1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.39.3 (SLES 10),
5 2.6.18-128.1.14.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.
18 Description: Do not start bulk transfer, if client tried to reconnect.
19 Details : Do not start bulk transfer, if client tried to reconnect,
20 but failed due to active rpcs. Check for exp_abort_active_req
25 Description: Prevent inconsistences between linux and lustre mount structures.
26 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
27 Make the sleep interruptible.
29 Severity : enhancement
31 Description: provide server to client comms path
32 Details : server to client communications path via new LDLM_SET_INFO rpc
36 Description: truncate starts GFP_FS allocation under transaction causing deadlock
37 Details : ldiskfs_truncate calls grab_cache_page which may start page
38 allocation under an open transaction. This may lead to
39 calling prune_icache with consequent lustre reentrance.
43 Description: send correctly lsm on open replay
44 Details : MDS is trust to LSM size on replay open, but client can set wrong size
49 Description: enable client interface failover
50 Details : When a child reconnects from another NID, properly update export
51 nid hash position and ldlm reverse import.
53 Severity : enhancement
55 Description: Communicate OST degraded/readonly state via statfs to MDS
56 Details : Flags in the statfs returned from OSTs indicate whether the
57 OST is in a degraded RAID state, or if the filesystem has
58 turned read-only after a filesystem error is detected.
63 Description: don't panic if EPROTO was hit when reading symlink
64 Details : correctly handling request reference in error cases.
66 Severity : enhancement
68 Description: Add LustreNetLink, a kernel-userspace communcation path.
70 Severity : enhancement
72 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
77 Description: access to llog context before init.
78 Details : move handling CATALOGS file at osc layer and forbid access to llog
82 Frequency : rare, only when client is writing to OST on local node
84 Description: Avoid deadlock for local client writes
85 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
86 memory freeing context. This allows OST threads to set the
87 PF_MEMALLOC flag on task structures in order to allocate memory
88 from reserved pools and complete IO.
89 Use GFP_HIGHUSER for OST allocations for non-local client writes,
90 so that the OST threads generate memory pressure and allow
91 inactive pages to be reclaimed.
93 Severity : enhancement
95 Description: Update kernel to RHEL5.3 2.6.18-128.1.14.el5.
100 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
101 Details : move ldlm namespace creation in setup phase to avoid grab
102 _lprocfs_lock with cli_sem held.
106 Description: Temporarily disable grant shrink.
107 Details : Disable the feature for debugging.
110 Frequency : only when formatting test filesystems
112 Description: Unable to run several mkfs.lustre on loop devices at the same time
113 Details : mkfs.lustre returns error 256 on the concurrent loop devices
114 formatting. The solution is to proper handle the error.
116 Severity : enhancement
118 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
120 Severity : enhancement
122 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
127 Description: resolve race between obd_disconnect and class_disconnect_exports
128 Details : if obd_disconnect will be called to already disconnected export he
129 forget release one reference and osc module can't unloaded.
131 Severity : enhancement
133 Description: Allow tuning service thread via /proc
134 Details : For each service a new
135 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
136 created that can be used to set min/max thread counts, and get the
137 current number of running threads.
142 Descriptoin: don't return error if have partially created objects for file.
143 Details : lov_update_create_set uses set->set_success as index for created
144 objects, so if some requests will be failed, they will have hole
145 at end of array and qos_shrink_lsm for allocate correct lsm.
147 Severity : enhancement
149 Description: Update OFED support to 1.4.1
154 Description: Reduce small size read RPC
155 Details : Set read-ahead limit for every file and only do read-ahead when
156 available read-ahead pages are bigger than 1M to avoid small size
159 Severity : enhancement
161 Description: move AT tunable parameters for more consistent usage
162 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
164 Severity : enhancement
166 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
167 Details : allow skip disconnected ost for send statfs request and hide error
173 Description: fix racy locking of mballoc block bitmaps causing BUG
174 Details : The locking of the mballoc buddy bitmap and the in-memory
175 block bitmap was using two different spin locks in some
176 cases. This made it possible to incorrectly access the
177 mballoc bitmap while another process was modifying it,
178 causing a sanity assertion to fail. While no on-disk corruption
179 was reported, there was some risk of this happening.
182 Frequency : rare, on llog test 6
184 Descriptoin: don't allow connect to already connected import
185 Details : allowing connect to already connected import is hide connecting problem.
190 Descriptoin: allow kill process which wait statahead result
191 Details : for some reasons 'ls' can stick in waiting result from statahead,
192 in this case need way for kill this process.
194 Severity : enhancement
196 Description: Add state history info file, enhance import info file
197 Details : Track import connection state changes in a new osc/mdc proc file;
198 add overview-type data to the osc/mdc import proc file.
200 Severity : enhancement
202 Description: MDS create should not wait for statfs RPC while holding DLM lock.
204 Severity : enhancement
206 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
211 Descriptoin: OST grant too much space to client even there are not enough space.
212 Details : Client will shrink its grant cache to OST if there are no write
213 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
214 this grant cache if there are already not enough avaible space
215 (left_space < total_clients * 32M).
218 Frequency : start MDS on uncleanly shutdowned MDS device
220 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
221 Details : stay in waiting mds<>ost recovery finished produce random bugs
222 due race between two ll_sync thread for one lov target. send
223 ACTIVATE event only if connect realy finished and import have
227 Frequency : rare, connect and disconnect target at same time
229 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
230 Details : don't call obd_disconnect under lov_lock. this long time
231 operation and can block ptlrpcd which answer to connect request.
236 Descriptoin: don't lose wakeup for imp_recovery_waitq
237 Details : recover_import_no_retry or invalidate_import and import_close can
238 both sleep on imp_recovery_waitq, but we was send only one wakeup
242 Frequency : always with long access acl
244 Descriptoin: mds can't pack reply with long acl.
245 Details : mds don't control size of acl but they limited by reint/getattr
248 Severity : enhancement
250 Description: Update to SLES10 kernel-2.6.16.60-0.33.
252 Severity : enhancement
254 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
257 Frequency : start MDS on uncleanly shutdowned MDS device
259 Descriptoin: aborting recovery hang on MDS
260 Details : don't throttle destroy RPCs for the MDT.
263 Frequency : on remount
265 Description: external journal device not working after the remount
266 Details : clear dev_rdonly flag for external journal devices in
272 Description: shutdown vs evict race
273 Details : client_disconnect_export vs connect request race.
274 if client will evicted at this time - we start invalidate
275 thread without referece to import and import can be freed
279 Frequency : rare, need acl's on inode.
281 Description: client can't handle ost additional correctly
282 Details : if ost was added after client connected to mds client can have
283 hit lnet_try_match_md ... to big messages to wide striped files.
284 in this case need teach client to handle config events about add
285 lov target and update client max ea size at that event.
287 Severity : enhancement
289 Description: Changelogs
290 Details : Changelogs are a lightweight mechanism to track filesystem
291 metadata and namespace changes. The changelog is recorded
292 permanently on the MDTs, and is periodically "consumed" / purged
293 when records are no longer needed.
295 Severity : enhancement
297 Description: compact fld format with extents
298 Details : Store range of seq rather than every seq in FLD. Seq
299 controller update FLD rather than clients. In Case of CMD, mdt0
300 has FLD, all other metadata server act as non persistent proxy
301 for FLD queries and cache fld entries in fld cache.
306 Description: don't skip ost target if they assigned to file
307 Details : Drop slow OSCs if we can, but not for requested start idx.
308 This means "if OSC is slow and it is not the requested
309 start OST, then it can be skipped, otherwise skip it only
310 if it is inactive/recovering/out-of-space.
314 Description: more cleanup in mds_lov
315 Details : not send LOV EA under replay, we can't know about they size at this
316 time. Don't allow client connect to mds before any ost connected,
317 for avoid problems with LOV EA size and returning EIO to client.
319 Severity : enhancement
321 Description: Interoperability at server side (Disk interoperability)
323 Severity : enhancement
325 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
327 Severity : enhancement
329 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
331 Severity : enhancement
333 Description: New client IO stack (CLIO).
335 Severity : enhancement
337 Description: Commit on sharing. Eliminate inter-client dependencies between
338 uncommitted transactions by doing transaction commits.
339 Thereby clients may recovery independently.
342 Frequency : Create a symlink file with a very long name
344 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
345 Details : If there is no extra space in the request for early cancels,
346 ldlm_req_handles_avail() returns 0 instead of a negative value.
348 Severity : enhancement
350 Description: Add /proc entry for import status
351 Details : The mdc, osc, and mgc import directories now have
352 an import directory that contains useful import data for debugging
355 Severity : enhancement
357 Description: Re-disable certain /proc logging
358 Details : Enable and disable client's offset_stats, extents_stats and
359 extents_stats_per_process stats logging on the fly.
362 Frequency : Only on FC kernels 2.6.22+
364 Description: oops in statahead
365 Details : Do not drop reference count for the dentry from VFS when lookup,
366 VFS will do that by itself.
368 Severity : enhancement
370 Description: Generic /proc file permissions
371 Details : Set /Proc file permissions in a more generic way to enable non-
372 root users operate on some /proc files.
376 Description: Hitting mdc_commit_close() ASSERTION
377 Details : Properly handle request reference release in
378 ll_release_openhandle().
382 Description: quota recovery deadlock during mds failover
383 Details : This patch includes att18982, att18236, att18237 in bz14840.
385 1. fix osts hang when mds does failover with quotaon
386 2. prevent watchdog storm when osts threads wait for the
391 Frequency : only patchless client
392 Description: add workaround for race between add/remove dentry from hash
394 Severity : enhancement
396 Description: Allow OST glimpses to return PW locks
400 Description: LBUG when llog conf file is full
401 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
405 Description: Prevent import from entering FULL state when server in recovery
409 Description: service mount cannot take device name with ":"
410 Details : Only when device name contains ":/" will mount treat it as
416 Description: replace ptlrpcd with the statahead thread to interpret the async
417 statahead RPC callback
421 Frequency : on recovery
422 Description: I/O failures after umount during fail back
423 Details : if client reconnected to restarted server we need join to recovery
424 instead of find server handler is changed and process self
425 eviction with cancel all locks.
427 Severity : enhancement
429 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
431 Severity : enhancement
433 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
435 Severity : enhancement
437 Description: Upcall on Lustre log has been dumped
438 Details : Allow for a user mode script to be called once a Lustre log has
439 been dumped. It passes the filename of the dumped log to the
440 script, the location of the script can be specified via
441 /proc/sys/lnet/debug_log_upcall.
446 Description: avoid idr_remove called for id which is not allocated.
447 Details : Move assigment s_dev for clustered nfs to end of initialization,
448 to avoid problem with error handling.
453 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
454 Details : When connection is reused this not moved from CONN_UNUSED_HASH
455 into CONN_USED_HASH and this prodice warning when put connection
456 again in unused hash.
459 Severity : enhancement
461 Description: File striping can now be set to use an arbitrary pool of OSTs.
463 Severity : enhancement
465 Description: Export bytes_read/bytes_write count on OSC/OST.
469 Description: Early reply size mismatch, MGC loses connection
470 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
471 the connect flags are properly negotiated.
476 Description: quotas are not honored with O_DIRECT
477 Details : all writes with the flag O_DIRECT will use grants which leads to
478 this problem. Now using OBD_BRW_SYNC to guard this.
482 Description: add quota statistics
483 Details : 1. sort out quota proc entries and proc code.
484 2. add quota statistics
486 Severity : enhancement
488 Description: enable quota support for HEAD.
492 Description: Properly propagate oinfo flags from lov to osc for statfs
493 Details : restore missing copy oi_flags to lov requests.
495 Severity : enhancement
497 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
498 : l_getgroups(8), lst(8), routerstat(8)
499 Details : included man pages for llobdstat(8), llstat(8),
500 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
502 Severity : enhancement
504 Description: configure's --enable-quota should check the
505 : kernel .config for CONFIG_QUOTA
506 Details : configure is terminated if --enable-quota is passed but
507 : no quota support is in kernel
511 Description: Remove portals compatibility
512 Details : Remove portals compatibility, not interoperable with releases
517 Description: Resolve device initialization race
518 Details : Prevent proc handler from accessing devices added to the
519 obd_devs array but yet be intialized.
521 Severity : enhancement
523 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
525 Severity : enhancement
527 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
532 Description: Using wrong pointer in osc_brw_prep_request
533 Details : Access to array[-1] can produce panic if kernel compiled with
534 CONFIG_PAGE_ALLOC enabled
538 Description: Client runs out of low memory
539 Details : Consider only lowmem when counting initial number of llap pages
543 Description: *optional* service tags registration
544 Details : if the "service tags" package is installed on a Lustre node
545 When the filesystem is mounted, a local-node service tag will
546 be created. See http://inventory.sun.com/ for more information
547 about the Service Tags asset management system.
551 Description: Kernel BUG tries to release flock
552 Details : Lustre does not destroy flock lock before last reference goes
553 away. So always drop flock locks when client is evicted and
554 perform unlock regardless of successfulness of speaking to MDS.
558 Description: add refcount protection for osc callbacks, avoid panic on shutdown
562 Description: sanity test 65a fails if stripecount of -1 is set
563 Details : handle -1 striping on filesystem in ll_dirstripe_verify
568 Description: ASSERTION(CheckWriteback(page,cmd)) failed
569 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
572 Severity : enhancement
574 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
578 Description: do not process already freed flock
579 Details : flock can possibly be freed by another thread before it reaches
580 to ldlm_flock_completion_ast.
584 Description: LBUG during stress test
585 Details : Need properly lock accesses the flock deadlock detection list.
589 Description: oops in page fault handler
590 Details : kernel page fault handler can return two special 'pages' in error
591 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
595 Description: timeout with invalidate import.
596 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
597 should be handled by ptlrpcd. This produce long age waiting and
598 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
600 Severity : enhancement
602 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
606 Description: enable MGS and MDT services start separately
607 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
608 a co-located MGS without starting the MGS, which is a complement
609 to 'nosvc' mount option.
613 Frequency : after recovery
614 Description: precreate to many object's after del orphan.
615 Details : del orphan st in oscc last_id == next_id and this triger growing
616 count of precreated objects. Set flag LOW to skip increase count
617 of precreated objects.
621 Frequency : rare, on clear nid stats
622 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
623 Details : when clean nid stats sometimes try destroy live entry,
624 and this produce panic in free.
628 Description: Stack overflow during MDS log replay
629 ease stack pressure by using a thread dealing llog_process.
633 Description: wait until IO finished before start new when do lock cancel.
634 Details : VM protocol want old IO finished before start new, in this case
635 need wait until PG_writeback is cleared until check dirty flag and
636 call writepages in lock cancel callback.
638 Severity : enhancement
640 Description: using special macro for print time and cleanup in includes.
644 Description: mds_mfd_close() ASSERTION(rc == 0)
645 Details : In mds_mfd_close(), we need protect inode's writecount change
646 within its orphan write semaphore to prevent possible races.
650 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
654 Frequency : rare, on shutdown ost
655 Description: don't hit live lock with umount ost.
656 Details : shrink_dcache_parent can be in long loop with destroy dentries,
657 use shrink_dcache_sb instead.
661 Description: don't panic with use echo client
662 Details : echo client pass NULL as client nid pointer and this produce null
667 Description: fix build on ppc32
668 Details : compile code with -m64 flag produce wrong object file for ppc32.
672 Description: add message levels for liblustreapi
676 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
677 Details : Move the 'good_osts' check before the 'total_bavail' check. This
678 will result in an -EAGAIN and in the exit call path we call
679 alloc_rr() which will with increasing aggressiveness attempt to
680 aquire precreated objects on the minimum number of required OSCs.
684 Description: Use old size assignment to avoid deadlock
685 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
686 the scheduling while holding a spinlock. We do not need locking
687 for size in ll_update_inode() because size is only updated from
688 the MDS for directories or files without objects, so there is no
689 other place to do the update, and concurrent access to such inodes
690 are protected by the inode lock.
694 Description: resolve "_IOWR redefined" build error on SLES10
698 Description: dump the memory debugging after all modules are unloaded to
699 suppress false negative in conf_sanity test 39
701 Severity : enhancement
703 Description: build kernel-ib packages for OFED 1.3 in our release cycle
708 Description: fix SLES kernel versioning
709 Details : the kernel version for our SLES 10 kernel did not include a "-"
710 before the "smp" at the end. while this was not a problem in
711 general, it did mean that software trying to use the kernel
712 version to try to detect a vendor specific kernel would fail.
713 this was most evident by the OFED build scripts.
717 Description: Don't update lov_desc members until making sure they are valid
718 Details : When updating lov_desc members via proc fs, need fix their
719 validities before doing the real update.
723 Description: don't put request into delay list while invalidate in flight.
724 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
725 import in flight. this produce timeout for invalidate and sometimes
726 can cause stale data.
730 Frequency : on ppc only
731 Description: not convert ost objects for directory because it's not exist.
732 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
736 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
738 Severity : enhancement
740 Description: Update to support 2.6.22.14 vanilla kernel.
744 Frequency : rare, on recovery
745 Description: read procfs can produce deadlock in some situation
746 Details : Holding lprocfs lock which send rpc can produce block for destroy
747 obd objects and this also block reconnect with -EALREADY. This
748 isn't fix all lprocfs bugs - but make it rare.
750 Severity : enhancement
752 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
755 Frequency : frequent on X2 node
757 Description: mdc_set_open_replay_data LBUG
758 Details : Set replay data for requests that are eligible for replay.
762 Description: lustre_mgs: operation 101 on unconnected MGS
763 Details : When MGC is disconnected from MGS long enough, MGS will evict the
764 MGC, and late on MGC cannot successfully connect to MGS and a lot
765 of the error messages complaining that MGS is not connected.
769 Frequency : on network error
770 Description: panic with double free request if network error
771 Details : mdc_finish_enqueue is finish request if any network error ocuring,
772 but it's true only for synchronus enqueue, for async enqueue
773 (via ptlrpcd) this incorrect and ptlrpcd want finish request
776 Severity : enhancement
778 Description: client-side metadata stat-ahead during readdir(directory readahead)
779 Details : perform client-side metadata stat-ahead when the client detects
780 readdir and sequential stat of dir entries therein
783 Frequency : on start mds
785 Description: Implement get_info(last_id) in obdfilter.
788 Frequency : occasional
790 Description: Correctly check stale fid, not start epoch if ost not support SOM
791 Details : open with flag O_CREATE need set old fid in op_fid3 because
792 op_fid2 was overwritten with new generated fid, but mds can answer
793 with one of these two fids and both is not stale. Setattr
794 incorrectly started an epoch and assume will be called
795 done_writing, but without SOM done_writing ever being called.
798 Frequency : rare, depends on device drivers and load
800 Description: MDS or OSS nodes crash due to stack overflow
801 Details : Code changes in 1.8.0 increased the stack usage of some functions.
802 In some cases, in conjunction with device drivers that use a lot
803 of stack the MDS (or possibly OSS) service threads could overflow
804 the stack. One change which was identified to consume additional
805 stack has been reworked to avoid the extra stack usage.
808 Frequency : occasional
810 Description: Do not fail import if osc_interpret_create gets -EAGAIN
811 Details : If osc_interpret_create got -EAGAIN it immediately exits and
812 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
813 oscc_has_objects and see OSC has no objests and call
814 oscc_internal_create to resend create request.
816 Severity : enhancement
818 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
820 Severity : enhancement
822 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
825 Frequency : very rare
827 Description: Fix replay if there is an un-replied request and open
828 Details : In some cases, older replay request will revert the
829 mcd->mcd_last_xid on MDS which is used to record the client's
832 Severity : enhancement
834 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
836 Severity : enhancement
838 Description: Add rhel5 support to HEAD.
840 Serverity : enhancement
842 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
847 Description: Don't allow skipping OSTs if index has been specified.
848 Details : Don't allow skipping OSTs if index has been specified, make
849 locking in internal create lots better.
853 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
854 Details : access to bitfield in structure is always rounded to long
855 and this produce problem with not atomic change any bit.
859 Description: Lustre make rpms failed.
860 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
861 builds Lustre rpms from tarball.
864 Frequency : rare on shutdown ost
866 Description: If llog cancel was not send before clean_exports phase, this can
867 produce deadlock in llog code.
868 Details : If llog thread has last reference to obd and call class_import_put
869 this produce deadlock because llog_cleanup_commit_master wait when
870 last llog_commit_thread exited, but this never success because was
871 called from llog_commit_thread.
875 Description: allow userland application know is lost one of stripes.
876 Details : fill lvb_blocks with error code on ost and return it to
877 application if error flag found.
881 Description: NULL lov_tgts causing MDS oops
882 Details : more safe checks for NULL lov_tgts for avoid oops.
884 Severity : enhancement
886 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
890 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
891 Details : If an OST is down the MDS will hang indefinitely in
892 obd_statfs_rqset() waiting for the statfs data. While for
893 MDS QOS usage of statfs, it should not stuck in waiting.
895 Severity : enhancement
897 Description: remote_acl support
898 Details : Support ACL-based permission check for remote user.
899 Support setfacl/getfacl for remote user with the utils
900 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
901 the system "{s,g}etfacl" utils.
903 Severity : enhancement
905 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
907 Severity : enhancement
909 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
913 Description: make mgs_setparam() handle fsname containing dash
914 Details : fsname containing a dash does not work with lctl conf_param
916 Severity : enhancement
918 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
920 Severity : enhancement
922 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
926 Frequency : rare, at shutdown
927 Description: access already free / zero obd_namespace.
928 Details : if client_disconnect_export was called without force flag set,
929 and exist connect request in flight, this can produce access to
930 NULL pointer (or already free pointer) when connect_interpret
931 store ocd flags in obd_namespace.
935 Frequency : only at startup
936 Description: not alloc memory with spinlock held.
937 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
938 if any spinlock held.
940 Severity : enhancement
942 Description: make lustre randomly fail allocating memory
943 Details : Make lustre randomly failed allocating memory for testing purpose.
945 Severity : enhancement
947 Description: lost problems with lov objid file
948 Details : Fixes some scability and access to not inited memory problems
949 in work with lov objdid file.
954 Description: lfs find does not continue on file error
955 Details : Continue other files processing when a file/dir is absent.
959 Description: Inconsistent usage of lustre_pack_reply()
960 Details : Standardize the usage of lustre_pack_reply() such that it
961 always generate a CERROR on failure.
964 Frequency : occasional
966 Description: MDS hang or stay in waiting lock
967 Details : If client receive lock with CBPENDING flag ldlm need send lock
968 cancel as separate rpc, to avoid situation when cancel request
969 can't processed due all i/o threads stay in wait lock.
973 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
977 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
981 Description: Update to SLES9 kernel-2.6.5-7.287.3.
985 Description: Update to RHEL4 latest kernel.
987 Severity : enhancement
989 Description: Build SLES10 patchless client fails
990 Details : The configure was broken by run ./configure with
991 --with-linux-obj=.... argument for patchless client. When the
992 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
993 can't search header adequately. Use absolute path such as
994 -I($LINUX)/include instead.
998 Description: interrupt oig_wait produce painc on resend.
999 Details : brw_redo_request can be used for resend requests from ptlrpcd and
1000 private set, and this produce situation when rq_ptlrpcd_data not
1001 copyed to new allocated request and triggered LBUG on assert
1002 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
1003 ptlrpcd set if request is changed and can be safety changed to use
1008 Description: LASSERT_{REQ,REP}SWAB macros are buggy
1009 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
1010 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
1011 these macros and replace them with their difinitions instead.
1015 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
1016 Details : Update which_patch & target file for SLES10 latest kernel.
1019 Description: Remove client patches from SLES 10 kernel.
1020 Details : This causes SLES 10 clients to behave as patchless clients
1021 even on a Lustre-patched (server) kernel.
1023 Severity : enhancement
1025 Description: self-adjustable client's lru lists
1026 Details : use adaptive algorithm for managing client cached locks lru
1027 lists according to current server load, other client's work
1028 pattern, memory activities, etc. Both, server and client
1029 side namespaces provide number of proc tunables for controlling
1032 Severity : enhancement
1034 Description: light-weight GSS support
1035 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
1036 and reduce performance overhead.
1038 Severity : enhancement
1040 Description: Linux keyring support
1041 Details : Support using service of Linux keyring for Lustre GSS internal
1042 context refresh/cache mechanism.
1046 Description: Fix errors in lfs documentation
1047 Details : Fixes man pages
1051 Description: don't use GFP_* in generic Lustre code.
1052 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
1056 Description: obdclass is limited by single OBD_ALLOC(idarray)
1057 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
1061 Description: warnings with build patchless client with vanila 2.6.19 and up
1062 Details : change old ctl_table style and replace ctl_table/ctl_table_header
1063 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1067 Frequency : Only for SLES
1068 Description: sanity_quota fail test_1
1069 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1070 file, which makes TSTID[2] inunique.
1074 Frequency : Only for SLES9
1075 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1076 Details : Update target/ChangeLog/which_patch .
1080 Frequency : Only for test_10 in sanity-quota.sh
1081 Description: a bug in quota test code
1082 Details : checking a test flag in a wrong place
1086 Frequency : Only for test_13 in sanity-quota.sh
1087 Description: a bug in quota test script
1088 Details : shouldn't check overfull allocated quota
1092 Description: jbd statistics
1093 Details : Port older jbd statistics patch for sles10
1095 Severity : enhancement
1097 Description: Add printing inode info into message about error in writepage.
1102 Description: reply_lock_interpret crash due to race with it and lock cancel.
1103 Details : Do not replay locks that are being cancelled. Do not reference
1104 locks by their address during replay, just by their handle.
1108 Frequency : When flocks are used.
1109 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1110 Details : Flock locks might destroy just granted lock if it could be merged
1111 with another existing flock, this is done in completion handler,
1112 so teach ldlm_cli_enquque_fini that this is a valid case for
1118 Description: Oops in read and write path when failing to allocate lock.
1119 Details : Check if lock allocation failed and return error back.
1123 Description: lstripe command fails for valid OST index
1124 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1125 instead of lov->desc.ld_active_tgt_count.
1129 Description: sanity.sh failed test 103
1130 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1135 Description: df doesn't work properly if diskfs blocksize != 4K
1136 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1140 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1141 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1143 Severity : enhancement
1145 Description: Adaptive timeouts
1146 Details : RPC timeouts adapt to changing server load and network
1147 conditions to reduce resend attempts and improve recovery time.
1151 Description: llapi_file_create() does not allow some changes
1152 Details : add llapi_file_open() that allows specifying the mode and
1153 open flags, and also returns an open file handle.
1157 Description: merge and cleanup kernel patches.
1158 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1162 Description: Add journal checksum support.(Kernel part)
1163 Details : The journal checksum feature adds two new flags i.e
1164 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1165 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1166 indicates that the commit block contains the checksum for
1167 the blocks described by the descriptor blocks. Now commit
1168 record can be sent to disk without waiting for descriptor
1169 blocks to be written to disk. This behavior is controlled
1170 using JBD2_FEATURE_ASYNC_COMMIT flag.
1174 Description: OSS needs mutliple precreate threads
1175 Details : Add ability to start more than one create thread per OSS.
1179 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1180 Details : execute lfs setstripe on client
1184 Description: mds_obd_create error creating tmp object
1185 Details : When the user sets quota on root, llog will be affected and can't
1186 create files and write files.
1189 Frequency : Always on ia64 patchless client, and possibly others.
1191 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1192 Details : This allows the patchless client to be loaded on architectures
1193 without this export.
1197 Description: RedHat Update kernel for RHEL5
1198 Details : Add the kernel config file for RHEL5.
1202 Description: RedHat Update kernel for RHEL5
1203 Details : Modify the kernel config file more closer RHEL5.
1207 Description: Build failure against Centos5 (RHEL5)
1208 Details : Define PAGE_SIZE when it isn't present.
1212 Description: test_7 on sanity-quota.sh can't be used on separate machines
1213 Details : add facet to handle it
1217 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1218 Details : with NFS, the anon dentry's parent was set to itself in
1219 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1220 corresponding dentry other than use rec->ur_name.
1222 Severity : enhancement
1224 Description: lfs setstripe enhancement
1225 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1229 Description: enable data checksumming by default
1230 Details : enable checksum by default, allow --disable-checksum
1231 configure option and "-o nochecksum" mount option. Checksums
1232 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1236 Description: lustre support for RHEL5
1237 Details : Add support for RHEL5.
1241 Description: Client eviction due to failover config
1242 Details : after a connection loss, the lustre client should attempt to
1243 reconnect to the last active server first before trying the
1244 other potential connections.
1248 Description: when mds and osts use different quota unit(32bit and 64bit),
1249 quota will be released repeatly.
1250 Details : void sending multiple quota reqs to mds, which will keep the
1251 status between the reqs.
1255 Description: osts not allocated evenly to files
1256 Details : change the condition to increase offset_idx
1259 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1261 Description: Data corruption for OSTs that are formatted larger than 2TB
1263 Details : When generating the bio request for lustre file writes the
1264 sector number would overflow a temporary variable before being
1265 used for the IO. The data reads correctly from Lustre (which
1266 will overflow in a similar manner) but other file data or
1267 filesystem metadata may be corrupted in some cases.
1271 Description: Tune the kernel for good SCSI performance.
1272 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1273 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1278 Description: rewrite ext2-derived code in obdclass/uuid.c
1279 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1280 from scratch preserving functionality.
1284 Description: "lfs find -obd UUID" prints directories
1285 Details : "lfs find -obd UUID" will return all directory names instead
1286 of just file names. It is incorrect because the directories
1287 do not reside on the OSTs.
1291 Description: use i_size_read and i_size_write in 2.6 port
1292 Details : replace inode->i_size access with i_size_read/write()
1295 Frequency : when removing large files
1297 Description: scheduling issue during removal of large Lustre files
1298 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1299 It causes scheduling issues when removing large files (17TB in the
1303 Frequency : only with liblustre clients on XT3
1305 Description: evictions taking too long
1306 Details : allow llrd to evict clients directly on OSTs
1309 Frequency : only on ppc
1311 Description: /proc/fs/lustre/devices broken on ppc
1312 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1313 We should make sure the type of 'index' is loff_t and then cast
1314 explicitly as needed below. Do not assign an explicitly cast
1318 Frequency : only for rhel5
1320 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1321 Details : Modify the target file & which_kernel.
1323 Severity : enhancement
1325 Description: omit set fsid for export NFS
1326 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1327 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1331 Frequency : Always, for kernels after 2.6.16
1332 Description: Fix warning idr_remove called for id=.. which is not allocated.
1333 Details : Last kernels save old s_dev before kill super and not allow
1334 to restore from callback - restore it before call kill_anon_super.
1338 Description: buffer overruns could theoretically occur
1339 Details : llapi_semantic_traverse() modifies the "path" argument by
1340 appending values to the end of the origin string, and a
1341 overrun may occur. Adding buffer overrun check in liblustreapi.
1345 Description: Fix error on 'ls .' at the top of the Lustre mount.
1346 Details : Don't revalidate dentry if it is a root dentry.
1350 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1351 Details : Modify vm-tunables-rhel4.patch.
1355 Description: Kernel config for 2.6.18-vanilla.
1356 Details : Modify targets/2.6-vanilla.target.in.
1357 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1358 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1359 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1360 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1364 Description: improve handling recoverable errors
1365 Details : if request processig with error which can be recoverable on server
1366 request should be resend, otherwise page released from cache and
1371 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1372 Details : Modify target file & which_patch.
1373 A flaw was found in the IA32 system call emulation provided
1374 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1375 value could be stored in the %RAX register, which could trigger an
1376 out-of-bounds system call table access. An untrusted local user
1377 could exploit this flaw to run code in the kernel
1378 (ie a root privilege escalation). (CVE-2007-4573).
1382 Description: O_DIRECT bypasses client statistics.
1383 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1384 read_ahead_stats were updated. copied the stats section from
1385 osc_send_oap_rpc() into async_internal().
1389 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1393 Description: change order of libsysio includes
1394 Details : '#include sysio.h' should always come before '#include xtio.h'
1398 Description: handle "serious error: objid * already exists" more gracefully
1399 Details : If LAST_ID value on disk is smaller than the objects existing in
1400 the O/0/d* directories, it indicates disk corruption and causes an
1401 LBUG(). If the object is 0-length, then we should use the existing
1402 object. This will help to avoid a full fsck in most cases.
1404 Severity : enhancement
1406 Description: adapt the lustre_config script to support the upgrade case
1407 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1408 server targets to 1.6 in parallel.
1412 Description: To avoid grant space > avaible space when the disk is almost
1413 full. Without this patch you might see the error "grant XXXX >
1414 available" or some LBUG about grant, when the disk is almost
1416 Details : In filter_check_grant, for non_grant cache write, we should
1417 check the left space by if (*left > ungranted + bytes), instead
1418 of (*left > ungranted), because only we are sure the left space
1419 is enough for another "bytes", then the ungrant space should be
1420 increase. In client, we should update cl_avail_grant only there
1421 is OBD_MD_FLGRANT in the reply.
1425 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1426 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1427 described in CVE-2007-4573. This problem could allow untrusted
1428 local users to gain root access.
1431 Frequency : when using O_DIRECT and quotas
1433 Description: Incorrect file ownership on O_DIRECT output files
1434 Details : block usage reported by 'lfs quota' does not take into account
1435 files that have been written with O_DIRECT.
1440 Description: touch file failed when fs is not full
1441 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1442 otherwise we can get ENOSP while fs is not full.
1446 Description: parallel lock callbacks
1447 Details : Instead of sending blocking and completion callbacks as separated
1448 requests, adding them to a set and sending in parallel.
1451 Frequency : only for Cray XT3
1452 Bugzilla : 12829/13455
1453 Description: Changing primary group doesn't change the group lustre assigns to
1455 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1456 overridden with the primary group provided by the group upcall,
1457 whereas the supplied fsgid can be trusted if it is in the list of
1458 supplementary groups returned by the group upcall.
1460 Severity : enhancement
1462 Description: Allow masking D_WARNING, D_ERROR messages from console
1463 Details : Console messages can now be disabled via lnet.printk.
1467 Description: User code with malformed file open parameter crashes client node
1468 Details : Before packing join_file req, all the related reference should be
1469 checked carefully in case some malformed flags cause fake
1470 join_file req on client.
1474 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1475 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1476 mode is equal to the granted mode. However, it should also check
1477 that LDLM_FL_CP_REQD is not set.
1481 Description: Heavy nfs access might result in deadlocks
1482 Details : After ELC code landed, it is now improper to enqueue any mds
1483 locks under och_sem, because enqueue might want to decide to
1484 cancel open locks for same inode we are holding och_sem for.
1488 Description: Client eviction while running blogbench
1489 Details : A lot of unlink operations with concurrent I/O can lead to a
1490 deadlock causing evictions. To address the problem, the number of
1491 oustanding OST_DESTROY requests is now throttled to
1492 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1493 callbacks are processed in priority.
1497 Description: enable ACLs on MDS by default
1498 Details : ACLs must be enabled on MDS by default.
1501 Frequency : PPC/PPC64 only
1503 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1504 Details : fix duplicated definitions between asm-ppc64/types.h and
1505 lustre_types.h on PPC.
1508 Frequency : PPC/PPC64 only
1510 Description: asm-ppc/segment.h does not exist
1511 Details : fix compile issue on PPC.
1515 Description: better handle error messages in extents code
1518 Frequency : RHEL4 only
1520 Description: mkfs is very slow on IA64/RHEL4
1521 Details : A performance regression has been discovered in the MPT Fusion
1522 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1523 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1524 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1526 Severity : enhancement
1528 Description: SNMP support enhancement
1529 Details : Adding total number of sampled request for an MDS node in snmp
1532 Severity : enhancement
1534 Description: Optimize ldlm waiting list processing for PR extent locks
1535 Details : When processing waiting list for read extent lock and meeting read
1536 lock that is same or wider to it that is not contended, skip
1537 processing rest of the list and immediatelly return current
1538 status of conflictness, since we are guaranteed there are no
1539 conflicting locks in the rest of the list.
1543 Description: Time out and refuse to reconnect
1544 Details : When the failover node is the primary node, it is possible
1545 to have two identical connections in imp_conn_list. We must
1546 compare not conn's pointers but NIDs, otherwise we can defeat
1547 connection throttling.
1551 Description: port llog fixes from b1_6 into HEAD
1552 Details : Port llog reference couting and some llog cleanups from b1_6
1553 (bug 10800) into HEAD, for protect from panic and access to already
1554 free llog structures.
1558 Description: Detect stride IO mode in read-ahead
1559 Details : When a client does stride read, read-ahead should detect that and
1560 read-ahead pages according to the detected stride pattern.
1564 Description: data checksumming impacts single node performance
1565 Details : add support for several checksum algorithm. Currently, only CRC32
1566 and Adler-32 are supported. The checksum type can be changed on
1567 the fly via /proc/fs/lustre/osc/*/checksum_type.
1571 Description: use adler32 for page checksums
1572 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1577 Description: build for x2 fails
1578 Details : fix compile issue on Cray systems.
1582 Description: Properly match for duplicate locks
1583 Details : Due to different lock order from skiplists code, we need to
1584 traverse entire list for now
1587 Frequency : only on PPC/SLES10
1589 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1590 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1591 be defined. Add a hack in mkfs_lustre.c to work around this compile
1596 Description: LASSERT on MDS when client holding flock lock dies
1597 Details : ldlm pool logic depends on number of granted locks equal to
1598 number of released locks which is not true for flock locks, so
1599 just exclude such locks from consideration.
1603 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1604 Details : Use fsfilt_sync() for both the whole filesystem sync and
1605 individual file sync to eliminate dangerous inode locking
1606 with I_LOCK that can lead to a deadlock.
1610 Description: performance in 1.6.3
1611 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1612 sure that 1MB requests are not fragmented by the block layer.
1614 Severity : enhancement
1616 Description: organize the server-side client stats on per-nid basis
1617 Details : Change the structure of stats under obdfilter and mds to
1625 The "uuid"s file would list the uuids of _active_ exports.
1626 And the clear entry is to clear all stats and stale nids.
1628 Severity : enhancement
1630 Description: eliminate client locks in face of contention
1631 Details : file contention detection and lockless i/o implementation
1632 for contended files.
1636 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1639 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1641 Description: MDS may lose file striping (and hence file data) in some cases
1642 Details : If there are additional extended attributes stored on the MDS,
1643 in particular ACLs, SELinux, or user attributes (if user_xattr
1644 is specified for the client mount options) then there is a risk
1645 of attribute loss. Additionally, the Lustre file striping
1646 needs to be larger than default (e.g. striped over all OSTs),
1647 and an additional attribute must be stored initially in the
1648 inode and then increase in size enough to be moved to the
1649 external attribute block (e.g. ACL growing in size) for file
1654 Description: skiplist implementation simplification
1655 Details : skiplists are used to group compatible locks on granted list
1656 that was implemented as tracking first and last lock of each
1657 lock group the patch changes that to using doubly linked lists
1661 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1662 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1663 because LASSERT on both the data supplied by a client, and the
1664 data on disk is dangerous and incorrect.
1666 Severity : enhancement
1668 Description: Slow truncate/writes to huge files at high offsets.
1669 Details : Directly associate cached pages to lock that protect those pages,
1670 this allows us to quickly find what pages to write and remove
1671 once lock callback is received.
1675 Description: more ldlm soft lockups
1676 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1677 starve other threads from the resource lock for a long time in
1678 case of long waiting queue, so change the debug level from
1679 D_OTHER to the less frequently used D_INFO.
1681 Severity : enhancement
1683 Description: add -gid, -group, -uid, -user options to lfs find
1687 Description: Hung threads in invalidate_inode_pages2_range
1688 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1689 once one is completed. As a result, some RPCs are stuck in the
1690 queue and are never sent.
1694 Description: filter threads hungs on waiting journal commit
1695 Details : Cleanup filter group llog code, then only filter group llog will
1696 be only created in the MDS/OST syncing process.
1700 Description: Procfs and llog threads access destoryed import sometimes.
1701 Details : Sync the import destoryed process with procfs and llog threads by
1702 the import refcount and semaphore.
1704 Severity : enhancement
1706 Description: openlock cache of b1_6 port to HEAD
1711 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1712 Details : If insertion of an extent fails, then discard the inode
1713 preallocation and free data blocks else it can lead to duplicate
1718 Description: don't always update ctime in ext3_xattr_set_handle()
1719 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1720 In some cases the ctime should not be updated, for example for
1721 2.0->1.8 compatibility it is necessary to delete an xattr and it
1722 should not update the ctime.
1726 Bugzilla : 15713/16362
1727 Description: Assertion in iopen_connect_dentry in 1.6.3
1728 Details : looking up an inode via iopen with the wrong generation number can
1729 populate the dcache with a disconneced dentry while the inode
1730 number is in the process of being reallocated. This causes an
1731 assertion failure in iopen since the inode's dentry list contains
1732 both a connected and disconnected dentry.
1736 Description: assertion failure in ldlm_handle2lock()
1737 Details : fix a race between class_handle_unhash() and class_handle2object()
1738 introduced in lustre 1.6.5 by bug 13622.
1743 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1744 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1749 Description: oops in sd_iostats_seq_show()
1750 Details : unloading/reloading the scsi low level driver triggers a kernel
1751 bug when trying to access the sd iostat file.
1756 Description: Kernel panics during QLogic driver reload
1757 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1758 patch, causing memory corruption.
1763 Description: journal_dev option does not work in b1_6
1764 Details : pass mount option during pre-mount.
1766 Severity : enhancement
1768 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1769 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1770 extent information of a file. It can be used to map logical blocks
1771 in a file to physical blocks in the block device.
1775 Description: LDLM soft lockups - improvement
1776 Details : It is be possible to send the lock handle along with each read
1777 or write request because the client is already doing a lock match
1778 itself so there isn't any reason the OST should have to re-do that
1782 Frequency : only on Cray X2
1784 Description: X2 build failures
1785 Details : fix build failures on Cray X2.
1789 Description: xid & resent requests
1790 Details : Initialize RPC XID from clock at startup (randomly if clock is
1793 Severity : enhancement
1795 Description: Add lustre_start utility to start or stop multiple Lustre servers
1800 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1801 Details : In case of memory pressure, list_del() can be called twice on
1802 req->rq_history_list, causing a kernel oops.
1806 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1807 Details : fix stack overflow in the distributed lock manager by defering
1808 export eviction after a failed AST to the elt thread instead of
1809 handling it in the dlm interpret routine.
1813 Description: Convert some comments to new format.
1814 Details : Update documenting comments to match doxygen conventions.
1818 Description: Grammar fixes.
1819 Details : A couple of trivial spelling fixes.
1823 Description: OSD_COUNTERS-mandatory
1824 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1825 to simplify the code.
1829 Description: simplify cmm_device freeing logic.
1830 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1834 Description: Add lockdep support to dt_object_operations locking interface.
1835 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
1836 parameter indicating lock ordering. Update mdd code to use new
1841 Description: Introduce failloc constants for lockless IO tests.
1842 Details : Add two new failloc constants to test lockless IO. Only one of
1843 them in implemented---another is checked in yet to be landed
1848 Description: Add lockdep support for inode mutex.
1849 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1850 in the situations where Lustre has to lock more than one inode
1855 Description: Add optional invariants checking support.
1856 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1857 configure switch. This macro is to be used for consistency and
1858 sanity checks that are too expensive to be left in `production'
1863 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1864 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1869 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1870 Details : Use spin_lock_nested() in (the only) situation where more than
1871 one ldlm_lock is locked simultaneously. Also, fix possible
1872 dead-lock in ldlm_lock_change_resource() by enforcing particular
1877 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1878 Details : Instead of specifying each ldlm_lock call-back through separate
1879 parameter, wrap them into struct ldlm_callback_suite.
1883 Description: Kill join_lru obd method and its callers.
1884 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
1885 in memory---a requirement implemented through ->o_join_lru() obd
1886 method in HEAD. Remove this method and its users.
1890 Description: Add asynchronous ldlm ENQUEUE completion handler.
1891 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
1892 case that stock ldlm_completion_ast() cannot handle as it waits
1893 until lock is granted. Introduce new ldlm_completion_ast_async()
1894 for this. Also comment ldlm_completion_ast().
1898 Description: ldlm_error <-> errno conversion.
1899 Details : Add functions to map (rather arbitrary) between LDLM error codes
1900 and standard errno values. CLIO needs this to prevent LDLM specific
1901 constants from escaping ldlm and osc.
1905 Description: Kill unused ldlm_handle2lock_ns() function.
1906 Details : Kill unused ldlm_handle2lock_ns() function.
1910 Description: Add lu_ref support to ldlm_lock
1911 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
1912 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
1913 and ldlm_resource. LDLM interface has to be changed, because code that
1914 releases a reference on a lock, has to "know" what reference this is.
1915 In the most frequent case
1917 lock = ldlm_handle2lock(handle);
1919 LDLM_LOCK_PUT(lock);
1921 no changes are required. When any other reference (received _not_ from
1922 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
1923 instead of LDLM_LOCK_PUT().
1925 Arguably, changes are pervasive, and interface requires some discipline
1926 for proper use. On the other hand, it was very instrumental in finding
1927 a few leaked lock references.
1931 Description: Add ldlm_lock_addref_try().
1932 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
1933 attempts to addref a lock that might be being canceled
1938 Description: Add ldlm_weigh_callback().
1939 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
1940 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
1941 hard-coded `number of pages' logic.
1945 Description: Add lockdep annotations to llog code.
1946 Details : Use appropriately tagged _nested() locking calls in the places
1947 where llog takes more than one ->lgh_lock lock.
1951 Description: Add loi_kms_set().
1952 Details : Wrap kms updates into a helper function.
1956 Description: Constify instances of struct lsm_operations.
1957 Details : Constify instances of struct lsm_operations.
1961 Description: lu_conf support.
1962 Details : On a server, a file system object is uniquely identified
1963 by a fid, which is sufficient to locate and load all object
1964 state (inode). On a client, on the other hand, more data are
1965 necessary instantiate an object. Change lu_object_find() and
1966 friends to take additional `lu_conf' argument describing object.
1967 Typically this includes layout information.
1971 Description: lu_context fixes.
1972 Details : Introduce new lu_context functions that are needed on the client
1973 side, where some system threads (ptlrpcd) are shared by multiple
1974 modules, and so cannot be stopped during module shutdown.
1978 Description: Add start and stop methods to lu_device_type_operations.
1979 Details : Introduce two new methods in lu_device_type_operations, that are
1980 invoked when first instance of a given type is created and last one
1981 is destroyed respectively. This is need by CLIO.
1985 Description: Add lu_ref support to struct lu_device.
1986 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
1987 track leaked references.
1991 Description: Introduce lu_kmem_descr.
1992 Details : lu_kmem_descr and its companion interface allow to create
1993 and destroy a number of kmem caches at once.
1997 Description: Fix lu_object finalization race.
1998 Details : Fix a race between lu_object_find() finding an object and its
1999 concurrent finalization. This race is (most likely) not possible
2000 on the server, but might happen on the client.
2004 Description: Introduce lu_ref interface.
2005 Details : lu_ref is a debugging module allowing to track references to
2006 a given object. It is quite cpu expensive, and has to be
2007 explicitly enabled with --enable-lu_ref. See usage description
2012 Description: Factor lu_site procfs stats into a separate function.
2013 Details : Separate lu_site stats printing code into a separate function
2014 that can be reused on a client.
2018 Description: Constify instances of struct {lu,dt,md}_device_operations.
2019 Details : Constify instances of struct {lu,dt,md}_device_operations.
2023 Description: Introduce struct md_site and move meta-data specific parts of
2024 struct lu_site here.
2025 Details : Move md-specific fields out of struct lu_site into special struct
2026 md_site, so that lu_site can be used on a client.
2030 Description: Kill mdd_lov_destroy().
2031 Details : Remove unused mdd code.
2035 Description: Add st_block checking to multistat.c.
2036 Details : Add st_block checking to multistat.c.
2040 Description: Add lu_ref support to struct obd_device.
2041 Details : Add lu_ref tracking to obd_device.
2045 Description: Kill obd_set_fail_loc().
2046 Details : Remove unused code.
2050 Description: Add special type for ptlrpc_request interpret functions.
2051 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
2052 there. Actual usage will be in CLIO.
2056 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
2057 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
2058 doesn't work with lockdep.
2062 Description: Add rwv.c test program.
2063 Details : New testing program exercising readv(2) and writev(2) (Qian).
2067 Description: Add sendfile.c test program.
2068 Details : New testing program exercising sendfile(2) (Jay).
2072 Description: Ratelimit a message that can be very frequent.
2073 Details : Ratelimit a memory allocation failure message that can
2078 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2079 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2083 Description: Liblustre build fixes.
2084 Details : Liblustre build fixes.
2088 Description: libcfs: add cfs_{need,cond}_resched() interface.
2089 Details : libcfs: add cfs_{need,cond}_resched() definition and
2090 implementations for Linux, NT, and liblustre.
2092 Severity : enhancement
2094 Description: More exported tunables for mballoc
2095 Details : Add support for tunable preallocation window and new tunables for
2096 large/small requests
2100 Description: Detect corruption of block bitmap and checking for preallocations
2101 Details : Checks validity of on-disk block bitmap. Also it does better
2102 checking of number of applied preallocations. When corruption is
2103 found, it turns filesystem readonly to prevent further corruptions.
2107 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2108 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2109 window to 0. If it does hit cache miss, and read-pattern is still
2110 stride-io mode, does not reset the stride window, but also does not
2111 increase the stride window length in this case.
2115 Frequency : only for big-endian servers
2116 Description: Check if system is big-endian while mounting fs with extents feature
2117 Details : Mounting a filesystem with extents feature will fail on big-endian
2118 systems since ext3-based ldiskfs is not supported on big-endian
2119 systems. This can be over-riden with "bigendian_extents" mount option.
2121 Severity : enhancement
2123 Description: The root squash functionality
2124 Details : A security feature, which is to prevent users from being able
2125 to mount lustre on their desktop, run as root, and delete
2126 all of the files in the filesystem. The goal is accomplished by
2127 remapping user id (UID) and group id (GID) of the root user to
2128 a UID and GID specified by the system administartor via Lustre
2129 configuration management server (MGS). The functionality also
2130 allows to specify sets of clients for which the remapping does
2135 Description: Excessive recovery window
2136 Details : With AT enabled, the recovery window can be excessively long (6000+
2137 seconds). To address this problem, we no longer use
2138 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2139 timeout no longer depends on the service time, it is set to
2140 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2141 time via pb_service_time.
2145 Description: Watchdog triggered on MDS failover
2146 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2147 the OSTs know that the MDS "UUID" can be reused for the same export
2148 from a different NID, so we do not need to wait for the export to be
2152 Frequency : rare, only if using MMP with Linux RAID
2154 Description: MMP doesn't work with Linux RAID
2155 Details : While using HA for Lustre servers with Linux RAID, it is possible
2156 that MMP will not detect multiple mounts. To make this work we
2157 need to unplug the device queue in RAID when the MMP block is being
2158 written. Also while reading the MMP block, we should read it from
2159 disk and not the cached one.
2161 Severity : enhancement
2163 Description: open file using fid
2164 Details : A file can be opened using just its fid, like
2165 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2168 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2171 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2172 Details : add proper locking for creation of jbd_4k slab cache
2176 Description: MMP check in ext3_remount() fails without displaying any error
2177 Details : When multiple mount protection fails during remount, proper error
2180 Severity : enhancement
2182 Description: Allow stripe size to be up to 4G-64k
2183 Details : Fix math logic to allow large stripe sizes.
2187 Description: add check for >8TB ldiskfs filesystems
2188 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2189 Don't allow >8TB ldiskfs filesystems to be mounted without
2190 force_over_8tb mount option
2195 Descriptoin: OSS complains "too many blocks" and crashes
2196 Details : In filter_commitrw_write():
2197 1. lquota_chkquota();
2198 2. obdo_from_inode();
2199 3. lquota_pending_commit();
2200 "2." has changed uid and gid in this case, then caused
2201 lquota_chkquota() and lquota_pending_commit() to use
2202 different uid and gid. That is the root of the bug.
2204 --------------------------------------------------------------------------------
2206 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2208 * Support for kernels:
2209 2.6.5-7.283 (SLES 9),
2210 2.6.9-55.EL (RHEL 4),
2211 2.6.16.46-0.14 (SLES 10),
2212 2.6.18.8 vanilla (kernel.org)
2213 * Client support for unpatched kernels:
2214 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2215 2.6.16 - 2.6.22 vanilla (kernel.org)
2216 * Due to recently discovered recovery problems, we do not recommend
2217 using patchless RHEL 4 clients with this or any earlier release.
2218 * Recommended e2fsprogs version: 1.39.cfs8
2219 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2220 * Starting with this release, the ldiskfs backing filesystem required
2221 by Lustre is now in its own package, lustre-ldiskfs. This package
2222 should be installed. It is versioned separately from Lustre and
2223 may be released separately in future.
2227 Description: block reactivating mgc import until all deactivates complete
2228 Details : Fix race when failing back MDT/MGS to itself (testing)
2231 Frequency : at statup only
2233 Description: mds_lov_synchronize race leads to various problems
2234 Details : simultaneous MDT->OST connections at startup can cause the
2235 sync to abort, leaving the OSC in a bad state.
2237 Severity : enhancement
2239 Description: add optional extra BUILD_VERSION info
2240 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2241 to override the lustre version.
2244 Frequency : 2.6.18 servers only
2246 Description: ll_kern_mount() doesn't release the module reference
2247 Details : The ldiskfs module reference count never drops down to 0
2248 because ll_kern_mount() doesn't release the module reference.
2253 Description: server LBUG when using old ost_num_threads parameter
2254 Details : Accept the old ost_num_threads parameter but warn that it
2255 is deprecated, and fix an off-by-one error that caused an LBUG.
2260 Description: Transient SCSI error results in persistent IO issue
2261 Details : iobuf->dr_error is not reinitialized to 0 between two
2265 Frequency : sometimes when underlying device returns I/O errors
2267 Description: OSTs not going read-only during write failures
2268 Details : OSTs are not remounted read-only when the journal commit threads
2269 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2270 instead of the ext3 wrappers.
2274 Description: poor connect scaling with increasing client count
2275 Details : Don't run filter_grant_sanity_check for more than 100 exports
2276 to improve scaling for large numbers of clients.
2279 Frequency : SLES10 only
2281 Description: sanity-quota.sh quotacheck failed: rc = -22
2282 Details : Quotas cannot be enabled on SLES10.
2285 Frequency : liblustre clients only
2287 Description: getdirentries does not give error when run on compute nodes
2288 Details : getdirentries does not fail when the size specified as an argument
2289 is too small to contain at least one entry
2291 Severity : enhancement
2293 Description: Add LNET router traceability for debug purposes
2294 Details : If a checksum failure occurs with a router as part of the
2295 IO path, the NID of the last router that forwarded the bulk data
2296 is printed so it can be identified.
2301 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2302 Details : Due to a race condition, liblustre clients were occasionally
2303 evicted incorrectly.
2305 Severity : enhancement
2307 Description: lfs setstripe use optional parameters instead of postional
2310 Severity : enhancement
2312 Description: Nanosecond timestamp support for ldiskfs
2313 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2314 resolution timestamps. There is not yet support for this at
2315 the Lustre filesystem level.
2318 Frequency : during server recovery
2320 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2321 Details : request with rq_no_resend flag not awake l_wait_event if they get
2325 Frequency : nfs export on patchless client
2327 Description: connectathon hang when test nfs export over patchless client
2328 Details : Disconnected dentry cannot be found with lookup, so we do not need
2329 to unhash it or make it invalid
2332 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2333 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2334 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2339 Description: open req refcounting wrong on reconnect
2340 Details : If reconnect happened between getting open reply from server and
2341 call to mdc_set_replay_data in ll_file_open, we will schedule
2342 replay for unreferenced request that we are about to free.
2343 Subsequent close will crash in variety of ways.
2344 Check that request is still eligible for replay in
2345 mdc_set_replay_data().
2350 Description: disable writes to filesystem when reading health_check file
2351 Details : the default for reading the health_check proc file has changed
2352 to NOT do a journal transaction and write to disk, because this
2353 can cause reads of the /proc file to hang and block HA state
2354 checking on a healthy but otherwise heavily loaded system. It
2355 is possible to return to the previous behaviour during configure
2356 with --enable-health-write.
2358 Severity : enhancement
2360 Description: 64-bit inode version
2361 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2362 to the inode. This will be required for version-based recovery.
2367 Description: MDS fails to start if a duplicate client export is detected
2368 Details : in some rare cases it was possible for a client to connect to
2369 an MDS multiple times. Upon recovery the MDS would detect this
2370 and fail during startup. Handle this more gracefully.
2372 Severity : enhancement
2374 Description: Add -o localflock option to simulate old noflock
2376 Details : This will achieve local-only flock/fcntl locks
2382 Description: log_commit_thread vs filter_destroy race leads to crash
2383 Details : Take import reference before releasing llog record semaphore
2388 Description: Wrong request locking in request set processing
2389 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2390 imp_delayed_list, in this place should be used imp_lock.
2393 Frequency : when reconnection
2395 Description: Grant Leak when osc reconnect to OST
2396 Details : When osc reconnect ost, OST(filter) should check whether it
2397 should grant more space to client by comparing fed_grant and
2398 cl_avail_grant, and return the granted space to client instead
2399 of "new granted" space, because client will call osc_init_grant
2400 to update the client grant space info.
2403 Frequency : when client reconnect to OST
2405 Description: Grant Leak when osc do resend and replay bulk write
2406 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2407 bulk write request, because the grant info while be sync between
2408 OSC and OST when reconnect, and we should ignore the grant info
2409 these of resend/replay write req.
2414 Description: Grant space more than avaiable left space sometimes.
2415 Details : When then OST is about to be full, if two bulk writing from
2416 different clients came to OST. Accord the avaliable space of the
2417 OST, the first req should be permitted, and the second one
2418 should be denied by ENOSPC. But if the seconde arrived before
2419 the first one is commited. The OST might wrongly permit second
2420 writing, which will cause grant space > avaiable space.
2423 Frequency : when client is evicted
2425 Description: Grant might be wrongly erased when osc is evicted by OST
2426 Details : when the import is evicted by server, it will fork another
2427 thread ptlrpc_invalidate_import_thread to invalidate the
2428 import, where the grant will be set to 0. While the original
2429 thread will update the grant it got when connecting. So if
2430 the former happened latter, the grant will be wrongly errased
2431 because of this race.
2436 Description: Checking Stale with correct fid
2437 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2438 to check whether it is stale, because sometimes, we want the
2439 enqueue happened anyway, and op_data.fid2 will not be initialized.
2441 Severity : enhancement
2443 Description: update patchless client
2444 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2447 Frequency : only with 2.4 kernel
2449 Description: random memory corruption
2450 Details : size of struct ll_inode_info is to big for union inode.u and this
2451 can be cause of random memory corruption.
2456 Description: Memory leak in recovery
2457 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2458 It should also check obd_fail before fsfilt_start, otherwise if
2459 fsfilt_start return -EROFS,(failover mds during mds recovery).
2460 then the req will return with repmsg->transno = 0 and rc = EROFS.
2461 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2462 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2463 be freed no matter whether fsfilt_commit success or not.
2466 Frequency : only with huge count clients
2468 Description: Prevents from taking the superblock lock in llap_from_page for
2470 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2471 for llap_from_page call in ll_removepage prevents from taking the
2472 superblock lock for a soon died page.
2477 Description: Not check open intent error before release open handle
2478 Details : in some rare cases, the open intent error is not checked before
2479 release open handle, which may cause
2480 ASSERTION(open_req->rq_transno != 0), because it tries to release
2481 the failed open handle.
2486 Description: Set cat log bitmap only after create log success.
2487 Details : in some rare cases, the cat log bitmap is set too early. and it
2488 should be set only after create log success.
2492 Description: Accessing a block bevice can re-enable I/O when Lustre is
2493 tearing down a device.
2494 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2499 Description: service threads may hog cpus when there are a lot of requests
2500 Details : Insert cond_resched to give other threads a chance to use some CPU
2505 Description: the cat log was not initialized in recovery
2506 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2507 unlink log on mds will not be initialized until mds post
2508 recovery. And also in mds post recovery, the unlink log will
2509 initialization will be done asynchronausly, so there will be race
2510 between add unlink log and unlink log initialization.
2514 Description: brw_stats were being printed incorrectly
2515 Details : brw_stats were being printed as log2 but all of them were not
2516 recorded as log2. Also remove some code duplication arising from
2517 filter_tally_{read,write}.
2521 Frequency : rare, only in recovery.
2522 Description: ASSERTION(req->rq_type != LI_POISON) failed
2523 Details : imp_lock should be held while iterating over imp_sending_list for
2524 prevent destroy request after get timeout in ptlrpc_queue_wait.
2528 Description: replay-single.sh test 52 fails
2529 Details : A lock's skiplist need to be cleanup when it being unlinked
2530 from its resource list.
2534 Description: Short directio read returns full requested size rather than
2536 Details : Direct I/O operations should return actual amount of bytes
2537 transferred rather than requested size.
2539 Severity : enhancement
2541 Description: metadata RPC reduction (e.g. for rm performance)
2542 Details : decrease the amount of synchronous RPC between clients and servers
2543 by canceling conflicing lock before the operation on the client
2544 and packing thier handles into the main operation RPC to server.
2546 Severity : enhancement
2548 Description: Async OSC create to avoid the blocking unnecessarily.
2549 Details : If a OST has no remain object, system will block on the creating
2550 when need to create a new object on this OST. Now, ways use
2551 pre-created objects when available, instead of blocking on an
2552 empty osc while others are not empty. If we must block, we block
2553 for the shortest possible period of time.
2557 Description: improve handling recoverable errors
2558 Details : if request processig with error which can be recoverable on server
2559 request should be resend, otherwise page released from cache and
2562 Severity : enhancement
2564 Description: refine locking for avoid write wrong info into lov_objid file
2565 Details : fix possible races with add new target and write/update data in
2568 --------------------------------------------------------------------------------
2570 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2575 Frequency : on some architectures
2577 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2578 Details : Uninitialized flags sometimes cause configuration commands to
2582 Frequency : patchless clients only
2584 Description: missing __iget() symbol export
2585 Details : The __iget() symbol export is missing. To avoid the need for
2586 this on patchless clients the deathrow inode reaper is turned
2587 off, and we depend on the VM to clean up old inodes. This
2588 dependency was during via the fix for bug 12181.
2590 --------------------------------------------------------------------------------
2592 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2594 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2595 INTEROPERATE with older versions automatically. In many cases a
2596 special upgrade step is needed. Please read the
2597 user documentation before upgrading any part of a live system.
2598 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2599 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2600 * WARNING: Lustre configuration and startup changes are required with
2601 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2605 Severity : enhancement
2607 Description: MountConf
2608 Details : Lustre configuration is now managed via mkfs and mount
2609 commands instead of lmc and lconf. New obd types (MGS, MGC)
2610 are added for dynamic configuration management. See
2611 https://mail.clusterfs.com/wikis/lustre/MountConf for
2614 Severity : enhancement
2616 Description: dynamic OST addition
2617 Details : OSTs can now be added to a live filesystem
2619 Severity : enhancement
2621 Description: startup order invariance
2622 Details : MDTs and OSTs can be started in any order. Clients only
2623 require the MDT to complete startup.
2625 Severity : enhancement
2627 Description: parallel, asynchronous orphan cleanup
2628 Details : orphan cleanup is now performed in separate threads for each
2629 OST, allowing parallel non-blocking operation.
2631 Severity : enhancement
2633 Description: optimized stripe assignment
2634 Details : stripe assignments are now made based on ost space available,
2635 ost previous usage, and OSS previous usage, in order to try
2636 to optimize storage space and networking resources.
2638 Severity : enhancement
2640 Description: Permanently set tunables
2641 Details : All writable /proc/fs/lustre tunables can now be permanently
2642 set on a per-server basis, at mkfs time or on a live system.
2644 Severity : enhancement
2646 Description: Lustre message v2
2647 Details : Add lustre message format v2.
2649 Severity : enhancement
2651 Description: client OST exclusion list
2652 Details : Clients can be started with a list of OSTs that should be
2653 declared "inactive" for known non-responsive OSTs.
2657 Description: SPEC SFS validation failure on NFS v2 over lustre.
2658 Details : Changes the blocksize for regular files to be 2x RPC size,
2659 and not depend on stripe size.
2661 Severity : enhancement
2663 Description: Multiple MD RPCs in flight.
2664 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2665 To avoid overly-overloading MDS, introduce a limit on number of
2666 MDS RPCs in flight for a single client and add /proc controls
2667 to adjust this limit.
2669 Severity : enhancement
2671 Description: client read/write statistics
2672 Details : Add client read/write call usage stats for performance
2673 analysis of user processes.
2674 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2675 file access. extents_stats shows chunk size distribution.
2676 extents_stats_per_process show chunk size distribution per
2679 Severity : enhancement
2681 Description: mds statistics
2682 Details : Add detailed mds operations statistics in
2683 /proc/fs/lustre/mds/*/stats.
2687 Description: Failure of copying files with lustre special EAs.
2688 Details : Client side always return success for setxattr call for lustre
2689 special xattr (currently only "trusted.lov").
2694 Description: Refcount LNET uuids
2695 Details : The global LNET uuid list grew linearly with every startup;
2696 refcount repeated list entries instead of always adding to
2699 Severity : enhancement
2701 Description: Dynamic service threads
2702 Details : Within a small range, start extra service threads
2703 automatically when the request queue builds up.
2706 Frequency : mixed-endian client/server environments
2708 Description: mixed-endian crashes
2709 Details : The new msg_v2 system had some failures in mixed-endian
2712 Severity : enhancement
2714 Description: Easy OST removal
2715 Details : OSTs can be permanently deactivated with e.g. 'lctl
2716 conf_param lustre-OST0001.osc.active=0'
2718 Severity : enhancement
2720 Description: MGS proc entries
2721 Details : Added basic proc entries for the MGS showing what filesystems
2724 Severity : enhancement
2726 Description: provide MGS failover
2727 Details : Added config lock reacquisition after MGS server failover.
2729 Severity : enhancement
2731 Description: add Linux 2.4 support
2732 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2736 Description: a large application tries to do I/O to the same resource and dies
2737 in the middle of it.
2738 Details : Check the req->rq_arrival time after the call to
2739 ost_brw_lock_get(), but before we do anything about
2740 processing it & sending the BULK transfer request. This
2741 should help move old stale pending locks off the queue as
2742 quickly as obd_timeout.
2745 Frequency : when an incorrect nid is specified during startup
2747 Description: ptlrpc connect to non-existant node causes kernel crash
2748 Details : LNET can't be re-entered from an event callback, which
2749 happened when we expire a message after the export has been
2750 cleaned up. Instead, hand the zombie cleanup off to another
2753 Severity : enhancement
2755 Description: plain/inodebits lock performance improvement
2756 Details : Grouping plain/inodebits in granted list by their request modes
2757 and bits policy, thus improving the performance of search through
2761 Frequency : only if OST filesystem is corrupted
2763 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2764 Details : for a short time RPCs with bulk IO are in the replay list,
2765 but replay of bulk IOs is unimplemented. If the OST filesystem
2766 is corrupted due to disk cache incoherency and then replay is
2767 started it is possible to trip an assertion. Avoid putting
2768 committed RPCs into the replay list at all to avoid this issue.
2771 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2773 Description: System hang on startup
2774 Details : This bug allowed the liblustre (e.g. catamount) client to
2775 return to the app before handling all startup RPCs. This
2776 could leave the node unresponsive to lustre network traffic
2777 and manifested as a server ptllnd timeout.
2779 Severity : enhancement
2781 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2782 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2783 boolean to control whether to print peer debug info when a
2784 client's RPC times out.
2787 Frequency : only for kernels with patches from Lustre below 1.4.3
2789 Description: Remove old rdonly API
2790 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2793 Frequency : only for devices with external journals
2795 Description: Set external device read-only also
2796 Details : During a commanded failover stop, we set the disk device
2797 read-only while the server shuts down. We now also set any
2798 external journal device read-only at the same time.
2801 Frequency : when upgrading from 1.4 while trying to change parameters
2803 Description: The wrong (new) MDC name was used when setting parameters for
2804 upgraded MDT's. Also allows changing of OSC (and MDC)
2805 parameters if --writeconf is specified at tunefs upgrade time.
2808 Frequency : when setting specific ost indicies
2810 Description: QOS code breaks on skipped indicies
2811 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2812 created with --index need not be sequential.
2814 Severity : enhancement
2816 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2817 Details : The uninit_groups feature works in conjunction with the kernel
2818 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2819 up the pass1 processing of e2fsck. This is a read-only feature
2820 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2821 mount filesystems that have had this feature enabled.
2823 Severity : enhancement
2825 Description: Improve multi-block allocation algorithm to avoid fragmentation
2826 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2827 allocation locality and avoid filesystem fragmentation.
2829 ------------------------------------------------------------------------------
2831 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2833 * Support for kernels:
2834 2.4.21-47.0.1.EL (RHEL 3)
2835 2.6.5-7.283 (SLES 9)
2836 2.6.9-42.0.10.EL (RHEL 4)
2837 2.6.12.6 vanilla (kernel.org)
2838 2.6.16.27-0.9 (SLES 10)
2839 * Recommended e2fsprogs version: 1.39.cfs5
2841 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2845 Frequency : occasional, depends on client load and configuration
2846 Bugzilla : 12181, 12203
2847 Description: data loss for recently-modified files
2849 Details : In some cases it is possible that recently written or created
2850 files may not be written to disk in a timely manner (this should
2851 normally be within 30s unless client IO load is very high).
2852 The problem appears as zero-length files or files that are a
2853 multiple of 1MB in size after a client crash or client eviction
2854 that are missing data at the end of the file.
2856 This problem is more likely to be hit on clients where files are
2857 repeatedly created and unlinked in the same directory, clients
2858 have a large amount of RAM, have many CPUs, the filesystem has
2859 many OSTs, the clients are rebooted frequently, and/or the files
2860 are not accessed by other nodes after being written.
2862 The presence of the problem can be detected by looking at
2863 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2864 smaller than the second (nr_unused) then dirty files will not
2865 be flushed automatically to disk. "sync; sleep 10" should be
2866 run several times on the node before unmounting it to update
2867 Lustre (this is also safe to run on nodes without this problem).
2869 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2870 that can cause this same problem, so customers using that kernel
2871 also need to update the kernel in addition to Lustre. In order
2872 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2874 It is normal that files written just before a client crash (less
2875 than 30s) may not yet have been flushed to disk, even for local
2879 Frequency : frequent on thin XT3 nodes
2881 Description: UUID collision on thin XT3 Linux nodes
2882 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2883 randomly, since we relied on an insufficiently-seeded PRNG.
2888 Description: OSS hangs after "All ost request buffers busy"
2889 Details : A deadlock between quota and journal operations caused OSS
2890 hangs after printing "All ost request buffers busy."
2893 Frequency : always on liblustre builds
2895 Description: Cleanup compiler warnings on liblustre
2898 Frequency : always on liblustre builds on XT3
2900 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2904 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2905 > 32000 subdirectories
2906 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2907 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2908 subdirectory count crosses 32000. This will aid e2fsck to
2909 correctly handle more than 32000 subdirectories.
2912 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2914 Description: System hang on startup
2915 Details : This bug allowed the liblustre (e.g. catamount) client to
2916 return to the app before handling all startup RPCs. This
2917 could leave the node unresponsive to lustre network traffic
2918 and manifested as a server ptllnd timeout.
2920 Severity : enhancement
2922 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2923 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2924 boolean to control whether to print peer debug info when a
2925 client's RPC times out.
2930 Description: make O_SYNC working on 2.6 kernels
2931 Details : 2.6 kernels use different method for mark pages for write,
2932 so need add a code to lustre for O_SYNC work.
2937 Description: Failure to close file and release space on NFS
2938 Details : Put inode details into lock acquired in ll_intent_file_open.
2939 Use mdc_intent_lock in ll_intent_open to properly
2940 detect all kind of errors unhandled by mdc_enqueue.
2945 Description: proc file read during shutdown sometimes raced obd removal,
2947 Details : Add lock to prevent obd access after proc file removal.
2950 Frequency : Only for files larger than 4GB on 32-bit clients.
2952 Description: improperly doing page alignment of locks
2953 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2954 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2959 Description: RPCs being resent when they shouldn't be
2960 Details : Some RPCs that should not be resent are being resent. This
2961 can cause inconsistencies in the RPC state machine. Do not
2962 resend such requests.
2965 Frequency : rare, only with NFS export
2967 Description: Crash on NFS re-export node
2968 Details : under very unusual load conditions an assertion is hit in
2969 ll_intent_file_open()
2972 Frequency : only if OST filesystem is corrupted
2974 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2975 Details : for a short time RPCs with bulk IO are in the replay list,
2976 but replay of bulk IOs is unimplemented. If the OST filesystem
2977 is corrupted due to disk cache incoherency and then replay is
2978 started it is possible to trip an assertion. Avoid putting
2979 committed RPCs into the replay list at all to avoid this issue.
2984 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2985 Details : Large single O_DIRECT read and write calls can fail to allocate
2986 a sufficiently large buffer to process the request. In case of
2987 allocation failure the allocation is retried with a smaller
2988 buffer and broken into smaller requests.
2990 Severity : enhancement
2992 Description: Add -o localflock option to simulate old noflock behaviour.
2993 Details : This will achieve local-only flock/fcntl locks coherentness.
2998 Description: versioning check is incomplete
2999 Details : Checking the version difference of client vs. server, report
3000 error if the gap is too big.
3005 Description: add support PG_writeback bit
3006 Details : add support for PG_writeback bit for Lustre, for more carefull
3007 work with page cache in 2.6 kernel. This also fix some deadlocks
3008 and remove hack for work O_SYNC with 2.6 kernel.
3010 Severity : enhancement
3012 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3013 Details : The uninit_groups feature works in conjunction with the kernel
3014 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3015 up the pass1 processing of e2fsck. This is a read-only feature
3016 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3017 mount filesystems that have had this feature enabled.
3019 Severity : enhancement
3021 Description: Improve multi-block allocation algorithm to avoid fragmentation
3022 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3023 allocation locality and avoid filesystem fragmentation.
3025 ------------------------------------------------------------------------------
3027 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
3029 * Support for kernels:
3030 2.6.9-42.0.3.EL (RHEL 4)
3031 2.6.5-7.276 (SLES 9)
3032 2.4.21-47.0.1.EL (RHEL 3)
3033 2.6.12.6 vanilla (kernel.org)
3034 2.6.16.21-0.8 (SLES10)
3035 * Recommended e2fsprogs version: 1.39.cfs2-0
3037 * The backwards-compatible /proc/sys/portals symlink has been removed
3038 in this release. Before upgrading, please ensure that you change
3039 any configuration scripts or /etc/sysctl.conf files that access
3040 /proc/sys/portals/* or sysctl portals.* to use the corresponding
3041 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
3042 in advance of the upgrade on any system running Lustre 1.4.6 or
3043 newer, since /proc/sys/lnet was added in that version.
3044 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3048 Frequency : only when quota is used
3050 Description: avoid scanning export list for quota master
3051 Details : Change the algorithms to avoid scanning export list in order
3052 to improve the efficiency.
3055 Frequency : MDS failover only, very rarely
3057 Description: "went back in time" messages on mds failover
3058 Details : The greatest transno may be lost when the current operation
3059 finishes with an error (transno==0) and the client's last_rcvd
3060 record is over-written. Save the greatest transno in the
3061 mds_last_transno for this case.
3064 Frequency : always for specific kernels and striping counts
3066 Description: client may get "Matching packet too big" without ACL support
3067 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
3068 when trying to access files in certain configurations. The
3069 clients should in fact be denied when mounting because they do
3070 not understand ACLs.
3073 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3075 Description: many clients connecting with IO in progress causes connect timeouts
3076 Details : Avoid synchronous journal commits to avoid delays caused by many
3077 clients connecting/disconnecting when bulk IO is in progress.
3078 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3079 OST_IO_PORTAL to avoid delays behind potentially many pending
3083 Frequency : occasionally with multiple writers to a single file
3085 Description: shared writes to file may result in wrong size reported by stat()
3086 Details : Allow growing of kms when extent lock is cancelled
3089 Frequency : always with random mmap IO to multi-striped file
3091 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3092 Details : Only if the hole is at the end of OST object so that kms is too
3093 small. Fix is to increase kms accordingly in ll_nopage.
3096 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3098 Description: writes to a missing object would leak memory on the OST
3099 Details : If there is an inconsistency between the MDS and OST filesystems,
3100 such that the MDS references an object that doesn't exist, writes
3101 to that object will leak memory due to incorrect cleanup in the
3102 error handling path, eventually running out of memory on the OST.
3107 Description: Creating too long symlink causes lustre errors
3108 Details : Check symlink and name lengths before sending requests to MDS.
3111 Frequency : only if flock is enabled (not on by default)
3113 Description: posix locks not released on fd closure on 2.6.9+
3114 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3115 kernels, this caused such locks not to be released on fd close and
3116 then assertions on fs unmount about still used locks.
3119 Frequency : MDS failover only, very rarely
3121 Description: clients may get ASSERTION(granted_lock != NULL)
3122 Details : When request was taking a long time, and a client was resending
3123 a getattr by name lock request. The were multiple lock requests
3124 with the same client lock handle and
3125 mds_getattr_name->fixup_handle_for_resent_request found one of the
3126 lock handles but later failed with ASSERTION(granted_lock != NULL).
3131 Description: handle->h_buffer_credits > 0, assertion failure
3132 Details : h_buffer_credits is zero after truncate, causing assertion
3133 failure. This patch extends the transaction or creates a new
3137 Frequency : NFS re-export or patchless client
3138 Bugzilla : 11179, 10796
3139 Description: Crash on NFS re-export node (__d_move)
3140 Details : We do not want to hash the dentry if we don't have a lock.
3141 But if this dentry is later used in d_move, we'd hit uninitialised
3142 list head d_hash, so we just do this to init d_hash field but
3143 leave dentry unhashed.
3146 Frequency : NFS re-export or patchless client
3148 Description: NFS exports has problem with symbolic link
3149 Details : lustre client didn't properly install dentry when re-exported
3150 to NFS or running patchless client.
3153 Frequency : NFS re-export or patchless client
3155 Description: Various nfs/patchless fixes.
3156 Details : fixes reuse disconected alias for lookup process - this fixes
3157 warning "find_exported_dentry: npd != pd",
3158 fix permission error with open files at nfs.
3159 fix apply umask when do revalidate.
3162 Frequency : occasional
3164 Description: Crash on NFS re-export node
3165 Details : calling clear_page() on the wrong pointer triggered oops in
3166 generic_mapping_read().
3169 Frequency : rarely, using O_DIRECT IO
3171 Description: unaligned directio crashes client with LASSERT
3172 Details : check for unaligned buffers before trying any requests.
3175 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3177 Description: stale data returned from RAID cache
3178 Details : If only a small amount of IO is done to the RAID device before
3179 reading it again it is possible to get stale data from the RAID
3180 cache instead of reading it from disk.
3183 Frequency : always for sles10 kernel
3185 Description: sles10 support
3186 Details : ll_follow_link: compile fixes and using of nd_set_link
3187 under newer kernels.
3190 Frequency : depends on arch, kernel and compiler version, always on sles10
3193 Description: recursive or deep enough symlinks cause stack overflow
3194 Details : getting rid of large stack-allocated variable in
3198 Frequency : depends on hardware
3200 Description: lustre write performance loss in the SLES10 kernel
3201 Details : the performance loss is caused by using of write barriers in the
3202 ext3 code. The SLES10 kernel turns barrier support on by
3203 default. The fix is to undo that change for ldiskfs.
3205 ------------------------------------------------------------------------------
3207 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3209 * Support for kernels:
3210 2.6.9-42.0.3EL (RHEL 4)
3211 2.6.5-7.276 (SLES 9)
3212 2.4.21-47.0.1.EL (RHEL 3)
3213 2.6.12.6 vanilla (kernel.org)
3217 Frequency : quota enabled and large files being deleted
3219 Description: releasing more than 4GB of quota at once hangs OST
3220 Details : If a user deletes more than 4GB of files on a single OST it
3221 will cause the OST to spin in an infinite loop. Release
3222 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3227 Description: statfs data retrieved from /proc may be stale or zero
3228 Details : When reading per-device statfs data from /proc, in the
3229 {kbytes,files}_{total,free,avail} files, it may appear
3230 as zero or be out of date.
3233 Frequency : systems with MD RAID1 external journal devices
3235 Description: lconf's call to blkid is confused by RAID1 journal devices
3236 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3237 of returning all block devices that match the journal UUID.
3240 Frequency : always, for aggregate stripe size over 4GB
3242 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3243 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3244 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3247 Frequency : always if "lfs find" used on a local file/directory
3249 Description: "lfs find" segfaults if used on a local file/directory
3250 Details : The case where a directory component was not specified wasn't
3251 handled correctly. Handle this properly.
3254 Frequency : always on ppc64
3256 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3257 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3258 perform endianness conversion.
3261 Frequency : rarely (truncate to non-zero file size after write under load)
3262 Bugzilla : 10730, 10687
3263 Description: Files padded with zeros to next 4K multiple
3264 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3265 it is possible that files that are truncated to a non-zero size
3266 immediately after being written are filled with zero bytes beyond
3267 the truncated size. No file data is lost.
3269 Severity : enhancement
3271 Description: Allow recovery/failover for liblustre clients.
3272 Details : liblustre clients were unaware of failover configurations until
3275 Severity : enhancement
3277 Description: user file locks should fail when not mounting with flock option
3278 Details : Set up an error-returning stub in ll_file_operations.lock field
3279 to prevent incorrect behaviour when client is mounted without
3280 flock option. Also, set up properly f_op->flock field for
3284 Frequency : always on ia64
3286 Description: "lfs df" loops on printing out MDS statfs information
3287 Details : The obd_ioctl_data was not initialized and in some systems
3288 this caused a failure during the ioctl that did not return
3289 an error. Initialize the struct and return an error on failure.
3292 Frequency : SLES 9 only
3294 Description: Error of copying files with lustre special EAs as root
3295 Details : Client side always return success for setxattr call for lustre
3296 special xattr (currently only "trusted.lov").
3299 Frequency : rarely on clusters with both ia64+i386 clients
3301 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3302 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3303 clients are doing concurrent writes to the same file it is
3304 possible that the ia64 clients may LASSERT because the OST
3305 extent locks are not PAGE_SIZE aligned. Ensure that grown
3306 locks are always aligned on the request boundary.
3309 Frequency : specific use, occasional
3311 Description: Overwriting in use executable truncates on-disk binary image
3312 Details : If one node attempts to overwrite an executable in use by
3313 another node, we now correctly return ETXTBSY instead of
3314 truncating the file.
3316 Severity : enhancement
3318 Description: Async OSC create to avoid the blocking unnecessarily.
3319 Details : If a OST has no remain object, system will block on the creating
3320 when need to create a new object on this OST. Now, ways use
3321 pre-created objects when available, instead of blocking on an
3322 empty osc while others are not empty. If we must block, we block
3323 for the shortest possible period of time.
3328 Description: chmod on Lustre root is propagated to other clients
3329 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3330 invalid by the follow_mount time.
3333 Frequency : liblustre clients only
3335 Description: Race in 'instant cancel' lock handling could lead to such locks
3336 never to be granted in case of SMP MDS
3337 Details : Do not destroy not yet granted but cbpending locks in
3341 Frequency : replay/resend of open
3343 Description: non null lock assetion failure in mds_intent_policy
3344 Details : Trying to replay/resend lockless open requests resulted in
3345 mds_open() returning 0 with no lock. Now it sets a flag if
3346 a lock is going to be returned.
3348 Severity : enhancement
3350 Description: Checksum enhancements
3351 Details : New checksum enhancements allow for resending RPCs that failed
3354 Severity : enhancement
3356 Description: Tunables on number of dirty pages in cacche
3357 Details : Allow to set limit on number of dirty pages cached.
3362 Description: client crash on unmount - lock still has references
3363 Details : In some error handling cases it was possible to leak a lock
3364 reference on a client while accessing a file. This was not
3365 harmful to the client during operation, but would cause the
3366 client to crash when the filesystem is unmounted.
3369 Frequency : specific case, rare
3371 Description: ETXTBSY on mds though file not in use by client
3372 Details : ETXTBSY is no longer incorrectly returned when attempting to
3373 chmod or chown a directory that the user previously tried to
3374 execute or a currently-executing binary.
3377 Frequency : extremely rare except on liblustre-based clients
3379 Description: Lustre space not freed when files are deleted
3380 Details : Clean up open-unlinked files after client eviction. Previously
3381 the unlink was skipped and the files remained as orphans.
3386 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3387 Details : In some cases it was possible to send a blocking callback to
3388 a client doing a glimpse, even though that client didn't get
3389 a lock granted. When the glimpse lock is cancelled on the OST
3390 the freed lock is left on the waiting list and corrupted the list.
3393 Frequency : all core dumps
3395 Description: Broke core dumps to lustre
3396 Details : Negative dentry may be unhashed if parent does not have UPDATE
3397 lock, but some callers, e.g. do_coredump, expect dentry to be
3398 hashed after successful create, hash it in ll_create_it.
3400 ------------------------------------------------------------------------------
3402 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3404 * Support for kernels:
3405 2.6.9-42.0.2.EL (RHEL 4)
3406 2.6.5-7.276 (SLES 9)
3407 2.4.21-40.EL (RHEL 3)
3408 2.6.12.6 vanilla (kernel.org)
3412 Frequency : always on RHEL 3
3414 Description: Number of open files grows over time
3415 Details : The number of open files grows over time, whether or not
3416 Lustre is started. This was due to a filp leak introduced
3417 by one of our kernel patches.
3419 ------------------------------------------------------------------------------
3421 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3423 * Support for kernels:
3424 2.6.9-42.EL (RHEL 4)
3425 2.6.5-7.276 (SLES 9)
3426 2.4.21-40.EL (RHEL 3)
3427 2.6.12.6 vanilla (kernel.org)
3432 Bugzilla : 5719, 9635, 9792, 9684
3433 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3434 Details : If a server is under heavy load and cannot reply to new
3435 connection requests before the client resends the (re)connect,
3436 the connection handling code can behave badly if two service
3437 threads are concurrently handing separate (re)connections from
3438 the same client. Add better locking to the connection handling
3439 code, and ensure that only a single connection will be processed
3440 for a given client UUID, even if the lock is dropped.
3442 Severity : enhancement
3444 Description: add TCP zero-copy support to kernel
3445 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3446 sends if the hardware supports scatter-gather and checksumming.
3447 This allows socklnd to do client-write and server-read more
3448 efficiently and reduce CPU utilization from skbuf copying.
3451 Frequency : only if NFS exporting from client
3453 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3454 Details : If setting attributes on a file created under NFS that had
3455 never been opened it would be possible to oops the client
3456 if the file had no objects.
3459 Frequency : always for liblustre
3461 Description: liblustre client does MDS+OSTs setattr RPC for each write
3462 Details : When doing a write from a liblustre client, the client
3463 incorrectly issued an RPC to the MDS and each OST the file was
3464 striped over in order to update the timestamps. When writing
3465 with small chunks and many clients this could overwhelm the MDS
3466 with RPCs. In all cases it would slow down the write because
3467 these RPCs are unnecessary.
3469 Severity : enhancement
3471 Description: allow number of MDS service threads to be changed at module load
3472 Details : It is now possible to change the number of MDS service threads
3473 running. Adding "options mds mds_num_threads={N}" to the MDS's
3474 /etc/modprobe.conf will set the number of threads for the next
3475 time Lustre is restarted (assuming the "mds" module is also
3476 reloaded at that time). The default number of threads will
3477 stay the same, 32 for most systems.
3482 Description: OST crash if filesystem is unformatted or corrupt
3483 Details : If an OST is started on a device that has never been formatted
3484 or if the filesystem is corrupt and cannot even mount then the
3485 error handling cleanup routines would dereference a NULL pointer.
3490 Description: NULL pointer deref in llap_from_page.
3491 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3492 incorrect) mapping if the page was truncated/reclaimed while it was
3493 searched for. Check for this condition and skip such pages when
3494 doing readahead. Introduce extra check to llap_from_page() to
3495 verify page->mapping->host is non-NULL (so page is not anonymous).
3498 Frequency : Sometimes when using sys_sendfile
3500 Description: "page not covered by a lock" warnings from ll_readpage
3501 Details : sendfile called ll_readpage without right page locks present.
3502 Now we introduced ll_file_sendfile that does necessary locking
3503 around call to generic_file_sendfile() much like we do in
3507 Frequency : with certain MDS communication failures at client mount time
3509 Description: NULL pointer deref after failed client mount
3510 Details : a client connection request may delayed by the network layer
3511 and not be sent until after the PTLRPC layer has timed out the
3512 request. If the client fails the mount immediately it will try
3513 to clean up before the network times out the request. Add a
3514 reference from the request import to the obd device and delay
3515 the cleanup until the network drops the request.
3518 Frequency : occasionally during client (re)connect
3520 Description: assertion failure during client (re)connect
3521 Details : processing a client connection request may be delayed by the
3522 client or server longer than the client connect timeout. This
3523 causes the client to resend the connection request. If the
3524 original connection request is replied in this interval, the
3525 client may trip an assertion failure in ptlrpc_connect_interpret()
3526 which thought it would be the only running connect process.
3529 Frequency : only with obd_echo servers and clients that are rebooted
3531 Description: kernel BUG accessing uninitialized data structure
3532 Details : When running an obd_echo server it did not start the ping_evictor
3533 thread, and when a client was evicted an uninitialized data
3534 structure was accessed. Start the ping_evictor in the RPC
3535 service startup instead of the OBD startup.
3537 Severity : enhancement
3538 Bugzilla : 10193 (patchless)
3539 Description: Remove dependency on various unexported kernel interfaces.
3540 Details : No longer need reparent_to_init, exit_mm, exit_files,
3541 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3544 Frequency : rare (only users of deprecated and unsupported LDAP config)
3546 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3547 Details : LDAP apparently contains 'lustreName' attributes instead of
3548 'name'. A simple remapping of the name is sufficient.
3551 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3553 Description: waiting_locks_callback trips kernel BUG if client is evicted
3554 Details : Running with the dump_on_timeout debug flag turned on makes
3555 it possible that the waiting_locks_callback() can try to dump
3556 the Lustre kernel debug logs from an interrupt handler. Defer
3557 this log dumping to the expired_lock_main() thread.
3559 Severity : enhancement
3561 Description: Support NFS exporting on 2.6 kernels.
3562 Details : Implement non-rawops metadata methods for NFS server to use without
3563 changing NFS server code.
3566 Frequency : very rare (synthetic metadata workload only)
3568 Description: two racing renames might cause an MDS thread to deadlock
3569 Details : Running the "racer" program may cause one MDS thread to rename
3570 a file from being the source of a rename to being the target of
3571 a rename at exactly the same time that another thread is doing
3572 so, and the second thread has already enqueued these locks after
3573 doing a lookup of the target and is trying to relock them in
3574 order. Ensure that we don't try to re-lock the same resource.
3577 Frequency : only very large systems with liblustre clients
3579 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3580 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3581 all OSTs in parallel. This allows the eviction of stale liblustre
3582 clients to proceed much faster than if they were done in series,
3583 and also offers similar improvements for other set_info RPCs.
3588 Description: excessive CPU usage during initial read phase on client
3589 Details : During the initial read phase on a client, it would agressively
3590 retry readahead on the file, consuming too much CPU and impacting
3591 performance (since 1.4.5.8). Improve the readahead algorithm
3592 to avoid this, and also improve some other common cases (read
3593 of small files in particular, where "small" is files smaller than
3594 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3599 Description: MDS crash when receiving packet with unknown intent.
3600 Details : Do not LBUG in unknown intent case, just return -EFAULT
3602 Severity : enhancement
3603 Bugzilla : 9293, 9385
3604 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3605 Details : Do not serialize getattr (non-intent version) and statfs.
3608 Frequency : occasional, when OST network is overloaded/intermittent
3610 Description: client evicted by OST after bulk IO timeout
3611 Details : If a client sends a bulk IO request (read or write) the OST
3612 may evict the client if it is unresposive to its data GET/PUT
3613 request. This is incorrect if the network is overloaded (takes
3614 too long to transfer the RPC data) or dropped the OST GET/PUT
3615 request. There is no need to evict the client at all, since
3616 the pinger and/or lock callbacks will handle this, and the
3617 client can restart the bulk request.
3620 Frequency : Always when mmapping file with no objects
3622 Description: client crashes when mmapping file with no objects
3623 Details : Check that we actually have objects in a file before doing any
3624 operations on objects in ll_vm_open, ll_vm_close and
3630 Description: Request leak when working with deleted CWD
3631 Details : Introduce advanced request refcount tracking for requests
3632 referenced from lustre intent.
3634 Severity : Enhancement
3636 Description: Cache open file handles on client.
3637 Details : MDS now will return special lock along with openhandle, if
3638 requested and client is allowed to hold openhandle, even if unused,
3639 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3640 closing files for every read/write openration.
3642 Severity : Enhancement
3644 Description: Cache open negative dentries on client when possible.
3645 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3646 negative dentries on lock revocation.
3651 Description: Remounting a client read-only wasn't possible with a zconf mount
3652 Details : It wasn't possible to remount a client read-only with llmount.
3654 Severity : enhancement
3655 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3656 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3657 MPICH is included to add Lustre-specific ADIO interfaces.
3658 This is based closely on the UFS ADIO layer and only differs
3659 in file creation, in order to allow the OST striping to be set.
3660 This is user-contributed code and not supported by CFS.
3665 Description: extended inode attributes (immutable, append-only) work improperly
3666 when 2.4 and 2.6 kernels are used on client/server or vice versa
3667 Details : Introduce kernel-independent values for these flags.
3669 Severity : enhancement
3672 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3673 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3674 granularity. Now, Fractional MB granularity is be supported,
3675 this is very useful for low memory system.
3677 Severity : enhancement
3679 Description: Getattr by fid
3680 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3681 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3682 have locks and avoids getattr rpc requests for it on every stat.
3685 Frequency : Always, for filesystems larger than 2TB
3687 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3688 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3689 filesystems larger than 2TB to be mounted with mballoc enabled.
3692 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3694 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3695 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3696 filesystems larger than 2TB is mounted then the kernel will
3697 silently corrupt the start of the filesystem. CONFIG_LBD is
3698 enabled for all CFS-supported kernels, but the possibility of
3699 this happening with a modified kernel config exists.
3701 Severity : enhancement
3703 Description: add client O_DIRECT support for 2.6 kernels
3704 Details : It is now possible to do O_DIRECT reads and writes to files
3705 in the Lustre client mountpoint on 2.6 kernel clients.
3707 Severity : enhancement
3709 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3710 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3711 parallel, not waiting for response from every OST before sending
3712 a rpc to the next OST.
3717 Description: setattr vs write race when updating file timestamps
3718 Details : Client processes that update a file timestamp into the past
3719 right after writing to the file (e.g. tar) it is possible that
3720 the updated file modification time can be reset to the current
3721 time due to a race between processing the setattr and write RPC.
3723 Severity : enhancement
3725 Description: Bring 'lfs find' closer in line with regular Linux find.
3726 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3727 -print0 options and obtains all the needed info through the lustre
3730 Severity : enhancement
3732 Description: support up to 1024 configured devices on one node
3733 Details : change obd_dev array from statically allocated to dynamically
3734 allocated structs as they are first used to reduce memory usage
3739 Description: Flush dirty partially truncated pages during truncate
3740 Details : Immediatelly flush partially truncated pages in filter_setattr,
3741 this way we completely avoid having any pages in page cache on OST
3742 and can retire ugly workarounds during writes to flush such pages.
3747 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3748 Details : protect inode from truncation within vfs_unlink() context
3749 just take a reference before calling vfs_unlink() and release it
3750 when parent's i_sem is free.
3755 Description: last_id value checked outside lock on OST caused LASSERT failure
3756 Details : If there were multiple MDS->OST object precreate requests in
3757 flight, it was possible that the OST's last object id was checked
3758 outside a lock and incorrectly tripped an assertion. Move checks
3759 inside locks, and discard old precreate requests.
3762 Frequency : always, if extents are used on OSTs
3764 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3765 Details : OSTs using the extents format would not zero the high 16 bits of
3766 the index physical block number. This is not a problem for any
3767 OST filesystems smaller than 16TB, and no kernels support ext3
3768 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3769 new/modified files) and can be fixed for existing filesystems
3770 with e2fsprogs-1.39-cfs1.
3775 Description: import connection selection may be incorrect if timer wraps
3776 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3777 import connections to be ignored if the 32-bit jiffies counter
3778 wraps. Use a 64-bit jiffies counter.
3781 Frequency : very large clusters immediately after boot
3783 Description: LNET request buffers exhausted under heavy short-term load
3784 Details : If a large number of client requests are generated on a service
3785 that has previously never seen so many requests it is possible
3786 that the request buffer growth cannot keep up with the spike in
3787 demand. Instead of dropping incoming requests, they are held in
3788 the LND until the RPC service can accept more requests.
3791 Frequency : Sometimes during replay
3793 Description: Assertion failure in ll_local_open after replay.
3794 Details : If replay happened on an open request reply before we were able
3795 to set replay handler, reply will become not swabbed tripping the
3796 assertion in ll_local_open. Now we set the handler right after
3797 recognising of open request
3800 Frequency : very rare
3802 Description: kernel reports "badness in vsnprintf"
3803 Details : Reading from the "recovery_status" /proc file in small chunks
3804 may cause a negative length in lprocfs_obd_rd_recovery_status()
3805 call to vsnprintf() (which is otherwise harmless). Exit early
3806 if there is no more space in the output buffer.
3808 Severity : enhancement
3810 Description: clear OBD RPC statistics by writing to them
3811 Details : It is now possible to clear the OBD RPC statistics by writing
3812 to the "stats" file.
3817 Description: Client mtime is not the same on different clients after utimes
3818 Details : In some cases, the client was using the utimes() syscall on
3819 a file cached on another node. The clients now validate the
3820 ctime from the MDS + OSTs to determine which one is right.
3825 Description: Inability to activate failout mode
3826 Details : lconf script incorrectly assumed that in pythong string's numeric
3827 value is used in comparisons.
3830 Frequency : always with multiple stripes per file
3832 Description: Inefficient object allocation for mutli-stripe files
3833 Details : When selecting which OSTs to stripe files over, for files with
3834 a stripe count that divides evenly into the number of OSTs,
3835 the MDS is always picking the same starting OST for each file.
3836 Return the OST selection heuristic to the original design.
3841 Description: mount failures may take full timeout to return an error
3842 Details : Under some heavy load conditions it is possible that a
3843 failed mount can wait for the full obd_timeout interval,
3844 possibly several minutes, before reporting an error.
3845 Instead return an error as soon as the status is known.
3847 Frequency : quota enabled and large files being deleted
3849 Description: releasing more than 4GB of quota at once hangs OST
3850 Details : If a user deletes more than 4GB of files on a single OST it
3851 will cause the OST to spin in an infinite loop. Release
3852 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3857 Description: statfs data retrieved from /proc may be stale or zero
3858 Details : When reading per-device statfs data from /proc, in the
3859 {kbytes,files}_{total,free,avail} files, it may appear
3860 as zero or be out of date.
3863 Frequency : systems with MD RAID1 external journal devices
3865 Description: lconf's call to blkid is confused by RAID1 journal devices
3866 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3867 of returning all block devices that match the journal UUID.
3870 Frequency : always, for aggregate stripe size over 4GB
3872 Description: assertion fails when trying to use 4GB stripe size
3873 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3874 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3877 Frequency : always on ppc64
3879 Description: the first write on an ext3 filesystem with mballoc got stuck
3880 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3881 perform endianness conversion.
3883 ------------------------------------------------------------------------------
3885 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3887 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3888 INTEROPERATE with older versions automatically. Please read the
3889 user documentation before upgrading any part of a live system.
3890 * WARNING: Lustre networking configuration changes are required with
3891 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3894 * Support for kernels:
3895 2.6.9-22.0.2.EL (RHEL 4)
3896 2.6.5-7.244 (SLES 9)
3897 2.6.12.6 vanilla (kernel.org)
3900 Severity : enhancement
3901 Bugzilla : 7981/8208
3902 Description: Introduced Lustre Networking (LNET)
3903 Details : LNET is new networking infrastructure for Lustre, it includes
3904 a reorganized network configuration mode (see the user
3905 documentation for full details) as well as support for routing
3906 between different network fabrics. Lustre Networking Devices
3907 (LNDs) for the supported network fabrics have also been
3908 created for this new infrastructure.
3910 Severity : enhancement
3911 Description: Introduced Access control lists
3912 Details : clients can set ACLs on files and directories in order to have
3913 more fine-grained permissions than the standard Unix UGO+RWX.
3914 The MDS must be started with the "-o acl" mount option.
3916 Severity : enhancement
3917 Description: Introduced filesystem quotas
3918 Details : Administrators may now establish per-user quotas on the
3921 Severity : enhancement
3923 Description: Configuration change for the XT3
3924 The PTLLND is now used to run Lustre over Portals on the XT3
3925 The configure option(s) --with-cray-portals are no longer used.
3926 Rather --with-portals=<path-to-portals-includes> is used to
3927 enable building on the XT3. In addition to enable XT3 specific
3928 features the option --enable-cray-xt3 must be used.
3933 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3934 Details : A race between allocating a new llcd and re-getting the llcd_lock
3935 allowed another thread to grab newly-allocated llcd.
3937 Severity : enhancement
3939 Description: 2.6 OST async journal commit and locking fix to improve performance
3940 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3941 2.6 kernels are now async as they already were in 2.4 kernels so
3942 that they can commit concurrently with the network bulk transfer.
3943 For block-allocated files the filter allocation semaphore is held
3944 to avoid filesystem fragmentation during allocation. BKL lock
3945 removed for 2.6 xattr operations where it is no longer needed.
3950 Description: lconf incorrectly determined whether two IP networks could talk
3951 Details : In some more complicated routing and multiple-network
3952 configurations, lconf will avoid trying to make a network
3953 connection to a disjoint part of the IP space. It was doing the
3954 math incorrectly for one set of cases.
3959 Description: Fix for potential infinite loop processing records in an llog.
3960 Details : If an llog record is corrupted/zeroed, it is possible to loop
3961 forever in llog_process(). Validate the llog record length
3962 and skip the remainder of the block on error.
3965 Frequency : occasional (liblustre only)
3967 Description: liblustre could not open files whose last component is a symlink
3968 Details : sysio_path_walk() would incorrectly pass the open intent to
3969 intermediate path components.
3972 Frequency : rare (liblustre only with non-standard tuning)
3973 Bugzilla : 7201 (7350)
3974 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3975 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3976 after it has been released, only do this for referenced locks
3978 Severity : enhancement
3980 Description: specifying an (invalid) directory default stripe_size of -1
3981 would reset the directory default striping
3982 Details : stripe_size -1 was used internally to signal directory stripe
3983 removal, now use "all default" to signal dir stripe removal
3984 as a directory striping of "all default" is not useful
3987 Frequency : common for large clusters running liblustre clients
3989 Description: doing an ls when liblustre clients are running is slow
3990 Details : sending a glimpse AST to a liblustre client waits for every AST
3991 to time out, as liblustre clients will not respond. Since they
3992 cannot cache data we refresh the OST lock LVB from disk instead.
3994 Severity : enhancement
3996 Description: doing an ls at the same time as file IO can be slow
3997 Details : enqueue and other "small" requests can be blocked behind many
3998 large IO requests. Create a new OST IO portal for non-IO
3999 requests so they can be processed faster.
4002 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
4004 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
4005 created file with mode 000
4006 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
4007 allow an ftruncate() as a non-root user to a file with mode 000.
4008 Lustre now respects this flag to disable mode checks when
4009 truncating a file owned by the user
4012 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
4015 Description: Revoking locks from clients that went dead or catatonic might take
4017 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
4018 cancellation of such locks instant on servers without waiting for
4019 any reply from clients. Clients drops these locks when cancel
4020 notification from server is received without replying.
4023 Frequency : liblustre-only, when liblustre client dies or becomes busy
4025 Description: Doing ls on Linux clients can take a long time with active
4027 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
4028 granting such locks to it in the first place if possible. Locks
4029 are taken by proxy on the OST during the read or write and
4030 dropped immediately afterward. Add connect flags handling, do
4031 not grant locks to liblustre clients for glimpse ASTs.
4033 Severity : enhancement
4035 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
4036 Details : Existing read-ahead algorithm is tuned for the case of streamlined
4037 sequential reads and behaves badly with applications doing random
4038 reads. Improve it by reading ahead at least read region, and
4039 avoiding excessive large RPC for small reads.
4041 Severity : enhancement
4043 Description: Creating more than 1000 files for a single job may cause a load
4044 imbalance on the OSTs if there are also a large number of OSTs.
4045 Details : qos_prep_create() uses an OST index reseed value that is an
4046 even multiple of the number of available OSTs so that if the
4047 reseed happens in the middle of the object allocation it will
4048 still utilize the OSTs as uniformly as possible.
4053 Description: OST or MDS may oops in ping_evictor_main()
4054 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
4055 but doesn't restart at beginning of obd_exports_timed list
4058 Severity : enhancement
4060 Description: improve by-nid export eviction on the MDS and OST
4061 Details : allow multiple exports with the same NID to be evicted at one
4062 time without re-searching the exports list.
4065 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
4067 Description: MDS may oops in groups_free()
4068 Details : in rare race conditions a newly allocated group_info struct is
4069 freed again, and this can be NULL. The 2.4 compatibility code
4070 for groups_free() checked for a NULL pointer, but 2.6 did not.
4073 Frequency : common for liblustre clients doing little filesystem IO
4074 Bugzilla : 9352, 7313
4075 Description: server may evict liblustre clients accessing contended locks
4076 Details : if a client is granted a lock or receives a completion AST
4077 with a blocking AST already set it would not reply to the AST
4078 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4081 Frequency : lfs setstripe, only systems with more than 160 OSTs
4083 Description: unable to set striping with a starting offset beyond OST 160
4084 Details : llapi_create_file() incorrectly limited the starting stripe
4085 index to the maximum single-file stripe count.
4088 Frequency : LDAP users only
4090 Description: lconf did not handle in-kernel recovery with LDAP properly
4091 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4093 Severity : enhancement
4095 Description: bind OST threads to NUMA nodes to improve performance
4096 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4097 node and do their allocations there to localize memory access
4099 Severity : enhancement
4101 Description: llmount can determine client NID directly from Myrinet (GM)
4102 Details : the client NID code from gmnalnid was moved directly into
4103 llmount, removing the need to use this or specifying the
4104 client NID explicitly when mounting GM clients with zeroconf
4107 Frequency : if client is started with down MDS
4109 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4110 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4111 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4112 ptlrpc_import_delayed_req() because we didn't check if the
4113 request was interrupted, and we also didn't break out of the
4114 event loop if there was no timeout
4119 Description: data loss during non-page-aligned writes to a single file from
4120 both multiple nodes and multiple threads on one node at same time
4121 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4122 inconsistency led to false short-reads, that were cached and later
4123 used by ->prepare_write() to fill in partially written page,
4124 leading to data loss.
4127 Frequency : always, if lconf --abort_recovery used
4129 Description: lconf --abort_recovery fails with 'Operation not supported'
4130 Details : lconf was attempting to abort recovery on the MDT device and not
4133 Severity : enhancement
4135 Description: remove cleanup logs
4136 Details : replace lconf-generated cleanup logs with lustre internal
4137 cleanup routines. Eliminates the need for client-cleanup and
4140 Severity : enhancement
4142 Description: add support for EAs (user and system) on lustre filesystems
4143 Details : it is now possible to store extended attributes in the Lustre
4144 client filesystem, and with the user_xattr mount option it
4145 is possible to allow users to store EAs on their files also
4147 Severity : enhancement
4149 Description: Add possibility (config option) to show minimal available OST free
4151 Details : When compiled with --enable-mindf configure option, statfs(2)
4152 (and so, df) will return least minimal free space available from
4153 all OSTs as amount of free space on FS, instead of summary of
4154 free spaces of all OSTs.
4156 Severity : enhancement
4158 Description: do not expand extent locks acquired on OST-side
4159 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4160 by server: they are not cached anyway.
4163 Frequency : when mmap is used/binaries executed from Lustre
4165 Description: Unmmap pages before throwing them away from read cache.
4166 Details : llap_shrink cache now attempts to unmap pages before discarding
4167 them (if unmapping failed - do not discard). SLES9 kernel has
4168 extra checks that trigger if this unmapping is not done first.
4173 Description: lconf didn't resolve symlinks before checking to see whether a
4174 given mountpoint was already in use
4177 Frequency : when migrating failover services
4178 Bugzilla : 6395, 9514
4179 Description: When migrating a subset of services from a node (e.g. failback
4180 from a failover service node) the remaining services would
4181 time out and evict clients.
4182 Details : lconf --force (implied by --failover) sets the global obd_timeout
4183 to 5 seconds in order to quickly disconnect, but this caused
4184 other RPCs to time out too quickly. Do not change the global
4185 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4187 Severity : enhancement
4188 Frequency : if MDS is started with down OST
4189 Bugzilla : 9439,5706
4190 Description: Allow startup/shutdown of an MDS without depending on the
4191 availability of the OSTs.
4192 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4193 Add appropriate locking and lov-osc refcounts for safe
4194 cleaning. Add osc abort_inflight calls in case the
4195 synchronize never started.
4198 Frequency : occasional (Cray XT3 only)
4200 Description: root not authorized to access files in CRAY_PORTALS environment
4201 Details : The client process capabilities were not honoured on the MDS in
4202 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4203 been accessed by an authorized user then root was able to access
4204 the file on the local client also. The root user capabilities
4205 are now allowed on the MDS, as this environment has secure UID.
4208 Frequency : occasional
4210 Description: ldiskfs "too long searching" message happens too often
4211 Details : A debugging message (otherwise harmless) prints too often on
4212 the OST console. This has been reduced to only happen when
4213 there are fragmentation problems on the filesystem.
4218 Description: Division by zero in statfs when all OSCs are inactive
4219 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4220 lov_statfs divides by value returned by lov_get_stripecnt().
4224 Bugzilla : 9489, 3273
4225 Description: First write from each client to each OST was only 4kB in size,
4226 to initialize client writeback cache, which caused sub-optimal
4227 RPCs and poor layout on disk for the first writen file.
4228 Details : Clients now request an initial cache grant at (re)connect time
4229 and so that they can start streaming writes to the cache right
4230 away and always do full-sized RPCs if there is enough data.
4231 If the OST is rebooted the client also re-establishes its grant
4232 so that client cached writes will be honoured under the grant.
4237 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4238 Details : Now I/O RPCs go to different portal number and (presumably) fast
4239 lock requests (and glimses) and other RPCs get their own service
4240 threads pool that should be able to service those RPCs
4243 Severity : enhancement
4245 Description: Ability to exchange lustre version between client and servers and
4246 issue warnings at client side if client is too old. Also for
4247 liblustre clients there is ability to refuse connection of too old
4249 Details : New 'version' field is added to connect data structure that is
4250 filled with version info. That info is later checked by server and
4254 Frequency : rare, liblustre only.
4255 Bugzilla : 9296, 9581
4256 Description: Two simultaneous writes from liblustre at offset within same page
4257 might proceed at the same time overwriting eachother with stale
4259 Details : I/O lock withing llu_file_prwv was released too early, before data
4260 actually was hitting the wire. Extended lock-holding time until
4261 server acknowledges receiving data.
4264 Frequency : extremely rare. Never observed in practice.
4266 Description: avoid generating lustre_handle cookie of 0.
4267 Details : class_handle_hash() generates handle cookies by incrementing
4268 global counter, and can hit 0 occasionaly (this is unlikely, but
4269 not impossible, because initial value of cookie counter is
4270 selected randonly). Value of 0 is used as a sentinel meaning
4271 "unassigned handle" --- avoid it. Also coalesce two critical
4272 sections in this function into one.
4274 Severity : enhancement
4276 Description: allow liblustre clients to delegate truncate locking to OST
4277 Details : To avoid overhead of locking, liblustre client instructs OST to
4278 take extent lock in ost_punch() on client's behalf. New connection
4279 flag is added to handle backward compatibility.
4281 Severity : enhancement
4282 Bugzilla : 4928, 7341, 9758
4283 Description: allow number of OST service threads to be specified
4284 Details : a module parameter allows the number of OST service threads
4285 to be specified via "options ost ost_num_threads={N}" in the
4286 OSS's /etc/modules.conf or /etc/modprobe.conf.
4290 Bugzilla : 6146, 9635, 9895
4291 Description: servers crash with bad pointer in target_handle_connect()
4292 Details : In rare cases when a client is reconnecting it was possible that
4293 the connection request was the last reference for that export.
4294 We would temporarily drop the export reference and get a new
4295 one, but this may have been the last reference and the export
4296 was just destroyed. Get new reference before dropping old one.
4298 Severity : enhancement
4299 Frequency : if client is started with failover MDS
4301 Description: Allow multiple MDS hostnames in the mount command
4302 Details : Try to read the configuration from all specified MDS
4303 hostnames during a client mount in case the "primary"
4306 Severity : enhancement
4308 Description: Stop sending data to evicted clients as soon as possible.
4309 Details : Check if the client we are about to send or are sending data to
4310 was evicted already. (Check is done every second of waiting,
4311 for which l_wait_event interface was extended to allow checking
4312 of exit condition at specified intervals).
4315 Frequency : rare, normally only when NFS exporting is done from client
4317 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4319 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4322 Frequency : always, if srand() is called before liblustre initialization
4324 Description: Liblustre uses system PRNG disturbing its usage by user application
4325 Details : Introduce internal to lustre fast and high-quality PRNG for
4326 lustre usage and make liblustre and some other places in generic
4327 lustre code to use it.
4329 Severity : enhancement
4330 Bugzilla : 9477, 9557, 9870
4331 Description: Verify that the MDS configuration logs are updated when xml is
4332 Details : Check if the .xml configuration logs are newer than the config
4333 logs stored on the MDS and report an error if this is the case.
4334 Request --write-conf, or allow starting with --old_conf.
4336 Severity : enhancement
4338 Description: Handle symlinks in the path when checking if Lustre is mounted.
4339 Details : Resolve intermediate symlinks when checking if a client has
4340 mounted a filesystem to avoid duplicate client mounts.
4345 Description: lconf can hit an error exception but still return success.
4346 Details : The lconf command catches the Command error exception at the top
4347 level script context and will exit with the associated exit
4348 status, but doesn't ensure that this exit status is non-zero.
4353 Description: failure of ptlrpc thread startup can cause oops
4354 Details : Starting a ptlrpc service thread can fail if there are a large
4355 number of threads or the server memory is very fragmented.
4356 Handle this without oopsing.
4359 Frequency : always, only if liblustre and non-default acceptor port was used
4361 Description: liblustre cannot connect to servers with non-default acceptor port
4362 Details : tcpnal_set_default_params() was not called and was therefore
4363 ignoring the environment varaible TCPNAL_PORT, as well as other
4364 TCPNAL_ environment variables
4369 Description: two objects could be created on the same OST for a single file
4370 Details : If an OST is down, in some cases it was possible to create two
4371 objects on a single OST for a single file. No problems other
4372 than potential performance impact and spurious error messages.
4376 Bugzilla : 5681, 9562
4377 Description: Client may oops in ll_unhash_aliases
4378 Details : Client dcache may become inconsistent in race condition.
4379 In some cases "getcwd" can fail if the current directory is
4385 Description: Inode refcounting problems in NFS export code
4386 Details : link_raw functions used to call d_instantiate without obtaining
4387 extra inode reference first.
4391 Bugzilla : 9942, 9903
4392 Description: Referencing freed requests leading to crash, memleaks with NFS.
4393 Details : We used to require that call to ll_revalidate_it was always
4394 followed by ll_lookup_it. Also with revalidate_special() it is
4395 possible to call ll_revalidate_it() twice for the same dentry
4396 even if first occurence returned success. This fix changes semantic
4397 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4398 reference on a request referred from the intent.
4399 ll_intent_release() then releases such a request.
4402 Frequency : rare, normally benchmark loads only
4404 Description: unlinked inodes were kept in memory on the client
4405 Details : If a client is repeatedly creating and unlinking files it
4406 can accumulate a lot of stale inodes in the inode slab cache.
4407 If there is no other client load running this can cause the
4408 client node to run out of memory. Instead flush old inodes
4409 from client cache that have the same inode number as a new inode.
4412 Frequency : SLES9 2.6.5 kernel and long filenames only
4413 Bugzilla : 9969, 10379
4414 Description: utime reports stale NFS file handle
4415 Details : SLES9 uses out-of-dentry names in some cases, which confused
4416 the lustre dentry revalidation. Change it to always use the
4420 Frequency : rare, unless heavy write-truncate concurrency is continuous
4421 Bugzilla : 4180, 6984, 7171, 9963, 9331
4422 Description: OST becomes very slow and/or deadlocked during object unlink
4423 Details : filter_destroy() was holding onto the parent directory lock
4424 while truncating+unlinking objects. For very large objects this
4425 may block other threads for a long time and slow overall OST
4426 responsiveness. It may also be possible to get a lock ordering
4427 deadlock in this case, or run out of journal credits because of
4428 the combined truncate+unlink. Solution is to do object truncate
4429 first in one transaction without parent lock, and then do the
4430 final unlink in a new transaction with the parent lock. This
4431 reduces the lock hold time dramatically.
4434 Frequency : rare, 2.4 kernels only
4436 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4437 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4438 before unmounting their filesystem in order to avoid deadlock.
4439 The kernel_locked() function in 2.4 kernels only checks whether
4440 the kernel lock is held, not whether it is this process that is
4441 holding it as 2.6 kernels do.
4446 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4447 Details : The client ptlrpc code may be trying to reconnect to a down
4448 server before a previous connection attempt has timed out.
4449 Increase the reconnect interval to be longer than the connection
4450 timeout interval to avoid sending duplicate connections to
4454 Frequency : echo_client brw_test command
4456 Description: fix echo_client to work with OST preallocated code
4457 Details : OST preallocation code (5137) didn't take echo_client IO path
4458 into account: echo_client calls filter methods outside of any
4459 OST thread and, hence, there is no per-thread preallocated
4460 pages and buffers to use. Solution: hijack pga pages for IO. As
4461 a byproduct, this avoids unnecessary data copying.
4465 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4466 Description: Client can oops in mdc_commit_close() after open replay
4467 Details : It was possible for the MDS to return an open request with no
4468 transaction number in mds_finish_transno() if the client was
4469 evicted, but without actually returning an error. Clients
4470 would later try to replay that open and may trip an assertion
4471 Simplify the client close codepath, and always return an error
4472 from the MDS in case the open is not successful.
4475 Frequency : rare, 2.6 OSTs only
4477 Description: OST may deadlock under high load on fragmented files
4478 Details : If there was a heavy load and highly-fragmented OST filesystems
4479 it was possible to have all the OST threads deadlock waiting on
4480 allocation of biovecs, because the biovecs were not released
4481 until the entire RPC IO was completed. Instead, release biovecs
4482 as soon as they are complete to ensure forward IO progress.
4484 Severity : enhancement
4486 Description: Support for specifying external journal device at mount
4487 Details : If an OST or MDS device is formatted with an external journal
4488 device, this device major/minor is stored in the ext3 superblock
4489 and may not be valid for failover. Allow detecting and
4490 specifying the external journal at mount time.
4495 Description: Mounting an MDS with pending unlinked files may cause oops
4496 Details : target_finish_recovery() calls mds_postrecov() which returned
4497 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4498 considers this an error and immediately begins cleaning up the
4499 lov, just after starting the mds_lov process
4501 Severity : enhancement
4503 Description: Implement 'lfs df' to report actual free space on per-OST basis
4504 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4505 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4506 usage of objects; '-h' to report in human readable format.
4508 ------------------------------------------------------------------------------
4510 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4517 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4518 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4519 at startup during orphan unlinking. Instead, skip buddy update
4520 before setup, as it will be regenerated after recovery is complete.
4525 Description: If an OST is inactive, its locks might reference stale inodes.
4526 Details : lov_change_cbdata() must iterate over all namespaces, even if
4527 they are inactive to clear inode references from the lock.
4529 Severity : enhancement
4530 Frequency : occasional, if non-standard max_dirty_mb used
4532 Description: Client will block write RPCs if not enough grant
4533 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4534 then the client will block writes while waiting for another RPC
4535 to complete instead of consuming its dirty limit. With change
4536 we get improved performance when max_dirty_mb is small.
4538 Severity : enhancement
4539 Bugzilla : 3389, 6253
4540 Description: Add support for supplementary groups on the MDS.
4541 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4542 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4543 lookups for user supplementary groups into a cache.
4547 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4548 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4549 for liblustre O_EXCL is enforced by MDS.
4553 Description: readdir never returns NULL in liblustre.
4554 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4555 dentry in struct dirent.
4559 Frequency : liblustre only, depends on application IO pattern
4560 Description: liblustre clients evicted if not contacting servers
4561 Details : Don't put liblustre clients into the ping_evictor list, so
4562 they will not be evicted by the pinger ever.
4564 Severity : enhancement
4566 Description: Add ability to evict clients by NID from MDS.
4567 Details : By echoing "nid:$NID" string into
4568 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4569 $NID would be instantly evicted from this MDS and from all active
4570 OSTs connected to it.
4574 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4575 Details : lookup_it_finish() used to query file size from OSTs that was not
4580 Description: service threads change working directory to that of init
4581 Details : Starting lustre service threads may pin the working directory
4582 of the parent thread, making that filesystem busy. Threads
4583 now change to the working directory of init to avoid this.
4587 Frequency : during shutdown only
4588 Description: shutdown with a failed MDS or OST can cause unmount to hang
4589 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4594 Frequency : 2.6 only
4595 Description: chmod/chown may include an extra supplementary group
4596 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4597 supplementary group and if none is specified this is used.
4600 Bugzilla : 5479 (6816)
4602 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4603 Details : If lookup is in progress on a file that is unlinked we might try
4604 to revalidate the inode and fail in revalidate after lookup is
4605 complete and ll_file_open() enqueues the open again but
4606 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4609 Frequency : always, if lconf --abort_recovery used
4611 Description: lconf --abort_recovery fails with 'Operation not supported'
4612 Details : lconf was attempting to abort recovery on the MDT device and not
4615 ------------------------------------------------------------------------------
4617 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4622 Frequency : rare (only unsupported configurations with a node running as an
4624 Bugzilla : 6514, 5137
4625 Description: Mounting a Lustre file system on a node running as an OST could
4627 Details : OSTs now preallocates memory needed to write out data at
4628 startup, instead of when needed, to avoid having to
4629 allocate memory in possibly low memory situations.
4630 Specifically, if the file system is mounted on on OST,
4631 memory pressure could force it to try to write out data,
4632 which it needed to allocate memory to do. Due to the low
4633 memory, it would be unable to do so and the node would
4634 become unresponsive.
4636 Severity : enhancement
4638 Description: Addition of lconf --service command line option
4639 Details : lconf now accepts a '--service <arg>' option, which is
4640 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4642 Severity : enhancement
4644 Description: Failover mode is now the default for OSTs.
4645 Details : By default, OSTs will now run in failover mode. To return to
4646 the old behaviour, add '--failout' to the lmc line for OSTs.
4648 Severity : enhancement
4650 Description: Health checks are now provided for MDS and OSTs
4651 Details : Additional detailed health check information on MSD and OSTs
4652 is now provided through the procfs health_check value.
4655 Frequency : occasional, depends on IO load
4657 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4658 after numerous create/delete cycles
4659 Details : The ext3 inode allocation policy would not allocate new inodes
4660 very well on the OSTs because there are no new directories
4661 being created. Instead we look for groups with free space if
4662 the parent directories are nearly full.
4667 Description: Network or server problems during mount may cause partially
4668 mounted clients instead of returning an error.
4669 Details : The config llog parsing code may overwrite the error return
4670 code during mount error handling, returning success instead
4676 Description: MDS can fail to allocate large reply buffers
4677 Details : After long uptimes the MDS can fail to allocate large reply
4678 buffers (e.g. zconf client mount config records) due to memory
4679 fragmentation or consumption by the buffer cache. Preallocate
4680 some large reply buffers so that these replies can be sent even
4681 under memory pressure.
4685 Frequency : rare (liblustre)
4686 Description: fsx running with liblustre complained that using truncate() to
4687 extend the file doesn't work. This patch corrects that issue.
4688 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4689 ATTR_SIZE and lsm use in llu_setattr_raw.
4693 Frequency : rare, only 2.6 kernels
4694 Description: Unusual file access patterns on the MDS may result in inode
4695 data being lost in very rare circumstances.
4696 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4697 code in the 2.6 kernel caused the inode and/or EA data not to
4698 be read from disk, causing single-file corruption.
4702 Frequency : rare, only 2.6 filesystems using extents
4703 Description: Heavy concurrent write and delete load may cause data corruption.
4704 Details : It was possible under high-load situations to have an extent
4705 metadata block in the block device cache from a just-unlinked
4706 file overwrite a newly-allocated data block. We now unmap any
4707 metadata buffers that alias just-allocated data blocks.
4711 Frequency : filesystems with default stripe_count larger than 77
4712 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4713 Details : lconf specifies an inode size of 4096 bytes when the default
4714 stripe_count is larger than 77. This conflicts with the default
4715 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4716 this case to avoid pinning too much memory for large EAs.
4718 ------------------------------------------------------------------------------
4720 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4725 Frequency : rare (extremely heavy IO load with hundreds of clients)
4727 Description: Client is evicted, gets IO error writing to file
4728 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4729 caused clients to be evicted for AST timeouts. The fixes in
4730 bug 5192 mean we no longer need to have such short AST timeouts
4731 so ldlm_timeout has been increased.
4734 Frequency : occasional during --force or --failover shutdown under load
4735 Bugzilla : 5949, 4834
4736 Description: Server oops/LBUG if stopped with --force or --failover under load
4737 Details : a collection of import/export refcount and cleanup ordering
4738 issues fixed for safer force cleanup
4741 Frequency : only filesystems larger than 120 OSTs
4742 Bugzilla : 5990, 6223
4743 Description: lfs getstripe would oops on a very large filesystem
4744 Details : lov_getconfig used kfree on vmalloc'd memory
4747 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4748 Bugzilla : 6242, 6243
4749 Description: reading from files that had been truncated to a non-zero size
4750 but never opened returned no data
4751 Details : ll_file_read() reads zeros from no-object files to EOF
4756 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4757 Details : There's little harm in aborting MDS/OSS recovery and letting it
4758 try again, so I removed the LASSERT and return an error instead.
4760 Severity : enhancement
4762 Description: New debugging infrastructure for tracking down data corruption
4763 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4764 (b) cover more of the client-side code path, and (c) try to narrow
4765 down where problems occurred
4769 Bugzilla : 3819, 4364, 4397, 6313
4770 Description: Racing close and eviction MDS could cause assertion in mds_close
4771 Details : It was possible to get multiple mfd references during close and
4772 client eviction, leading to one thread referencing a freed mfd.
4774 Severity: : enhancement
4775 Bugzilla : 3262, 6359
4776 Description: Attempts to reconnect to servers are now more aggressive.
4777 Details : This builds on the enhanced upcall-less recovery that was added
4778 in 1.4.2. When trying to reconnect to servers, clients will
4779 now try each server in the failover group every 10 seconds. By
4780 default, clients would previously try one server every 25 seconds.
4785 Description: After recovery, certain operations trigger a failed
4786 assertion on a client.
4787 Details : Failing over an mds, using lconf -d --failover, while a
4788 client was doing a readdir() call would cause the client to
4789 LBUG after recovery completed and the readdir() was resent.
4791 Severity : enhancement
4793 Description: Default groups are now added by lconf
4794 Details : You can now run lconf --group <servicename> without having to
4795 manually add groups with lmc.
4798 Frequency : occasional
4800 Description: Nodes with an elan id of 0 trigger a failed assertion
4803 Frequency : always when accessing e.g. tty/console device nodes
4805 Description: tty and some other devices nodes cannot be used on lustre
4806 Details : file's private_data field is used by device data and lustre
4807 values in there got lost. New field was added to struct file to
4808 store fs-specific private data.
4811 Frequency : when exporting Lustre via NFS
4813 Description: NFSD failed occasionally when looking up a path component
4814 Details : NFSD is looking up ".." which was broken in ext3 directories
4815 that had grown large enough to become hashed.
4818 Frequency : Clusters with multiple interfaces not on the same subnet
4820 Description: Nodes will repeatedly try to reconnect to an interface which it
4821 cannot reach and report an error to the log.
4822 Details : Extra peer list entries will be created by lconf with some peers
4823 unreachable. lconf now validates the peer before adding it.
4826 Frequency : Only if a default stripe is set on the filesystem root.
4828 Description: Setting a default stripe on the filesystem root prevented the
4829 filesystem from being remounted.
4830 Details : The client was sending extra request flags in the root getattr
4831 request and did not allocate a reply buffer for the dir EA.
4834 Frequency : occasional, higher if lots of files are accessed by one client
4835 Bugzilla : 6159, 6097
4836 Description: Client trips assertion regarding lsm mismatch/magic
4837 Details : While revalidating inodes the VFS looks up inodes with ifind()
4838 and in rare cases can find an inode that is being freed.
4839 The ll_test_inode() code will free the lsm during ifind()
4840 when it finds an existing inode and then the VFS later attaches
4841 this free lsm to a new inode.
4845 Bugzilla : 6422, 7030
4846 Description: MDS deadlock between mkdir and client eviction
4847 Details : Creating a new file via mkdir or mknod (starting a transaction
4848 and getting the ns lock) can deadlock with client eviction
4849 (gets ns lock and trying to finish a synchronous transaction).
4852 Frequency : occasional
4853 Description: While starting a server, the fsfilt_ext3 module could not be
4855 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4856 kernels. Previously, lconf would still use the ext3 name
4857 when trying to load modules. Now, it will correctly use
4858 ext3 on 2.4 and ldiskfs on 2.6.
4860 Severity : enhancement
4861 Description: The default stripe count has been changed to 1
4862 Details : The interpretation of the default stripe count (0, to lfs
4863 or lmc) has been changed to mean striping across a single
4864 OST, rather than all available. For general usage we have
4865 found a stripe count of 1 or 2 works best.
4867 Severity : enhancement
4868 Description: Add support for compiling against Cray portals.
4869 Details : Conditional compiling for some areas that are different
4873 Frequency : occasional
4874 Bugzilla : 6409, 6834
4875 Description: Creating files with an explicit stripe count may lead to
4876 a failed assertion on the MDS
4877 Details : If some OSTs are full or unavailable, creating files may
4878 trigger a failed assertion on the MDS. Now, Lustre will
4879 try to use other servers or return an error to the
4883 Frequency : occasional
4885 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4886 caused lock timeout during readahead (since 1.4.2).
4887 Details : Processes doing readahead might match a lock that hasn't been
4888 granted yet if there are overlapping and conflicting lock
4889 requests. The readahead process waits on ungranted lock
4890 (original lock is CBPENDING), while OST waits for that process
4891 to cancel CBPENDING read lock and eventually evicts client.
4893 Severity : enhancement
4895 Description: Initial enabling of flock support for clients
4896 Details : Implements fcntl advisory locking and file status functions.
4897 This feature is provided as an optional mount flag (default
4898 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4899 locking are implemented yet, and those that are are not guaranteed
4900 to be completely correct in production environments.
4901 mount -t lustre -o [flock|noflock] ...
4904 Frequency : occasional
4906 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4907 assertion in the ext3 JBD (journaling) layer.
4908 Details : The b_committed_data struct is protected by the big kernel lock
4909 in 2.4 kernels, serializing journal_commit_transaction() and
4910 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4911 access to this struct. In 2.6 kernels there is finer grained
4912 locking to improve SMP performance of the JBD layer.
4916 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4921 Description: merge and cleanup kernel patches.
4924 -----------------------------------------------------------------------------
4926 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4928 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4929 versions of Lustre. Please update all servers and clients to
4930 version 1.4.2 or later at the same time. You must also run
4931 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4932 to update the configuration logs.
4934 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4935 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4936 - swab llog records of type '0' so we get proper header size/idx (5861)
4937 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4938 - fix rename of one directory over another leaking an inode (5953)
4939 - avoid SetPageDirty on 2.6 (5981)
4940 - don't re-add just-being-destroyed locks to the waiting list (5653)
4941 - when creating new directories, inherit the parent's custom
4942 striping settings if present parent (3048)
4943 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4944 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4945 - don't hold client locks on temporary worklist from l_lru (5666)
4946 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4947 - automatically evict dead clients (5921)
4948 - Update file size properly in create+truncate+fstat case (6196)
4949 - Do not unhash mountpoint dentries, do not allow removal of
4951 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4952 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4953 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4954 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4955 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4956 - Correctly handle reads of files with no objects (6243)
4957 - lctl recover will also mark a device active if deactivate used (5933)
4959 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4960 - allow --write-conf on an MDS with different nettype than client (5619)
4961 - don't write config llogs to MDS for mounts not from that MDS (5617)
4962 - lconf should create multiple TCP connections from a client (5201)
4963 - init scripts are now turned off by default; run chkconfig --on
4964 lustre and chkconfig --on lustrefs to use them
4965 - upcalls are no longer needed for clients to recover to failover
4967 - add --abort-recovery option to lconf to abort recovery on device
4969 - add support for an arbitrary number of OSTs (3026)
4970 - Quota support protocol changes.
4971 - forward compatibility changes to wire structs (6007)
4972 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4973 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4974 - improved "lustre status" script
4975 - initialize blocksize for non-regular files (6062)
4976 - added --disable-server and --disable-client configure options (5782)
4977 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4978 - Vanilla 2.4.29 support
4979 - increase maximum number of obd devices to 520 (6242)
4980 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4981 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4982 SLES 9 2.6.5 kernels
4983 - limit stripes per file to 160 (the maximum EA size) (6093)
4985 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4988 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4989 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4990 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4991 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4992 - small ext3 extents cleanups and fixes (5733)
4993 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4994 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4995 - avoid needless client->OST connect, fix handle mismatch (5317)
4996 - fix DLM error path that led to out-of-sync client, long delays (5779)
4997 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4998 - fix several locking issues related to i_size (5492,5624,5654,5672)
4999 - don't move pending lock onto export if it is already evicted (5683)
5000 - fix kernel oops when creating .foo in unlinked directory (5548)
5001 - fix deadlock in obdfilter statistics vs. object create (5811)
5002 - use time_{before,after} to avoid timer jiffies wrap (5882)
5003 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
5004 - Do not leak request if server was not able to process it (5154)
5005 - If mds_open unable to find parent dir, make that negative lookup(5154)
5006 - don't create new directories with extent-mapping (5909, 5936)
5008 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
5009 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
5010 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
5011 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
5012 - support for RHEL4 kernel on i686 (5773)
5013 - provide error messages when incompatible logs are encountered (5898)
5015 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
5016 * version 1.4.0.10 (1.4.1 release candidate 1)
5018 - don't keep a lock reference when lock is not granted (4238)
5019 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5020 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
5021 - drop import inflight refcount on signal_completed_replay error (5255)
5022 - unlock page after async write error during send (3677)
5023 - handle missing objects in filter_preprw_read properly (5265)
5024 - no transno return for symlink open, don't save no-trasno open (3440)
5025 - don't try to complete elan receive that already failed (4012)
5026 - free RPC server reply state on error (5406)
5027 - clean up thread from ptlrpc_start_thread() on error (5160)
5028 - readahead could read extra page into cache that wasn't ejected (5388)
5029 - prevent races in class_attach/setup/cleanup/detach (5260)
5030 - don't dereference de->d_inode after l_dput of de (5458)
5031 - use "int" for stripe value returned from lock_to_stripe (5544)
5032 - mballoc allocation and error-checking fixes in 2.6 (5504)
5033 - block device patches to fix I/O request sizes in 2.6 (5482)
5034 - look up hostnames for IB nals (5602)
5035 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
5036 - don't start multiple acceptors for the same port (5277)
5037 - fix incorrect LASSERT in mds_getattr_name (5635)
5038 - export a proc file for general "ping" checking (5628)
5039 - fix "lfs check" to not block when the MDS is down (5628)
5041 - service request history (4965)
5042 - put {ll,lov,osc}_async_page structs in a single slab (4699)
5043 - create an "evict_client" /proc entry on OSTs, like the MDS has
5044 - fix mount usage message, return errors per mount(8) (5168)
5045 - change grep [] to grep "[]" in tests so they work in more UMLs
5046 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
5047 - remove ancient LOV_MAGIC_V0 EA support (5047)
5048 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
5049 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
5050 - allow readahead allocations to fail when low on memory (5383)
5051 - mmap locking landed again, after considerable improvement (2828)
5052 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
5054 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
5057 - send OST transaction number in read/write reply to free req (4966)
5058 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5059 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5060 - the watchdog thread now runs as interruptible (5246)
5061 - flock/lockf fixes (but it's still disabled, pending 5135)
5062 - don't use EXT3 constants in llite code (5094)
5063 - memory shortage at startup could cause assertion (5176)
5065 - reorganization of lov code
5066 - single portals codebase
5068 - add extents/mballoc support (5025)
5069 - direct I/O reads in the obdfilter (4048)
5070 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5072 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5075 - send OST transaction number in read/write reply to free req (4966)
5076 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5077 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5078 - the watchdog thread now runs as interruptible (5246)
5079 - handle missing objects in filter_preprw_read properly (5265)
5080 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5081 - drop import inflight refcount on signal_completed_replay error (5255)
5082 - unlock page after async write error during send (3677)
5083 - return original error code on reconstructed replies (3761)
5084 - no transno return for symlink open, don't save no-trasno open (3440)
5086 - add pid to ldlm debugging output (4922)
5087 - bump the watchdog timeouts -- we can't handle 30sec yet
5088 - extra debugging for orphan dentry/inode bug (5259)
5090 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5093 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5094 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5095 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5096 - fix NULL dereference in /proc/sys/portals/routes (4827)
5097 - allow failed mdc_close() operations to be interrupted (4561)
5098 - stop precreate on OST before MDS would time out on it (4778)
5099 - don't send partial-page writes before EOF from client (4410)
5100 - discard client grant for sub-page writes on large-page clients (4520)
5101 - don't free dentries not owned by NFS code, check generation (4806)
5102 - fix lsm leak if mds_create_objects() fails (4801)
5103 - limit debug_daemon file size, always print CERROR messages (4789)
5104 - use transno after validating reply (3892)
5105 - process timed out requests if import state changes (3754)
5106 - update mtime on OST during writes, return in glimpse (4829)
5107 - add mkfsoptions to LDAP (4679)
5108 - use ->max_readahead method instead of zapping global ra (5039)
5109 - don't interrupt __l_wait_event() during strace
5111 - add software watchdogs to catch hung threads quickly (4941)
5112 - make lustrefs init script start after nfs is mounted
5113 - fix CWARN/ERROR duplication (4930)
5114 - return async write errors to application if possible (2248)
5115 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5116 - print NAL number in %x format (4645)
5117 - update barely-supported suse-2.4.21-171 series (4842)
5118 - support for sles 9 %post scripts
5119 - support for building 2.6 kernel-source packages
5120 - support for sles km_* packages
5122 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5125 - ignore -ENOENT errors in osc_destroy (3639)
5126 - notify osc create thread that OSC is being cleaned up (4600)
5127 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5128 - reconstruct ost_handle() like mds_handle() (4657)
5129 - create a new thread to do import eviction to avoid deadlock (3969)
5130 - let lconf resolve symlinked-to devices (4629)
5131 - don't unlink "objects" from directory with default EA (4554)
5132 - hold socknal file ref over connect in case target is down (4394)
5133 - allow more than 32000 subdirectories in a single directory (3244)
5134 - fix blocks count for O_DIRECT writes (3751)
5135 - OST returns ENOSPC from object create when no space left (4539)
5136 - don't send truncate RPC if file size isn't changing (4410)
5137 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5138 - bind to privileged port in socknal and tcpnal (3689)
5140 - rate limit CERROR/CWARN console message to avoid overload (4519)
5141 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5142 - basic mmap support (3918)
5143 - kernel patch series update from b1_4 (4711)
5145 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5148 - avoid crash during MDS cleanup with OST shut down (2775)
5149 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5150 - don't use bad inodes on the MDS (3744)
5151 - dynamic object preallocation to improve recovery speed (4236)
5152 - don't hold spinlock over lock dumping or change debug flags (4401)
5153 - don't zero obd_dev when it is force cleaned (3651)
5154 - print grants to console if they go negative (4431)
5155 - "lctl deactivate" will stop automatic recovery attempts (3406)
5156 - look for existing locks in ldlm_handle_enqueue() (3764)
5157 - don't resolve lock handle twice in recovery avoiding race (4401)
5158 - revalidate should check working dir is a directory (4134)
5160 - don't always mark "slow" obdfilter messages as errors (4418)
5162 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5165 - don't close LustreDB during write_conf until it is done (3860)
5166 - fix typo in lconf for_each_profile (3821)
5167 - allow dumping logs from multiple threads at one time (3820)
5168 - don't allow multiple threads in OSC recovery (3812)
5169 - fix debug_size parameters (3864)
5170 - fix mds_postrecov to initialize import for llog ctxt (3121)
5171 - replace config semaphore with spinlock (3306)
5172 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5173 - don't allow enqueue to complete on a destroyed export (3822)
5174 - down write_lock before checking llog header bitmap (3825)
5175 - recover from lock replay timeout (3764)
5176 - up llog sem before sending rpc (3652)
5177 - reduce ns lock hold times when setting kms (3267)
5178 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5179 - fix NULL deref and obd_dev leak on setup error (3312)
5180 - replace some LBUG about llog ops with error handling (3841)
5181 - don't match INVALID dentries from d_lookup and spin (3784)
5182 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5183 - fix invalid assertion in ptlrpc_set_wait (3880)
5185 - add libwrap support for the TCP acceptor (3996)
5186 - add /proc/sys/portals/routes for non-root route listing (3994)
5187 - allow setting MDS UUID in .xml (2580)
5188 - print the stack of a process that LBUGs (4228)
5190 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5193 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5194 - make sure to unset replay flag from failed open requests (3440)
5195 - if default stripe count is 0, use OST count for inode size (3636)
5196 - update parent mtime/ctime on client for create/unlink (2611)
5197 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5198 - free recovery state on server during a forced cleanup (3571)
5199 - unregister_reply for resent reqs (3063)
5200 - loop back devices mounting and status check on 2.6 (3563)
5201 - fix resource-creation race that can provoke i_size == 0 (3513)
5202 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5203 - more debugging for page-accounting assertion (3746)
5204 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5205 - avoid deadlock after precreation failure (3758)
5206 - fix race and lock order deadlock in orphan handling (3450, 3750)
5207 - add validity checks when grabbing inodes from l_ast_data (3599)
5209 - add /proc/.../recovery_status to obdfilter (3428)
5210 - lightweight CDEBUG infrastructure, debug daemon (3668)
5211 - change default OSC RPC parameters to be better on small clusters
5212 - turn off OST read cache for files smaller than 32MB
5213 - install man pages and include them in rpms (3100)
5214 - add new init script for (un)mounting lustre filesystems (2593)
5215 - run chkconfig in %post for init scripts (3701)
5216 - drop scimac NAL (unmaintained)
5218 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5221 - clean kiobufs before and after use (3485)
5222 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5223 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5224 - change init script to fail more clearly if not run as root (1528)
5225 - allow clients to reconnect during replay (1742)
5226 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5227 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5228 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5229 - fix cond_resched (3554)
5230 - only evict unfinished clients after recovery (3515)
5231 - allow bulk resend, prevent data loss (3570)
5232 - dynamic ptlrpc request buffer allocation (2102)
5233 - don't allow unlinking open directory if it isn't empty (2904)
5234 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5235 - remove extraneous obd dereference causing LASSERT failure (3334)
5236 - don't use get_cycles() when creating temp. files on the mds (3156)
5237 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5238 - handle EEXIST for set-stripe, set proper directory name (3336)
5240 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5241 - fix ksocknal_fmb_callback() error messages (2918)
5243 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5246 - don't copy lvb into (possibly NULL) reply on error (2983)
5247 - don't deref dentry after dput, don't free lvb on error (2922)
5248 - use the kms to determine writeback rpc length (2947)
5249 - increment oti_logcookies when osc is inactive (2948)
5250 - update client's i_blocks count via lvb messages (2543)
5251 - handle intent open/close of special files properly (1557)
5252 - mount MDS with errors=remount-ro, like obdfilter (2009)
5253 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5254 - don't use cancelling-locks' kms values (2947)
5255 - use highest lock extent for kms, not last one (2925)
5256 - don't dereference ERR_PTR() dentry in error handling path (3107)
5257 - fix thread race in portals_debug_dumplog() (3122)
5258 - create lprocfs device entries at setup instead of at attach (1519)
5259 - common AST error handler, don't evict client on completion race (3145)
5260 - zero nameidata in detach_mnt in 2.6 (3118)
5261 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5262 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5263 - initialize RPC timeout timer earlier for 2.6 (3219)
5264 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5265 - print nal/nid for unknown nid (3258)
5266 - additional checks for oscc recovery before doing precreate (3284)
5267 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5268 - don't crash in mdc_close for bad permissions on open (3285)
5269 - zero i_rdev for non-device files (3147)
5270 - clear page->private before handing to FS, better assertion (3119)
5271 - tune the read pipeline (3236)
5272 - fix incorrect decref of invalidated dentry (2350)
5273 - provide read-ahead stats and refine rpc in flight stats (3328)
5274 - don't hold journal transaction open across create RPC (3313)
5275 - update atime on MDS at close time (3265)
5276 - close LDAP connection when recovering to avoid server load (3315)
5277 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5278 - don't leak open file on MDS after open resend (3325)
5279 - serialize filter_precreate and filter_destroy_precreated (3329)
5280 - loop device shouldn't call sync_dev() for nul device (3092)
5281 - clear page cache after eviction (2766)
5282 - resynchronize MDS->OST in background (2824)
5283 - refuse to mount the same filesystem twice on same mountpoint (3394)
5284 - allow llmount to create routes for mounting behind routers (3320)
5285 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5286 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5287 - fix rare problem with rename on htree directories (3417)
5289 - allow default OST striping configuration per directory (1414)
5290 - fix compilation for qswnal for 2.6 kernels (3125)
5291 - increase maximum number of MDS request buffers for large systems
5292 - change liblustreapi to be useful for external progs like lfsck (3098)
5293 - increase local configuration timeout for slow disks (3353)
5294 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5296 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5299 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5300 - don't overwrite extent policy data in reply if lock was blocked (2901)
5301 - drop filter export grants atomically with removal from device (2663)
5302 - del obd_self_export from work_list in class_disconnect_exports (2908)
5303 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5304 - swab reply message in mdc_close, other PPC fixes (2464)
5305 - fix destroying of named logs (2325)
5306 - overwrite old logs when running lconf --write_conf (2264)
5307 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5308 - fix race in target_handle_connect (2898)
5309 - mds_reint_create() should take same inode create lock (2926)
5310 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5311 - don't close files for self_export to avoid uninitialized obd (2936)
5312 - allow MDS with the same name as client node (2939)
5313 - hold dentry reference for closed log files for unlink (2325)
5314 - reserve space for all logs during transactions (2059)
5315 - don't evict page beyond end of stripe extent (2925)
5316 - don't oops on a deleted current working directory (2399)
5317 - handle hard links to targets without a parent properly (2517)
5318 - don't dereference NULL lock when racing during eviction (2867)
5319 - don't grow lock extents when lots of conflicting locks (2919)
5321 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5324 - account for cache space usage on clients to avoid data loss (974)
5325 - lfsck support in lustre kernel code (2349)
5326 - reduce journal credits needed for BRW writes (2370)
5327 - orphan handling to avoid losing space on client/server crashes
5328 - ptlrpcd can be blocked, stopping ALL progress (2477)
5329 - use lock value blocks to assist in proper KMS, faster stat (1021)
5330 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5331 - recovery for initial connections (2355)
5332 - fixes for mds_cleanup_orphans (1934)
5333 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5334 - block all file creations until orphan recovery completes (1901)
5335 - client remove rq_connection from request struct (2423)
5336 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5337 - recovery timer race (2670)
5338 - mdc_close recovey bug (2532)
5339 - ptlrpc cleanup bug (2710)
5340 - mds timeout on local locks (2588)
5341 - namespace lock held during RPCs (2431)
5342 - handle interrupted sync write properly (2503)
5343 - don't try to handle a message that hasn't been replied to (2699)
5344 - client assert failure during cleanup after abort recovery (2701)
5345 - leak mdc device after failed mount (2712)
5346 - ptlrpc_check_set allows timedout requests to complete (2714)
5347 - wait for inflight reqs when ptlrpcd finishes (2710)
5348 - make sure unregistered services are removed from the srv_list
5349 - reset bulk XID's when resending them (caught by 1138 test)
5350 - unregister_bulk after timeout
5351 - fix lconf error (2694)
5352 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5353 - readahead locks pages, leaves pending causing memory pressure (2673)
5354 - increase OST request buffers to 4096 on large machines (2729)
5355 - fix up permission of existing directories in simple_mkdir (2661)
5356 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5357 - don't assign transno to errored transactions (2742)
5358 - don't delete objects on OST if given a bogus objid from MDS (2751)
5359 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5360 - if rq_no_resend, then timeout request after recovery (2432)
5361 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5362 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5363 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5364 - MDS recovery shouldn't delete objects that it has given out (2730)
5365 - if enqueue arrives after completion, don't clobber LVB (2819)
5366 - don't unlock pages twice when trigger_group_io returns error (2814)
5367 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5368 - don't write pages to disk if there was an error (1450)
5369 - don't ping imports that have recovery disabled (2676)
5370 - take buffered bytes into account when balancing socknal conn (2817)
5371 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5372 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5373 - remove little-endian swabbing of llog records (1987)
5374 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5375 - retry reposting request buffers if they fail (1191)
5376 - grow extent at grant time to avoid granting a revoked lock (2809)
5377 - lock revoke doesn't evict page if covered by a second lock (2765)
5378 - disable VM readahead to avoid reading outside lock extents (2805)
5380 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5381 - updated kernel patches for hp-2.4.20 kernel (2681)
5383 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5386 - fix truncated write corruption (2366)
5387 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5389 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5390 - ensure len(uuid) < 37 in lmc (1171)
5391 - fix ia64 OOPS in llog_test (2255)
5392 - zero end of page at obdfilter for partial page writes (2648)
5393 - don't leave stale dentries around after renames (bug 2428)
5394 - fix timeouts when evicting a client with a single lock held (2642)
5395 - set deadline for the initial HELLO message to drain (2634)
5396 - print out dotted-quad IP addresses in the socknal (2302)
5398 - additional debugging for MDS client eviction problem (2443)
5399 - fix mkfsoptions support for osts (2603, 2604)
5401 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5404 - add series for the vanilla 2.6.0 kernel
5405 - add series for the vanilla 2.4.24 kernel
5406 - add series for a cray x86/64 UL kernel drop
5407 - fix xattr patches for the vanilla 2.4.19 series
5409 - generate true UUIDs in lmc (1171)
5410 - have portals stack dumping break in UML (2466)
5411 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5412 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5413 - wake callback waiting threads on client eviction (2460)
5414 - Add --ptldebug and --subsystem to lmc (1719)
5415 - update assertion to allow safe interrupt allocation
5416 - set rq_no_resend for cancel requests (2432)
5417 - recalculate ptlrpcd timeout after resend (2494)
5418 - call vfs_rmdir when removing pending directories (2368)
5419 - fix renaming a file to itself (2429)
5420 - lmc creates a default one-stripe lov (2454)
5421 - expand procfs space to handle large clusters (2326)
5422 - increase UML stack to avoid overflow
5423 - update lconf's list of debug and subsystem masks
5424 - fix lfs find --obd (2510)
5425 - /proc tunable for disabling filter read caching (2591)
5426 - stop rpm packages from altering slapd.conf (2301)
5427 - disable nagle in the socknal under 0conf (2578)
5428 - choose mds inode size based on stripe count (2572)
5429 - fix kernel-source rpm problems (2516)
5431 - add --disable-doc to avoid pdf generation (2421)
5432 - update documentation, tests, type-os, comments
5433 - avoid format warnings on ia64
5434 - remove the TOE NAL
5435 - tiny code cleanups by removing unused fields
5437 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5440 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5441 - remove the most blatant lies from BUILDING file (2371)
5442 - change default debug level to reasonable production setting
5443 - reduce client side cache size to reduce cache flush time
5444 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5445 - make TCP zerocopy and pinger support enabled by default (2476)
5446 - sync writes completed after process exits caused crashes (2319)
5447 - maintain correct mount count on the MDS (2356)
5448 - backout 1557, because 2316 wasn't really fixed
5449 - better file I/O statistics gathering in /proc
5450 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5451 - another kernel patch to fix zero-copy TCP function export
5452 - don't take duplicate lock when processing re-sent getattr (2420)
5453 - lctl uses obd_self_export instead of creating new conn (2353)
5454 - MDS/OST recovery case which requires object creation asserted (2425)
5455 - move lfs from /usr/sbin to /usr/bin in packages
5456 - fix race between mds_client_add and mds_client_free (2417)
5457 - use kmalloc instead of slabs in portals (2430)
5458 - don't create duplicate records when a failover MDS is present (2442)
5459 - remove unnecessary mount age check (2332)
5460 - don't remove directory inodes from locks prematurely (2451)
5461 - don't break if MDS service name is the same as hostname (2103)
5462 - fix races in client write RPC generation when cache full (2482)
5464 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5467 - remove now-unused request->rq_obd (278)
5468 - if an allocation fails, print out how much memory we've used (1933)
5469 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5470 - add the "configurable stack size" patch to most series files (1256)
5471 - ability to write large log records, for 100+ OST configs (2306)
5472 - fix NULL deref when filter_prep fails (2314)
5473 - fix operator precedence error in filter_sync
5474 - dynamic allocation of socknal TX descriptors (2315)
5475 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5476 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5477 - ensure that CWARN messages in Portals always get to the syslog
5478 - __init/__exit are not for prototype decls (ldlm_init/exit)
5479 - x86-64 compile warning fixes
5480 - fix gateway LMC keyword conflict (2318)
5481 - fix MDS lock inversions in getattr/reint paths (1844)
5482 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5483 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5484 - DLM race condition prevented some lock evictions (2328)
5485 - ENOMEM detection and retry on socknal sends (2230)
5486 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5487 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5488 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5489 - better error messages when a client is rejected during recovery (1505)
5490 - avoid cancelling locks which were never granted, after failure (2330)
5491 - fix i_sem/journal inversion in mds_client_add (2333)
5492 - fix truncate/getattr lock cycle deadlock (2334)
5493 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5494 - fix two rare exit paths which could leak an l_lock() ref (2321)
5495 - fencepost error in MDS/OST orphan recovery (2226)
5496 - make log record alignment 8 bytes (1988)
5497 - lstripe now fails when requested offset > ost_count (2237)
5498 - ensure that all kernel series have a complete list.h (1607)
5499 - fix crashes in special-file operations (2316)
5500 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5502 - allow configurable automake binary, for testing new versions
5503 - small update to the lfs documentation
5505 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5507 * fix negative export reference count in fsfilt_sync (2312)
5509 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5510 * release candidate 0.9.1
5512 - orphans are moved into the PENDING directory for possible recovery
5513 - replayed opens now open by fid for orphan/rename safety (1042)
5514 - last close of an orphan inode generates a transno (683)
5515 - chdir() and mount() now pin the directory entry (1020)
5516 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5517 - discard very old requests without processing them (1502)
5518 - remove some common, well-understood CERRORs (1505)
5519 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5520 - clear "grant" flags in OST replies until OST grant code lands (1644)
5521 - fix read performance by not clobbering i_blksize on client (1598)
5522 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5523 - make LRU size a /proc tunable, clears locks when reduced (707)
5524 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5525 - server locks take references on exports now (1558)
5526 - build fixes for 2.4.20-rh trees (1663)
5527 - return an error from lov_create if all OSCs are inactive (1751)
5528 - fix import levels when a reconnect happens without a timeout (1597)
5529 - exit early from mds_open if we get a lookup error (1749)
5530 - partial page read at EOF wouldn't wait for disk before sending (1642)
5531 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5532 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5533 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5534 - write/truncate lock inversion (1639)
5535 - Don't auto-load obdclass, portals modules during cleanup (1495)
5536 - fix timestamps from jumping to "now" (1763)
5537 - extra journal assertions (1648)
5538 - add an extra multiunlink test (1771)
5539 - fix read_record/write_record API (1776)
5540 - fix leak of offset_extent, possible incorrect i_size later (1772)
5541 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5542 - Debugging for the kqswnal_get_idle_tx problems (1820)
5543 - Allow recovery to be attempted multiple times (1536)
5544 - Write out MDS last_rcvd file after it is first created (1600)
5545 - Fix tx_descriptor leak in failed transmit situations (1827)
5546 - ext3 journaling fixes for assertion failure after IO error (1871)
5547 - class_export_put() on freed export after completion AST error (1896)
5548 - Fix revalidate looping in VFS (1322)
5549 - Don't access a freed export during MDS_REINT timeout (1521)
5550 - Add open-unlink recovery support on the MDS (1673,1764)
5551 - Return an error if no MDS data was read from last_rcvd (1946)
5552 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5553 - Don't setup a disk device that doesn't match exported UUID (317)
5554 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5555 - avoid committing NULL handle in force close
5556 - local.sh is now a one-stripe LOV configuration
5557 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5558 - fix ext3 htree duplicate directory entry corruption (1516)
5559 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5560 - update to new LOV EA format (2097)
5561 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5563 2003-06-15 Phil Schwan <phil@clusterfs.com>
5566 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5567 - per-import/export recovery handling (958, 931, 959)
5568 - multiple last-rcvd slots, for serving multiple FSes (949)
5569 - connections are again shared between multiple imp/exports (963, 964)
5570 - "umount -f" would hang if any requests needed to be sent (393, 978)
5571 - avoid pinning large req buffer by copying for queued messages (989)
5572 - add "uuid" to "lctl device" command to help upcalls (991)
5573 - "open" RPCs with transnos would confuse recovery counters (1037)
5574 - do proper endian conversion of all wire messages (288, 340, 891)
5575 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5576 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5577 - fix ptlrpc_connection leak in target_handle_connect (1174)
5578 - fix import refcounting bug in OST and MDS cleanup (1134)
5579 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5580 - fix very unlikely obd_types race condition (501)
5581 - remove osc_open hack for echo_client (1187)
5582 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5583 - a failure in read_inode2 leads to deadlock (1139)
5584 - cancel ack-locks as soon as transaction is committed (1072)
5585 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5586 - make sure to commitrw after any preprw to avoid deadlock (1162)
5587 - failing to execute a file in a lustre FS would lock inode (1203)
5588 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5589 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5590 - obd_uuid2tgt would walk off the end of the list (1255)
5591 - on IA64 the osc would give portals incorrect bulk size (1258)
5592 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5593 - fix lock inversion caused by new llite matching code (1282)
5594 - limit the number of dirty pages on a client to 10MB (1286)
5595 - timed out locks were not being corrected cancelled (1289)
5596 - fix O_DIRECT above 4GB on IA-32 (1292)
5597 * major user-visible changes
5598 - fail out/fail over policy now controlled by the upcall (993)
5600 - add OBD_PING to check server availability and failure (954)
5601 - lustre messages are now sent in sending host order (288, 340, 891)
5602 - add eadatalen to MDS getattr reply (340)
5603 - OST read replies may contain second buffer, with per-page status (593)
5605 2003-03-11 Phil Schwan <phil@clusterfs.com>
5608 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5609 - failed open()s could cause deadlock; fixed (867, 869)
5610 - stop cancelling OST locks when files are closed (481)
5611 - overlapping XID spaces caused network corruption (851, 853)
5612 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5613 - setattr_raw added, to do single-RPC, server-side setattrs
5614 - lmc/lconf syntax change for OST UUIDs
5615 - fix crashy race condition between ptlrpc_free_req and osc_close
5616 - don't use request in mdc_enqueue if we hit a timeout (889)
5617 - don't set the inode i_size for regular files from the MDS (896)
5618 - handle out of order completion AST (842)
5619 - don't LBUG if a lock request times out after receiving AST (913)
5620 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5621 - if a bad lock AST arrives, send an error instead of dropping entirely
5622 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5623 - fix leak in bulk IO when only partially completed (899, 900, 926)
5624 - fix O_DIRECT for ia64 (55)
5625 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5626 - C-z after timeout could hang a process forever; fixed (977)
5628 - client-side I/O cache (678, 924, 929, 941, 970)
5630 - READPAGE and SETATTRs which don't take server-side locks get
5633 2003-02-11 Phil Schwan <phil@clusterfs.com>
5636 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5637 - Fix fsfilt_extN_readpage() to read a full page of directory
5638 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5639 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5640 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5641 - Most intent operations were removed, in favour of a new RPC mode
5642 that does a single RPC to the server and bypasses most of the VFS
5643 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5644 - Aggressively cancel local locks on DLM servers
5645 - mds_reint_unlink sends EA to the client if it's the last nlink.
5646 client uses that EA to unlink OST objects.
5647 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5648 - recursive symlinks were fixed (439)
5649 - fixed NULL deref in DEBUG_REQ
5650 - filter_update_lastobjid no longer calls sync, which annoyed extN
5651 - fixed multi-client small-writes to a single file problem (445)
5652 - fixed mtime updates during file writes (607)
5653 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5654 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5655 - fixed Linux OST size reporting problem (444, 656)
5656 - OST now updates object mtime with writes or setattr (607, 619)
5657 - client verifies file size before zeroing page past EOF (445)
5658 - OST now writes last allocated objid to disk with allocation (108)
5659 - LOV on echo now works (409)
5661 - mds_reint_unlink sends a new buffer, with the EA included. this
5662 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5663 set if a regular file was being unlinked, and it was the last link
5664 - use PtlGet from the target for bulk writes (315)
5665 - OST now updates object mtime with writes or setattr (607, 619)
5666 - LDLM now has a grant-time callback to revalidate locked items, if
5668 - Many MDS operations were reorganized to combat race conditions
5670 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5671 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5673 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5676 - Fully reactivate OST imports after reconnection (512, others)
5677 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5678 - More graceful error handling for truncating on dead OST (515)
5679 - Don't error out unless we're actually accessing dead stripes (474)
5680 - Fix garbage sizes when stripes are missing (410)
5681 - LRU counters were broken, causing constant lock purge (433, 432)
5682 - garbage on read from stripes with failed OSTs (441)
5683 - mark OSCs as active before reconnecting during recovery (438)
5684 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5685 - lfind did not preserve OST order in output (443)
5686 - symlinks cause hung clients, incorrect data (439)
5687 - stop dereferencing request after dropping refcount (457)
5688 - don't LASSERT(spin_is_locked) on non-SMP (455)
5689 - fixes for many rename() bugs
5690 - fstat didn't correctly synchronize attributes (399)
5691 - server must handle lock cancellation during blocking AST prep (487)
5692 - bulk descriptors were free()d too soon (511)
5693 - fix paths in lconf, which would load incorrect modules (451, 507)
5694 - fix confusing lconf 'host not found' error message (386)
5695 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5696 - fix race condition in mdc_blocking_ast() for inode access (526)
5697 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5698 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5699 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5700 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5701 - fix lstripe in several ways (kernel side) (527)
5702 - fix request leak in ldlm_cli_enqueue (262)
5703 - incorrect OSC was marked inactive after OST failure
5704 - call mds_fs_cleanup before unmounting filesystem (524)
5705 - fix races between taking ns_lock and ldlm_lock_change_resource
5706 - fix races updating LOV export open file list
5707 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5708 - fix recovery NULL deref in ldlm_cli_cancel_unused
5709 - fix some DLM races by using new hash table for lock handles (419)
5710 - permit the client to specify desired inodes, at replay
5711 - duplicate requests when we queue them for replay reintegration
5712 - fix last_rcvd offset calculation
5713 - sync after each recovered transaction, so we always make progress
5714 - never, not always, ERESTART requests without transnos
5715 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5717 - skip replay if the MDS says that the client is already connected
5718 - don't check for a recovery-enabled export to match lctl's UUID
5719 - don't INC_USE_COUNT for phantom exports
5720 - don't crash when cleaning up phantom exports (567)
5721 - don't double-finish or set replay data for errored mdc_open requests
5722 - abort requests when they time out, so we don't get old replies
5723 - send/receive replies for AST messages again
5724 - if the client says that it doesn't have the lock, cancel it on the
5726 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5727 mark it as destroyed, it will all work out when decref is called
5728 - fix module use counts (22, 581)
5730 - ASTs now expect a reply (server cancels lock on error reply)
5732 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5735 - fix many simultaneous client startup (392)
5736 - fix dentry->d_it clobbering
5737 - credentials weren't being shipped for readdir/getattr operations
5738 - remove invalid assertions triggered during some concurrent MD
5740 - proper Lustre versions added (336, 389)
5741 - fix memory leak for create error case (398)
5742 - fix LOV locking bug that would get cli/srv out of sync
5743 - fix echo client over LOV (409)
5744 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5745 - fix double-O_EXCL intent crash (424)
5746 - avoid sending multiple lock CANCELs (352)
5748 - MDS can do multi-client recovery (modulo bugs in new code)
5750 - many updates, edits, cleanups
5752 2002-11-18 Phil Schwan <phil@clusterfs.com>
5755 - fix null d_it dereference (346)
5756 - fix full OST/dbench hang (333)
5757 - fix permission problem with file removal (286)
5758 - fix removal of OSCs from LOV when they fail
5759 - fix NULL deref during bulk timeout (214)
5760 - fix problems related to multiple filesystems on one MDS (241)
5761 - fixed serious subtle metadata locking bugs
5762 - free locks on clients when inodes are removed due to memory
5764 - fix inode pointer in lock data (285)
5765 - partial support for multiple MDS on a single host (241)
5766 - data locks weren't cancelled at clear_inode time (290, 311)
5767 - intent locks could lead to unbounded lock growth (205)
5768 - added a maximum lock count, an LRU list, and a flusher
5769 - fix multiple rename (365)
5770 - properly abstracted the echo client
5771 - OSC locked 1 byte too many; fixed
5772 - rewrote brw callback code:
5773 - fixed recovery bugs related to LOVs (306)
5774 - fixed too-many-pages-in-one-write crash (191)
5775 - fixed (again) crash in sync_io_timeout (214)
5776 - probably fixed callback-related race (385)
5778 - Add capability to MDS protocol
5779 - LDLM cancellations and callbacks on different portals
5781 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5784 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5785 - defer open object destruction to close time (601981, 138)
5786 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5787 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5788 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5789 - don't call obd_setattr in ll_file_release for destroyed objects
5790 * protocol change to lustre_msg: move |version| and add |flags|
5791 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5792 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5793 * added replay of create, unlink, link and rename operations during
5794 MDS failover; recovery should be much more robust now
5795 * remove failed OSCs from LOVs (only lov_create uses this so far)
5796 * the lustre-HOWTO was brought (more) up to date (582544)
5798 2002-10-23 Phil Schwan <phil@clusterfs.com>
5801 - in-use dentries weren't being reused properly (617851)
5802 - prevent multiple LDLM setup (599178)
5803 - fix LOV size calculations for truncate (617853)
5804 - fix client handling of MDS intent errors (POSIX)
5805 - fix permission bug in lovstripe.c test (624321)
5806 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5807 - truncate past end of file could corrupt data
5808 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5809 - a race in recovery could return ETIMEDOUT to apps (623947)
5810 - building outside the source directory was fixed
5811 * the lustre-HOWTO was brought (more) up to date (582544)
5812 * major progress was made on recovery functionality
5814 2002-10-10 Phil Schwan <phil@clusterfs.com>
5817 - recovery deadlock fix
5818 - rm -rf causes LBUG fix (617817)
5819 - file open by multiple tasks fix (618962)
5820 - directory permissions bugs (602707 and 620007)
5821 - journal_stop fixed with locking (611313)
5822 - O_APPEND failures resolved (618273, perhaps 614459)
5823 - lconf PATH fix (619770)
5824 - IA64 build fix (621450)
5825 - RPC buffer sizes scale with amount of memory
5827 2002-10-01 Phil Schwan <phil@clusterfs.com>
5830 - locks would be cancelled without throwing away data pages,
5831 resulting in inconsistent data (605627)
5832 - inode attributes were not always being refreshed (605627, 612449)
5833 - lconf now continues to cleanup after lctl reports an error
5834 - MDS now enforces user permissions (602707)
5835 - lprocfs cleanup fixed, but not yet enabled (614157)
5836 - fixed infinite server hang, should a client not respond to an AST
5837 - avoid going into recovery if user calls readlink() with a buffer
5838 that's too small (613941)
5839 - AST RPCs no longer require replies (614867) -- this may be changed
5840 - don't crash server if client sends an IOV that's too big (611336)
5841 - fixed lock conversion deadlock (611892)
5842 - fixed the following of symlinks (614622)
5843 * recovery: the server can remove locks from a client that dies, other
5844 clients can make progress
5845 * more extN patch fixes
5846 * compile-time configurable ptlrpc buffer allocations
5848 - collaborative read cache document
5849 - Lustre Lite Performance CDR document-in-progress
5851 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5854 - fix typo in patch-2.4.18
5856 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5859 - clear ptlrpc request each time in handle_incoming_request()
5860 - unlink of files now destroys the object on the OST
5862 2002-09-19 Peter Braam <braam@clusterfs.com>
5864 * add hard link support
5865 * change obdfile creation method
5866 * kernel patch changed
5868 2002-09-19 Peter Braam <braam@clusterfs.com>
5871 - stack overflow bug in extN fixed
5873 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5875 * documentation updates
5876 - add man pages for config tools
5877 - update tests/README to describe testing with new config tools
5878 - finish metadata API descriptions
5879 * bug fixes and cleanups
5880 - statfs workaround for 16TB limit
5881 - LOV stripe allocation improved, can stripe on subset of OSTs
5882 - LOV file size/IO offset was wrong for files > 4GB in size
5883 - object EA data was being dropped, caused files to be unreadable
5884 - memory overflow with non-LOV OST caused memory corruption
5885 - fixed regression tests to work with new config tools, obdfilter
5886 - fixed bug when directory size became larger than 1 block
5887 - fixed bug (for single client case) when PWD was deleted
5888 - invalidate local directory pages when doing intent-based ops
5889 - avoid LDLM oops when lock callback contained bad data
5891 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5893 * documentation updates
5894 * bug fixes and cleanups
5895 - configuration tools
5898 - 64-bit compile warnings
5899 - 64-bit internal statfs data
5901 * test_brw on persistent OST devices
5903 * lprocfs (disabled)
5905 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5907 * documentation updates
5908 * bug fixes and cleanups
5909 * configuration tools
5911 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5914 * many small fixes to 0_5_4
5915 * io/network handling
5916 * thinkos in MDS operations
5918 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5921 * crucial basic fixes to 0.5.3
5922 * IOR, Iozone work over Elan
5925 2002-08-07 Phil Schwan <phil@clusterfs.com>
5926 * version 0_5_3, our first alpha
5927 * we use the new Portals iovs
5928 * documentation updates
5929 * bug fixes and cleanups
5930 * small changes in the DLM wire protocol
5932 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5933 * version 0_5_1 with some initial stability,
5934 * locking on MD and file I/O.
5935 * documentation updates
5936 * several bug fixes since 0.5.0
5937 * small changes in wire protocol
5939 2002-07-18 Phil Schwan <phil@clusterfs.com>
5941 * delivered as Lustre Light Alpha
5942 * fixed a crash after handling invalid MDS requests
5943 * fixed directory pages for architectures with non-4k pages sizes
5945 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5946 * release version v0_4_4
5947 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5949 * Fixes a number of interruption problems with OST operations.
5950 * Update documentation for portals header changes
5951 * Move all wire protocol structs/defines to lustre_idl.h
5952 * Fixes symlink length bug.
5953 * Add tcpdump to repository.
5955 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5956 * release version v0_4_3
5957 * Fixes statfs for inodes on extN.
5958 * Fixes bug in runtests which would delete /etc/hosts.
5959 * Use 64-bit object IDs wherever possible (not into VFS though)
5960 Remove ost_get_info, which is unused by lustre, and out of date.
5962 2002-07-03 Peter Braam <braam@clusterfs.com>
5963 * release version v0_4_2 Fixes a lookup error (type not passed)
5964 * move forward to head of Portals
5965 * move forward to latest Lustre kernel
5967 2002-06-25 Peter Braam <braam@clusterfs.com>
5968 * release version v0_4_1. Hopefully stable on single node use.