ptlrpc_request interpret functions.
1967 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
1968 there. Actual usage will be in CLIO.
1972 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
1973 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
1974 doesn't work with lockdep.
1978 Description: Add rwv.c test program.
1979 Details : New testing program exercising readv(2) and writev(2) (Qian).
1983 Description: Add sendfile.c test program.
1984 Details : New testing program exercising sendfile(2) (Jay).
1988 Description: Ratelimit a message that can be very frequent.
1989 Details : Ratelimit a memory allocation failure message that can
1994 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
1995 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
1999 Description: Liblustre build fixes.
2000 Details : Liblustre build fixes.
2004 Description: libcfs: add cfs_{need,cond}_resched() interface.
2005 Details : libcfs: add cfs_{need,cond}_resched() definition and
2006 implementations for Linux, NT, and liblustre.
2008 Severity : enhancement
2010 Description: More exported tunables for mballoc
2011 Details : Add support for tunable preallocation window and new tunables for
2012 large/small requests
2016 Description: Detect corruption of block bitmap and checking for preallocations
2017 Details : Checks validity of on-disk block bitmap. Also it does better
2018 checking of number of applied preallocations. When corruption is
2019 found, it turns filesystem readonly to prevent further corruptions.
2023 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2024 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2025 window to 0. If it does hit cache miss, and read-pattern is still
2026 stride-io mode, does not reset the stride window, but also does not
2027 increase the stride window length in this case.
2031 Frequency : only for big-endian servers
2032 Description: Check if system is big-endian while mounting fs with extents feature
2033 Details : Mounting a filesystem with extents feature will fail on big-endian
2034 systems since ext3-based ldiskfs is not supported on big-endian
2035 systems. This can be over-riden with "bigendian_extents" mount option.
2037 Severity : enhancement
2039 Description: The root squash functionality
2040 Details : A security feature, which is to prevent users from being able
2041 to mount lustre on their desktop, run as root, and delete
2042 all of the files in the filesystem. The goal is accomplished by
2043 remapping user id (UID) and group id (GID) of the root user to
2044 a UID and GID specified by the system administartor via Lustre
2045 configuration management server (MGS). The functionality also
2046 allows to specify sets of clients for which the remapping does
2051 Description: Excessive recovery window
2052 Details : With AT enabled, the recovery window can be excessively long (6000+
2053 seconds). To address this problem, we no longer use
2054 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2055 timeout no longer depends on the service time, it is set to
2056 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2057 time via pb_service_time.
2061 Description: Watchdog triggered on MDS failover
2062 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2063 the OSTs know that the MDS "UUID" can be reused for the same export
2064 from a different NID, so we do not need to wait for the export to be
2068 Frequency : rare, only if using MMP with Linux RAID
2070 Description: MMP doesn't work with Linux RAID
2071 Details : While using HA for Lustre servers with Linux RAID, it is possible
2072 that MMP will not detect multiple mounts. To make this work we
2073 need to unplug the device queue in RAID when the MMP block is being
2074 written. Also while reading the MMP block, we should read it from
2075 disk and not the cached one.
2077 Severity : enhancement
2079 Description: open file using fid
2080 Details : A file can be opened using just its fid, like
2081 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2084 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2087 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2088 Details : add proper locking for creation of jbd_4k slab cache
2092 Description: MMP check in ext3_remount() fails without displaying any error
2093 Details : When multiple mount protection fails during remount, proper error
2096 Severity : enhancement
2098 Description: Allow stripe size to be up to 4G-64k
2099 Details : Fix math logic to allow large stripe sizes.
2103 Description: add check for >8TB ldiskfs filesystems
2104 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2105 Don't allow >8TB ldiskfs filesystems to be mounted without
2106 force_over_8tb mount option
2108 --------------------------------------------------------------------------------
2110 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2112 * Support for kernels:
2113 2.6.5-7.283 (SLES 9),
2114 2.6.9-55.EL (RHEL 4),
2115 2.6.16.46-0.14 (SLES 10),
2116 2.6.18.8 vanilla (kernel.org)
2117 * Client support for unpatched kernels:
2118 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2119 2.6.16 - 2.6.22 vanilla (kernel.org)
2120 * Due to recently discovered recovery problems, we do not recommend
2121 using patchless RHEL 4 clients with this or any earlier release.
2122 * Recommended e2fsprogs version: 1.39.cfs8
2123 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2124 * Starting with this release, the ldiskfs backing filesystem required
2125 by Lustre is now in its own package, lustre-ldiskfs. This package
2126 should be installed. It is versioned separately from Lustre and
2127 may be released separately in future.
2131 Description: block reactivating mgc import until all deactivates complete
2132 Details : Fix race when failing back MDT/MGS to itself (testing)
2135 Frequency : at statup only
2137 Description: mds_lov_synchronize race leads to various problems
2138 Details : simultaneous MDT->OST connections at startup can cause the
2139 sync to abort, leaving the OSC in a bad state.
2141 Severity : enhancement
2143 Description: add optional extra BUILD_VERSION info
2144 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2145 to override the lustre version.
2148 Frequency : 2.6.18 servers only
2150 Description: ll_kern_mount() doesn't release the module reference
2151 Details : The ldiskfs module reference count never drops down to 0
2152 because ll_kern_mount() doesn't release the module reference.
2157 Description: server LBUG when using old ost_num_threads parameter
2158 Details : Accept the old ost_num_threads parameter but warn that it
2159 is deprecated, and fix an off-by-one error that caused an LBUG.
2164 Description: Transient SCSI error results in persistent IO issue
2165 Details : iobuf->dr_error is not reinitialized to 0 between two
2169 Frequency : sometimes when underlying device returns I/O errors
2171 Description: OSTs not going read-only during write failures
2172 Details : OSTs are not remounted read-only when the journal commit threads
2173 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2174 instead of the ext3 wrappers.
2178 Description: poor connect scaling with increasing client count
2179 Details : Don't run filter_grant_sanity_check for more than 100 exports
2180 to improve scaling for large numbers of clients.
2183 Frequency : SLES10 only
2185 Description: sanity-quota.sh quotacheck failed: rc = -22
2186 Details : Quotas cannot be enabled on SLES10.
2189 Frequency : liblustre clients only
2191 Description: getdirentries does not give error when run on compute nodes
2192 Details : getdirentries does not fail when the size specified as an argument
2193 is too small to contain at least one entry
2195 Severity : enhancement
2197 Description: Add LNET router traceability for debug purposes
2198 Details : If a checksum failure occurs with a router as part of the
2199 IO path, the NID of the last router that forwarded the bulk data
2200 is printed so it can be identified.
2205 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2206 Details : Due to a race condition, liblustre clients were occasionally
2207 evicted incorrectly.
2209 Severity : enhancement
2211 Description: lfs setstripe use optional parameters instead of postional
2214 Severity : enhancement
2216 Description: Nanosecond timestamp support for ldiskfs
2217 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2218 resolution timestamps. There is not yet support for this at
2219 the Lustre filesystem level.
2222 Frequency : during server recovery
2224 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2225 Details : request with rq_no_resend flag not awake l_wait_event if they get
2229 Frequency : nfs export on patchless client
2231 Description: connectathon hang when test nfs export over patchless client
2232 Details : Disconnected dentry cannot be found with lookup, so we do not need
2233 to unhash it or make it invalid
2236 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2237 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2238 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2243 Description: open req refcounting wrong on reconnect
2244 Details : If reconnect happened between getting open reply from server and
2245 call to mdc_set_replay_data in ll_file_open, we will schedule
2246 replay for unreferenced request that we are about to free.
2247 Subsequent close will crash in variety of ways.
2248 Check that request is still eligible for replay in
2249 mdc_set_replay_data().
2254 Description: disable writes to filesystem when reading health_check file
2255 Details : the default for reading the health_check proc file has changed
2256 to NOT do a journal transaction and write to disk, because this
2257 can cause reads of the /proc file to hang and block HA state
2258 checking on a healthy but otherwise heavily loaded system. It
2259 is possible to return to the previous behaviour during configure
2260 with --enable-health-write.
2262 Severity : enhancement
2264 Description: 64-bit inode version
2265 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2266 to the inode. This will be required for version-based recovery.
2271 Description: MDS fails to start if a duplicate client export is detected
2272 Details : in some rare cases it was possible for a client to connect to
2273 an MDS multiple times. Upon recovery the MDS would detect this
2274 and fail during startup. Handle this more gracefully.
2276 Severity : enhancement
2278 Description: Add -o localflock option to simulate old noflock
2280 Details : This will achieve local-only flock/fcntl locks
2286 Description: log_commit_thread vs filter_destroy race leads to crash
2287 Details : Take import reference before releasing llog record semaphore
2292 Description: Wrong request locking in request set processing
2293 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2294 imp_delayed_list, in this place should be used imp_lock.
2297 Frequency : when reconnection
2299 Description: Grant Leak when osc reconnect to OST
2300 Details : When osc reconnect ost, OST(filter) should check whether it
2301 should grant more space to client by comparing fed_grant and
2302 cl_avail_grant, and return the granted space to client instead
2303 of "new granted" space, because client will call osc_init_grant
2304 to update the client grant space info.
2307 Frequency : when client reconnect to OST
2309 Description: Grant Leak when osc do resend and replay bulk write
2310 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2311 bulk write request, because the grant info while be sync between
2312 OSC and OST when reconnect, and we should ignore the grant info
2313 these of resend/replay write req.
2318 Description: Grant space more than avaiable left space sometimes.
2319 Details : When then OST is about to be full, if two bulk writing from
2320 different clients came to OST. Accord the avaliable space of the
2321 OST, the first req should be permitted, and the second one
2322 should be denied by ENOSPC. But if the seconde arrived before
2323 the first one is commited. The OST might wrongly permit second
2324 writing, which will cause grant space > avaiable space.
2327 Frequency : when client is evicted
2329 Description: Grant might be wrongly erased when osc is evicted by OST
2330 Details : when the import is evicted by server, it will fork another
2331 thread ptlrpc_invalidate_import_thread to invalidate the
2332 import, where the grant will be set to 0. While the original
2333 thread will update the grant it got when connecting. So if
2334 the former happened latter, the grant will be wrongly errased
2335 because of this race.
2340 Description: Checking Stale with correct fid
2341 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2342 to check whether it is stale, because sometimes, we want the
2343 enqueue happened anyway, and op_data.fid2 will not be initialized.
2345 Severity : enhancement
2347 Description: update patchless client
2348 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2351 Frequency : only with 2.4 kernel
2353 Description: random memory corruption
2354 Details : size of struct ll_inode_info is to big for union inode.u and this
2355 can be cause of random memory corruption.
2360 Description: Memory leak in recovery
2361 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2362 It should also check obd_fail before fsfilt_start, otherwise if
2363 fsfilt_start return -EROFS,(failover mds during mds recovery).
2364 then the req will return with repmsg->transno = 0 and rc = EROFS.
2365 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2366 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2367 be freed no matter whether fsfilt_commit success or not.
2370 Frequency : only with huge count clients
2372 Description: Prevents from taking the superblock lock in llap_from_page for
2374 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2375 for llap_from_page call in ll_removepage prevents from taking the
2376 superblock lock for a soon died page.
2381 Description: Not check open intent error before release open handle
2382 Details : in some rare cases, the open intent error is not checked before
2383 release open handle, which may cause
2384 ASSERTION(open_req->rq_transno != 0), because it tries to release
2385 the failed open handle.
2390 Description: Set cat log bitmap only after create log success.
2391 Details : in some rare cases, the cat log bitmap is set too early. and it
2392 should be set only after create log success.
2396 Description: Accessing a block bevice can re-enable I/O when Lustre is
2397 tearing down a device.
2398 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2403 Description: service threads may hog cpus when there are a lot of requests
2404 Details : Insert cond_resched to give other threads a chance to use some CPU
2409 Description: the cat log was not initialized in recovery
2410 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2411 unlink log on mds will not be initialized until mds post
2412 recovery. And also in mds post recovery, the unlink log will
2413 initialization will be done asynchronausly, so there will be race
2414 between add unlink log and unlink log initialization.
2418 Description: brw_stats were being printed incorrectly
2419 Details : brw_stats were being printed as log2 but all of them were not
2420 recorded as log2. Also remove some code duplication arising from
2421 filter_tally_{read,write}.
2425 Frequency : rare, only in recovery.
2426 Description: ASSERTION(req->rq_type != LI_POISON) failed
2427 Details : imp_lock should be held while iterating over imp_sending_list for
2428 prevent destroy request after get timeout in ptlrpc_queue_wait.
2432 Description: replay-single.sh test 52 fails
2433 Details : A lock's skiplist need to be cleanup when it being unlinked
2434 from its resource list.
2438 Description: Short directio read returns full requested size rather than
2440 Details : Direct I/O operations should return actual amount of bytes
2441 transferred rather than requested size.
2443 Severity : enhancement
2445 Description: metadata RPC reduction (e.g. for rm performance)
2446 Details : decrease the amount of synchronous RPC between clients and servers
2447 by canceling conflicing lock before the operation on the client
2448 and packing thier handles into the main operation RPC to server.
2450 Severity : enhancement
2452 Description: Async OSC create to avoid the blocking unnecessarily.
2453 Details : If a OST has no remain object, system will block on the creating
2454 when need to create a new object on this OST. Now, ways use
2455 pre-created objects when available, instead of blocking on an
2456 empty osc while others are not empty. If we must block, we block
2457 for the shortest possible period of time.
2461 Description: improve handling recoverable errors
2462 Details : if request processig with error which can be recoverable on server
2463 request should be resend, otherwise page released from cache and
2466 Severity : enhancement
2468 Description: refine locking for avoid write wrong info into lov_objid file
2469 Details : fix possible races with add new target and write/update data in
2472 --------------------------------------------------------------------------------
2474 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2479 Frequency : on some architectures
2481 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2482 Details : Uninitialized flags sometimes cause configuration commands to
2486 Frequency : patchless clients only
2488 Description: missing __iget() symbol export
2489 Details : The __iget() symbol export is missing. To avoid the need for
2490 this on patchless clients the deathrow inode reaper is turned
2491 off, and we depend on the VM to clean up old inodes. This
2492 dependency was during via the fix for bug 12181.
2494 --------------------------------------------------------------------------------
2496 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2498 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2499 INTEROPERATE with older versions automatically. In many cases a
2500 special upgrade step is needed. Please read the
2501 user documentation before upgrading any part of a live system.
2502 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2503 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2504 * WARNING: Lustre configuration and startup changes are required with
2505 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2509 Severity : enhancement
2511 Description: MountConf
2512 Details : Lustre configuration is now managed via mkfs and mount
2513 commands instead of lmc and lconf. New obd types (MGS, MGC)
2514 are added for dynamic configuration management. See
2515 https://mail.clusterfs.com/wikis/lustre/MountConf for
2518 Severity : enhancement
2520 Description: dynamic OST addition
2521 Details : OSTs can now be added to a live filesystem
2523 Severity : enhancement
2525 Description: startup order invariance
2526 Details : MDTs and OSTs can be started in any order. Clients only
2527 require the MDT to complete startup.
2529 Severity : enhancement
2531 Description: parallel, asynchronous orphan cleanup
2532 Details : orphan cleanup is now performed in separate threads for each
2533 OST, allowing parallel non-blocking operation.
2535 Severity : enhancement
2537 Description: optimized stripe assignment
2538 Details : stripe assignments are now made based on ost space available,
2539 ost previous usage, and OSS previous usage, in order to try
2540 to optimize storage space and networking resources.
2542 Severity : enhancement
2544 Description: Permanently set tunables
2545 Details : All writable /proc/fs/lustre tunables can now be permanently
2546 set on a per-server basis, at mkfs time or on a live system.
2548 Severity : enhancement
2550 Description: Lustre message v2
2551 Details : Add lustre message format v2.
2553 Severity : enhancement
2555 Description: client OST exclusion list
2556 Details : Clients can be started with a list of OSTs that should be
2557 declared "inactive" for known non-responsive OSTs.
2561 Description: SPEC SFS validation failure on NFS v2 over lustre.
2562 Details : Changes the blocksize for regular files to be 2x RPC size,
2563 and not depend on stripe size.
2565 Severity : enhancement
2567 Description: Multiple MD RPCs in flight.
2568 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2569 To avoid overly-overloading MDS, introduce a limit on number of
2570 MDS RPCs in flight for a single client and add /proc controls
2571 to adjust this limit.
2573 Severity : enhancement
2575 Description: client read/write statistics
2576 Details : Add client read/write call usage stats for performance
2577 analysis of user processes.
2578 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2579 file access. extents_stats shows chunk size distribution.
2580 extents_stats_per_process show chunk size distribution per
2583 Severity : enhancement
2585 Description: mds statistics
2586 Details : Add detailed mds operations statistics in
2587 /proc/fs/lustre/mds/*/stats.
2591 Description: Failure of copying files with lustre special EAs.
2592 Details : Client side always return success for setxattr call for lustre
2593 special xattr (currently only "trusted.lov").
2598 Description: Refcount LNET uuids
2599 Details : The global LNET uuid list grew linearly with every startup;
2600 refcount repeated list entries instead of always adding to
2603 Severity : enhancement
2605 Description: Dynamic service threads
2606 Details : Within a small range, start extra service threads
2607 automatically when the request queue builds up.
2610 Frequency : mixed-endian client/server environments
2612 Description: mixed-endian crashes
2613 Details : The new msg_v2 system had some failures in mixed-endian
2616 Severity : enhancement
2618 Description: Easy OST removal
2619 Details : OSTs can be permanently deactivated with e.g. 'lctl
2620 conf_param lustre-OST0001.osc.active=0'
2622 Severity : enhancement
2624 Description: MGS proc entries
2625 Details : Added basic proc entries for the MGS showing what filesystems
2628 Severity : enhancement
2630 Description: provide MGS failover
2631 Details : Added config lock reacquisition after MGS server failover.
2633 Severity : enhancement
2635 Description: add Linux 2.4 support
2636 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2640 Description: a large application tries to do I/O to the same resource and dies
2641 in the middle of it.
2642 Details : Check the req->rq_arrival time after the call to
2643 ost_brw_lock_get(), but before we do anything about
2644 processing it & sending the BULK transfer request. This
2645 should help move old stale pending locks off the queue as
2646 quickly as obd_timeout.
2649 Frequency : when an incorrect nid is specified during startup
2651 Description: ptlrpc connect to non-existant node causes kernel crash
2652 Details : LNET can't be re-entered from an event callback, which
2653 happened when we expire a message after the export has been
2654 cleaned up. Instead, hand the zombie cleanup off to another
2657 Severity : enhancement
2659 Description: plain/inodebits lock performance improvement
2660 Details : Grouping plain/inodebits in granted list by their request modes
2661 and bits policy, thus improving the performance of search through
2665 Frequency : only if OST filesystem is corrupted
2667 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2668 Details : for a short time RPCs with bulk IO are in the replay list,
2669 but replay of bulk IOs is unimplemented. If the OST filesystem
2670 is corrupted due to disk cache incoherency and then replay is
2671 started it is possible to trip an assertion. Avoid putting
2672 committed RPCs into the replay list at all to avoid this issue.
2675 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2677 Description: System hang on startup
2678 Details : This bug allowed the liblustre (e.g. catamount) client to
2679 return to the app before handling all startup RPCs. This
2680 could leave the node unresponsive to lustre network traffic
2681 and manifested as a server ptllnd timeout.
2683 Severity : enhancement
2685 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2686 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2687 boolean to control whether to print peer debug info when a
2688 client's RPC times out.
2691 Frequency : only for kernels with patches from Lustre below 1.4.3
2693 Description: Remove old rdonly API
2694 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2697 Frequency : only for devices with external journals
2699 Description: Set external device read-only also
2700 Details : During a commanded failover stop, we set the disk device
2701 read-only while the server shuts down. We now also set any
2702 external journal device read-only at the same time.
2705 Frequency : when upgrading from 1.4 while trying to change parameters
2707 Description: The wrong (new) MDC name was used when setting parameters for
2708 upgraded MDT's. Also allows changing of OSC (and MDC)
2709 parameters if --writeconf is specified at tunefs upgrade time.
2712 Frequency : when setting specific ost indicies
2714 Description: QOS code breaks on skipped indicies
2715 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2716 created with --index need not be sequential.
2718 Severity : enhancement
2720 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2721 Details : The uninit_groups feature works in conjunction with the kernel
2722 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2723 up the pass1 processing of e2fsck. This is a read-only feature
2724 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2725 mount filesystems that have had this feature enabled.
2727 Severity : enhancement
2729 Description: Improve multi-block allocation algorithm to avoid fragmentation
2730 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2731 allocation locality and avoid filesystem fragmentation.
2733 ------------------------------------------------------------------------------
2735 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2737 * Support for kernels:
2738 2.4.21-47.0.1.EL (RHEL 3)
2739 2.6.5-7.283 (SLES 9)
2740 2.6.9-42.0.10.EL (RHEL 4)
2741 2.6.12.6 vanilla (kernel.org)
2742 2.6.16.27-0.9 (SLES 10)
2743 * Recommended e2fsprogs version: 1.39.cfs5
2745 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2749 Frequency : occasional, depends on client load and configuration
2750 Bugzilla : 12181, 12203
2751 Description: data loss for recently-modified files
2753 Details : In some cases it is possible that recently written or created
2754 files may not be written to disk in a timely manner (this should
2755 normally be within 30s unless client IO load is very high).
2756 The problem appears as zero-length files or files that are a
2757 multiple of 1MB in size after a client crash or client eviction
2758 that are missing data at the end of the file.
2760 This problem is more likely to be hit on clients where files are
2761 repeatedly created and unlinked in the same directory, clients
2762 have a large amount of RAM, have many CPUs, the filesystem has
2763 many OSTs, the clients are rebooted frequently, and/or the files
2764 are not accessed by other nodes after being written.
2766 The presence of the problem can be detected by looking at
2767 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2768 smaller than the second (nr_unused) then dirty files will not
2769 be flushed automatically to disk. "sync; sleep 10" should be
2770 run several times on the node before unmounting it to update
2771 Lustre (this is also safe to run on nodes without this problem).
2773 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2774 that can cause this same problem, so customers using that kernel
2775 also need to update the kernel in addition to Lustre. In order
2776 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2778 It is normal that files written just before a client crash (less
2779 than 30s) may not yet have been flushed to disk, even for local
2783 Frequency : frequent on thin XT3 nodes
2785 Description: UUID collision on thin XT3 Linux nodes
2786 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2787 randomly, since we relied on an insufficiently-seeded PRNG.
2792 Description: OSS hangs after "All ost request buffers busy"
2793 Details : A deadlock between quota and journal operations caused OSS
2794 hangs after printing "All ost request buffers busy."
2797 Frequency : always on liblustre builds
2799 Description: Cleanup compiler warnings on liblustre
2802 Frequency : always on liblustre builds on XT3
2804 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2808 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2809 > 32000 subdirectories
2810 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2811 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2812 subdirectory count crosses 32000. This will aid e2fsck to
2813 correctly handle more than 32000 subdirectories.
2816 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2818 Description: System hang on startup
2819 Details : This bug allowed the liblustre (e.g. catamount) client to
2820 return to the app before handling all startup RPCs. This
2821 could leave the node unresponsive to lustre network traffic
2822 and manifested as a server ptllnd timeout.
2824 Severity : enhancement
2826 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2827 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2828 boolean to control whether to print peer debug info when a
2829 client's RPC times out.
2834 Description: make O_SYNC working on 2.6 kernels
2835 Details : 2.6 kernels use different method for mark pages for write,
2836 so need add a code to lustre for O_SYNC work.
2841 Description: Failure to close file and release space on NFS
2842 Details : Put inode details into lock acquired in ll_intent_file_open.
2843 Use mdc_intent_lock in ll_intent_open to properly
2844 detect all kind of errors unhandled by mdc_enqueue.
2849 Description: proc file read during shutdown sometimes raced obd removal,
2851 Details : Add lock to prevent obd access after proc file removal.
2854 Frequency : Only for files larger than 4GB on 32-bit clients.
2856 Description: improperly doing page alignment of locks
2857 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2858 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2863 Description: RPCs being resent when they shouldn't be
2864 Details : Some RPCs that should not be resent are being resent. This
2865 can cause inconsistencies in the RPC state machine. Do not
2866 resend such requests.
2869 Frequency : rare, only with NFS export
2871 Description: Crash on NFS re-export node
2872 Details : under very unusual load conditions an assertion is hit in
2873 ll_intent_file_open()
2876 Frequency : only if OST filesystem is corrupted
2878 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2879 Details : for a short time RPCs with bulk IO are in the replay list,
2880 but replay of bulk IOs is unimplemented. If the OST filesystem
2881 is corrupted due to disk cache incoherency and then replay is
2882 started it is possible to trip an assertion. Avoid putting
2883 committed RPCs into the replay list at all to avoid this issue.
2888 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2889 Details : Large single O_DIRECT read and write calls can fail to allocate
2890 a sufficiently large buffer to process the request. In case of
2891 allocation failure the allocation is retried with a smaller
2892 buffer and broken into smaller requests.
2894 Severity : enhancement
2896 Description: Add -o localflock option to simulate old noflock behaviour.
2897 Details : This will achieve local-only flock/fcntl locks coherentness.
2902 Description: versioning check is incomplete
2903 Details : Checking the version difference of client vs. server, report
2904 error if the gap is too big.
2909 Description: add support PG_writeback bit
2910 Details : add support for PG_writeback bit for Lustre, for more carefull
2911 work with page cache in 2.6 kernel. This also fix some deadlocks
2912 and remove hack for work O_SYNC with 2.6 kernel.
2914 Severity : enhancement
2916 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2917 Details : The uninit_groups feature works in conjunction with the kernel
2918 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2919 up the pass1 processing of e2fsck. This is a read-only feature
2920 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2921 mount filesystems that have had this feature enabled.
2923 Severity : enhancement
2925 Description: Improve multi-block allocation algorithm to avoid fragmentation
2926 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2927 allocation locality and avoid filesystem fragmentation.
2929 ------------------------------------------------------------------------------
2931 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2933 * Support for kernels:
2934 2.6.9-42.0.3.EL (RHEL 4)
2935 2.6.5-7.276 (SLES 9)
2936 2.4.21-47.0.1.EL (RHEL 3)
2937 2.6.12.6 vanilla (kernel.org)
2938 2.6.16.21-0.8 (SLES10)
2939 * Recommended e2fsprogs version: 1.39.cfs2-0
2941 * The backwards-compatible /proc/sys/portals symlink has been removed
2942 in this release. Before upgrading, please ensure that you change
2943 any configuration scripts or /etc/sysctl.conf files that access
2944 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2945 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2946 in advance of the upgrade on any system running Lustre 1.4.6 or
2947 newer, since /proc/sys/lnet was added in that version.
2948 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2952 Frequency : only when quota is used
2954 Description: avoid scanning export list for quota master
2955 Details : Change the algorithms to avoid scanning export list in order
2956 to improve the efficiency.
2959 Frequency : MDS failover only, very rarely
2961 Description: "went back in time" messages on mds failover
2962 Details : The greatest transno may be lost when the current operation
2963 finishes with an error (transno==0) and the client's last_rcvd
2964 record is over-written. Save the greatest transno in the
2965 mds_last_transno for this case.
2968 Frequency : always for specific kernels and striping counts
2970 Description: client may get "Matching packet too big" without ACL support
2971 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2972 when trying to access files in certain configurations. The
2973 clients should in fact be denied when mounting because they do
2974 not understand ACLs.
2977 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2979 Description: many clients connecting with IO in progress causes connect timeouts
2980 Details : Avoid synchronous journal commits to avoid delays caused by many
2981 clients connecting/disconnecting when bulk IO is in progress.
2982 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2983 OST_IO_PORTAL to avoid delays behind potentially many pending
2987 Frequency : occasionally with multiple writers to a single file
2989 Description: shared writes to file may result in wrong size reported by stat()
2990 Details : Allow growing of kms when extent lock is cancelled
2993 Frequency : always with random mmap IO to multi-striped file
2995 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2996 Details : Only if the hole is at the end of OST object so that kms is too
2997 small. Fix is to increase kms accordingly in ll_nopage.
3000 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3002 Description: writes to a missing object would leak memory on the OST
3003 Details : If there is an inconsistency between the MDS and OST filesystems,
3004 such that the MDS references an object that doesn't exist, writes
3005 to that object will leak memory due to incorrect cleanup in the
3006 error handling path, eventually running out of memory on the OST.
3011 Description: Creating too long symlink causes lustre errors
3012 Details : Check symlink and name lengths before sending requests to MDS.
3015 Frequency : only if flock is enabled (not on by default)
3017 Description: posix locks not released on fd closure on 2.6.9+
3018 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3019 kernels, this caused such locks not to be released on fd close and
3020 then assertions on fs unmount about still used locks.
3023 Frequency : MDS failover only, very rarely
3025 Description: clients may get ASSERTION(granted_lock != NULL)
3026 Details : When request was taking a long time, and a client was resending
3027 a getattr by name lock request. The were multiple lock requests
3028 with the same client lock handle and
3029 mds_getattr_name->fixup_handle_for_resent_request found one of the
3030 lock handles but later failed with ASSERTION(granted_lock != NULL).
3035 Description: handle->h_buffer_credits > 0, assertion failure
3036 Details : h_buffer_credits is zero after truncate, causing assertion
3037 failure. This patch extends the transaction or creates a new
3041 Frequency : NFS re-export or patchless client
3042 Bugzilla : 11179, 10796
3043 Description: Crash on NFS re-export node (__d_move)
3044 Details : We do not want to hash the dentry if we don't have a lock.
3045 But if this dentry is later used in d_move, we'd hit uninitialised
3046 list head d_hash, so we just do this to init d_hash field but
3047 leave dentry unhashed.
3050 Frequency : NFS re-export or patchless client
3052 Description: NFS exports has problem with symbolic link
3053 Details : lustre client didn't properly install dentry when re-exported
3054 to NFS or running patchless client.
3057 Frequency : NFS re-export or patchless client
3059 Description: Various nfs/patchless fixes.
3060 Details : fixes reuse disconected alias for lookup process - this fixes
3061 warning "find_exported_dentry: npd != pd",
3062 fix permission error with open files at nfs.
3063 fix apply umask when do revalidate.
3066 Frequency : occasional
3068 Description: Crash on NFS re-export node
3069 Details : calling clear_page() on the wrong pointer triggered oops in
3070 generic_mapping_read().
3073 Frequency : rarely, using O_DIRECT IO
3075 Description: unaligned directio crashes client with LASSERT
3076 Details : check for unaligned buffers before trying any requests.
3079 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3081 Description: stale data returned from RAID cache
3082 Details : If only a small amount of IO is done to the RAID device before
3083 reading it again it is possible to get stale data from the RAID
3084 cache instead of reading it from disk.
3087 Frequency : always for sles10 kernel
3089 Description: sles10 support
3090 Details : ll_follow_link: compile fixes and using of nd_set_link
3091 under newer kernels.
3094 Frequency : depends on arch, kernel and compiler version, always on sles10
3097 Description: recursive or deep enough symlinks cause stack overflow
3098 Details : getting rid of large stack-allocated variable in
3102 Frequency : depends on hardware
3104 Description: lustre write performance loss in the SLES10 kernel
3105 Details : the performance loss is caused by using of write barriers in the
3106 ext3 code. The SLES10 kernel turns barrier support on by
3107 default. The fix is to undo that change for ldiskfs.
3109 ------------------------------------------------------------------------------
3111 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3113 * Support for kernels:
3114 2.6.9-42.0.3EL (RHEL 4)
3115 2.6.5-7.276 (SLES 9)
3116 2.4.21-47.0.1.EL (RHEL 3)
3117 2.6.12.6 vanilla (kernel.org)
3121 Frequency : quota enabled and large files being deleted
3123 Description: releasing more than 4GB of quota at once hangs OST
3124 Details : If a user deletes more than 4GB of files on a single OST it
3125 will cause the OST to spin in an infinite loop. Release
3126 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3131 Description: statfs data retrieved from /proc may be stale or zero
3132 Details : When reading per-device statfs data from /proc, in the
3133 {kbytes,files}_{total,free,avail} files, it may appear
3134 as zero or be out of date.
3137 Frequency : systems with MD RAID1 external journal devices
3139 Description: lconf's call to blkid is confused by RAID1 journal devices
3140 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3141 of returning all block devices that match the journal UUID.
3144 Frequency : always, for aggregate stripe size over 4GB
3146 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3147 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3148 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3151 Frequency : always if "lfs find" used on a local file/directory
3153 Description: "lfs find" segfaults if used on a local file/directory
3154 Details : The case where a directory component was not specified wasn't
3155 handled correctly. Handle this properly.
3158 Frequency : always on ppc64
3160 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3161 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3162 perform endianness conversion.
3165 Frequency : rarely (truncate to non-zero file size after write under load)
3166 Bugzilla : 10730, 10687
3167 Description: Files padded with zeros to next 4K multiple
3168 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3169 it is possible that files that are truncated to a non-zero size
3170 immediately after being written are filled with zero bytes beyond
3171 the truncated size. No file data is lost.
3173 Severity : enhancement
3175 Description: Allow recovery/failover for liblustre clients.
3176 Details : liblustre clients were unaware of failover configurations until
3179 Severity : enhancement
3181 Description: user file locks should fail when not mounting with flock option
3182 Details : Set up an error-returning stub in ll_file_operations.lock field
3183 to prevent incorrect behaviour when client is mounted without
3184 flock option. Also, set up properly f_op->flock field for
3188 Frequency : always on ia64
3190 Description: "lfs df" loops on printing out MDS statfs information
3191 Details : The obd_ioctl_data was not initialized and in some systems
3192 this caused a failure during the ioctl that did not return
3193 an error. Initialize the struct and return an error on failure.
3196 Frequency : SLES 9 only
3198 Description: Error of copying files with lustre special EAs as root
3199 Details : Client side always return success for setxattr call for lustre
3200 special xattr (currently only "trusted.lov").
3203 Frequency : rarely on clusters with both ia64+i386 clients
3205 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3206 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3207 clients are doing concurrent writes to the same file it is
3208 possible that the ia64 clients may LASSERT because the OST
3209 extent locks are not PAGE_SIZE aligned. Ensure that grown
3210 locks are always aligned on the request boundary.
3213 Frequency : specific use, occasional
3215 Description: Overwriting in use executable truncates on-disk binary image
3216 Details : If one node attempts to overwrite an executable in use by
3217 another node, we now correctly return ETXTBSY instead of
3218 truncating the file.
3220 Severity : enhancement
3222 Description: Async OSC create to avoid the blocking unnecessarily.
3223 Details : If a OST has no remain object, system will block on the creating
3224 when need to create a new object on this OST. Now, ways use
3225 pre-created objects when available, instead of blocking on an
3226 empty osc while others are not empty. If we must block, we block
3227 for the shortest possible period of time.
3232 Description: chmod on Lustre root is propagated to other clients
3233 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3234 invalid by the follow_mount time.
3237 Frequency : liblustre clients only
3239 Description: Race in 'instant cancel' lock handling could lead to such locks
3240 never to be granted in case of SMP MDS
3241 Details : Do not destroy not yet granted but cbpending locks in
3245 Frequency : replay/resend of open
3247 Description: non null lock assetion failure in mds_intent_policy
3248 Details : Trying to replay/resend lockless open requests resulted in
3249 mds_open() returning 0 with no lock. Now it sets a flag if
3250 a lock is going to be returned.
3252 Severity : enhancement
3254 Description: Checksum enhancements
3255 Details : New checksum enhancements allow for resending RPCs that failed
3258 Severity : enhancement
3260 Description: Tunables on number of dirty pages in cacche
3261 Details : Allow to set limit on number of dirty pages cached.
3266 Description: client crash on unmount - lock still has references
3267 Details : In some error handling cases it was possible to leak a lock
3268 reference on a client while accessing a file. This was not
3269 harmful to the client during operation, but would cause the
3270 client to crash when the filesystem is unmounted.
3273 Frequency : specific case, rare
3275 Description: ETXTBSY on mds though file not in use by client
3276 Details : ETXTBSY is no longer incorrectly returned when attempting to
3277 chmod or chown a directory that the user previously tried to
3278 execute or a currently-executing binary.
3281 Frequency : extremely rare except on liblustre-based clients
3283 Description: Lustre space not freed when files are deleted
3284 Details : Clean up open-unlinked files after client eviction. Previously
3285 the unlink was skipped and the files remained as orphans.
3290 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3291 Details : In some cases it was possible to send a blocking callback to
3292 a client doing a glimpse, even though that client didn't get
3293 a lock granted. When the glimpse lock is cancelled on the OST
3294 the freed lock is left on the waiting list and corrupted the list.
3297 Frequency : all core dumps
3299 Description: Broke core dumps to lustre
3300 Details : Negative dentry may be unhashed if parent does not have UPDATE
3301 lock, but some callers, e.g. do_coredump, expect dentry to be
3302 hashed after successful create, hash it in ll_create_it.
3304 ------------------------------------------------------------------------------
3306 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3308 * Support for kernels:
3309 2.6.9-42.0.2.EL (RHEL 4)
3310 2.6.5-7.276 (SLES 9)
3311 2.4.21-40.EL (RHEL 3)
3312 2.6.12.6 vanilla (kernel.org)
3316 Frequency : always on RHEL 3
3318 Description: Number of open files grows over time
3319 Details : The number of open files grows over time, whether or not
3320 Lustre is started. This was due to a filp leak introduced
3321 by one of our kernel patches.
3323 ------------------------------------------------------------------------------
3325 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3327 * Support for kernels:
3328 2.6.9-42.EL (RHEL 4)
3329 2.6.5-7.276 (SLES 9)
3330 2.4.21-40.EL (RHEL 3)
3331 2.6.12.6 vanilla (kernel.org)
3336 Bugzilla : 5719, 9635, 9792, 9684
3337 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3338 Details : If a server is under heavy load and cannot reply to new
3339 connection requests before the client resends the (re)connect,
3340 the connection handling code can behave badly if two service
3341 threads are concurrently handing separate (re)connections from
3342 the same client. Add better locking to the connection handling
3343 code, and ensure that only a single connection will be processed
3344 for a given client UUID, even if the lock is dropped.
3346 Severity : enhancement
3348 Description: add TCP zero-copy support to kernel
3349 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3350 sends if the hardware supports scatter-gather and checksumming.
3351 This allows socklnd to do client-write and server-read more
3352 efficiently and reduce CPU utilization from skbuf copying.
3355 Frequency : only if NFS exporting from client
3357 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3358 Details : If setting attributes on a file created under NFS that had
3359 never been opened it would be possible to oops the client
3360 if the file had no objects.
3363 Frequency : always for liblustre
3365 Description: liblustre client does MDS+OSTs setattr RPC for each write
3366 Details : When doing a write from a liblustre client, the client
3367 incorrectly issued an RPC to the MDS and each OST the file was
3368 striped over in order to update the timestamps. When writing
3369 with small chunks and many clients this could overwhelm the MDS
3370 with RPCs. In all cases it would slow down the write because
3371 these RPCs are unnecessary.
3373 Severity : enhancement
3375 Description: allow number of MDS service threads to be changed at module load
3376 Details : It is now possible to change the number of MDS service threads
3377 running. Adding "options mds mds_num_threads={N}" to the MDS's
3378 /etc/modprobe.conf will set the number of threads for the next
3379 time Lustre is restarted (assuming the "mds" module is also
3380 reloaded at that time). The default number of threads will
3381 stay the same, 32 for most systems.
3386 Description: OST crash if filesystem is unformatted or corrupt
3387 Details : If an OST is started on a device that has never been formatted
3388 or if the filesystem is corrupt and cannot even mount then the
3389 error handling cleanup routines would dereference a NULL pointer.
3394 Description: NULL pointer deref in llap_from_page.
3395 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3396 incorrect) mapping if the page was truncated/reclaimed while it was
3397 searched for. Check for this condition and skip such pages when
3398 doing readahead. Introduce extra check to llap_from_page() to
3399 verify page->mapping->host is non-NULL (so page is not anonymous).
3402 Frequency : Sometimes when using sys_sendfile
3404 Description: "page not covered by a lock" warnings from ll_readpage
3405 Details : sendfile called ll_readpage without right page locks present.
3406 Now we introduced ll_file_sendfile that does necessary locking
3407 around call to generic_file_sendfile() much like we do in
3411 Frequency : with certain MDS communication failures at client mount time
3413 Description: NULL pointer deref after failed client mount
3414 Details : a client connection request may delayed by the network layer
3415 and not be sent until after the PTLRPC layer has timed out the
3416 request. If the client fails the mount immediately it will try
3417 to clean up before the network times out the request. Add a
3418 reference from the request import to the obd device and delay
3419 the cleanup until the network drops the request.
3422 Frequency : occasionally during client (re)connect
3424 Description: assertion failure during client (re)connect
3425 Details : processing a client connection request may be delayed by the
3426 client or server longer than the client connect timeout. This
3427 causes the client to resend the connection request. If the
3428 original connection request is replied in this interval, the
3429 client may trip an assertion failure in ptlrpc_connect_interpret()
3430 which thought it would be the only running connect process.
3433 Frequency : only with obd_echo servers and clients that are rebooted
3435 Description: kernel BUG accessing uninitialized data structure
3436 Details : When running an obd_echo server it did not start the ping_evictor
3437 thread, and when a client was evicted an uninitialized data
3438 structure was accessed. Start the ping_evictor in the RPC
3439 service startup instead of the OBD startup.
3441 Severity : enhancement
3442 Bugzilla : 10193 (patchless)
3443 Description: Remove dependency on various unexported kernel interfaces.
3444 Details : No longer need reparent_to_init, exit_mm, exit_files,
3445 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3448 Frequency : rare (only users of deprecated and unsupported LDAP config)
3450 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3451 Details : LDAP apparently contains 'lustreName' attributes instead of
3452 'name'. A simple remapping of the name is sufficient.
3455 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3457 Description: waiting_locks_callback trips kernel BUG if client is evicted
3458 Details : Running with the dump_on_timeout debug flag turned on makes
3459 it possible that the waiting_locks_callback() can try to dump
3460 the Lustre kernel debug logs from an interrupt handler. Defer
3461 this log dumping to the expired_lock_main() thread.
3463 Severity : enhancement
3465 Description: Support NFS exporting on 2.6 kernels.
3466 Details : Implement non-rawops metadata methods for NFS server to use without
3467 changing NFS server code.
3470 Frequency : very rare (synthetic metadata workload only)
3472 Description: two racing renames might cause an MDS thread to deadlock
3473 Details : Running the "racer" program may cause one MDS thread to rename
3474 a file from being the source of a rename to being the target of
3475 a rename at exactly the same time that another thread is doing
3476 so, and the second thread has already enqueued these locks after
3477 doing a lookup of the target and is trying to relock them in
3478 order. Ensure that we don't try to re-lock the same resource.
3481 Frequency : only very large systems with liblustre clients
3483 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3484 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3485 all OSTs in parallel. This allows the eviction of stale liblustre
3486 clients to proceed much faster than if they were done in series,
3487 and also offers similar improvements for other set_info RPCs.
3492 Description: excessive CPU usage during initial read phase on client
3493 Details : During the initial read phase on a client, it would agressively
3494 retry readahead on the file, consuming too much CPU and impacting
3495 performance (since 1.4.5.8). Improve the readahead algorithm
3496 to avoid this, and also improve some other common cases (read
3497 of small files in particular, where "small" is files smaller than
3498 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3503 Description: MDS crash when receiving packet with unknown intent.
3504 Details : Do not LBUG in unknown intent case, just return -EFAULT
3506 Severity : enhancement
3507 Bugzilla : 9293, 9385
3508 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3509 Details : Do not serialize getattr (non-intent version) and statfs.
3512 Frequency : occasional, when OST network is overloaded/intermittent
3514 Description: client evicted by OST after bulk IO timeout
3515 Details : If a client sends a bulk IO request (read or write) the OST
3516 may evict the client if it is unresposive to its data GET/PUT
3517 request. This is incorrect if the network is overloaded (takes
3518 too long to transfer the RPC data) or dropped the OST GET/PUT
3519 request. There is no need to evict the client at all, since
3520 the pinger and/or lock callbacks will handle this, and the
3521 client can restart the bulk request.
3524 Frequency : Always when mmapping file with no objects
3526 Description: client crashes when mmapping file with no objects
3527 Details : Check that we actually have objects in a file before doing any
3528 operations on objects in ll_vm_open, ll_vm_close and
3534 Description: Request leak when working with deleted CWD
3535 Details : Introduce advanced request refcount tracking for requests
3536 referenced from lustre intent.
3538 Severity : Enhancement
3540 Description: Cache open file handles on client.
3541 Details : MDS now will return special lock along with openhandle, if
3542 requested and client is allowed to hold openhandle, even if unused,
3543 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3544 closing files for every read/write openration.
3546 Severity : Enhancement
3548 Description: Cache open negative dentries on client when possible.
3549 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3550 negative dentries on lock revocation.
3555 Description: Remounting a client read-only wasn't possible with a zconf mount
3556 Details : It wasn't possible to remount a client read-only with llmount.
3558 Severity : enhancement
3559 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3560 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3561 MPICH is included to add Lustre-specific ADIO interfaces.
3562 This is based closely on the UFS ADIO layer and only differs
3563 in file creation, in order to allow the OST striping to be set.
3564 This is user-contributed code and not supported by CFS.
3569 Description: extended inode attributes (immutable, append-only) work improperly
3570 when 2.4 and 2.6 kernels are used on client/server or vice versa
3571 Details : Introduce kernel-independent values for these flags.
3573 Severity : enhancement
3576 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3577 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3578 granularity. Now, Fractional MB granularity is be supported,
3579 this is very useful for low memory system.
3581 Severity : enhancement
3583 Description: Getattr by fid
3584 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3585 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3586 have locks and avoids getattr rpc requests for it on every stat.
3589 Frequency : Always, for filesystems larger than 2TB
3591 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3592 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3593 filesystems larger than 2TB to be mounted with mballoc enabled.
3596 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3598 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3599 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3600 filesystems larger than 2TB is mounted then the kernel will
3601 silently corrupt the start of the filesystem. CONFIG_LBD is
3602 enabled for all CFS-supported kernels, but the possibility of
3603 this happening with a modified kernel config exists.
3605 Severity : enhancement
3607 Description: add client O_DIRECT support for 2.6 kernels
3608 Details : It is now possible to do O_DIRECT reads and writes to files
3609 in the Lustre client mountpoint on 2.6 kernel clients.
3611 Severity : enhancement
3613 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3614 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3615 parallel, not waiting for response from every OST before sending
3616 a rpc to the next OST.
3621 Description: setattr vs write race when updating file timestamps
3622 Details : Client processes that update a file timestamp into the past
3623 right after writing to the file (e.g. tar) it is possible that
3624 the updated file modification time can be reset to the current
3625 time due to a race between processing the setattr and write RPC.
3627 Severity : enhancement
3629 Description: Bring 'lfs find' closer in line with regular Linux find.
3630 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3631 -print0 options and obtains all the needed info through the lustre
3634 Severity : enhancement
3636 Description: support up to 1024 configured devices on one node
3637 Details : change obd_dev array from statically allocated to dynamically
3638 allocated structs as they are first used to reduce memory usage
3643 Description: Flush dirty partially truncated pages during truncate
3644 Details : Immediatelly flush partially truncated pages in filter_setattr,
3645 this way we completely avoid having any pages in page cache on OST
3646 and can retire ugly workarounds during writes to flush such pages.
3651 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3652 Details : protect inode from truncation within vfs_unlink() context
3653 just take a reference before calling vfs_unlink() and release it
3654 when parent's i_sem is free.
3659 Description: last_id value checked outside lock on OST caused LASSERT failure
3660 Details : If there were multiple MDS->OST object precreate requests in
3661 flight, it was possible that the OST's last object id was checked
3662 outside a lock and incorrectly tripped an assertion. Move checks
3663 inside locks, and discard old precreate requests.
3666 Frequency : always, if extents are used on OSTs
3668 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3669 Details : OSTs using the extents format would not zero the high 16 bits of
3670 the index physical block number. This is not a problem for any
3671 OST filesystems smaller than 16TB, and no kernels support ext3
3672 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3673 new/modified files) and can be fixed for existing filesystems
3674 with e2fsprogs-1.39-cfs1.
3679 Description: import connection selection may be incorrect if timer wraps
3680 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3681 import connections to be ignored if the 32-bit jiffies counter
3682 wraps. Use a 64-bit jiffies counter.
3685 Frequency : very large clusters immediately after boot
3687 Description: LNET request buffers exhausted under heavy short-term load
3688 Details : If a large number of client requests are generated on a service
3689 that has previously never seen so many requests it is possible
3690 that the request buffer growth cannot keep up with the spike in
3691 demand. Instead of dropping incoming requests, they are held in
3692 the LND until the RPC service can accept more requests.
3695 Frequency : Sometimes during replay
3697 Description: Assertion failure in ll_local_open after replay.
3698 Details : If replay happened on an open request reply before we were able
3699 to set replay handler, reply will become not swabbed tripping the
3700 assertion in ll_local_open. Now we set the handler right after
3701 recognising of open request
3704 Frequency : very rare
3706 Description: kernel reports "badness in vsnprintf"
3707 Details : Reading from the "recovery_status" /proc file in small chunks
3708 may cause a negative length in lprocfs_obd_rd_recovery_status()
3709 call to vsnprintf() (which is otherwise harmless). Exit early
3710 if there is no more space in the output buffer.
3712 Severity : enhancement
3714 Description: clear OBD RPC statistics by writing to them
3715 Details : It is now possible to clear the OBD RPC statistics by writing
3716 to the "stats" file.
3721 Description: Client mtime is not the same on different clients after utimes
3722 Details : In some cases, the client was using the utimes() syscall on
3723 a file cached on another node. The clients now validate the
3724 ctime from the MDS + OSTs to determine which one is right.
3729 Description: Inability to activate failout mode
3730 Details : lconf script incorrectly assumed that in pythong string's numeric
3731 value is used in comparisons.
3734 Frequency : always with multiple stripes per file
3736 Description: Inefficient object allocation for mutli-stripe files
3737 Details : When selecting which OSTs to stripe files over, for files with
3738 a stripe count that divides evenly into the number of OSTs,
3739 the MDS is always picking the same starting OST for each file.
3740 Return the OST selection heuristic to the original design.
3745 Description: mount failures may take full timeout to return an error
3746 Details : Under some heavy load conditions it is possible that a
3747 failed mount can wait for the full obd_timeout interval,
3748 possibly several minutes, before reporting an error.
3749 Instead return an error as soon as the status is known.
3751 Frequency : quota enabled and large files being deleted
3753 Description: releasing more than 4GB of quota at once hangs OST
3754 Details : If a user deletes more than 4GB of files on a single OST it
3755 will cause the OST to spin in an infinite loop. Release
3756 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3761 Description: statfs data retrieved from /proc may be stale or zero
3762 Details : When reading per-device statfs data from /proc, in the
3763 {kbytes,files}_{total,free,avail} files, it may appear
3764 as zero or be out of date.
3767 Frequency : systems with MD RAID1 external journal devices
3769 Description: lconf's call to blkid is confused by RAID1 journal devices
3770 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3771 of returning all block devices that match the journal UUID.
3774 Frequency : always, for aggregate stripe size over 4GB
3776 Description: assertion fails when trying to use 4GB stripe size
3777 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3778 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3781 Frequency : always on ppc64
3783 Description: the first write on an ext3 filesystem with mballoc got stuck
3784 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3785 perform endianness conversion.
3787 ------------------------------------------------------------------------------
3789 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3791 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3792 INTEROPERATE with older versions automatically. Please read the
3793 user documentation before upgrading any part of a live system.
3794 * WARNING: Lustre networking configuration changes are required with
3795 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3798 * Support for kernels:
3799 2.6.9-22.0.2.EL (RHEL 4)
3800 2.6.5-7.244 (SLES 9)
3801 2.6.12.6 vanilla (kernel.org)
3804 Severity : enhancement
3805 Bugzilla : 7981/8208
3806 Description: Introduced Lustre Networking (LNET)
3807 Details : LNET is new networking infrastructure for Lustre, it includes
3808 a reorganized network configuration mode (see the user
3809 documentation for full details) as well as support for routing
3810 between different network fabrics. Lustre Networking Devices
3811 (LNDs) for the supported network fabrics have also been
3812 created for this new infrastructure.
3814 Severity : enhancement
3815 Description: Introduced Access control lists
3816 Details : clients can set ACLs on files and directories in order to have
3817 more fine-grained permissions than the standard Unix UGO+RWX.
3818 The MDS must be started with the "-o acl" mount option.
3820 Severity : enhancement
3821 Description: Introduced filesystem quotas
3822 Details : Administrators may now establish per-user quotas on the
3825 Severity : enhancement
3827 Description: Configuration change for the XT3
3828 The PTLLND is now used to run Lustre over Portals on the XT3
3829 The configure option(s) --with-cray-portals are no longer used.
3830 Rather --with-portals=<path-to-portals-includes> is used to
3831 enable building on the XT3. In addition to enable XT3 specific
3832 features the option --enable-cray-xt3 must be used.
3837 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3838 Details : A race between allocating a new llcd and re-getting the llcd_lock
3839 allowed another thread to grab newly-allocated llcd.
3841 Severity : enhancement
3843 Description: 2.6 OST async journal commit and locking fix to improve performance
3844 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3845 2.6 kernels are now async as they already were in 2.4 kernels so
3846 that they can commit concurrently with the network bulk transfer.
3847 For block-allocated files the filter allocation semaphore is held
3848 to avoid filesystem fragmentation during allocation. BKL lock
3849 removed for 2.6 xattr operations where it is no longer needed.
3854 Description: lconf incorrectly determined whether two IP networks could talk
3855 Details : In some more complicated routing and multiple-network
3856 configurations, lconf will avoid trying to make a network
3857 connection to a disjoint part of the IP space. It was doing the
3858 math incorrectly for one set of cases.
3863 Description: Fix for potential infinite loop processing records in an llog.
3864 Details : If an llog record is corrupted/zeroed, it is possible to loop
3865 forever in llog_process(). Validate the llog record length
3866 and skip the remainder of the block on error.
3869 Frequency : occasional (liblustre only)
3871 Description: liblustre could not open files whose last component is a symlink
3872 Details : sysio_path_walk() would incorrectly pass the open intent to
3873 intermediate path components.
3876 Frequency : rare (liblustre only with non-standard tuning)
3877 Bugzilla : 7201 (7350)
3878 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3879 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3880 after it has been released, only do this for referenced locks
3882 Severity : enhancement
3884 Description: specifying an (invalid) directory default stripe_size of -1
3885 would reset the directory default striping
3886 Details : stripe_size -1 was used internally to signal directory stripe
3887 removal, now use "all default" to signal dir stripe removal
3888 as a directory striping of "all default" is not useful
3891 Frequency : common for large clusters running liblustre clients
3893 Description: doing an ls when liblustre clients are running is slow
3894 Details : sending a glimpse AST to a liblustre client waits for every AST
3895 to time out, as liblustre clients will not respond. Since they
3896 cannot cache data we refresh the OST lock LVB from disk instead.
3898 Severity : enhancement
3900 Description: doing an ls at the same time as file IO can be slow
3901 Details : enqueue and other "small" requests can be blocked behind many
3902 large IO requests. Create a new OST IO portal for non-IO
3903 requests so they can be processed faster.
3906 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3908 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3909 created file with mode 000
3910 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3911 allow an ftruncate() as a non-root user to a file with mode 000.
3912 Lustre now respects this flag to disable mode checks when
3913 truncating a file owned by the user
3916 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3919 Description: Revoking locks from clients that went dead or catatonic might take
3921 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3922 cancellation of such locks instant on servers without waiting for
3923 any reply from clients. Clients drops these locks when cancel
3924 notification from server is received without replying.
3927 Frequency : liblustre-only, when liblustre client dies or becomes busy
3929 Description: Doing ls on Linux clients can take a long time with active
3931 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3932 granting such locks to it in the first place if possible. Locks
3933 are taken by proxy on the OST during the read or write and
3934 dropped immediately afterward. Add connect flags handling, do
3935 not grant locks to liblustre clients for glimpse ASTs.
3937 Severity : enhancement
3939 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3940 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3941 sequential reads and behaves badly with applications doing random
3942 reads. Improve it by reading ahead at least read region, and
3943 avoiding excessive large RPC for small reads.
3945 Severity : enhancement
3947 Description: Creating more than 1000 files for a single job may cause a load
3948 imbalance on the OSTs if there are also a large number of OSTs.
3949 Details : qos_prep_create() uses an OST index reseed value that is an
3950 even multiple of the number of available OSTs so that if the
3951 reseed happens in the middle of the object allocation it will
3952 still utilize the OSTs as uniformly as possible.
3957 Description: OST or MDS may oops in ping_evictor_main()
3958 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3959 but doesn't restart at beginning of obd_exports_timed list
3962 Severity : enhancement
3964 Description: improve by-nid export eviction on the MDS and OST
3965 Details : allow multiple exports with the same NID to be evicted at one
3966 time without re-searching the exports list.
3969 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3971 Description: MDS may oops in groups_free()
3972 Details : in rare race conditions a newly allocated group_info struct is
3973 freed again, and this can be NULL. The 2.4 compatibility code
3974 for groups_free() checked for a NULL pointer, but 2.6 did not.
3977 Frequency : common for liblustre clients doing little filesystem IO
3978 Bugzilla : 9352, 7313
3979 Description: server may evict liblustre clients accessing contended locks
3980 Details : if a client is granted a lock or receives a completion AST
3981 with a blocking AST already set it would not reply to the AST
3982 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3985 Frequency : lfs setstripe, only systems with more than 160 OSTs
3987 Description: unable to set striping with a starting offset beyond OST 160
3988 Details : llapi_create_file() incorrectly limited the starting stripe
3989 index to the maximum single-file stripe count.
3992 Frequency : LDAP users only
3994 Description: lconf did not handle in-kernel recovery with LDAP properly
3995 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3997 Severity : enhancement
3999 Description: bind OST threads to NUMA nodes to improve performance
4000 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4001 node and do their allocations there to localize memory access
4003 Severity : enhancement
4005 Description: llmount can determine client NID directly from Myrinet (GM)
4006 Details : the client NID code from gmnalnid was moved directly into
4007 llmount, removing the need to use this or specifying the
4008 client NID explicitly when mounting GM clients with zeroconf
4011 Frequency : if client is started with down MDS
4013 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4014 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4015 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4016 ptlrpc_import_delayed_req() because we didn't check if the
4017 request was interrupted, and we also didn't break out of the
4018 event loop if there was no timeout
4023 Description: data loss during non-page-aligned writes to a single file from
4024 both multiple nodes and multiple threads on one node at same time
4025 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4026 inconsistency led to false short-reads, that were cached and later
4027 used by ->prepare_write() to fill in partially written page,
4028 leading to data loss.
4031 Frequency : always, if lconf --abort_recovery used
4033 Description: lconf --abort_recovery fails with 'Operation not supported'
4034 Details : lconf was attempting to abort recovery on the MDT device and not
4037 Severity : enhancement
4039 Description: remove cleanup logs
4040 Details : replace lconf-generated cleanup logs with lustre internal
4041 cleanup routines. Eliminates the need for client-cleanup and
4044 Severity : enhancement
4046 Description: add support for EAs (user and system) on lustre filesystems
4047 Details : it is now possible to store extended attributes in the Lustre
4048 client filesystem, and with the user_xattr mount option it
4049 is possible to allow users to store EAs on their files also
4051 Severity : enhancement
4053 Description: Add possibility (config option) to show minimal available OST free
4055 Details : When compiled with --enable-mindf configure option, statfs(2)
4056 (and so, df) will return least minimal free space available from
4057 all OSTs as amount of free space on FS, instead of summary of
4058 free spaces of all OSTs.
4060 Severity : enhancement
4062 Description: do not expand extent locks acquired on OST-side
4063 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4064 by server: they are not cached anyway.
4067 Frequency : when mmap is used/binaries executed from Lustre
4069 Description: Unmmap pages before throwing them away from read cache.
4070 Details : llap_shrink cache now attempts to unmap pages before discarding
4071 them (if unmapping failed - do not discard). SLES9 kernel has
4072 extra checks that trigger if this unmapping is not done first.
4077 Description: lconf didn't resolve symlinks before checking to see whether a
4078 given mountpoint was already in use
4081 Frequency : when migrating failover services
4082 Bugzilla : 6395, 9514
4083 Description: When migrating a subset of services from a node (e.g. failback
4084 from a failover service node) the remaining services would
4085 time out and evict clients.
4086 Details : lconf --force (implied by --failover) sets the global obd_timeout
4087 to 5 seconds in order to quickly disconnect, but this caused
4088 other RPCs to time out too quickly. Do not change the global
4089 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4091 Severity : enhancement
4092 Frequency : if MDS is started with down OST
4093 Bugzilla : 9439,5706
4094 Description: Allow startup/shutdown of an MDS without depending on the
4095 availability of the OSTs.
4096 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4097 Add appropriate locking and lov-osc refcounts for safe
4098 cleaning. Add osc abort_inflight calls in case the
4099 synchronize never started.
4102 Frequency : occasional (Cray XT3 only)
4104 Description: root not authorized to access files in CRAY_PORTALS environment
4105 Details : The client process capabilities were not honoured on the MDS in
4106 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4107 been accessed by an authorized user then root was able to access
4108 the file on the local client also. The root user capabilities
4109 are now allowed on the MDS, as this environment has secure UID.
4112 Frequency : occasional
4114 Description: ldiskfs "too long searching" message happens too often
4115 Details : A debugging message (otherwise harmless) prints too often on
4116 the OST console. This has been reduced to only happen when
4117 there are fragmentation problems on the filesystem.
4122 Description: Division by zero in statfs when all OSCs are inactive
4123 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4124 lov_statfs divides by value returned by lov_get_stripecnt().
4128 Bugzilla : 9489, 3273
4129 Description: First write from each client to each OST was only 4kB in size,
4130 to initialize client writeback cache, which caused sub-optimal
4131 RPCs and poor layout on disk for the first writen file.
4132 Details : Clients now request an initial cache grant at (re)connect time
4133 and so that they can start streaming writes to the cache right
4134 away and always do full-sized RPCs if there is enough data.
4135 If the OST is rebooted the client also re-establishes its grant
4136 so that client cached writes will be honoured under the grant.
4141 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4142 Details : Now I/O RPCs go to different portal number and (presumably) fast
4143 lock requests (and glimses) and other RPCs get their own service
4144 threads pool that should be able to service those RPCs
4147 Severity : enhancement
4149 Description: Ability to exchange lustre version between client and servers and
4150 issue warnings at client side if client is too old. Also for
4151 liblustre clients there is ability to refuse connection of too old
4153 Details : New 'version' field is added to connect data structure that is
4154 filled with version info. That info is later checked by server and
4158 Frequency : rare, liblustre only.
4159 Bugzilla : 9296, 9581
4160 Description: Two simultaneous writes from liblustre at offset within same page
4161 might proceed at the same time overwriting eachother with stale
4163 Details : I/O lock withing llu_file_prwv was released too early, before data
4164 actually was hitting the wire. Extended lock-holding time until
4165 server acknowledges receiving data.
4168 Frequency : extremely rare. Never observed in practice.
4170 Description: avoid generating lustre_handle cookie of 0.
4171 Details : class_handle_hash() generates handle cookies by incrementing
4172 global counter, and can hit 0 occasionaly (this is unlikely, but
4173 not impossible, because initial value of cookie counter is
4174 selected randonly). Value of 0 is used as a sentinel meaning
4175 "unassigned handle" --- avoid it. Also coalesce two critical
4176 sections in this function into one.
4178 Severity : enhancement
4180 Description: allow liblustre clients to delegate truncate locking to OST
4181 Details : To avoid overhead of locking, liblustre client instructs OST to
4182 take extent lock in ost_punch() on client's behalf. New connection
4183 flag is added to handle backward compatibility.
4185 Severity : enhancement
4186 Bugzilla : 4928, 7341, 9758
4187 Description: allow number of OST service threads to be specified
4188 Details : a module parameter allows the number of OST service threads
4189 to be specified via "options ost ost_num_threads={N}" in the
4190 OSS's /etc/modules.conf or /etc/modprobe.conf.
4194 Bugzilla : 6146, 9635, 9895
4195 Description: servers crash with bad pointer in target_handle_connect()
4196 Details : In rare cases when a client is reconnecting it was possible that
4197 the connection request was the last reference for that export.
4198 We would temporarily drop the export reference and get a new
4199 one, but this may have been the last reference and the export
4200 was just destroyed. Get new reference before dropping old one.
4202 Severity : enhancement
4203 Frequency : if client is started with failover MDS
4205 Description: Allow multiple MDS hostnames in the mount command
4206 Details : Try to read the configuration from all specified MDS
4207 hostnames during a client mount in case the "primary"
4210 Severity : enhancement
4212 Description: Stop sending data to evicted clients as soon as possible.
4213 Details : Check if the client we are about to send or are sending data to
4214 was evicted already. (Check is done every second of waiting,
4215 for which l_wait_event interface was extended to allow checking
4216 of exit condition at specified intervals).
4219 Frequency : rare, normally only when NFS exporting is done from client
4221 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4223 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4226 Frequency : always, if srand() is called before liblustre initialization
4228 Description: Liblustre uses system PRNG disturbing its usage by user application
4229 Details : Introduce internal to lustre fast and high-quality PRNG for
4230 lustre usage and make liblustre and some other places in generic
4231 lustre code to use it.
4233 Severity : enhancement
4234 Bugzilla : 9477, 9557, 9870
4235 Description: Verify that the MDS configuration logs are updated when xml is
4236 Details : Check if the .xml configuration logs are newer than the config
4237 logs stored on the MDS and report an error if this is the case.
4238 Request --write-conf, or allow starting with --old_conf.
4240 Severity : enhancement
4242 Description: Handle symlinks in the path when checking if Lustre is mounted.
4243 Details : Resolve intermediate symlinks when checking if a client has
4244 mounted a filesystem to avoid duplicate client mounts.
4249 Description: lconf can hit an error exception but still return success.
4250 Details : The lconf command catches the Command error exception at the top
4251 level script context and will exit with the associated exit
4252 status, but doesn't ensure that this exit status is non-zero.
4257 Description: failure of ptlrpc thread startup can cause oops
4258 Details : Starting a ptlrpc service thread can fail if there are a large
4259 number of threads or the server memory is very fragmented.
4260 Handle this without oopsing.
4263 Frequency : always, only if liblustre and non-default acceptor port was used
4265 Description: liblustre cannot connect to servers with non-default acceptor port
4266 Details : tcpnal_set_default_params() was not called and was therefore
4267 ignoring the environment varaible TCPNAL_PORT, as well as other
4268 TCPNAL_ environment variables
4273 Description: two objects could be created on the same OST for a single file
4274 Details : If an OST is down, in some cases it was possible to create two
4275 objects on a single OST for a single file. No problems other
4276 than potential performance impact and spurious error messages.
4280 Bugzilla : 5681, 9562
4281 Description: Client may oops in ll_unhash_aliases
4282 Details : Client dcache may become inconsistent in race condition.
4283 In some cases "getcwd" can fail if the current directory is
4289 Description: Inode refcounting problems in NFS export code
4290 Details : link_raw functions used to call d_instantiate without obtaining
4291 extra inode reference first.
4295 Bugzilla : 9942, 9903
4296 Description: Referencing freed requests leading to crash, memleaks with NFS.
4297 Details : We used to require that call to ll_revalidate_it was always
4298 followed by ll_lookup_it. Also with revalidate_special() it is
4299 possible to call ll_revalidate_it() twice for the same dentry
4300 even if first occurence returned success. This fix changes semantic
4301 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4302 reference on a request referred from the intent.
4303 ll_intent_release() then releases such a request.
4306 Frequency : rare, normally benchmark loads only
4308 Description: unlinked inodes were kept in memory on the client
4309 Details : If a client is repeatedly creating and unlinking files it
4310 can accumulate a lot of stale inodes in the inode slab cache.
4311 If there is no other client load running this can cause the
4312 client node to run out of memory. Instead flush old inodes
4313 from client cache that have the same inode number as a new inode.
4316 Frequency : SLES9 2.6.5 kernel and long filenames only
4317 Bugzilla : 9969, 10379
4318 Description: utime reports stale NFS file handle
4319 Details : SLES9 uses out-of-dentry names in some cases, which confused
4320 the lustre dentry revalidation. Change it to always use the
4324 Frequency : rare, unless heavy write-truncate concurrency is continuous
4325 Bugzilla : 4180, 6984, 7171, 9963, 9331
4326 Description: OST becomes very slow and/or deadlocked during object unlink
4327 Details : filter_destroy() was holding onto the parent directory lock
4328 while truncating+unlinking objects. For very large objects this
4329 may block other threads for a long time and slow overall OST
4330 responsiveness. It may also be possible to get a lock ordering
4331 deadlock in this case, or run out of journal credits because of
4332 the combined truncate+unlink. Solution is to do object truncate
4333 first in one transaction without parent lock, and then do the
4334 final unlink in a new transaction with the parent lock. This
4335 reduces the lock hold time dramatically.
4338 Frequency : rare, 2.4 kernels only
4340 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4341 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4342 before unmounting their filesystem in order to avoid deadlock.
4343 The kernel_locked() function in 2.4 kernels only checks whether
4344 the kernel lock is held, not whether it is this process that is
4345 holding it as 2.6 kernels do.
4350 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4351 Details : The client ptlrpc code may be trying to reconnect to a down
4352 server before a previous connection attempt has timed out.
4353 Increase the reconnect interval to be longer than the connection
4354 timeout interval to avoid sending duplicate connections to
4358 Frequency : echo_client brw_test command
4360 Description: fix echo_client to work with OST preallocated code
4361 Details : OST preallocation code (5137) didn't take echo_client IO path
4362 into account: echo_client calls filter methods outside of any
4363 OST thread and, hence, there is no per-thread preallocated
4364 pages and buffers to use. Solution: hijack pga pages for IO. As
4365 a byproduct, this avoids unnecessary data copying.
4369 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4370 Description: Client can oops in mdc_commit_close() after open replay
4371 Details : It was possible for the MDS to return an open request with no
4372 transaction number in mds_finish_transno() if the client was
4373 evicted, but without actually returning an error. Clients
4374 would later try to replay that open and may trip an assertion
4375 Simplify the client close codepath, and always return an error
4376 from the MDS in case the open is not successful.
4379 Frequency : rare, 2.6 OSTs only
4381 Description: OST may deadlock under high load on fragmented files
4382 Details : If there was a heavy load and highly-fragmented OST filesystems
4383 it was possible to have all the OST threads deadlock waiting on
4384 allocation of biovecs, because the biovecs were not released
4385 until the entire RPC IO was completed. Instead, release biovecs
4386 as soon as they are complete to ensure forward IO progress.
4388 Severity : enhancement
4390 Description: Support for specifying external journal device at mount
4391 Details : If an OST or MDS device is formatted with an external journal
4392 device, this device major/minor is stored in the ext3 superblock
4393 and may not be valid for failover. Allow detecting and
4394 specifying the external journal at mount time.
4399 Description: Mounting an MDS with pending unlinked files may cause oops
4400 Details : target_finish_recovery() calls mds_postrecov() which returned
4401 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4402 considers this an error and immediately begins cleaning up the
4403 lov, just after starting the mds_lov process
4405 Severity : enhancement
4407 Description: Implement 'lfs df' to report actual free space on per-OST basis
4408 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4409 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4410 usage of objects; '-h' to report in human readable format.
4412 ------------------------------------------------------------------------------
4414 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4421 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4422 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4423 at startup during orphan unlinking. Instead, skip buddy update
4424 before setup, as it will be regenerated after recovery is complete.
4429 Description: If an OST is inactive, its locks might reference stale inodes.
4430 Details : lov_change_cbdata() must iterate over all namespaces, even if
4431 they are inactive to clear inode references from the lock.
4433 Severity : enhancement
4434 Frequency : occasional, if non-standard max_dirty_mb used
4436 Description: Client will block write RPCs if not enough grant
4437 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4438 then the client will block writes while waiting for another RPC
4439 to complete instead of consuming its dirty limit. With change
4440 we get improved performance when max_dirty_mb is small.
4442 Severity : enhancement
4443 Bugzilla : 3389, 6253
4444 Description: Add support for supplementary groups on the MDS.
4445 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4446 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4447 lookups for user supplementary groups into a cache.
4451 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4452 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4453 for liblustre O_EXCL is enforced by MDS.
4457 Description: readdir never returns NULL in liblustre.
4458 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4459 dentry in struct dirent.
4463 Frequency : liblustre only, depends on application IO pattern
4464 Description: liblustre clients evicted if not contacting servers
4465 Details : Don't put liblustre clients into the ping_evictor list, so
4466 they will not be evicted by the pinger ever.
4468 Severity : enhancement
4470 Description: Add ability to evict clients by NID from MDS.
4471 Details : By echoing "nid:$NID" string into
4472 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4473 $NID would be instantly evicted from this MDS and from all active
4474 OSTs connected to it.
4478 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4479 Details : lookup_it_finish() used to query file size from OSTs that was not
4484 Description: service threads change working directory to that of init
4485 Details : Starting lustre service threads may pin the working directory
4486 of the parent thread, making that filesystem busy. Threads
4487 now change to the working directory of init to avoid this.
4491 Frequency : during shutdown only
4492 Description: shutdown with a failed MDS or OST can cause unmount to hang
4493 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4498 Frequency : 2.6 only
4499 Description: chmod/chown may include an extra supplementary group
4500 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4501 supplementary group and if none is specified this is used.
4504 Bugzilla : 5479 (6816)
4506 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4507 Details : If lookup is in progress on a file that is unlinked we might try
4508 to revalidate the inode and fail in revalidate after lookup is
4509 complete and ll_file_open() enqueues the open again but
4510 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4513 Frequency : always, if lconf --abort_recovery used
4515 Description: lconf --abort_recovery fails with 'Operation not supported'
4516 Details : lconf was attempting to abort recovery on the MDT device and not
4519 ------------------------------------------------------------------------------
4521 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4526 Frequency : rare (only unsupported configurations with a node running as an
4528 Bugzilla : 6514, 5137
4529 Description: Mounting a Lustre file system on a node running as an OST could
4531 Details : OSTs now preallocates memory needed to write out data at
4532 startup, instead of when needed, to avoid having to
4533 allocate memory in possibly low memory situations.
4534 Specifically, if the file system is mounted on on OST,
4535 memory pressure could force it to try to write out data,
4536 which it needed to allocate memory to do. Due to the low
4537 memory, it would be unable to do so and the node would
4538 become unresponsive.
4540 Severity : enhancement
4542 Description: Addition of lconf --service command line option
4543 Details : lconf now accepts a '--service <arg>' option, which is
4544 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4546 Severity : enhancement
4548 Description: Failover mode is now the default for OSTs.
4549 Details : By default, OSTs will now run in failover mode. To return to
4550 the old behaviour, add '--failout' to the lmc line for OSTs.
4552 Severity : enhancement
4554 Description: Health checks are now provided for MDS and OSTs
4555 Details : Additional detailed health check information on MSD and OSTs
4556 is now provided through the procfs health_check value.
4559 Frequency : occasional, depends on IO load
4561 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4562 after numerous create/delete cycles
4563 Details : The ext3 inode allocation policy would not allocate new inodes
4564 very well on the OSTs because there are no new directories
4565 being created. Instead we look for groups with free space if
4566 the parent directories are nearly full.
4571 Description: Network or server problems during mount may cause partially
4572 mounted clients instead of returning an error.
4573 Details : The config llog parsing code may overwrite the error return
4574 code during mount error handling, returning success instead
4580 Description: MDS can fail to allocate large reply buffers
4581 Details : After long uptimes the MDS can fail to allocate large reply
4582 buffers (e.g. zconf client mount config records) due to memory
4583 fragmentation or consumption by the buffer cache. Preallocate
4584 some large reply buffers so that these replies can be sent even
4585 under memory pressure.
4589 Frequency : rare (liblustre)
4590 Description: fsx running with liblustre complained that using truncate() to
4591 extend the file doesn't work. This patch corrects that issue.
4592 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4593 ATTR_SIZE and lsm use in llu_setattr_raw.
4597 Frequency : rare, only 2.6 kernels
4598 Description: Unusual file access patterns on the MDS may result in inode
4599 data being lost in very rare circumstances.
4600 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4601 code in the 2.6 kernel caused the inode and/or EA data not to
4602 be read from disk, causing single-file corruption.
4606 Frequency : rare, only 2.6 filesystems using extents
4607 Description: Heavy concurrent write and delete load may cause data corruption.
4608 Details : It was possible under high-load situations to have an extent
4609 metadata block in the block device cache from a just-unlinked
4610 file overwrite a newly-allocated data block. We now unmap any
4611 metadata buffers that alias just-allocated data blocks.
4615 Frequency : filesystems with default stripe_count larger than 77
4616 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4617 Details : lconf specifies an inode size of 4096 bytes when the default
4618 stripe_count is larger than 77. This conflicts with the default
4619 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4620 this case to avoid pinning too much memory for large EAs.
4622 ------------------------------------------------------------------------------
4624 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4629 Frequency : rare (extremely heavy IO load with hundreds of clients)
4631 Description: Client is evicted, gets IO error writing to file
4632 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4633 caused clients to be evicted for AST timeouts. The fixes in
4634 bug 5192 mean we no longer need to have such short AST timeouts
4635 so ldlm_timeout has been increased.
4638 Frequency : occasional during --force or --failover shutdown under load
4639 Bugzilla : 5949, 4834
4640 Description: Server oops/LBUG if stopped with --force or --failover under load
4641 Details : a collection of import/export refcount and cleanup ordering
4642 issues fixed for safer force cleanup
4645 Frequency : only filesystems larger than 120 OSTs
4646 Bugzilla : 5990, 6223
4647 Description: lfs getstripe would oops on a very large filesystem
4648 Details : lov_getconfig used kfree on vmalloc'd memory
4651 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4652 Bugzilla : 6242, 6243
4653 Description: reading from files that had been truncated to a non-zero size
4654 but never opened returned no data
4655 Details : ll_file_read() reads zeros from no-object files to EOF
4660 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4661 Details : There's little harm in aborting MDS/OSS recovery and letting it
4662 try again, so I removed the LASSERT and return an error instead.
4664 Severity : enhancement
4666 Description: New debugging infrastructure for tracking down data corruption
4667 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4668 (b) cover more of the client-side code path, and (c) try to narrow
4669 down where problems occurred
4673 Bugzilla : 3819, 4364, 4397, 6313
4674 Description: Racing close and eviction MDS could cause assertion in mds_close
4675 Details : It was possible to get multiple mfd references during close and
4676 client eviction, leading to one thread referencing a freed mfd.
4678 Severity: : enhancement
4679 Bugzilla : 3262, 6359
4680 Description: Attempts to reconnect to servers are now more aggressive.
4681 Details : This builds on the enhanced upcall-less recovery that was added
4682 in 1.4.2. When trying to reconnect to servers, clients will
4683 now try each server in the failover group every 10 seconds. By
4684 default, clients would previously try one server every 25 seconds.
4689 Description: After recovery, certain operations trigger a failed
4690 assertion on a client.
4691 Details : Failing over an mds, using lconf -d --failover, while a
4692 client was doing a readdir() call would cause the client to
4693 LBUG after recovery completed and the readdir() was resent.
4695 Severity : enhancement
4697 Description: Default groups are now added by lconf
4698 Details : You can now run lconf --group <servicename> without having to
4699 manually add groups with lmc.
4702 Frequency : occasional
4704 Description: Nodes with an elan id of 0 trigger a failed assertion
4707 Frequency : always when accessing e.g. tty/console device nodes
4709 Description: tty and some other devices nodes cannot be used on lustre
4710 Details : file's private_data field is used by device data and lustre
4711 values in there got lost. New field was added to struct file to
4712 store fs-specific private data.
4715 Frequency : when exporting Lustre via NFS
4717 Description: NFSD failed occasionally when looking up a path component
4718 Details : NFSD is looking up ".." which was broken in ext3 directories
4719 that had grown large enough to become hashed.
4722 Frequency : Clusters with multiple interfaces not on the same subnet
4724 Description: Nodes will repeatedly try to reconnect to an interface which it
4725 cannot reach and report an error to the log.
4726 Details : Extra peer list entries will be created by lconf with some peers
4727 unreachable. lconf now validates the peer before adding it.
4730 Frequency : Only if a default stripe is set on the filesystem root.
4732 Description: Setting a default stripe on the filesystem root prevented the
4733 filesystem from being remounted.
4734 Details : The client was sending extra request flags in the root getattr
4735 request and did not allocate a reply buffer for the dir EA.
4738 Frequency : occasional, higher if lots of files are accessed by one client
4739 Bugzilla : 6159, 6097
4740 Description: Client trips assertion regarding lsm mismatch/magic
4741 Details : While revalidating inodes the VFS looks up inodes with ifind()
4742 and in rare cases can find an inode that is being freed.
4743 The ll_test_inode() code will free the lsm during ifind()
4744 when it finds an existing inode and then the VFS later attaches
4745 this free lsm to a new inode.
4749 Bugzilla : 6422, 7030
4750 Description: MDS deadlock between mkdir and client eviction
4751 Details : Creating a new file via mkdir or mknod (starting a transaction
4752 and getting the ns lock) can deadlock with client eviction
4753 (gets ns lock and trying to finish a synchronous transaction).
4756 Frequency : occasional
4757 Description: While starting a server, the fsfilt_ext3 module could not be
4759 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4760 kernels. Previously, lconf would still use the ext3 name
4761 when trying to load modules. Now, it will correctly use
4762 ext3 on 2.4 and ldiskfs on 2.6.
4764 Severity : enhancement
4765 Description: The default stripe count has been changed to 1
4766 Details : The interpretation of the default stripe count (0, to lfs
4767 or lmc) has been changed to mean striping across a single
4768 OST, rather than all available. For general usage we have
4769 found a stripe count of 1 or 2 works best.
4771 Severity : enhancement
4772 Description: Add support for compiling against Cray portals.
4773 Details : Conditional compiling for some areas that are different
4777 Frequency : occasional
4778 Bugzilla : 6409, 6834
4779 Description: Creating files with an explicit stripe count may lead to
4780 a failed assertion on the MDS
4781 Details : If some OSTs are full or unavailable, creating files may
4782 trigger a failed assertion on the MDS. Now, Lustre will
4783 try to use other servers or return an error to the
4787 Frequency : occasional
4789 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4790 caused lock timeout during readahead (since 1.4.2).
4791 Details : Processes doing readahead might match a lock that hasn't been
4792 granted yet if there are overlapping and conflicting lock
4793 requests. The readahead process waits on ungranted lock
4794 (original lock is CBPENDING), while OST waits for that process
4795 to cancel CBPENDING read lock and eventually evicts client.
4797 Severity : enhancement
4799 Description: Initial enabling of flock support for clients
4800 Details : Implements fcntl advisory locking and file status functions.
4801 This feature is provided as an optional mount flag (default
4802 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4803 locking are implemented yet, and those that are are not guaranteed
4804 to be completely correct in production environments.
4805 mount -t lustre -o [flock|noflock] ...
4808 Frequency : occasional
4810 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4811 assertion in the ext3 JBD (journaling) layer.
4812 Details : The b_committed_data struct is protected by the big kernel lock
4813 in 2.4 kernels, serializing journal_commit_transaction() and
4814 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4815 access to this struct. In 2.6 kernels there is finer grained
4816 locking to improve SMP performance of the JBD layer.
4820 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4825 Description: merge and cleanup kernel patches.
4828 -----------------------------------------------------------------------------
4830 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4832 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4833 versions of Lustre. Please update all servers and clients to
4834 version 1.4.2 or later at the same time. You must also run
4835 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4836 to update the configuration logs.
4838 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4839 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4840 - swab llog records of type '0' so we get proper header size/idx (5861)
4841 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4842 - fix rename of one directory over another leaking an inode (5953)
4843 - avoid SetPageDirty on 2.6 (5981)
4844 - don't re-add just-being-destroyed locks to the waiting list (5653)
4845 - when creating new directories, inherit the parent's custom
4846 striping settings if present parent (3048)
4847 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4848 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4849 - don't hold client locks on temporary worklist from l_lru (5666)
4850 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4851 - automatically evict dead clients (5921)
4852 - Update file size properly in create+truncate+fstat case (6196)
4853 - Do not unhash mountpoint dentries, do not allow removal of
4855 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4856 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4857 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4858 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4859 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4860 - Correctly handle reads of files with no objects (6243)
4861 - lctl recover will also mark a device active if deactivate used (5933)
4863 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4864 - allow --write-conf on an MDS with different nettype than client (5619)
4865 - don't write config llogs to MDS for mounts not from that MDS (5617)
4866 - lconf should create multiple TCP connections from a client (5201)
4867 - init scripts are now turned off by default; run chkconfig --on
4868 lustre and chkconfig --on lustrefs to use them
4869 - upcalls are no longer needed for clients to recover to failover
4871 - add --abort-recovery option to lconf to abort recovery on device
4873 - add support for an arbitrary number of OSTs (3026)
4874 - Quota support protocol changes.
4875 - forward compatibility changes to wire structs (6007)
4876 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4877 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4878 - improved "lustre status" script
4879 - initialize blocksize for non-regular files (6062)
4880 - added --disable-server and --disable-client configure options (5782)
4881 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4882 - Vanilla 2.4.29 support
4883 - increase maximum number of obd devices to 520 (6242)
4884 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4885 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4886 SLES 9 2.6.5 kernels
4887 - limit stripes per file to 160 (the maximum EA size) (6093)
4889 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4892 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4893 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4894 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4895 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4896 - small ext3 extents cleanups and fixes (5733)
4897 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4898 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4899 - avoid needless client->OST connect, fix handle mismatch (5317)
4900 - fix DLM error path that led to out-of-sync client, long delays (5779)
4901 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4902 - fix several locking issues related to i_size (5492,5624,5654,5672)
4903 - don't move pending lock onto export if it is already evicted (5683)
4904 - fix kernel oops when creating .foo in unlinked directory (5548)
4905 - fix deadlock in obdfilter statistics vs. object create (5811)
4906 - use time_{before,after} to avoid timer jiffies wrap (5882)
4907 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4908 - Do not leak request if server was not able to process it (5154)
4909 - If mds_open unable to find parent dir, make that negative lookup(5154)
4910 - don't create new directories with extent-mapping (5909, 5936)
4912 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4913 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4914 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4915 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4916 - support for RHEL4 kernel on i686 (5773)
4917 - provide error messages when incompatible logs are encountered (5898)
4919 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4920 * version 1.4.0.10 (1.4.1 release candidate 1)
4922 - don't keep a lock reference when lock is not granted (4238)
4923 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4924 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4925 - drop import inflight refcount on signal_completed_replay error (5255)
4926 - unlock page after async write error during send (3677)
4927 - handle missing objects in filter_preprw_read properly (5265)
4928 - no transno return for symlink open, don't save no-trasno open (3440)
4929 - don't try to complete elan receive that already failed (4012)
4930 - free RPC server reply state on error (5406)
4931 - clean up thread from ptlrpc_start_thread() on error (5160)
4932 - readahead could read extra page into cache that wasn't ejected (5388)
4933 - prevent races in class_attach/setup/cleanup/detach (5260)
4934 - don't dereference de->d_inode after l_dput of de (5458)
4935 - use "int" for stripe value returned from lock_to_stripe (5544)
4936 - mballoc allocation and error-checking fixes in 2.6 (5504)
4937 - block device patches to fix I/O request sizes in 2.6 (5482)
4938 - look up hostnames for IB nals (5602)
4939 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4940 - don't start multiple acceptors for the same port (5277)
4941 - fix incorrect LASSERT in mds_getattr_name (5635)
4942 - export a proc file for general "ping" checking (5628)
4943 - fix "lfs check" to not block when the MDS is down (5628)
4945 - service request history (4965)
4946 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4947 - create an "evict_client" /proc entry on OSTs, like the MDS has
4948 - fix mount usage message, return errors per mount(8) (5168)
4949 - change grep [] to grep "[]" in tests so they work in more UMLs
4950 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4951 - remove ancient LOV_MAGIC_V0 EA support (5047)
4952 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4953 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4954 - allow readahead allocations to fail when low on memory (5383)
4955 - mmap locking landed again, after considerable improvement (2828)
4956 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4958 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4961 - send OST transaction number in read/write reply to free req (4966)
4962 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4963 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4964 - the watchdog thread now runs as interruptible (5246)
4965 - flock/lockf fixes (but it's still disabled, pending 5135)
4966 - don't use EXT3 constants in llite code (5094)
4967 - memory shortage at startup could cause assertion (5176)
4969 - reorganization of lov code
4970 - single portals codebase
4972 - add extents/mballoc support (5025)
4973 - direct I/O reads in the obdfilter (4048)
4974 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4976 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4979 - send OST transaction number in read/write reply to free req (4966)
4980 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4981 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4982 - the watchdog thread now runs as interruptible (5246)
4983 - handle missing objects in filter_preprw_read properly (5265)
4984 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4985 - drop import inflight refcount on signal_completed_replay error (5255)
4986 - unlock page after async write error during send (3677)
4987 - return original error code on reconstructed replies (3761)
4988 - no transno return for symlink open, don't save no-trasno open (3440)
4990 - add pid to ldlm debugging output (4922)
4991 - bump the watchdog timeouts -- we can't handle 30sec yet
4992 - extra debugging for orphan dentry/inode bug (5259)
4994 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4997 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4998 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4999 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5000 - fix NULL dereference in /proc/sys/portals/routes (4827)
5001 - allow failed mdc_close() operations to be interrupted (4561)
5002 - stop precreate on OST before MDS would time out on it (4778)
5003 - don't send partial-page writes before EOF from client (4410)
5004 - discard client grant for sub-page writes on large-page clients (4520)
5005 - don't free dentries not owned by NFS code, check generation (4806)
5006 - fix lsm leak if mds_create_objects() fails (4801)
5007 - limit debug_daemon file size, always print CERROR messages (4789)
5008 - use transno after validating reply (3892)
5009 - process timed out requests if import state changes (3754)
5010 - update mtime on OST during writes, return in glimpse (4829)
5011 - add mkfsoptions to LDAP (4679)
5012 - use ->max_readahead method instead of zapping global ra (5039)
5013 - don't interrupt __l_wait_event() during strace
5015 - add software watchdogs to catch hung threads quickly (4941)
5016 - make lustrefs init script start after nfs is mounted
5017 - fix CWARN/ERROR duplication (4930)
5018 - return async write errors to application if possible (2248)
5019 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5020 - print NAL number in %x format (4645)
5021 - update barely-supported suse-2.4.21-171 series (4842)
5022 - support for sles 9 %post scripts
5023 - support for building 2.6 kernel-source packages
5024 - support for sles km_* packages
5026 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5029 - ignore -ENOENT errors in osc_destroy (3639)
5030 - notify osc create thread that OSC is being cleaned up (4600)
5031 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5032 - reconstruct ost_handle() like mds_handle() (4657)
5033 - create a new thread to do import eviction to avoid deadlock (3969)
5034 - let lconf resolve symlinked-to devices (4629)
5035 - don't unlink "objects" from directory with default EA (4554)
5036 - hold socknal file ref over connect in case target is down (4394)
5037 - allow more than 32000 subdirectories in a single directory (3244)
5038 - fix blocks count for O_DIRECT writes (3751)
5039 - OST returns ENOSPC from object create when no space left (4539)
5040 - don't send truncate RPC if file size isn't changing (4410)
5041 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5042 - bind to privileged port in socknal and tcpnal (3689)
5044 - rate limit CERROR/CWARN console message to avoid overload (4519)
5045 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5046 - basic mmap support (3918)
5047 - kernel patch series update from b1_4 (4711)
5049 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5052 - avoid crash during MDS cleanup with OST shut down (2775)
5053 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5054 - don't use bad inodes on the MDS (3744)
5055 - dynamic object preallocation to improve recovery speed (4236)
5056 - don't hold spinlock over lock dumping or change debug flags (4401)
5057 - don't zero obd_dev when it is force cleaned (3651)
5058 - print grants to console if they go negative (4431)
5059 - "lctl deactivate" will stop automatic recovery attempts (3406)
5060 - look for existing locks in ldlm_handle_enqueue() (3764)
5061 - don't resolve lock handle twice in recovery avoiding race (4401)
5062 - revalidate should check working dir is a directory (4134)
5064 - don't always mark "slow" obdfilter messages as errors (4418)
5066 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5069 - don't close LustreDB during write_conf until it is done (3860)
5070 - fix typo in lconf for_each_profile (3821)
5071 - allow dumping logs from multiple threads at one time (3820)
5072 - don't allow multiple threads in OSC recovery (3812)
5073 - fix debug_size parameters (3864)
5074 - fix mds_postrecov to initialize import for llog ctxt (3121)
5075 - replace config semaphore with spinlock (3306)
5076 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5077 - don't allow enqueue to complete on a destroyed export (3822)
5078 - down write_lock before checking llog header bitmap (3825)
5079 - recover from lock replay timeout (3764)
5080 - up llog sem before sending rpc (3652)
5081 - reduce ns lock hold times when setting kms (3267)
5082 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5083 - fix NULL deref and obd_dev leak on setup error (3312)
5084 - replace some LBUG about llog ops with error handling (3841)
5085 - don't match INVALID dentries from d_lookup and spin (3784)
5086 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5087 - fix invalid assertion in ptlrpc_set_wait (3880)
5089 - add libwrap support for the TCP acceptor (3996)
5090 - add /proc/sys/portals/routes for non-root route listing (3994)
5091 - allow setting MDS UUID in .xml (2580)
5092 - print the stack of a process that LBUGs (4228)
5094 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5097 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5098 - make sure to unset replay flag from failed open requests (3440)
5099 - if default stripe count is 0, use OST count for inode size (3636)
5100 - update parent mtime/ctime on client for create/unlink (2611)
5101 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5102 - free recovery state on server during a forced cleanup (3571)
5103 - unregister_reply for resent reqs (3063)
5104 - loop back devices mounting and status check on 2.6 (3563)
5105 - fix resource-creation race that can provoke i_size == 0 (3513)
5106 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5107 - more debugging for page-accounting assertion (3746)
5108 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5109 - avoid deadlock after precreation failure (3758)
5110 - fix race and lock order deadlock in orphan handling (3450, 3750)
5111 - add validity checks when grabbing inodes from l_ast_data (3599)
5113 - add /proc/.../recovery_status to obdfilter (3428)
5114 - lightweight CDEBUG infrastructure, debug daemon (3668)
5115 - change default OSC RPC parameters to be better on small clusters
5116 - turn off OST read cache for files smaller than 32MB
5117 - install man pages and include them in rpms (3100)
5118 - add new init script for (un)mounting lustre filesystems (2593)
5119 - run chkconfig in %post for init scripts (3701)
5120 - drop scimac NAL (unmaintained)
5122 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5125 - clean kiobufs before and after use (3485)
5126 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5127 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5128 - change init script to fail more clearly if not run as root (1528)
5129 - allow clients to reconnect during replay (1742)
5130 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5131 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5132 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5133 - fix cond_resched (3554)
5134 - only evict unfinished clients after recovery (3515)
5135 - allow bulk resend, prevent data loss (3570)
5136 - dynamic ptlrpc request buffer allocation (2102)
5137 - don't allow unlinking open directory if it isn't empty (2904)
5138 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5139 - remove extraneous obd dereference causing LASSERT failure (3334)
5140 - don't use get_cycles() when creating temp. files on the mds (3156)
5141 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5142 - handle EEXIST for set-stripe, set proper directory name (3336)
5144 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5145 - fix ksocknal_fmb_callback() error messages (2918)
5147 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5150 - don't copy lvb into (possibly NULL) reply on error (2983)
5151 - don't deref dentry after dput, don't free lvb on error (2922)
5152 - use the kms to determine writeback rpc length (2947)
5153 - increment oti_logcookies when osc is inactive (2948)
5154 - update client's i_blocks count via lvb messages (2543)
5155 - handle intent open/close of special files properly (1557)
5156 - mount MDS with errors=remount-ro, like obdfilter (2009)
5157 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5158 - don't use cancelling-locks' kms values (2947)
5159 - use highest lock extent for kms, not last one (2925)
5160 - don't dereference ERR_PTR() dentry in error handling path (3107)
5161 - fix thread race in portals_debug_dumplog() (3122)
5162 - create lprocfs device entries at setup instead of at attach (1519)
5163 - common AST error handler, don't evict client on completion race (3145)
5164 - zero nameidata in detach_mnt in 2.6 (3118)
5165 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5166 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5167 - initialize RPC timeout timer earlier for 2.6 (3219)
5168 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5169 - print nal/nid for unknown nid (3258)
5170 - additional checks for oscc recovery before doing precreate (3284)
5171 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5172 - don't crash in mdc_close for bad permissions on open (3285)
5173 - zero i_rdev for non-device files (3147)
5174 - clear page->private before handing to FS, better assertion (3119)
5175 - tune the read pipeline (3236)
5176 - fix incorrect decref of invalidated dentry (2350)
5177 - provide read-ahead stats and refine rpc in flight stats (3328)
5178 - don't hold journal transaction open across create RPC (3313)
5179 - update atime on MDS at close time (3265)
5180 - close LDAP connection when recovering to avoid server load (3315)
5181 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5182 - don't leak open file on MDS after open resend (3325)
5183 - serialize filter_precreate and filter_destroy_precreated (3329)
5184 - loop device shouldn't call sync_dev() for nul device (3092)
5185 - clear page cache after eviction (2766)
5186 - resynchronize MDS->OST in background (2824)
5187 - refuse to mount the same filesystem twice on same mountpoint (3394)
5188 - allow llmount to create routes for mounting behind routers (3320)
5189 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5190 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5191 - fix rare problem with rename on htree directories (3417)
5193 - allow default OST striping configuration per directory (1414)
5194 - fix compilation for qswnal for 2.6 kernels (3125)
5195 - increase maximum number of MDS request buffers for large systems
5196 - change liblustreapi to be useful for external progs like lfsck (3098)
5197 - increase local configuration timeout for slow disks (3353)
5198 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5200 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5203 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5204 - don't overwrite extent policy data in reply if lock was blocked (2901)
5205 - drop filter export grants atomically with removal from device (2663)
5206 - del obd_self_export from work_list in class_disconnect_exports (2908)
5207 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5208 - swab reply message in mdc_close, other PPC fixes (2464)
5209 - fix destroying of named logs (2325)
5210 - overwrite old logs when running lconf --write_conf (2264)
5211 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5212 - fix race in target_handle_connect (2898)
5213 - mds_reint_create() should take same inode create lock (2926)
5214 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5215 - don't close files for self_export to avoid uninitialized obd (2936)
5216 - allow MDS with the same name as client node (2939)
5217 - hold dentry reference for closed log files for unlink (2325)
5218 - reserve space for all logs during transactions (2059)
5219 - don't evict page beyond end of stripe extent (2925)
5220 - don't oops on a deleted current working directory (2399)
5221 - handle hard links to targets without a parent properly (2517)
5222 - don't dereference NULL lock when racing during eviction (2867)
5223 - don't grow lock extents when lots of conflicting locks (2919)
5225 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5228 - account for cache space usage on clients to avoid data loss (974)
5229 - lfsck support in lustre kernel code (2349)
5230 - reduce journal credits needed for BRW writes (2370)
5231 - orphan handling to avoid losing space on client/server crashes
5232 - ptlrpcd can be blocked, stopping ALL progress (2477)
5233 - use lock value blocks to assist in proper KMS, faster stat (1021)
5234 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5235 - recovery for initial connections (2355)
5236 - fixes for mds_cleanup_orphans (1934)
5237 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5238 - block all file creations until orphan recovery completes (1901)
5239 - client remove rq_connection from request struct (2423)
5240 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5241 - recovery timer race (2670)
5242 - mdc_close recovey bug (2532)
5243 - ptlrpc cleanup bug (2710)
5244 - mds timeout on local locks (2588)
5245 - namespace lock held during RPCs (2431)
5246 - handle interrupted sync write properly (2503)
5247 - don't try to handle a message that hasn't been replied to (2699)
5248 - client assert failure during cleanup after abort recovery (2701)
5249 - leak mdc device after failed mount (2712)
5250 - ptlrpc_check_set allows timedout requests to complete (2714)
5251 - wait for inflight reqs when ptlrpcd finishes (2710)
5252 - make sure unregistered services are removed from the srv_list
5253 - reset bulk XID's when resending them (caught by 1138 test)
5254 - unregister_bulk after timeout
5255 - fix lconf error (2694)
5256 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5257 - readahead locks pages, leaves pending causing memory pressure (2673)
5258 - increase OST request buffers to 4096 on large machines (2729)
5259 - fix up permission of existing directories in simple_mkdir (2661)
5260 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5261 - don't assign transno to errored transactions (2742)
5262 - don't delete objects on OST if given a bogus objid from MDS (2751)
5263 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5264 - if rq_no_resend, then timeout request after recovery (2432)
5265 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5266 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5267 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5268 - MDS recovery shouldn't delete objects that it has given out (2730)
5269 - if enqueue arrives after completion, don't clobber LVB (2819)
5270 - don't unlock pages twice when trigger_group_io returns error (2814)
5271 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5272 - don't write pages to disk if there was an error (1450)
5273 - don't ping imports that have recovery disabled (2676)
5274 - take buffered bytes into account when balancing socknal conn (2817)
5275 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5276 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5277 - remove little-endian swabbing of llog records (1987)
5278 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5279 - retry reposting request buffers if they fail (1191)
5280 - grow extent at grant time to avoid granting a revoked lock (2809)
5281 - lock revoke doesn't evict page if covered by a second lock (2765)
5282 - disable VM readahead to avoid reading outside lock extents (2805)
5284 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5285 - updated kernel patches for hp-2.4.20 kernel (2681)
5287 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5290 - fix truncated write corruption (2366)
5291 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5293 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5294 - ensure len(uuid) < 37 in lmc (1171)
5295 - fix ia64 OOPS in llog_test (2255)
5296 - zero end of page at obdfilter for partial page writes (2648)
5297 - don't leave stale dentries around after renames (bug 2428)
5298 - fix timeouts when evicting a client with a single lock held (2642)
5299 - set deadline for the initial HELLO message to drain (2634)
5300 - print out dotted-quad IP addresses in the socknal (2302)
5302 - additional debugging for MDS client eviction problem (2443)
5303 - fix mkfsoptions support for osts (2603, 2604)
5305 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5308 - add series for the vanilla 2.6.0 kernel
5309 - add series for the vanilla 2.4.24 kernel
5310 - add series for a cray x86/64 UL kernel drop
5311 - fix xattr patches for the vanilla 2.4.19 series
5313 - generate true UUIDs in lmc (1171)
5314 - have portals stack dumping break in UML (2466)
5315 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5316 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5317 - wake callback waiting threads on client eviction (2460)
5318 - Add --ptldebug and --subsystem to lmc (1719)
5319 - update assertion to allow safe interrupt allocation
5320 - set rq_no_resend for cancel requests (2432)
5321 - recalculate ptlrpcd timeout after resend (2494)
5322 - call vfs_rmdir when removing pending directories (2368)
5323 - fix renaming a file to itself (2429)
5324 - lmc creates a default one-stripe lov (2454)
5325 - expand procfs space to handle large clusters (2326)
5326 - increase UML stack to avoid overflow
5327 - update lconf's list of debug and subsystem masks
5328 - fix lfs find --obd (2510)
5329 - /proc tunable for disabling filter read caching (2591)
5330 - stop rpm packages from altering slapd.conf (2301)
5331 - disable nagle in the socknal under 0conf (2578)
5332 - choose mds inode size based on stripe count (2572)
5333 - fix kernel-source rpm problems (2516)
5335 - add --disable-doc to avoid pdf generation (2421)
5336 - update documentation, tests, type-os, comments
5337 - avoid format warnings on ia64
5338 - remove the TOE NAL
5339 - tiny code cleanups by removing unused fields
5341 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5344 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5345 - remove the most blatant lies from BUILDING file (2371)
5346 - change default debug level to reasonable production setting
5347 - reduce client side cache size to reduce cache flush time
5348 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5349 - make TCP zerocopy and pinger support enabled by default (2476)
5350 - sync writes completed after process exits caused crashes (2319)
5351 - maintain correct mount count on the MDS (2356)
5352 - backout 1557, because 2316 wasn't really fixed
5353 - better file I/O statistics gathering in /proc
5354 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5355 - another kernel patch to fix zero-copy TCP function export
5356 - don't take duplicate lock when processing re-sent getattr (2420)
5357 - lctl uses obd_self_export instead of creating new conn (2353)
5358 - MDS/OST recovery case which requires object creation asserted (2425)
5359 - move lfs from /usr/sbin to /usr/bin in packages
5360 - fix race between mds_client_add and mds_client_free (2417)
5361 - use kmalloc instead of slabs in portals (2430)
5362 - don't create duplicate records when a failover MDS is present (2442)
5363 - remove unnecessary mount age check (2332)
5364 - don't remove directory inodes from locks prematurely (2451)
5365 - don't break if MDS service name is the same as hostname (2103)
5366 - fix races in client write RPC generation when cache full (2482)
5368 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5371 - remove now-unused request->rq_obd (278)
5372 - if an allocation fails, print out how much memory we've used (1933)
5373 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5374 - add the "configurable stack size" patch to most series files (1256)
5375 - ability to write large log records, for 100+ OST configs (2306)
5376 - fix NULL deref when filter_prep fails (2314)
5377 - fix operator precedence error in filter_sync
5378 - dynamic allocation of socknal TX descriptors (2315)
5379 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5380 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5381 - ensure that CWARN messages in Portals always get to the syslog
5382 - __init/__exit are not for prototype decls (ldlm_init/exit)
5383 - x86-64 compile warning fixes
5384 - fix gateway LMC keyword conflict (2318)
5385 - fix MDS lock inversions in getattr/reint paths (1844)
5386 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5387 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5388 - DLM race condition prevented some lock evictions (2328)
5389 - ENOMEM detection and retry on socknal sends (2230)
5390 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5391 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5392 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5393 - better error messages when a client is rejected during recovery (1505)
5394 - avoid cancelling locks which were never granted, after failure (2330)
5395 - fix i_sem/journal inversion in mds_client_add (2333)
5396 - fix truncate/getattr lock cycle deadlock (2334)
5397 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5398 - fix two rare exit paths which could leak an l_lock() ref (2321)
5399 - fencepost error in MDS/OST orphan recovery (2226)
5400 - make log record alignment 8 bytes (1988)
5401 - lstripe now fails when requested offset > ost_count (2237)
5402 - ensure that all kernel series have a complete list.h (1607)
5403 - fix crashes in special-file operations (2316)
5404 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5406 - allow configurable automake binary, for testing new versions
5407 - small update to the lfs documentation
5409 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5411 * fix negative export reference count in fsfilt_sync (2312)
5413 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5414 * release candidate 0.9.1
5416 - orphans are moved into the PENDING directory for possible recovery
5417 - replayed opens now open by fid for orphan/rename safety (1042)
5418 - last close of an orphan inode generates a transno (683)
5419 - chdir() and mount() now pin the directory entry (1020)
5420 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5421 - discard very old requests without processing them (1502)
5422 - remove some common, well-understood CERRORs (1505)
5423 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5424 - clear "grant" flags in OST replies until OST grant code lands (1644)
5425 - fix read performance by not clobbering i_blksize on client (1598)
5426 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5427 - make LRU size a /proc tunable, clears locks when reduced (707)
5428 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5429 - server locks take references on exports now (1558)
5430 - build fixes for 2.4.20-rh trees (1663)
5431 - return an error from lov_create if all OSCs are inactive (1751)
5432 - fix import levels when a reconnect happens without a timeout (1597)
5433 - exit early from mds_open if we get a lookup error (1749)
5434 - partial page read at EOF wouldn't wait for disk before sending (1642)
5435 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5436 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5437 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5438 - write/truncate lock inversion (1639)
5439 - Don't auto-load obdclass, portals modules during cleanup (1495)
5440 - fix timestamps from jumping to "now" (1763)
5441 - extra journal assertions (1648)
5442 - add an extra multiunlink test (1771)
5443 - fix read_record/write_record API (1776)
5444 - fix leak of offset_extent, possible incorrect i_size later (1772)
5445 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5446 - Debugging for the kqswnal_get_idle_tx problems (1820)
5447 - Allow recovery to be attempted multiple times (1536)
5448 - Write out MDS last_rcvd file after it is first created (1600)
5449 - Fix tx_descriptor leak in failed transmit situations (1827)
5450 - ext3 journaling fixes for assertion failure after IO error (1871)
5451 - class_export_put() on freed export after completion AST error (1896)
5452 - Fix revalidate looping in VFS (1322)
5453 - Don't access a freed export during MDS_REINT timeout (1521)
5454 - Add open-unlink recovery support on the MDS (1673,1764)
5455 - Return an error if no MDS data was read from last_rcvd (1946)
5456 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5457 - Don't setup a disk device that doesn't match exported UUID (317)
5458 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5459 - avoid committing NULL handle in force close
5460 - local.sh is now a one-stripe LOV configuration
5461 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5462 - fix ext3 htree duplicate directory entry corruption (1516)
5463 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5464 - update to new LOV EA format (2097)
5465 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5467 2003-06-15 Phil Schwan <phil@clusterfs.com>
5470 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5471 - per-import/export recovery handling (958, 931, 959)
5472 - multiple last-rcvd slots, for serving multiple FSes (949)
5473 - connections are again shared between multiple imp/exports (963, 964)
5474 - "umount -f" would hang if any requests needed to be sent (393, 978)
5475 - avoid pinning large req buffer by copying for queued messages (989)
5476 - add "uuid" to "lctl device" command to help upcalls (991)
5477 - "open" RPCs with transnos would confuse recovery counters (1037)
5478 - do proper endian conversion of all wire messages (288, 340, 891)
5479 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5480 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5481 - fix ptlrpc_connection leak in target_handle_connect (1174)
5482 - fix import refcounting bug in OST and MDS cleanup (1134)
5483 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5484 - fix very unlikely obd_types race condition (501)
5485 - remove osc_open hack for echo_client (1187)
5486 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5487 - a failure in read_inode2 leads to deadlock (1139)
5488 - cancel ack-locks as soon as transaction is committed (1072)
5489 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5490 - make sure to commitrw after any preprw to avoid deadlock (1162)
5491 - failing to execute a file in a lustre FS would lock inode (1203)
5492 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5493 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5494 - obd_uuid2tgt would walk off the end of the list (1255)
5495 - on IA64 the osc would give portals incorrect bulk size (1258)
5496 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5497 - fix lock inversion caused by new llite matching code (1282)
5498 - limit the number of dirty pages on a client to 10MB (1286)
5499 - timed out locks were not being corrected cancelled (1289)
5500 - fix O_DIRECT above 4GB on IA-32 (1292)
5501 * major user-visible changes
5502 - fail out/fail over policy now controlled by the upcall (993)
5504 - add OBD_PING to check server availability and failure (954)
5505 - lustre messages are now sent in sending host order (288, 340, 891)
5506 - add eadatalen to MDS getattr reply (340)
5507 - OST read replies may contain second buffer, with per-page status (593)
5509 2003-03-11 Phil Schwan <phil@clusterfs.com>
5512 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5513 - failed open()s could cause deadlock; fixed (867, 869)
5514 - stop cancelling OST locks when files are closed (481)
5515 - overlapping XID spaces caused network corruption (851, 853)
5516 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5517 - setattr_raw added, to do single-RPC, server-side setattrs
5518 - lmc/lconf syntax change for OST UUIDs
5519 - fix crashy race condition between ptlrpc_free_req and osc_close
5520 - don't use request in mdc_enqueue if we hit a timeout (889)
5521 - don't set the inode i_size for regular files from the MDS (896)
5522 - handle out of order completion AST (842)
5523 - don't LBUG if a lock request times out after receiving AST (913)
5524 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5525 - if a bad lock AST arrives, send an error instead of dropping entirely
5526 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5527 - fix leak in bulk IO when only partially completed (899, 900, 926)
5528 - fix O_DIRECT for ia64 (55)
5529 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5530 - C-z after timeout could hang a process forever; fixed (977)
5532 - client-side I/O cache (678, 924, 929, 941, 970)
5534 - READPAGE and SETATTRs which don't take server-side locks get
5537 2003-02-11 Phil Schwan <phil@clusterfs.com>
5540 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5541 - Fix fsfilt_extN_readpage() to read a full page of directory
5542 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5543 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5544 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5545 - Most intent operations were removed, in favour of a new RPC mode
5546 that does a single RPC to the server and bypasses most of the VFS
5547 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5548 - Aggressively cancel local locks on DLM servers
5549 - mds_reint_unlink sends EA to the client if it's the last nlink.
5550 client uses that EA to unlink OST objects.
5551 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5552 - recursive symlinks were fixed (439)
5553 - fixed NULL deref in DEBUG_REQ
5554 - filter_update_lastobjid no longer calls sync, which annoyed extN
5555 - fixed multi-client small-writes to a single file problem (445)
5556 - fixed mtime updates during file writes (607)
5557 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5558 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5559 - fixed Linux OST size reporting problem (444, 656)
5560 - OST now updates object mtime with writes or setattr (607, 619)
5561 - client verifies file size before zeroing page past EOF (445)
5562 - OST now writes last allocated objid to disk with allocation (108)
5563 - LOV on echo now works (409)
5565 - mds_reint_unlink sends a new buffer, with the EA included. this
5566 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5567 set if a regular file was being unlinked, and it was the last link
5568 - use PtlGet from the target for bulk writes (315)
5569 - OST now updates object mtime with writes or setattr (607, 619)
5570 - LDLM now has a grant-time callback to revalidate locked items, if
5572 - Many MDS operations were reorganized to combat race conditions
5574 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5575 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5577 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5580 - Fully reactivate OST imports after reconnection (512, others)
5581 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5582 - More graceful error handling for truncating on dead OST (515)
5583 - Don't error out unless we're actually accessing dead stripes (474)
5584 - Fix garbage sizes when stripes are missing (410)
5585 - LRU counters were broken, causing constant lock purge (433, 432)
5586 - garbage on read from stripes with failed OSTs (441)
5587 - mark OSCs as active before reconnecting during recovery (438)
5588 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5589 - lfind did not preserve OST order in output (443)
5590 - symlinks cause hung clients, incorrect data (439)
5591 - stop dereferencing request after dropping refcount (457)
5592 - don't LASSERT(spin_is_locked) on non-SMP (455)
5593 - fixes for many rename() bugs
5594 - fstat didn't correctly synchronize attributes (399)
5595 - server must handle lock cancellation during blocking AST prep (487)
5596 - bulk descriptors were free()d too soon (511)
5597 - fix paths in lconf, which would load incorrect modules (451, 507)
5598 - fix confusing lconf 'host not found' error message (386)
5599 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5600 - fix race condition in mdc_blocking_ast() for inode access (526)
5601 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5602 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5603 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5604 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5605 - fix lstripe in several ways (kernel side) (527)
5606 - fix request leak in ldlm_cli_enqueue (262)
5607 - incorrect OSC was marked inactive after OST failure
5608 - call mds_fs_cleanup before unmounting filesystem (524)
5609 - fix races between taking ns_lock and ldlm_lock_change_resource
5610 - fix races updating LOV export open file list
5611 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5612 - fix recovery NULL deref in ldlm_cli_cancel_unused
5613 - fix some DLM races by using new hash table for lock handles (419)
5614 - permit the client to specify desired inodes, at replay
5615 - duplicate requests when we queue them for replay reintegration
5616 - fix last_rcvd offset calculation
5617 - sync after each recovered transaction, so we always make progress
5618 - never, not always, ERESTART requests without transnos
5619 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5621 - skip replay if the MDS says that the client is already connected
5622 - don't check for a recovery-enabled export to match lctl's UUID
5623 - don't INC_USE_COUNT for phantom exports
5624 - don't crash when cleaning up phantom exports (567)
5625 - don't double-finish or set replay data for errored mdc_open requests
5626 - abort requests when they time out, so we don't get old replies
5627 - send/receive replies for AST messages again
5628 - if the client says that it doesn't have the lock, cancel it on the
5630 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5631 mark it as destroyed, it will all work out when decref is called
5632 - fix module use counts (22, 581)
5634 - ASTs now expect a reply (server cancels lock on error reply)
5636 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5639 - fix many simultaneous client startup (392)
5640 - fix dentry->d_it clobbering
5641 - credentials weren't being shipped for readdir/getattr operations
5642 - remove invalid assertions triggered during some concurrent MD
5644 - proper Lustre versions added (336, 389)
5645 - fix memory leak for create error case (398)
5646 - fix LOV locking bug that would get cli/srv out of sync
5647 - fix echo client over LOV (409)
5648 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5649 - fix double-O_EXCL intent crash (424)
5650 - avoid sending multiple lock CANCELs (352)
5652 - MDS can do multi-client recovery (modulo bugs in new code)
5654 - many updates, edits, cleanups
5656 2002-11-18 Phil Schwan <phil@clusterfs.com>
5659 - fix null d_it dereference (346)
5660 - fix full OST/dbench hang (333)
5661 - fix permission problem with file removal (286)
5662 - fix removal of OSCs from LOV when they fail
5663 - fix NULL deref during bulk timeout (214)
5664 - fix problems related to multiple filesystems on one MDS (241)
5665 - fixed serious subtle metadata locking bugs
5666 - free locks on clients when inodes are removed due to memory
5668 - fix inode pointer in lock data (285)
5669 - partial support for multiple MDS on a single host (241)
5670 - data locks weren't cancelled at clear_inode time (290, 311)
5671 - intent locks could lead to unbounded lock growth (205)
5672 - added a maximum lock count, an LRU list, and a flusher
5673 - fix multiple rename (365)
5674 - properly abstracted the echo client
5675 - OSC locked 1 byte too many; fixed
5676 - rewrote brw callback code:
5677 - fixed recovery bugs related to LOVs (306)
5678 - fixed too-many-pages-in-one-write crash (191)
5679 - fixed (again) crash in sync_io_timeout (214)
5680 - probably fixed callback-related race (385)
5682 - Add capability to MDS protocol
5683 - LDLM cancellations and callbacks on different portals
5685 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5688 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5689 - defer open object destruction to close time (601981, 138)
5690 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5691 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5692 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5693 - don't call obd_setattr in ll_file_release for destroyed objects
5694 * protocol change to lustre_msg: move |version| and add |flags|
5695 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5696 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5697 * added replay of create, unlink, link and rename operations during
5698 MDS failover; recovery should be much more robust now
5699 * remove failed OSCs from LOVs (only lov_create uses this so far)
5700 * the lustre-HOWTO was brought (more) up to date (582544)
5702 2002-10-23 Phil Schwan <phil@clusterfs.com>
5705 - in-use dentries weren't being reused properly (617851)
5706 - prevent multiple LDLM setup (599178)
5707 - fix LOV size calculations for truncate (617853)
5708 - fix client handling of MDS intent errors (POSIX)
5709 - fix permission bug in lovstripe.c test (624321)
5710 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5711 - truncate past end of file could corrupt data
5712 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5713 - a race in recovery could return ETIMEDOUT to apps (623947)
5714 - building outside the source directory was fixed
5715 * the lustre-HOWTO was brought (more) up to date (582544)
5716 * major progress was made on recovery functionality
5718 2002-10-10 Phil Schwan <phil@clusterfs.com>
5721 - recovery deadlock fix
5722 - rm -rf causes LBUG fix (617817)
5723 - file open by multiple tasks fix (618962)
5724 - directory permissions bugs (602707 and 620007)
5725 - journal_stop fixed with locking (611313)
5726 - O_APPEND failures resolved (618273, perhaps 614459)
5727 - lconf PATH fix (619770)
5728 - IA64 build fix (621450)
5729 - RPC buffer sizes scale with amount of memory
5731 2002-10-01 Phil Schwan <phil@clusterfs.com>
5734 - locks would be cancelled without throwing away data pages,
5735 resulting in inconsistent data (605627)
5736 - inode attributes were not always being refreshed (605627, 612449)
5737 - lconf now continues to cleanup after lctl reports an error
5738 - MDS now enforces user permissions (602707)
5739 - lprocfs cleanup fixed, but not yet enabled (614157)
5740 - fixed infinite server hang, should a client not respond to an AST
5741 - avoid going into recovery if user calls readlink() with a buffer
5742 that's too small (613941)
5743 - AST RPCs no longer require replies (614867) -- this may be changed
5744 - don't crash server if client sends an IOV that's too big (611336)
5745 - fixed lock conversion deadlock (611892)
5746 - fixed the following of symlinks (614622)
5747 * recovery: the server can remove locks from a client that dies, other
5748 clients can make progress
5749 * more extN patch fixes
5750 * compile-time configurable ptlrpc buffer allocations
5752 - collaborative read cache document
5753 - Lustre Lite Performance CDR document-in-progress
5755 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5758 - fix typo in patch-2.4.18
5760 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5763 - clear ptlrpc request each time in handle_incoming_request()
5764 - unlink of files now destroys the object on the OST
5766 2002-09-19 Peter Braam <braam@clusterfs.com>
5768 * add hard link support
5769 * change obdfile creation method
5770 * kernel patch changed
5772 2002-09-19 Peter Braam <braam@clusterfs.com>
5775 - stack overflow bug in extN fixed
5777 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5779 * documentation updates
5780 - add man pages for config tools
5781 - update tests/README to describe testing with new config tools
5782 - finish metadata API descriptions
5783 * bug fixes and cleanups
5784 - statfs workaround for 16TB limit
5785 - LOV stripe allocation improved, can stripe on subset of OSTs
5786 - LOV file size/IO offset was wrong for files > 4GB in size
5787 - object EA data was being dropped, caused files to be unreadable
5788 - memory overflow with non-LOV OST caused memory corruption
5789 - fixed regression tests to work with new config tools, obdfilter
5790 - fixed bug when directory size became larger than 1 block
5791 - fixed bug (for single client case) when PWD was deleted
5792 - invalidate local directory pages when doing intent-based ops
5793 - avoid LDLM oops when lock callback contained bad data
5795 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5797 * documentation updates
5798 * bug fixes and cleanups
5799 - configuration tools
5802 - 64-bit compile warnings
5803 - 64-bit internal statfs data
5805 * test_brw on persistent OST devices
5807 * lprocfs (disabled)
5809 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5811 * documentation updates
5812 * bug fixes and cleanups
5813 * configuration tools
5815 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5818 * many small fixes to 0_5_4
5819 * io/network handling
5820 * thinkos in MDS operations
5822 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5825 * crucial basic fixes to 0.5.3
5826 * IOR, Iozone work over Elan
5829 2002-08-07 Phil Schwan <phil@clusterfs.com>
5830 * version 0_5_3, our first alpha
5831 * we use the new Portals iovs
5832 * documentation updates
5833 * bug fixes and cleanups
5834 * small changes in the DLM wire protocol
5836 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5837 * version 0_5_1 with some initial stability,
5838 * locking on MD and file I/O.
5839 * documentation updates
5840 * several bug fixes since 0.5.0
5841 * small changes in wire protocol
5843 2002-07-18 Phil Schwan <phil@clusterfs.com>
5845 * delivered as Lustre Light Alpha
5846 * fixed a crash after handling invalid MDS requests
5847 * fixed directory pages for architectures with non-4k pages sizes
5849 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5850 * release version v0_4_4
5851 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5853 * Fixes a number of interruption problems with OST operations.
5854 * Update documentation for portals header changes
5855 * Move all wire protocol structs/defines to lustre_idl.h
5856 * Fixes symlink length bug.
5857 * Add tcpdump to repository.
5859 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5860 * release version v0_4_3
5861 * Fixes statfs for inodes on extN.
5862 * Fixes bug in runtests which would delete /etc/hosts.
5863 * Use 64-bit object IDs wherever possible (not into VFS though)
5864 Remove ost_get_info, which is unused by lustre, and out of date.
5866 2002-07-03 Peter Braam <braam@clusterfs.com>
5867 * release version v0_4_2 Fixes a lookup error (type not passed)
5868 * move forward to head of Portals
5869 * move forward to latest Lustre kernel
5871 2002-06-25 Peter Braam <braam@clusterfs.com>
5872 * release version v0_4_1. Hopefully stable on single node use.