1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.37 (SLES 10),
5 2.6.18-128.1.6.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.41.6.sun1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
14 * File join has been disabled in this release, refer to Bugzilla 16929.
18 Description: Avoid deadlock for local client writes
19 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
20 memory freeing context. This allows OST threads to set the
21 PF_MEMALLOC flag on task structures in order to allocate memory
22 from reserved pools and complete IO.
23 Use GFP_HIGHUSER for OST allocations for non-local client writes,
24 so that the OST threads generate memory pressure and allow
25 inactive pages to be reclaimed.
30 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
31 Details : move ldlm namespace creation in setup phase to avoid grab
32 _lprocfs_lock with cli_sem held.
36 Description: Temporarily disable grant shrink.
37 Details : Disable the feature for debugging.
41 Description: Unable to run several mkfs.lustre on loop devices at the same
43 Details : mkfs.lustre returns error 256 on the concurrent loop devices
44 formatting. The solution is to proper handle the error.
46 Severity : enhancement
48 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
50 Severity : enhancement
52 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
56 Description: resolve race between obd_disconnect and class_disconnect_exports
57 Details : if obd_disconnect will be called to already disconnected export he
58 forget release one reference and osc module can't unloaded.
60 Severity : enhancement
62 Description: Allow tuning service thread via /proc
63 Details : For each service a new
64 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
65 created that can be used to set min/max thread counts, and get the
66 current number of running threads.
70 Descriptoin: don't return error if have particaly created objects for file.
71 Details : lov_update_create_set uses set->set_success as index for created objects,
72 so if some requests will be failed, they will have hole at end of
73 array and we can use qos_shrink_lsm for allocate correct lsm.
75 Severity : enhancement
77 Description: Update OFED support to 1.4.1
81 Description: Reduce small size read RPC
82 Details : Set read-ahead limite for every file and only do read-ahead when
83 available read-ahead pages are bigger than 1M to avoid small size
86 Severity : enhancement
88 Description: move AT tunable parameters for more consistent usage
89 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
91 Severity : enhancement
93 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
94 Details : allow skip disconnected ost for send statfs request and hide error
100 Description: fix racy locking of mballoc block bitmaps causing BUG
101 Details : The locking of the mballoc buddy bitmap and the in-memory
102 block bitmap was using two different spin locks in some
103 cases. This made it possible to incorrectly access the
104 mballoc bitmap while another process was modifying it,
105 causing a sanity assertion to fail. While no on-disk corruption
106 was reported, there was some risk of this happening.
109 Frequency : rare, on llog test 6
111 Descriptoin: don't allow connect to already connected import
112 Details : allowing connect to already connected import is hide connecting problem.
117 Descriptoin: allow kill process which wait statahead result
118 Details : for some reasons 'ls' can stick in waiting result from statahead,
119 in this case need way for kill this process.
121 Severity : enhancement
123 Description: Add state history info file, enhance import info file
124 Details : Track import connection state changes in a new osc/mdc proc file;
125 add overview-type data to the osc/mdc import proc file.
127 Severity : enhancement
129 Description: MDS create should not wait for statfs RPC while holding DLM lock.
131 Severity : enhancement
133 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
138 Descriptoin: OST grant too much space to client even there are not enough space.
139 Details : Client will shrink its grant cache to OST if there are no write
140 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
141 this grant cache if there are already not enough avaible space
142 (left_space < total_clients * 32M).
145 Frequency : start MDS on uncleanly shutdowned MDS device
147 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
148 Details : stay in waiting mds<>ost recovery finished produce random bugs
149 due race between two ll_sync thread for one lov target. send
150 ACTIVATE event only if connect realy finished and import have
154 Frequency : rare, connect and disconnect target at same time
156 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
157 Details : don't call obd_disconnect under lov_lock. this long time
158 operation and can block ptlrpcd which answer to connect request.
163 Descriptoin: don't lose wakeup for imp_recovery_waitq
164 Details : recover_import_no_retry or invalidate_import and import_close can
165 both sleep on imp_recovery_waitq, but we was send only one wakeup
169 Frequency : always with long access acl
171 Descriptoin: mds can't pack reply with long acl.
172 Details : mds don't control size of acl but they limited by reint/getattr
175 Severity : enhancement
177 Description: Update to SLES10 kernel-2.6.16.60-0.33.
179 Severity : enhancement
181 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
184 Frequency : start MDS on uncleanly shutdowned MDS device
186 Descriptoin: aborting recovery hang on MDS
187 Details : don't throttle destroy RPCs for the MDT.
190 Frequency : on remount
192 Description: external journal device not working after the remount
193 Details : clear dev_rdonly flag for external journal devices in
199 Description: shutdown vs evict race
200 Details : client_disconnect_export vs connect request race.
201 if client will evicted at this time - we start invalidate
202 thread without referece to import and import can be freed
206 Frequency : rare, need acl's on inode.
208 Description: client can't handle ost additional correctly
209 Details : if ost was added after client connected to mds client can have
210 hit lnet_try_match_md ... to big messages to wide striped files.
211 in this case need teach client to handle config events about add
212 lov target and update client max ea size at that event.
214 Severity : enhancement
216 Description: Changelogs
217 Details : Changelogs are a lightweight mechanism to track filesystem
218 metadata and namespace changes. The changelog is recorded
219 permanently on the MDTs, and is periodically "consumed" / purged
220 when records are no longer needed.
222 Severity : enhancement
224 Description: compact fld format with extents
225 Details : Store range of seq rather than every seq in FLD. Seq
226 controller update FLD rather than clients. In Case of CMD, mdt0
227 has FLD, all other metadata server act as non persistent proxy
228 for FLD queries and cache fld entries in fld cache.
233 Description: don't skip ost target if they assigned to file
234 Details : Drop slow OSCs if we can, but not for requested start idx.
235 This means "if OSC is slow and it is not the requested
236 start OST, then it can be skipped, otherwise skip it only
237 if it is inactive/recovering/out-of-space.
241 Description: more cleanup in mds_lov
242 Details : not send LOV EA under replay, we can't know about they size at this
243 time. Don't allow client connect to mds before any ost connected,
244 for avoid problems with LOV EA size and returning EIO to client.
246 Severity : enhancement
248 Description: Interoperability at server side (Disk interoperability)
250 Severity : enhancement
252 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
254 Severity : enhancement
256 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
258 Severity : enhancement
260 Description: New client IO stack (CLIO).
262 Severity : enhancement
264 Description: Commit on sharing. Eliminate inter-client dependencies between
265 uncommitted transactions by doing transaction commits.
266 Thereby clients may recovery independently.
269 Frequency : Create a symlink file with a very long name
271 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
272 Details : If there is no extra space in the request for early cancels,
273 ldlm_req_handles_avail() returns 0 instead of a negative value.
275 Severity : enhancement
277 Description: Add /proc entry for import status
278 Details : The mdc, osc, and mgc import directories now have
279 an import directory that contains useful import data for debugging
282 Severity : enhancement
284 Description: Re-disable certain /proc logging
285 Details : Enable and disable client's offset_stats, extents_stats and
286 extents_stats_per_process stats logging on the fly.
289 Frequency : Only on FC kernels 2.6.22+
291 Description: oops in statahead
292 Details : Do not drop reference count for the dentry from VFS when lookup,
293 VFS will do that by itself.
295 Severity : enhancement
297 Description: Generic /proc file permissions
298 Details : Set /Proc file permissions in a more generic way to enable non-
299 root users operate on some /proc files.
303 Description: Hitting mdc_commit_close() ASSERTION
304 Details : Properly handle request reference release in
305 ll_release_openhandle().
309 Description: quota recovery deadlock during mds failover
310 Details : This patch includes att18982, att18236, att18237 in bz14840.
312 1. fix osts hang when mds does failover with quotaon
313 2. prevent watchdog storm when osts threads wait for the
318 Frequency : only patchless client
319 Description: add workaround for race between add/remove dentry from hash
321 Severity : enhancement
323 Description: Allow OST glimpses to return PW locks
327 Description: LBUG when llog conf file is full
328 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
332 Description: Prevent import from entering FULL state when server in recovery
336 Description: service mount cannot take device name with ":"
337 Details : Only when device name contains ":/" will mount treat it as
343 Description: replace ptlrpcd with the statahead thread to interpret the async
344 statahead RPC callback
348 Frequency : on recovery
349 Description: I/O failures after umount during fail back
350 Details : if client reconnected to restarted server we need join to recovery
351 instead of find server handler is changed and process self
352 eviction with cancel all locks.
354 Severity : enhancement
356 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
358 Severity : enhancement
360 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
362 Severity : enhancement
364 Description: Upcall on Lustre log has been dumped
365 Details : Allow for a user mode script to be called once a Lustre log has
366 been dumped. It passes the filename of the dumped log to the
367 script, the location of the script can be specified via
368 /proc/sys/lnet/debug_log_upcall.
373 Description: avoid idr_remove called for id which is not allocated.
374 Details : Move assigment s_dev for clustered nfs to end of initialization,
375 to avoid problem with error handling.
380 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
381 Details : When connection is reused this not moved from CONN_UNUSED_HASH
382 into CONN_USED_HASH and this prodice warning when put connection
383 again in unused hash.
386 Severity : enhancement
388 Description: File striping can now be set to use an arbitrary pool of OSTs.
390 Severity : enhancement
392 Description: Export bytes_read/bytes_write count on OSC/OST.
396 Description: Early reply size mismatch, MGC loses connection
397 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
398 the connect flags are properly negotiated.
403 Description: quotas are not honored with O_DIRECT
404 Details : all writes with the flag O_DIRECT will use grants which leads to
405 this problem. Now using OBD_BRW_SYNC to guard this.
409 Description: add quota statistics
410 Details : 1. sort out quota proc entries and proc code.
411 2. add quota statistics
413 Severity : enhancement
415 Description: enable quota support for HEAD.
419 Description: Properly propagate oinfo flags from lov to osc for statfs
420 Details : restore missing copy oi_flags to lov requests.
422 Severity : enhancement
424 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
425 : l_getgroups(8), lst(8), routerstat(8)
426 Details : included man pages for llobdstat(8), llstat(8),
427 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
429 Severity : enhancement
431 Description: configure's --enable-quota should check the
432 : kernel .config for CONFIG_QUOTA
433 Details : configure is terminated if --enable-quota is passed but
434 : no quota support is in kernel
438 Description: Remove portals compatibility
439 Details : Remove portals compatibility, not interoperable with releases
444 Description: Resolve device initialization race
445 Details : Prevent proc handler from accessing devices added to the
446 obd_devs array but yet be intialized.
448 Severity : enhancement
450 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
452 Severity : enhancement
454 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
459 Description: Using wrong pointer in osc_brw_prep_request
460 Details : Access to array[-1] can produce panic if kernel compiled with
461 CONFIG_PAGE_ALLOC enabled
465 Description: Client runs out of low memory
466 Details : Consider only lowmem when counting initial number of llap pages
470 Description: *optional* service tags registration
471 Details : if the "service tags" package is installed on a Lustre node
472 When the filesystem is mounted, a local-node service tag will
473 be created. See http://inventory.sun.com/ for more information
474 about the Service Tags asset management system.
478 Description: Kernel BUG tries to release flock
479 Details : Lustre does not destroy flock lock before last reference goes
480 away. So always drop flock locks when client is evicted and
481 perform unlock regardless of successfulness of speaking to MDS.
485 Description: add refcount protection for osc callbacks, avoid panic on shutdown
489 Description: sanity test 65a fails if stripecount of -1 is set
490 Details : handle -1 striping on filesystem in ll_dirstripe_verify
495 Description: ASSERTION(CheckWriteback(page,cmd)) failed
496 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
499 Severity : enhancement
501 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
505 Description: do not process already freed flock
506 Details : flock can possibly be freed by another thread before it reaches
507 to ldlm_flock_completion_ast.
511 Description: LBUG during stress test
512 Details : Need properly lock accesses the flock deadlock detection list.
516 Description: oops in page fault handler
517 Details : kernel page fault handler can return two special 'pages' in error
518 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
522 Description: timeout with invalidate import.
523 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
524 should be handled by ptlrpcd. This produce long age waiting and
525 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
527 Severity : enhancement
529 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
533 Description: enable MGS and MDT services start separately
534 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
535 a co-located MGS without starting the MGS, which is a complement
536 to 'nosvc' mount option.
540 Frequency : after recovery
541 Description: precreate to many object's after del orphan.
542 Details : del orphan st in oscc last_id == next_id and this triger growing
543 count of precreated objects. Set flag LOW to skip increase count
544 of precreated objects.
548 Frequency : rare, on clear nid stats
549 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
550 Details : when clean nid stats sometimes try destroy live entry,
551 and this produce panic in free.
555 Description: Stack overflow during MDS log replay
556 ease stack pressure by using a thread dealing llog_process.
560 Description: wait until IO finished before start new when do lock cancel.
561 Details : VM protocol want old IO finished before start new, in this case
562 need wait until PG_writeback is cleared until check dirty flag and
563 call writepages in lock cancel callback.
565 Severity : enhancement
567 Description: using special macro for print time and cleanup in includes.
571 Description: mds_mfd_close() ASSERTION(rc == 0)
572 Details : In mds_mfd_close(), we need protect inode's writecount change
573 within its orphan write semaphore to prevent possible races.
577 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
581 Frequency : rare, on shutdown ost
582 Description: don't hit live lock with umount ost.
583 Details : shrink_dcache_parent can be in long loop with destroy dentries,
584 use shrink_dcache_sb instead.
588 Description: don't panic with use echo client
589 Details : echo client pass NULL as client nid pointer and this produce null
594 Description: fix build on ppc32
595 Details : compile code with -m64 flag produce wrong object file for ppc32.
599 Description: add message levels for liblustreapi
603 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
604 Details : Move the 'good_osts' check before the 'total_bavail' check. This
605 will result in an -EAGAIN and in the exit call path we call
606 alloc_rr() which will with increasing aggressiveness attempt to
607 aquire precreated objects on the minimum number of required OSCs.
611 Description: Use old size assignment to avoid deadlock
612 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
613 the scheduling while holding a spinlock. We do not need locking
614 for size in ll_update_inode() because size is only updated from
615 the MDS for directories or files without objects, so there is no
616 other place to do the update, and concurrent access to such inodes
617 are protected by the inode lock.
621 Description: resolve "_IOWR redefined" build error on SLES10
625 Description: dump the memory debugging after all modules are unloaded to
626 suppress false negative in conf_sanity test 39
628 Severity : enhancement
630 Description: build kernel-ib packages for OFED 1.3 in our release cycle
635 Description: fix SLES kernel versioning
636 Details : the kernel version for our SLES 10 kernel did not include a "-"
637 before the "smp" at the end. while this was not a problem in
638 general, it did mean that software trying to use the kernel
639 version to try to detect a vendor specific kernel would fail.
640 this was most evident by the OFED build scripts.
644 Description: Don't update lov_desc members until making sure they are valid
645 Details : When updating lov_desc members via proc fs, need fix their
646 validities before doing the real update.
650 Description: don't put request into delay list while invalidate in flight.
651 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
652 import in flight. this produce timeout for invalidate and sometimes
653 can cause stale data.
657 Frequency : on ppc only
658 Description: not convert ost objects for directory because it's not exist.
659 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
663 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
665 Severity : enhancement
667 Description: Update to support 2.6.22.14 vanilla kernel.
671 Frequency : rare, on recovery
672 Description: read procfs can produce deadlock in some situation
673 Details : Holding lprocfs lock which send rpc can produce block for destroy
674 obd objects and this also block reconnect with -EALREADY. This
675 isn't fix all lprocfs bugs - but make it rare.
677 Severity : enhancement
679 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
682 Frequency : frequent on X2 node
684 Description: mdc_set_open_replay_data LBUG
685 Details : Set replay data for requests that are eligible for replay.
689 Description: lustre_mgs: operation 101 on unconnected MGS
690 Details : When MGC is disconnected from MGS long enough, MGS will evict the
691 MGC, and late on MGC cannot successfully connect to MGS and a lot
692 of the error messages complaining that MGS is not connected.
696 Frequency : on network error
697 Description: panic with double free request if network error
698 Details : mdc_finish_enqueue is finish request if any network error ocuring,
699 but it's true only for synchronus enqueue, for async enqueue
700 (via ptlrpcd) this incorrect and ptlrpcd want finish request
703 Severity : enhancement
705 Description: client-side metadata stat-ahead during readdir(directory readahead)
706 Details : perform client-side metadata stat-ahead when the client detects
707 readdir and sequential stat of dir entries therein
710 Frequency : on start mds
712 Description: Implement get_info(last_id) in obdfilter.
715 Frequency : occasional
717 Description: Correctly check stale fid, not start epoch if ost not support SOM
718 Details : open with flag O_CREATE need set old fid in op_fid3 because
719 op_fid2 was overwritten with new generated fid, but mds can answer
720 with one of these two fids and both is not stale. Setattr
721 incorrectly started an epoch and assume will be called
722 done_writing, but without SOM done_writing ever being called.
725 Frequency : rare, depends on device drivers and load
727 Description: MDS or OSS nodes crash due to stack overflow
728 Details : Code changes in 1.8.0 increased the stack usage of some functions.
729 In some cases, in conjunction with device drivers that use a lot
730 of stack the MDS (or possibly OSS) service threads could overflow
731 the stack. One change which was identified to consume additional
732 stack has been reworked to avoid the extra stack usage.
735 Frequency : occasional
737 Description: Do not fail import if osc_interpret_create gets -EAGAIN
738 Details : If osc_interpret_create got -EAGAIN it immediately exits and
739 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
740 oscc_has_objects and see OSC has no objests and call
741 oscc_internal_create to resend create request.
743 Severity : enhancement
745 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
747 Severity : enhancement
749 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
752 Frequency : very rare
754 Description: Fix replay if there is an un-replied request and open
755 Details : In some cases, older replay request will revert the
756 mcd->mcd_last_xid on MDS which is used to record the client's
759 Severity : enhancement
761 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
763 Severity : enhancement
765 Description: Add rhel5 support to HEAD.
767 Serverity : enhancement
769 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
774 Description: Don't allow skipping OSTs if index has been specified.
775 Details : Don't allow skipping OSTs if index has been specified, make
776 locking in internal create lots better.
780 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
781 Details : access to bitfield in structure is always rounded to long
782 and this produce problem with not atomic change any bit.
786 Description: Lustre make rpms failed.
787 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
788 builds Lustre rpms from tarball.
791 Frequency : rare on shutdown ost
793 Description: If llog cancel was not send before clean_exports phase, this can
794 produce deadlock in llog code.
795 Details : If llog thread has last reference to obd and call class_import_put
796 this produce deadlock because llog_cleanup_commit_master wait when
797 last llog_commit_thread exited, but this never success because was
798 called from llog_commit_thread.
802 Description: allow userland application know is lost one of stripes.
803 Details : fill lvb_blocks with error code on ost and return it to
804 application if error flag found.
808 Description: NULL lov_tgts causing MDS oops
809 Details : more safe checks for NULL lov_tgts for avoid oops.
811 Severity : enhancement
813 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
817 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
818 Details : If an OST is down the MDS will hang indefinitely in
819 obd_statfs_rqset() waiting for the statfs data. While for
820 MDS QOS usage of statfs, it should not stuck in waiting.
822 Severity : enhancement
824 Description: remote_acl support
825 Details : Support ACL-based permission check for remote user.
826 Support setfacl/getfacl for remote user with the utils
827 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
828 the system "{s,g}etfacl" utils.
830 Severity : enhancement
832 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
834 Severity : enhancement
836 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
840 Description: make mgs_setparam() handle fsname containing dash
841 Details : fsname containing a dash does not work with lctl conf_param
843 Severity : enhancement
845 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
847 Severity : enhancement
849 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
853 Frequency : rare, at shutdown
854 Description: access already free / zero obd_namespace.
855 Details : if client_disconnect_export was called without force flag set,
856 and exist connect request in flight, this can produce access to
857 NULL pointer (or already free pointer) when connect_interpret
858 store ocd flags in obd_namespace.
862 Frequency : only at startup
863 Description: not alloc memory with spinlock held.
864 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
865 if any spinlock held.
867 Severity : enhancement
869 Description: make lustre randomly fail allocating memory
870 Details : Make lustre randomly failed allocating memory for testing purpose.
872 Severity : enhancement
874 Description: lost problems with lov objid file
875 Details : Fixes some scability and access to not inited memory problems
876 in work with lov objdid file.
881 Description: lfs find does not continue on file error
882 Details : Continue other files processing when a file/dir is absent.
886 Description: Inconsistent usage of lustre_pack_reply()
887 Details : Standardize the usage of lustre_pack_reply() such that it
888 always generate a CERROR on failure.
891 Frequency : occasional
893 Description: MDS hang or stay in waiting lock
894 Details : If client receive lock with CBPENDING flag ldlm need send lock
895 cancel as separate rpc, to avoid situation when cancel request
896 can't processed due all i/o threads stay in wait lock.
900 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
904 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
908 Description: Update to SLES9 kernel-2.6.5-7.287.3.
912 Description: Update to RHEL4 latest kernel.
914 Severity : enhancement
916 Description: Build SLES10 patchless client fails
917 Details : The configure was broken by run ./configure with
918 --with-linux-obj=.... argument for patchless client. When the
919 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
920 can't search header adequately. Use absolute path such as
921 -I($LINUX)/include instead.
925 Description: interrupt oig_wait produce painc on resend.
926 Details : brw_redo_request can be used for resend requests from ptlrpcd and
927 private set, and this produce situation when rq_ptlrpcd_data not
928 copyed to new allocated request and triggered LBUG on assert
929 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
930 ptlrpcd set if request is changed and can be safety changed to use
935 Description: LASSERT_{REQ,REP}SWAB macros are buggy
936 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
937 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
938 these macros and replace them with their difinitions instead.
942 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
943 Details : Update which_patch & target file for SLES10 latest kernel.
946 Description: Remove client patches from SLES 10 kernel.
947 Details : This causes SLES 10 clients to behave as patchless clients
948 even on a Lustre-patched (server) kernel.
950 Severity : enhancement
952 Description: self-adjustable client's lru lists
953 Details : use adaptive algorithm for managing client cached locks lru
954 lists according to current server load, other client's work
955 pattern, memory activities, etc. Both, server and client
956 side namespaces provide number of proc tunables for controlling
959 Severity : enhancement
961 Description: light-weight GSS support
962 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
963 and reduce performance overhead.
965 Severity : enhancement
967 Description: Linux keyring support
968 Details : Support using service of Linux keyring for Lustre GSS internal
969 context refresh/cache mechanism.
973 Description: Fix errors in lfs documentation
974 Details : Fixes man pages
978 Description: don't use GFP_* in generic Lustre code.
979 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
983 Description: obdclass is limited by single OBD_ALLOC(idarray)
984 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
988 Description: warnings with build patchless client with vanila 2.6.19 and up
989 Details : change old ctl_table style and replace ctl_table/ctl_table_header
990 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
994 Frequency : Only for SLES
995 Description: sanity_quota fail test_1
996 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
997 file, which makes TSTID[2] inunique.
1001 Frequency : Only for SLES9
1002 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1003 Details : Update target/ChangeLog/which_patch .
1007 Frequency : Only for test_10 in sanity-quota.sh
1008 Description: a bug in quota test code
1009 Details : checking a test flag in a wrong place
1013 Frequency : Only for test_13 in sanity-quota.sh
1014 Description: a bug in quota test script
1015 Details : shouldn't check overfull allocated quota
1019 Description: jbd statistics
1020 Details : Port older jbd statistics patch for sles10
1022 Severity : enhancement
1024 Description: Add printing inode info into message about error in writepage.
1029 Description: reply_lock_interpret crash due to race with it and lock cancel.
1030 Details : Do not replay locks that are being cancelled. Do not reference
1031 locks by their address during replay, just by their handle.
1035 Frequency : When flocks are used.
1036 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1037 Details : Flock locks might destroy just granted lock if it could be merged
1038 with another existing flock, this is done in completion handler,
1039 so teach ldlm_cli_enquque_fini that this is a valid case for
1045 Description: Oops in read and write path when failing to allocate lock.
1046 Details : Check if lock allocation failed and return error back.
1050 Description: lstripe command fails for valid OST index
1051 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1052 instead of lov->desc.ld_active_tgt_count.
1056 Description: sanity.sh failed test 103
1057 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1062 Description: df doesn't work properly if diskfs blocksize != 4K
1063 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1067 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1068 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1070 Severity : enhancement
1072 Description: Adaptive timeouts
1073 Details : RPC timeouts adapt to changing server load and network
1074 conditions to reduce resend attempts and improve recovery time.
1078 Description: llapi_file_create() does not allow some changes
1079 Details : add llapi_file_open() that allows specifying the mode and
1080 open flags, and also returns an open file handle.
1084 Description: merge and cleanup kernel patches.
1085 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1089 Description: Add journal checksum support.(Kernel part)
1090 Details : The journal checksum feature adds two new flags i.e
1091 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1092 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1093 indicates that the commit block contains the checksum for
1094 the blocks described by the descriptor blocks. Now commit
1095 record can be sent to disk without waiting for descriptor
1096 blocks to be written to disk. This behavior is controlled
1097 using JBD2_FEATURE_ASYNC_COMMIT flag.
1101 Description: OSS needs mutliple precreate threads
1102 Details : Add ability to start more than one create thread per OSS.
1106 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1107 Details : execute lfs setstripe on client
1111 Description: mds_obd_create error creating tmp object
1112 Details : When the user sets quota on root, llog will be affected and can't
1113 create files and write files.
1116 Frequency : Always on ia64 patchless client, and possibly others.
1118 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1119 Details : This allows the patchless client to be loaded on architectures
1120 without this export.
1124 Description: RedHat Update kernel for RHEL5
1125 Details : Add the kernel config file for RHEL5.
1129 Description: RedHat Update kernel for RHEL5
1130 Details : Modify the kernel config file more closer RHEL5.
1134 Description: Build failure against Centos5 (RHEL5)
1135 Details : Define PAGE_SIZE when it isn't present.
1139 Description: test_7 on sanity-quota.sh can't be used on separate machines
1140 Details : add facet to handle it
1144 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1145 Details : with NFS, the anon dentry's parent was set to itself in
1146 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1147 corresponding dentry other than use rec->ur_name.
1149 Severity : enhancement
1151 Description: lfs setstripe enhancement
1152 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1156 Description: enable data checksumming by default
1157 Details : enable checksum by default, allow --disable-checksum
1158 configure option and "-o nochecksum" mount option. Checksums
1159 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1163 Description: lustre support for RHEL5
1164 Details : Add support for RHEL5.
1168 Description: Client eviction due to failover config
1169 Details : after a connection loss, the lustre client should attempt to
1170 reconnect to the last active server first before trying the
1171 other potential connections.
1175 Description: when mds and osts use different quota unit(32bit and 64bit),
1176 quota will be released repeatly.
1177 Details : void sending multiple quota reqs to mds, which will keep the
1178 status between the reqs.
1182 Description: osts not allocated evenly to files
1183 Details : change the condition to increase offset_idx
1186 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1188 Description: Data corruption for OSTs that are formatted larger than 2TB
1190 Details : When generating the bio request for lustre file writes the
1191 sector number would overflow a temporary variable before being
1192 used for the IO. The data reads correctly from Lustre (which
1193 will overflow in a similar manner) but other file data or
1194 filesystem metadata may be corrupted in some cases.
1198 Description: Tune the kernel for good SCSI performance.
1199 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1200 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1205 Description: rewrite ext2-derived code in obdclass/uuid.c
1206 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1207 from scratch preserving functionality.
1211 Description: "lfs find -obd UUID" prints directories
1212 Details : "lfs find -obd UUID" will return all directory names instead
1213 of just file names. It is incorrect because the directories
1214 do not reside on the OSTs.
1218 Description: use i_size_read and i_size_write in 2.6 port
1219 Details : replace inode->i_size access with i_size_read/write()
1222 Frequency : when removing large files
1224 Description: scheduling issue during removal of large Lustre files
1225 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1226 It causes scheduling issues when removing large files (17TB in the
1230 Frequency : only with liblustre clients on XT3
1232 Description: evictions taking too long
1233 Details : allow llrd to evict clients directly on OSTs
1236 Frequency : only on ppc
1238 Description: /proc/fs/lustre/devices broken on ppc
1239 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1240 We should make sure the type of 'index' is loff_t and then cast
1241 explicitly as needed below. Do not assign an explicitly cast
1245 Frequency : only for rhel5
1247 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1248 Details : Modify the target file & which_kernel.
1250 Severity : enhancement
1252 Description: omit set fsid for export NFS
1253 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1254 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1258 Frequency : Always, for kernels after 2.6.16
1259 Description: Fix warning idr_remove called for id=.. which is not allocated.
1260 Details : Last kernels save old s_dev before kill super and not allow
1261 to restore from callback - restore it before call kill_anon_super.
1265 Description: buffer overruns could theoretically occur
1266 Details : llapi_semantic_traverse() modifies the "path" argument by
1267 appending values to the end of the origin string, and a
1268 overrun may occur. Adding buffer overrun check in liblustreapi.
1272 Description: Fix error on 'ls .' at the top of the Lustre mount.
1273 Details : Don't revalidate dentry if it is a root dentry.
1277 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1278 Details : Modify vm-tunables-rhel4.patch.
1282 Description: Kernel config for 2.6.18-vanilla.
1283 Details : Modify targets/2.6-vanilla.target.in.
1284 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1285 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1286 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1287 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1291 Description: improve handling recoverable errors
1292 Details : if request processig with error which can be recoverable on server
1293 request should be resend, otherwise page released from cache and
1298 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1299 Details : Modify target file & which_patch.
1300 A flaw was found in the IA32 system call emulation provided
1301 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1302 value could be stored in the %RAX register, which could trigger an
1303 out-of-bounds system call table access. An untrusted local user
1304 could exploit this flaw to run code in the kernel
1305 (ie a root privilege escalation). (CVE-2007-4573).
1309 Description: O_DIRECT bypasses client statistics.
1310 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1311 read_ahead_stats were updated. copied the stats section from
1312 osc_send_oap_rpc() into async_internal().
1316 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1320 Description: change order of libsysio includes
1321 Details : '#include sysio.h' should always come before '#include xtio.h'
1325 Description: handle "serious error: objid * already exists" more gracefully
1326 Details : If LAST_ID value on disk is smaller than the objects existing in
1327 the O/0/d* directories, it indicates disk corruption and causes an
1328 LBUG(). If the object is 0-length, then we should use the existing
1329 object. This will help to avoid a full fsck in most cases.
1331 Severity : enhancement
1333 Description: adapt the lustre_config script to support the upgrade case
1334 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1335 server targets to 1.6 in parallel.
1339 Description: To avoid grant space > avaible space when the disk is almost
1340 full. Without this patch you might see the error "grant XXXX >
1341 available" or some LBUG about grant, when the disk is almost
1343 Details : In filter_check_grant, for non_grant cache write, we should
1344 check the left space by if (*left > ungranted + bytes), instead
1345 of (*left > ungranted), because only we are sure the left space
1346 is enough for another "bytes", then the ungrant space should be
1347 increase. In client, we should update cl_avail_grant only there
1348 is OBD_MD_FLGRANT in the reply.
1352 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1353 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1354 described in CVE-2007-4573. This problem could allow untrusted
1355 local users to gain root access.
1358 Frequency : when using O_DIRECT and quotas
1360 Description: Incorrect file ownership on O_DIRECT output files
1361 Details : block usage reported by 'lfs quota' does not take into account
1362 files that have been written with O_DIRECT.
1367 Description: touch file failed when fs is not full
1368 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1369 otherwise we can get ENOSP while fs is not full.
1373 Description: parallel lock callbacks
1374 Details : Instead of sending blocking and completion callbacks as separated
1375 requests, adding them to a set and sending in parallel.
1378 Frequency : only for Cray XT3
1379 Bugzilla : 12829/13455
1380 Description: Changing primary group doesn't change the group lustre assigns to
1382 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1383 overridden with the primary group provided by the group upcall,
1384 whereas the supplied fsgid can be trusted if it is in the list of
1385 supplementary groups returned by the group upcall.
1387 Severity : enhancement
1389 Description: Allow masking D_WARNING, D_ERROR messages from console
1390 Details : Console messages can now be disabled via lnet.printk.
1394 Description: User code with malformed file open parameter crashes client node
1395 Details : Before packing join_file req, all the related reference should be
1396 checked carefully in case some malformed flags cause fake
1397 join_file req on client.
1401 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1402 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1403 mode is equal to the granted mode. However, it should also check
1404 that LDLM_FL_CP_REQD is not set.
1408 Description: Heavy nfs access might result in deadlocks
1409 Details : After ELC code landed, it is now improper to enqueue any mds
1410 locks under och_sem, because enqueue might want to decide to
1411 cancel open locks for same inode we are holding och_sem for.
1415 Description: Client eviction while running blogbench
1416 Details : A lot of unlink operations with concurrent I/O can lead to a
1417 deadlock causing evictions. To address the problem, the number of
1418 oustanding OST_DESTROY requests is now throttled to
1419 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1420 callbacks are processed in priority.
1424 Description: enable ACLs on MDS by default
1425 Details : ACLs must be enabled on MDS by default.
1428 Frequency : PPC/PPC64 only
1430 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1431 Details : fix duplicated definitions between asm-ppc64/types.h and
1432 lustre_types.h on PPC.
1435 Frequency : PPC/PPC64 only
1437 Description: asm-ppc/segment.h does not exist
1438 Details : fix compile issue on PPC.
1442 Description: better handle error messages in extents code
1445 Frequency : RHEL4 only
1447 Description: mkfs is very slow on IA64/RHEL4
1448 Details : A performance regression has been discovered in the MPT Fusion
1449 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1450 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1451 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1453 Severity : enhancement
1455 Description: SNMP support enhancement
1456 Details : Adding total number of sampled request for an MDS node in snmp
1459 Severity : enhancement
1461 Description: Optimize ldlm waiting list processing for PR extent locks
1462 Details : When processing waiting list for read extent lock and meeting read
1463 lock that is same or wider to it that is not contended, skip
1464 processing rest of the list and immediatelly return current
1465 status of conflictness, since we are guaranteed there are no
1466 conflicting locks in the rest of the list.
1470 Description: Time out and refuse to reconnect
1471 Details : When the failover node is the primary node, it is possible
1472 to have two identical connections in imp_conn_list. We must
1473 compare not conn's pointers but NIDs, otherwise we can defeat
1474 connection throttling.
1478 Description: port llog fixes from b1_6 into HEAD
1479 Details : Port llog reference couting and some llog cleanups from b1_6
1480 (bug 10800) into HEAD, for protect from panic and access to already
1481 free llog structures.
1485 Description: Detect stride IO mode in read-ahead
1486 Details : When a client does stride read, read-ahead should detect that and
1487 read-ahead pages according to the detected stride pattern.
1491 Description: data checksumming impacts single node performance
1492 Details : add support for several checksum algorithm. Currently, only CRC32
1493 and Adler-32 are supported. The checksum type can be changed on
1494 the fly via /proc/fs/lustre/osc/*/checksum_type.
1498 Description: use adler32 for page checksums
1499 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1504 Description: build for x2 fails
1505 Details : fix compile issue on Cray systems.
1509 Description: Properly match for duplicate locks
1510 Details : Due to different lock order from skiplists code, we need to
1511 traverse entire list for now
1514 Frequency : only on PPC/SLES10
1516 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1517 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1518 be defined. Add a hack in mkfs_lustre.c to work around this compile
1523 Description: LASSERT on MDS when client holding flock lock dies
1524 Details : ldlm pool logic depends on number of granted locks equal to
1525 number of released locks which is not true for flock locks, so
1526 just exclude such locks from consideration.
1530 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1531 Details : Use fsfilt_sync() for both the whole filesystem sync and
1532 individual file sync to eliminate dangerous inode locking
1533 with I_LOCK that can lead to a deadlock.
1537 Description: performance in 1.6.3
1538 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1539 sure that 1MB requests are not fragmented by the block layer.
1541 Severity : enhancement
1543 Description: organize the server-side client stats on per-nid basis
1544 Details : Change the structure of stats under obdfilter and mds to
1552 The "uuid"s file would list the uuids of _active_ exports.
1553 And the clear entry is to clear all stats and stale nids.
1555 Severity : enhancement
1557 Description: eliminate client locks in face of contention
1558 Details : file contention detection and lockless i/o implementation
1559 for contended files.
1563 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1566 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1568 Description: MDS may lose file striping (and hence file data) in some cases
1569 Details : If there are additional extended attributes stored on the MDS,
1570 in particular ACLs, SELinux, or user attributes (if user_xattr
1571 is specified for the client mount options) then there is a risk
1572 of attribute loss. Additionally, the Lustre file striping
1573 needs to be larger than default (e.g. striped over all OSTs),
1574 and an additional attribute must be stored initially in the
1575 inode and then increase in size enough to be moved to the
1576 external attribute block (e.g. ACL growing in size) for file
1581 Description: skiplist implementation simplification
1582 Details : skiplists are used to group compatible locks on granted list
1583 that was implemented as tracking first and last lock of each
1584 lock group the patch changes that to using doubly linked lists
1588 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1589 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1590 because LASSERT on both the data supplied by a client, and the
1591 data on disk is dangerous and incorrect.
1593 Severity : enhancement
1595 Description: Slow truncate/writes to huge files at high offsets.
1596 Details : Directly associate cached pages to lock that protect those pages,
1597 this allows us to quickly find what pages to write and remove
1598 once lock callback is received.
1602 Description: more ldlm soft lockups
1603 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1604 starve other threads from the resource lock for a long time in
1605 case of long waiting queue, so change the debug level from
1606 D_OTHER to the less frequently used D_INFO.
1608 Severity : enhancement
1610 Description: add -gid, -group, -uid, -user options to lfs find
1614 Description: Hung threads in invalidate_inode_pages2_range
1615 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1616 once one is completed. As a result, some RPCs are stuck in the
1617 queue and are never sent.
1621 Description: filter threads hungs on waiting journal commit
1622 Details : Cleanup filter group llog code, then only filter group llog will
1623 be only created in the MDS/OST syncing process.
1627 Description: Procfs and llog threads access destoryed import sometimes.
1628 Details : Sync the import destoryed process with procfs and llog threads by
1629 the import refcount and semaphore.
1631 Severity : enhancement
1633 Description: openlock cache of b1_6 port to HEAD
1638 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1639 Details : If insertion of an extent fails, then discard the inode
1640 preallocation and free data blocks else it can lead to duplicate
1645 Description: don't always update ctime in ext3_xattr_set_handle()
1646 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1647 In some cases the ctime should not be updated, for example for
1648 2.0->1.8 compatibility it is necessary to delete an xattr and it
1649 should not update the ctime.
1653 Bugzilla : 15713/16362
1654 Description: Assertion in iopen_connect_dentry in 1.6.3
1655 Details : looking up an inode via iopen with the wrong generation number can
1656 populate the dcache with a disconneced dentry while the inode
1657 number is in the process of being reallocated. This causes an
1658 assertion failure in iopen since the inode's dentry list contains
1659 both a connected and disconnected dentry.
1663 Description: assertion failure in ldlm_handle2lock()
1664 Details : fix a race between class_handle_unhash() and class_handle2object()
1665 introduced in lustre 1.6.5 by bug 13622.
1670 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1671 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1676 Description: oops in sd_iostats_seq_show()
1677 Details : unloading/reloading the scsi low level driver triggers a kernel
1678 bug when trying to access the sd iostat file.
1683 Description: Kernel panics during QLogic driver reload
1684 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1685 patch, causing memory corruption.
1690 Description: journal_dev option does not work in b1_6
1691 Details : pass mount option during pre-mount.
1693 Severity : enhancement
1695 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1696 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1697 extent information of a file. It can be used to map logical blocks
1698 in a file to physical blocks in the block device.
1702 Description: LDLM soft lockups - improvement
1703 Details : It is be possible to send the lock handle along with each read
1704 or write request because the client is already doing a lock match
1705 itself so there isn't any reason the OST should have to re-do that
1709 Frequency : only on Cray X2
1711 Description: X2 build failures
1712 Details : fix build failures on Cray X2.
1716 Description: xid & resent requests
1717 Details : Initialize RPC XID from clock at startup (randomly if clock is
1720 Severity : enhancement
1722 Description: Add lustre_start utility to start or stop multiple Lustre servers
1727 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1728 Details : In case of memory pressure, list_del() can be called twice on
1729 req->rq_history_list, causing a kernel oops.
1733 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1734 Details : fix stack overflow in the distributed lock manager by defering
1735 export eviction after a failed AST to the elt thread instead of
1736 handling it in the dlm interpret routine.
1740 Description: Convert some comments to new format.
1741 Details : Update documenting comments to match doxygen conventions.
1745 Description: Grammar fixes.
1746 Details : A couple of trivial spelling fixes.
1750 Description: OSD_COUNTERS-mandatory
1751 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1752 to simplify the code.
1756 Description: simplify cmm_device freeing logic.
1757 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1761 Description: Add lockdep support to dt_object_operations locking interface.
1762 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
1763 parameter indicating lock ordering. Update mdd code to use new
1768 Description: Introduce failloc constants for lockless IO tests.
1769 Details : Add two new failloc constants to test lockless IO. Only one of
1770 them in implemented---another is checked in yet to be landed
1775 Description: Add lockdep support for inode mutex.
1776 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1777 in the situations where Lustre has to lock more than one inode
1782 Description: Add optional invariants checking support.
1783 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1784 configure switch. This macro is to be used for consistency and
1785 sanity checks that are too expensive to be left in `production'
1790 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1791 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1796 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1797 Details : Use spin_lock_nested() in (the only) situation where more than
1798 one ldlm_lock is locked simultaneously. Also, fix possible
1799 dead-lock in ldlm_lock_change_resource() by enforcing particular
1804 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1805 Details : Instead of specifying each ldlm_lock call-back through separate
1806 parameter, wrap them into struct ldlm_callback_suite.
1810 Description: Kill join_lru obd method and its callers.
1811 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
1812 in memory---a requirement implemented through ->o_join_lru() obd
1813 method in HEAD. Remove this method and its users.
1817 Description: Add asynchronous ldlm ENQUEUE completion handler.
1818 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
1819 case that stock ldlm_completion_ast() cannot handle as it waits
1820 until lock is granted. Introduce new ldlm_completion_ast_async()
1821 for this. Also comment ldlm_completion_ast().
1825 Description: ldlm_error <-> errno conversion.
1826 Details : Add functions to map (rather arbitrary) between LDLM error codes
1827 and standard errno values. CLIO needs this to prevent LDLM specific
1828 constants from escaping ldlm and osc.
1832 Description: Kill unused ldlm_handle2lock_ns() function.
1833 Details : Kill unused ldlm_handle2lock_ns() function.
1837 Description: Add lu_ref support to ldlm_lock
1838 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
1839 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
1840 and ldlm_resource. LDLM interface has to be changed, because code that
1841 releases a reference on a lock, has to "know" what reference this is.
1842 In the most frequent case
1844 lock = ldlm_handle2lock(handle);
1846 LDLM_LOCK_PUT(lock);
1848 no changes are required. When any other reference (received _not_ from
1849 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
1850 instead of LDLM_LOCK_PUT().
1852 Arguably, changes are pervasive, and interface requires some discipline
1853 for proper use. On the other hand, it was very instrumental in finding
1854 a few leaked lock references.
1858 Description: Add ldlm_lock_addref_try().
1859 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
1860 attempts to addref a lock that might be being canceled
1865 Description: Add ldlm_weigh_callback().
1866 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
1867 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
1868 hard-coded `number of pages' logic.
1872 Description: Add lockdep annotations to llog code.
1873 Details : Use appropriately tagged _nested() locking calls in the places
1874 where llog takes more than one ->lgh_lock lock.
1878 Description: Add loi_kms_set().
1879 Details : Wrap kms updates into a helper function.
1883 Description: Constify instances of struct lsm_operations.
1884 Details : Constify instances of struct lsm_operations.
1888 Description: lu_conf support.
1889 Details : On a server, a file system object is uniquely identified
1890 by a fid, which is sufficient to locate and load all object
1891 state (inode). On a client, on the other hand, more data are
1892 necessary instantiate an object. Change lu_object_find() and
1893 friends to take additional `lu_conf' argument describing object.
1894 Typically this includes layout information.
1898 Description: lu_context fixes.
1899 Details : Introduce new lu_context functions that are needed on the client
1900 side, where some system threads (ptlrpcd) are shared by multiple
1901 modules, and so cannot be stopped during module shutdown.
1905 Description: Add start and stop methods to lu_device_type_operations.
1906 Details : Introduce two new methods in lu_device_type_operations, that are
1907 invoked when first instance of a given type is created and last one
1908 is destroyed respectively. This is need by CLIO.
1912 Description: Add lu_ref support to struct lu_device.
1913 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
1914 track leaked references.
1918 Description: Introduce lu_kmem_descr.
1919 Details : lu_kmem_descr and its companion interface allow to create
1920 and destroy a number of kmem caches at once.
1924 Description: Fix lu_object finalization race.
1925 Details : Fix a race between lu_object_find() finding an object and its
1926 concurrent finalization. This race is (most likely) not possible
1927 on the server, but might happen on the client.
1931 Description: Introduce lu_ref interface.
1932 Details : lu_ref is a debugging module allowing to track references to
1933 a given object. It is quite cpu expensive, and has to be
1934 explicitly enabled with --enable-lu_ref. See usage description
1939 Description: Factor lu_site procfs stats into a separate function.
1940 Details : Separate lu_site stats printing code into a separate function
1941 that can be reused on a client.
1945 Description: Constify instances of struct {lu,dt,md}_device_operations.
1946 Details : Constify instances of struct {lu,dt,md}_device_operations.
1950 Description: Introduce struct md_site and move meta-data specific parts of
1951 struct lu_site here.
1952 Details : Move md-specific fields out of struct lu_site into special struct
1953 md_site, so that lu_site can be used on a client.
1957 Description: Kill mdd_lov_destroy().
1958 Details : Remove unused mdd code.
1962 Description: Add st_block checking to multistat.c.
1963 Details : Add st_block checking to multistat.c.
1967 Description: Add lu_ref support to struct obd_device.
1968 Details : Add lu_ref tracking to obd_device.
1972 Description: Kill obd_set_fail_loc().
1973 Details : Remove unused code.
1977 Description: Add special type for ptlrpc_request interpret functions.
1978 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
1979 there. Actual usage will be in CLIO.
1983 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
1984 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
1985 doesn't work with lockdep.
1989 Description: Add rwv.c test program.
1990 Details : New testing program exercising readv(2) and writev(2) (Qian).
1994 Description: Add sendfile.c test program.
1995 Details : New testing program exercising sendfile(2) (Jay).
1999 Description: Ratelimit a message that can be very frequent.
2000 Details : Ratelimit a memory allocation failure message that can
2005 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2006 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2010 Description: Liblustre build fixes.
2011 Details : Liblustre build fixes.
2015 Description: libcfs: add cfs_{need,cond}_resched() interface.
2016 Details : libcfs: add cfs_{need,cond}_resched() definition and
2017 implementations for Linux, NT, and liblustre.
2019 Severity : enhancement
2021 Description: More exported tunables for mballoc
2022 Details : Add support for tunable preallocation window and new tunables for
2023 large/small requests
2027 Description: Detect corruption of block bitmap and checking for preallocations
2028 Details : Checks validity of on-disk block bitmap. Also it does better
2029 checking of number of applied preallocations. When corruption is
2030 found, it turns filesystem readonly to prevent further corruptions.
2034 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2035 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2036 window to 0. If it does hit cache miss, and read-pattern is still
2037 stride-io mode, does not reset the stride window, but also does not
2038 increase the stride window length in this case.
2042 Frequency : only for big-endian servers
2043 Description: Check if system is big-endian while mounting fs with extents feature
2044 Details : Mounting a filesystem with extents feature will fail on big-endian
2045 systems since ext3-based ldiskfs is not supported on big-endian
2046 systems. This can be over-riden with "bigendian_extents" mount option.
2048 Severity : enhancement
2050 Description: The root squash functionality
2051 Details : A security feature, which is to prevent users from being able
2052 to mount lustre on their desktop, run as root, and delete
2053 all of the files in the filesystem. The goal is accomplished by
2054 remapping user id (UID) and group id (GID) of the root user to
2055 a UID and GID specified by the system administartor via Lustre
2056 configuration management server (MGS). The functionality also
2057 allows to specify sets of clients for which the remapping does
2062 Description: Excessive recovery window
2063 Details : With AT enabled, the recovery window can be excessively long (6000+
2064 seconds). To address this problem, we no longer use
2065 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2066 timeout no longer depends on the service time, it is set to
2067 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2068 time via pb_service_time.
2072 Description: Watchdog triggered on MDS failover
2073 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2074 the OSTs know that the MDS "UUID" can be reused for the same export
2075 from a different NID, so we do not need to wait for the export to be
2079 Frequency : rare, only if using MMP with Linux RAID
2081 Description: MMP doesn't work with Linux RAID
2082 Details : While using HA for Lustre servers with Linux RAID, it is possible
2083 that MMP will not detect multiple mounts. To make this work we
2084 need to unplug the device queue in RAID when the MMP block is being
2085 written. Also while reading the MMP block, we should read it from
2086 disk and not the cached one.
2088 Severity : enhancement
2090 Description: open file using fid
2091 Details : A file can be opened using just its fid, like
2092 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2095 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2098 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2099 Details : add proper locking for creation of jbd_4k slab cache
2103 Description: MMP check in ext3_remount() fails without displaying any error
2104 Details : When multiple mount protection fails during remount, proper error
2107 Severity : enhancement
2109 Description: Allow stripe size to be up to 4G-64k
2110 Details : Fix math logic to allow large stripe sizes.
2114 Description: add check for >8TB ldiskfs filesystems
2115 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2116 Don't allow >8TB ldiskfs filesystems to be mounted without
2117 force_over_8tb mount option
2119 --------------------------------------------------------------------------------
2121 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2123 * Support for kernels:
2124 2.6.5-7.283 (SLES 9),
2125 2.6.9-55.EL (RHEL 4),
2126 2.6.16.46-0.14 (SLES 10),
2127 2.6.18.8 vanilla (kernel.org)
2128 * Client support for unpatched kernels:
2129 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2130 2.6.16 - 2.6.22 vanilla (kernel.org)
2131 * Due to recently discovered recovery problems, we do not recommend
2132 using patchless RHEL 4 clients with this or any earlier release.
2133 * Recommended e2fsprogs version: 1.39.cfs8
2134 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2135 * Starting with this release, the ldiskfs backing filesystem required
2136 by Lustre is now in its own package, lustre-ldiskfs. This package
2137 should be installed. It is versioned separately from Lustre and
2138 may be released separately in future.
2142 Description: block reactivating mgc import until all deactivates complete
2143 Details : Fix race when failing back MDT/MGS to itself (testing)
2146 Frequency : at statup only
2148 Description: mds_lov_synchronize race leads to various problems
2149 Details : simultaneous MDT->OST connections at startup can cause the
2150 sync to abort, leaving the OSC in a bad state.
2152 Severity : enhancement
2154 Description: add optional extra BUILD_VERSION info
2155 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2156 to override the lustre version.
2159 Frequency : 2.6.18 servers only
2161 Description: ll_kern_mount() doesn't release the module reference
2162 Details : The ldiskfs module reference count never drops down to 0
2163 because ll_kern_mount() doesn't release the module reference.
2168 Description: server LBUG when using old ost_num_threads parameter
2169 Details : Accept the old ost_num_threads parameter but warn that it
2170 is deprecated, and fix an off-by-one error that caused an LBUG.
2175 Description: Transient SCSI error results in persistent IO issue
2176 Details : iobuf->dr_error is not reinitialized to 0 between two
2180 Frequency : sometimes when underlying device returns I/O errors
2182 Description: OSTs not going read-only during write failures
2183 Details : OSTs are not remounted read-only when the journal commit threads
2184 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2185 instead of the ext3 wrappers.
2189 Description: poor connect scaling with increasing client count
2190 Details : Don't run filter_grant_sanity_check for more than 100 exports
2191 to improve scaling for large numbers of clients.
2194 Frequency : SLES10 only
2196 Description: sanity-quota.sh quotacheck failed: rc = -22
2197 Details : Quotas cannot be enabled on SLES10.
2200 Frequency : liblustre clients only
2202 Description: getdirentries does not give error when run on compute nodes
2203 Details : getdirentries does not fail when the size specified as an argument
2204 is too small to contain at least one entry
2206 Severity : enhancement
2208 Description: Add LNET router traceability for debug purposes
2209 Details : If a checksum failure occurs with a router as part of the
2210 IO path, the NID of the last router that forwarded the bulk data
2211 is printed so it can be identified.
2216 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2217 Details : Due to a race condition, liblustre clients were occasionally
2218 evicted incorrectly.
2220 Severity : enhancement
2222 Description: lfs setstripe use optional parameters instead of postional
2225 Severity : enhancement
2227 Description: Nanosecond timestamp support for ldiskfs
2228 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2229 resolution timestamps. There is not yet support for this at
2230 the Lustre filesystem level.
2233 Frequency : during server recovery
2235 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2236 Details : request with rq_no_resend flag not awake l_wait_event if they get
2240 Frequency : nfs export on patchless client
2242 Description: connectathon hang when test nfs export over patchless client
2243 Details : Disconnected dentry cannot be found with lookup, so we do not need
2244 to unhash it or make it invalid
2247 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2248 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2249 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2254 Description: open req refcounting wrong on reconnect
2255 Details : If reconnect happened between getting open reply from server and
2256 call to mdc_set_replay_data in ll_file_open, we will schedule
2257 replay for unreferenced request that we are about to free.
2258 Subsequent close will crash in variety of ways.
2259 Check that request is still eligible for replay in
2260 mdc_set_replay_data().
2265 Description: disable writes to filesystem when reading health_check file
2266 Details : the default for reading the health_check proc file has changed
2267 to NOT do a journal transaction and write to disk, because this
2268 can cause reads of the /proc file to hang and block HA state
2269 checking on a healthy but otherwise heavily loaded system. It
2270 is possible to return to the previous behaviour during configure
2271 with --enable-health-write.
2273 Severity : enhancement
2275 Description: 64-bit inode version
2276 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2277 to the inode. This will be required for version-based recovery.
2282 Description: MDS fails to start if a duplicate client export is detected
2283 Details : in some rare cases it was possible for a client to connect to
2284 an MDS multiple times. Upon recovery the MDS would detect this
2285 and fail during startup. Handle this more gracefully.
2287 Severity : enhancement
2289 Description: Add -o localflock option to simulate old noflock
2291 Details : This will achieve local-only flock/fcntl locks
2297 Description: log_commit_thread vs filter_destroy race leads to crash
2298 Details : Take import reference before releasing llog record semaphore
2303 Description: Wrong request locking in request set processing
2304 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2305 imp_delayed_list, in this place should be used imp_lock.
2308 Frequency : when reconnection
2310 Description: Grant Leak when osc reconnect to OST
2311 Details : When osc reconnect ost, OST(filter) should check whether it
2312 should grant more space to client by comparing fed_grant and
2313 cl_avail_grant, and return the granted space to client instead
2314 of "new granted" space, because client will call osc_init_grant
2315 to update the client grant space info.
2318 Frequency : when client reconnect to OST
2320 Description: Grant Leak when osc do resend and replay bulk write
2321 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2322 bulk write request, because the grant info while be sync between
2323 OSC and OST when reconnect, and we should ignore the grant info
2324 these of resend/replay write req.
2329 Description: Grant space more than avaiable left space sometimes.
2330 Details : When then OST is about to be full, if two bulk writing from
2331 different clients came to OST. Accord the avaliable space of the
2332 OST, the first req should be permitted, and the second one
2333 should be denied by ENOSPC. But if the seconde arrived before
2334 the first one is commited. The OST might wrongly permit second
2335 writing, which will cause grant space > avaiable space.
2338 Frequency : when client is evicted
2340 Description: Grant might be wrongly erased when osc is evicted by OST
2341 Details : when the import is evicted by server, it will fork another
2342 thread ptlrpc_invalidate_import_thread to invalidate the
2343 import, where the grant will be set to 0. While the original
2344 thread will update the grant it got when connecting. So if
2345 the former happened latter, the grant will be wrongly errased
2346 because of this race.
2351 Description: Checking Stale with correct fid
2352 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2353 to check whether it is stale, because sometimes, we want the
2354 enqueue happened anyway, and op_data.fid2 will not be initialized.
2356 Severity : enhancement
2358 Description: update patchless client
2359 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2362 Frequency : only with 2.4 kernel
2364 Description: random memory corruption
2365 Details : size of struct ll_inode_info is to big for union inode.u and this
2366 can be cause of random memory corruption.
2371 Description: Memory leak in recovery
2372 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2373 It should also check obd_fail before fsfilt_start, otherwise if
2374 fsfilt_start return -EROFS,(failover mds during mds recovery).
2375 then the req will return with repmsg->transno = 0 and rc = EROFS.
2376 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2377 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2378 be freed no matter whether fsfilt_commit success or not.
2381 Frequency : only with huge count clients
2383 Description: Prevents from taking the superblock lock in llap_from_page for
2385 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2386 for llap_from_page call in ll_removepage prevents from taking the
2387 superblock lock for a soon died page.
2392 Description: Not check open intent error before release open handle
2393 Details : in some rare cases, the open intent error is not checked before
2394 release open handle, which may cause
2395 ASSERTION(open_req->rq_transno != 0), because it tries to release
2396 the failed open handle.
2401 Description: Set cat log bitmap only after create log success.
2402 Details : in some rare cases, the cat log bitmap is set too early. and it
2403 should be set only after create log success.
2407 Description: Accessing a block bevice can re-enable I/O when Lustre is
2408 tearing down a device.
2409 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2414 Description: service threads may hog cpus when there are a lot of requests
2415 Details : Insert cond_resched to give other threads a chance to use some CPU
2420 Description: the cat log was not initialized in recovery
2421 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2422 unlink log on mds will not be initialized until mds post
2423 recovery. And also in mds post recovery, the unlink log will
2424 initialization will be done asynchronausly, so there will be race
2425 between add unlink log and unlink log initialization.
2429 Description: brw_stats were being printed incorrectly
2430 Details : brw_stats were being printed as log2 but all of them were not
2431 recorded as log2. Also remove some code duplication arising from
2432 filter_tally_{read,write}.
2436 Frequency : rare, only in recovery.
2437 Description: ASSERTION(req->rq_type != LI_POISON) failed
2438 Details : imp_lock should be held while iterating over imp_sending_list for
2439 prevent destroy request after get timeout in ptlrpc_queue_wait.
2443 Description: replay-single.sh test 52 fails
2444 Details : A lock's skiplist need to be cleanup when it being unlinked
2445 from its resource list.
2449 Description: Short directio read returns full requested size rather than
2451 Details : Direct I/O operations should return actual amount of bytes
2452 transferred rather than requested size.
2454 Severity : enhancement
2456 Description: metadata RPC reduction (e.g. for rm performance)
2457 Details : decrease the amount of synchronous RPC between clients and servers
2458 by canceling conflicing lock before the operation on the client
2459 and packing thier handles into the main operation RPC to server.
2461 Severity : enhancement
2463 Description: Async OSC create to avoid the blocking unnecessarily.
2464 Details : If a OST has no remain object, system will block on the creating
2465 when need to create a new object on this OST. Now, ways use
2466 pre-created objects when available, instead of blocking on an
2467 empty osc while others are not empty. If we must block, we block
2468 for the shortest possible period of time.
2472 Description: improve handling recoverable errors
2473 Details : if request processig with error which can be recoverable on server
2474 request should be resend, otherwise page released from cache and
2477 Severity : enhancement
2479 Description: refine locking for avoid write wrong info into lov_objid file
2480 Details : fix possible races with add new target and write/update data in
2483 --------------------------------------------------------------------------------
2485 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2490 Frequency : on some architectures
2492 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2493 Details : Uninitialized flags sometimes cause configuration commands to
2497 Frequency : patchless clients only
2499 Description: missing __iget() symbol export
2500 Details : The __iget() symbol export is missing. To avoid the need for
2501 this on patchless clients the deathrow inode reaper is turned
2502 off, and we depend on the VM to clean up old inodes. This
2503 dependency was during via the fix for bug 12181.
2505 --------------------------------------------------------------------------------
2507 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2509 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2510 INTEROPERATE with older versions automatically. In many cases a
2511 special upgrade step is needed. Please read the
2512 user documentation before upgrading any part of a live system.
2513 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2514 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2515 * WARNING: Lustre configuration and startup changes are required with
2516 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2520 Severity : enhancement
2522 Description: MountConf
2523 Details : Lustre configuration is now managed via mkfs and mount
2524 commands instead of lmc and lconf. New obd types (MGS, MGC)
2525 are added for dynamic configuration management. See
2526 https://mail.clusterfs.com/wikis/lustre/MountConf for
2529 Severity : enhancement
2531 Description: dynamic OST addition
2532 Details : OSTs can now be added to a live filesystem
2534 Severity : enhancement
2536 Description: startup order invariance
2537 Details : MDTs and OSTs can be started in any order. Clients only
2538 require the MDT to complete startup.
2540 Severity : enhancement
2542 Description: parallel, asynchronous orphan cleanup
2543 Details : orphan cleanup is now performed in separate threads for each
2544 OST, allowing parallel non-blocking operation.
2546 Severity : enhancement
2548 Description: optimized stripe assignment
2549 Details : stripe assignments are now made based on ost space available,
2550 ost previous usage, and OSS previous usage, in order to try
2551 to optimize storage space and networking resources.
2553 Severity : enhancement
2555 Description: Permanently set tunables
2556 Details : All writable /proc/fs/lustre tunables can now be permanently
2557 set on a per-server basis, at mkfs time or on a live system.
2559 Severity : enhancement
2561 Description: Lustre message v2
2562 Details : Add lustre message format v2.
2564 Severity : enhancement
2566 Description: client OST exclusion list
2567 Details : Clients can be started with a list of OSTs that should be
2568 declared "inactive" for known non-responsive OSTs.
2572 Description: SPEC SFS validation failure on NFS v2 over lustre.
2573 Details : Changes the blocksize for regular files to be 2x RPC size,
2574 and not depend on stripe size.
2576 Severity : enhancement
2578 Description: Multiple MD RPCs in flight.
2579 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2580 To avoid overly-overloading MDS, introduce a limit on number of
2581 MDS RPCs in flight for a single client and add /proc controls
2582 to adjust this limit.
2584 Severity : enhancement
2586 Description: client read/write statistics
2587 Details : Add client read/write call usage stats for performance
2588 analysis of user processes.
2589 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2590 file access. extents_stats shows chunk size distribution.
2591 extents_stats_per_process show chunk size distribution per
2594 Severity : enhancement
2596 Description: mds statistics
2597 Details : Add detailed mds operations statistics in
2598 /proc/fs/lustre/mds/*/stats.
2602 Description: Failure of copying files with lustre special EAs.
2603 Details : Client side always return success for setxattr call for lustre
2604 special xattr (currently only "trusted.lov").
2609 Description: Refcount LNET uuids
2610 Details : The global LNET uuid list grew linearly with every startup;
2611 refcount repeated list entries instead of always adding to
2614 Severity : enhancement
2616 Description: Dynamic service threads
2617 Details : Within a small range, start extra service threads
2618 automatically when the request queue builds up.
2621 Frequency : mixed-endian client/server environments
2623 Description: mixed-endian crashes
2624 Details : The new msg_v2 system had some failures in mixed-endian
2627 Severity : enhancement
2629 Description: Easy OST removal
2630 Details : OSTs can be permanently deactivated with e.g. 'lctl
2631 conf_param lustre-OST0001.osc.active=0'
2633 Severity : enhancement
2635 Description: MGS proc entries
2636 Details : Added basic proc entries for the MGS showing what filesystems
2639 Severity : enhancement
2641 Description: provide MGS failover
2642 Details : Added config lock reacquisition after MGS server failover.
2644 Severity : enhancement
2646 Description: add Linux 2.4 support
2647 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2651 Description: a large application tries to do I/O to the same resource and dies
2652 in the middle of it.
2653 Details : Check the req->rq_arrival time after the call to
2654 ost_brw_lock_get(), but before we do anything about
2655 processing it & sending the BULK transfer request. This
2656 should help move old stale pending locks off the queue as
2657 quickly as obd_timeout.
2660 Frequency : when an incorrect nid is specified during startup
2662 Description: ptlrpc connect to non-existant node causes kernel crash
2663 Details : LNET can't be re-entered from an event callback, which
2664 happened when we expire a message after the export has been
2665 cleaned up. Instead, hand the zombie cleanup off to another
2668 Severity : enhancement
2670 Description: plain/inodebits lock performance improvement
2671 Details : Grouping plain/inodebits in granted list by their request modes
2672 and bits policy, thus improving the performance of search through
2676 Frequency : only if OST filesystem is corrupted
2678 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2679 Details : for a short time RPCs with bulk IO are in the replay list,
2680 but replay of bulk IOs is unimplemented. If the OST filesystem
2681 is corrupted due to disk cache incoherency and then replay is
2682 started it is possible to trip an assertion. Avoid putting
2683 committed RPCs into the replay list at all to avoid this issue.
2686 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2688 Description: System hang on startup
2689 Details : This bug allowed the liblustre (e.g. catamount) client to
2690 return to the app before handling all startup RPCs. This
2691 could leave the node unresponsive to lustre network traffic
2692 and manifested as a server ptllnd timeout.
2694 Severity : enhancement
2696 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2697 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2698 boolean to control whether to print peer debug info when a
2699 client's RPC times out.
2702 Frequency : only for kernels with patches from Lustre below 1.4.3
2704 Description: Remove old rdonly API
2705 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2708 Frequency : only for devices with external journals
2710 Description: Set external device read-only also
2711 Details : During a commanded failover stop, we set the disk device
2712 read-only while the server shuts down. We now also set any
2713 external journal device read-only at the same time.
2716 Frequency : when upgrading from 1.4 while trying to change parameters
2718 Description: The wrong (new) MDC name was used when setting parameters for
2719 upgraded MDT's. Also allows changing of OSC (and MDC)
2720 parameters if --writeconf is specified at tunefs upgrade time.
2723 Frequency : when setting specific ost indicies
2725 Description: QOS code breaks on skipped indicies
2726 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2727 created with --index need not be sequential.
2729 Severity : enhancement
2731 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2732 Details : The uninit_groups feature works in conjunction with the kernel
2733 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2734 up the pass1 processing of e2fsck. This is a read-only feature
2735 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2736 mount filesystems that have had this feature enabled.
2738 Severity : enhancement
2740 Description: Improve multi-block allocation algorithm to avoid fragmentation
2741 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2742 allocation locality and avoid filesystem fragmentation.
2744 ------------------------------------------------------------------------------
2746 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2748 * Support for kernels:
2749 2.4.21-47.0.1.EL (RHEL 3)
2750 2.6.5-7.283 (SLES 9)
2751 2.6.9-42.0.10.EL (RHEL 4)
2752 2.6.12.6 vanilla (kernel.org)
2753 2.6.16.27-0.9 (SLES 10)
2754 * Recommended e2fsprogs version: 1.39.cfs5
2756 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2760 Frequency : occasional, depends on client load and configuration
2761 Bugzilla : 12181, 12203
2762 Description: data loss for recently-modified files
2764 Details : In some cases it is possible that recently written or created
2765 files may not be written to disk in a timely manner (this should
2766 normally be within 30s unless client IO load is very high).
2767 The problem appears as zero-length files or files that are a
2768 multiple of 1MB in size after a client crash or client eviction
2769 that are missing data at the end of the file.
2771 This problem is more likely to be hit on clients where files are
2772 repeatedly created and unlinked in the same directory, clients
2773 have a large amount of RAM, have many CPUs, the filesystem has
2774 many OSTs, the clients are rebooted frequently, and/or the files
2775 are not accessed by other nodes after being written.
2777 The presence of the problem can be detected by looking at
2778 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2779 smaller than the second (nr_unused) then dirty files will not
2780 be flushed automatically to disk. "sync; sleep 10" should be
2781 run several times on the node before unmounting it to update
2782 Lustre (this is also safe to run on nodes without this problem).
2784 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2785 that can cause this same problem, so customers using that kernel
2786 also need to update the kernel in addition to Lustre. In order
2787 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2789 It is normal that files written just before a client crash (less
2790 than 30s) may not yet have been flushed to disk, even for local
2794 Frequency : frequent on thin XT3 nodes
2796 Description: UUID collision on thin XT3 Linux nodes
2797 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2798 randomly, since we relied on an insufficiently-seeded PRNG.
2803 Description: OSS hangs after "All ost request buffers busy"
2804 Details : A deadlock between quota and journal operations caused OSS
2805 hangs after printing "All ost request buffers busy."
2808 Frequency : always on liblustre builds
2810 Description: Cleanup compiler warnings on liblustre
2813 Frequency : always on liblustre builds on XT3
2815 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2819 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2820 > 32000 subdirectories
2821 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2822 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2823 subdirectory count crosses 32000. This will aid e2fsck to
2824 correctly handle more than 32000 subdirectories.
2827 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2829 Description: System hang on startup
2830 Details : This bug allowed the liblustre (e.g. catamount) client to
2831 return to the app before handling all startup RPCs. This
2832 could leave the node unresponsive to lustre network traffic
2833 and manifested as a server ptllnd timeout.
2835 Severity : enhancement
2837 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2838 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2839 boolean to control whether to print peer debug info when a
2840 client's RPC times out.
2845 Description: make O_SYNC working on 2.6 kernels
2846 Details : 2.6 kernels use different method for mark pages for write,
2847 so need add a code to lustre for O_SYNC work.
2852 Description: Failure to close file and release space on NFS
2853 Details : Put inode details into lock acquired in ll_intent_file_open.
2854 Use mdc_intent_lock in ll_intent_open to properly
2855 detect all kind of errors unhandled by mdc_enqueue.
2860 Description: proc file read during shutdown sometimes raced obd removal,
2862 Details : Add lock to prevent obd access after proc file removal.
2865 Frequency : Only for files larger than 4GB on 32-bit clients.
2867 Description: improperly doing page alignment of locks
2868 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2869 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2874 Description: RPCs being resent when they shouldn't be
2875 Details : Some RPCs that should not be resent are being resent. This
2876 can cause inconsistencies in the RPC state machine. Do not
2877 resend such requests.
2880 Frequency : rare, only with NFS export
2882 Description: Crash on NFS re-export node
2883 Details : under very unusual load conditions an assertion is hit in
2884 ll_intent_file_open()
2887 Frequency : only if OST filesystem is corrupted
2889 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2890 Details : for a short time RPCs with bulk IO are in the replay list,
2891 but replay of bulk IOs is unimplemented. If the OST filesystem
2892 is corrupted due to disk cache incoherency and then replay is
2893 started it is possible to trip an assertion. Avoid putting
2894 committed RPCs into the replay list at all to avoid this issue.
2899 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2900 Details : Large single O_DIRECT read and write calls can fail to allocate
2901 a sufficiently large buffer to process the request. In case of
2902 allocation failure the allocation is retried with a smaller
2903 buffer and broken into smaller requests.
2905 Severity : enhancement
2907 Description: Add -o localflock option to simulate old noflock behaviour.
2908 Details : This will achieve local-only flock/fcntl locks coherentness.
2913 Description: versioning check is incomplete
2914 Details : Checking the version difference of client vs. server, report
2915 error if the gap is too big.
2920 Description: add support PG_writeback bit
2921 Details : add support for PG_writeback bit for Lustre, for more carefull
2922 work with page cache in 2.6 kernel. This also fix some deadlocks
2923 and remove hack for work O_SYNC with 2.6 kernel.
2925 Severity : enhancement
2927 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2928 Details : The uninit_groups feature works in conjunction with the kernel
2929 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2930 up the pass1 processing of e2fsck. This is a read-only feature
2931 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2932 mount filesystems that have had this feature enabled.
2934 Severity : enhancement
2936 Description: Improve multi-block allocation algorithm to avoid fragmentation
2937 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2938 allocation locality and avoid filesystem fragmentation.
2940 ------------------------------------------------------------------------------
2942 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2944 * Support for kernels:
2945 2.6.9-42.0.3.EL (RHEL 4)
2946 2.6.5-7.276 (SLES 9)
2947 2.4.21-47.0.1.EL (RHEL 3)
2948 2.6.12.6 vanilla (kernel.org)
2949 2.6.16.21-0.8 (SLES10)
2950 * Recommended e2fsprogs version: 1.39.cfs2-0
2952 * The backwards-compatible /proc/sys/portals symlink has been removed
2953 in this release. Before upgrading, please ensure that you change
2954 any configuration scripts or /etc/sysctl.conf files that access
2955 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2956 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2957 in advance of the upgrade on any system running Lustre 1.4.6 or
2958 newer, since /proc/sys/lnet was added in that version.
2959 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2963 Frequency : only when quota is used
2965 Description: avoid scanning export list for quota master
2966 Details : Change the algorithms to avoid scanning export list in order
2967 to improve the efficiency.
2970 Frequency : MDS failover only, very rarely
2972 Description: "went back in time" messages on mds failover
2973 Details : The greatest transno may be lost when the current operation
2974 finishes with an error (transno==0) and the client's last_rcvd
2975 record is over-written. Save the greatest transno in the
2976 mds_last_transno for this case.
2979 Frequency : always for specific kernels and striping counts
2981 Description: client may get "Matching packet too big" without ACL support
2982 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2983 when trying to access files in certain configurations. The
2984 clients should in fact be denied when mounting because they do
2985 not understand ACLs.
2988 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2990 Description: many clients connecting with IO in progress causes connect timeouts
2991 Details : Avoid synchronous journal commits to avoid delays caused by many
2992 clients connecting/disconnecting when bulk IO is in progress.
2993 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2994 OST_IO_PORTAL to avoid delays behind potentially many pending
2998 Frequency : occasionally with multiple writers to a single file
3000 Description: shared writes to file may result in wrong size reported by stat()
3001 Details : Allow growing of kms when extent lock is cancelled
3004 Frequency : always with random mmap IO to multi-striped file
3006 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3007 Details : Only if the hole is at the end of OST object so that kms is too
3008 small. Fix is to increase kms accordingly in ll_nopage.
3011 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3013 Description: writes to a missing object would leak memory on the OST
3014 Details : If there is an inconsistency between the MDS and OST filesystems,
3015 such that the MDS references an object that doesn't exist, writes
3016 to that object will leak memory due to incorrect cleanup in the
3017 error handling path, eventually running out of memory on the OST.
3022 Description: Creating too long symlink causes lustre errors
3023 Details : Check symlink and name lengths before sending requests to MDS.
3026 Frequency : only if flock is enabled (not on by default)
3028 Description: posix locks not released on fd closure on 2.6.9+
3029 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3030 kernels, this caused such locks not to be released on fd close and
3031 then assertions on fs unmount about still used locks.
3034 Frequency : MDS failover only, very rarely
3036 Description: clients may get ASSERTION(granted_lock != NULL)
3037 Details : When request was taking a long time, and a client was resending
3038 a getattr by name lock request. The were multiple lock requests
3039 with the same client lock handle and
3040 mds_getattr_name->fixup_handle_for_resent_request found one of the
3041 lock handles but later failed with ASSERTION(granted_lock != NULL).
3046 Description: handle->h_buffer_credits > 0, assertion failure
3047 Details : h_buffer_credits is zero after truncate, causing assertion
3048 failure. This patch extends the transaction or creates a new
3052 Frequency : NFS re-export or patchless client
3053 Bugzilla : 11179, 10796
3054 Description: Crash on NFS re-export node (__d_move)
3055 Details : We do not want to hash the dentry if we don't have a lock.
3056 But if this dentry is later used in d_move, we'd hit uninitialised
3057 list head d_hash, so we just do this to init d_hash field but
3058 leave dentry unhashed.
3061 Frequency : NFS re-export or patchless client
3063 Description: NFS exports has problem with symbolic link
3064 Details : lustre client didn't properly install dentry when re-exported
3065 to NFS or running patchless client.
3068 Frequency : NFS re-export or patchless client
3070 Description: Various nfs/patchless fixes.
3071 Details : fixes reuse disconected alias for lookup process - this fixes
3072 warning "find_exported_dentry: npd != pd",
3073 fix permission error with open files at nfs.
3074 fix apply umask when do revalidate.
3077 Frequency : occasional
3079 Description: Crash on NFS re-export node
3080 Details : calling clear_page() on the wrong pointer triggered oops in
3081 generic_mapping_read().
3084 Frequency : rarely, using O_DIRECT IO
3086 Description: unaligned directio crashes client with LASSERT
3087 Details : check for unaligned buffers before trying any requests.
3090 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3092 Description: stale data returned from RAID cache
3093 Details : If only a small amount of IO is done to the RAID device before
3094 reading it again it is possible to get stale data from the RAID
3095 cache instead of reading it from disk.
3098 Frequency : always for sles10 kernel
3100 Description: sles10 support
3101 Details : ll_follow_link: compile fixes and using of nd_set_link
3102 under newer kernels.
3105 Frequency : depends on arch, kernel and compiler version, always on sles10
3108 Description: recursive or deep enough symlinks cause stack overflow
3109 Details : getting rid of large stack-allocated variable in
3113 Frequency : depends on hardware
3115 Description: lustre write performance loss in the SLES10 kernel
3116 Details : the performance loss is caused by using of write barriers in the
3117 ext3 code. The SLES10 kernel turns barrier support on by
3118 default. The fix is to undo that change for ldiskfs.
3120 ------------------------------------------------------------------------------
3122 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3124 * Support for kernels:
3125 2.6.9-42.0.3EL (RHEL 4)
3126 2.6.5-7.276 (SLES 9)
3127 2.4.21-47.0.1.EL (RHEL 3)
3128 2.6.12.6 vanilla (kernel.org)
3132 Frequency : quota enabled and large files being deleted
3134 Description: releasing more than 4GB of quota at once hangs OST
3135 Details : If a user deletes more than 4GB of files on a single OST it
3136 will cause the OST to spin in an infinite loop. Release
3137 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3142 Description: statfs data retrieved from /proc may be stale or zero
3143 Details : When reading per-device statfs data from /proc, in the
3144 {kbytes,files}_{total,free,avail} files, it may appear
3145 as zero or be out of date.
3148 Frequency : systems with MD RAID1 external journal devices
3150 Description: lconf's call to blkid is confused by RAID1 journal devices
3151 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3152 of returning all block devices that match the journal UUID.
3155 Frequency : always, for aggregate stripe size over 4GB
3157 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3158 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3159 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3162 Frequency : always if "lfs find" used on a local file/directory
3164 Description: "lfs find" segfaults if used on a local file/directory
3165 Details : The case where a directory component was not specified wasn't
3166 handled correctly. Handle this properly.
3169 Frequency : always on ppc64
3171 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3172 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3173 perform endianness conversion.
3176 Frequency : rarely (truncate to non-zero file size after write under load)
3177 Bugzilla : 10730, 10687
3178 Description: Files padded with zeros to next 4K multiple
3179 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3180 it is possible that files that are truncated to a non-zero size
3181 immediately after being written are filled with zero bytes beyond
3182 the truncated size. No file data is lost.
3184 Severity : enhancement
3186 Description: Allow recovery/failover for liblustre clients.
3187 Details : liblustre clients were unaware of failover configurations until
3190 Severity : enhancement
3192 Description: user file locks should fail when not mounting with flock option
3193 Details : Set up an error-returning stub in ll_file_operations.lock field
3194 to prevent incorrect behaviour when client is mounted without
3195 flock option. Also, set up properly f_op->flock field for
3199 Frequency : always on ia64
3201 Description: "lfs df" loops on printing out MDS statfs information
3202 Details : The obd_ioctl_data was not initialized and in some systems
3203 this caused a failure during the ioctl that did not return
3204 an error. Initialize the struct and return an error on failure.
3207 Frequency : SLES 9 only
3209 Description: Error of copying files with lustre special EAs as root
3210 Details : Client side always return success for setxattr call for lustre
3211 special xattr (currently only "trusted.lov").
3214 Frequency : rarely on clusters with both ia64+i386 clients
3216 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3217 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3218 clients are doing concurrent writes to the same file it is
3219 possible that the ia64 clients may LASSERT because the OST
3220 extent locks are not PAGE_SIZE aligned. Ensure that grown
3221 locks are always aligned on the request boundary.
3224 Frequency : specific use, occasional
3226 Description: Overwriting in use executable truncates on-disk binary image
3227 Details : If one node attempts to overwrite an executable in use by
3228 another node, we now correctly return ETXTBSY instead of
3229 truncating the file.
3231 Severity : enhancement
3233 Description: Async OSC create to avoid the blocking unnecessarily.
3234 Details : If a OST has no remain object, system will block on the creating
3235 when need to create a new object on this OST. Now, ways use
3236 pre-created objects when available, instead of blocking on an
3237 empty osc while others are not empty. If we must block, we block
3238 for the shortest possible period of time.
3243 Description: chmod on Lustre root is propagated to other clients
3244 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3245 invalid by the follow_mount time.
3248 Frequency : liblustre clients only
3250 Description: Race in 'instant cancel' lock handling could lead to such locks
3251 never to be granted in case of SMP MDS
3252 Details : Do not destroy not yet granted but cbpending locks in
3256 Frequency : replay/resend of open
3258 Description: non null lock assetion failure in mds_intent_policy
3259 Details : Trying to replay/resend lockless open requests resulted in
3260 mds_open() returning 0 with no lock. Now it sets a flag if
3261 a lock is going to be returned.
3263 Severity : enhancement
3265 Description: Checksum enhancements
3266 Details : New checksum enhancements allow for resending RPCs that failed
3269 Severity : enhancement
3271 Description: Tunables on number of dirty pages in cacche
3272 Details : Allow to set limit on number of dirty pages cached.
3277 Description: client crash on unmount - lock still has references
3278 Details : In some error handling cases it was possible to leak a lock
3279 reference on a client while accessing a file. This was not
3280 harmful to the client during operation, but would cause the
3281 client to crash when the filesystem is unmounted.
3284 Frequency : specific case, rare
3286 Description: ETXTBSY on mds though file not in use by client
3287 Details : ETXTBSY is no longer incorrectly returned when attempting to
3288 chmod or chown a directory that the user previously tried to
3289 execute or a currently-executing binary.
3292 Frequency : extremely rare except on liblustre-based clients
3294 Description: Lustre space not freed when files are deleted
3295 Details : Clean up open-unlinked files after client eviction. Previously
3296 the unlink was skipped and the files remained as orphans.
3301 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3302 Details : In some cases it was possible to send a blocking callback to
3303 a client doing a glimpse, even though that client didn't get
3304 a lock granted. When the glimpse lock is cancelled on the OST
3305 the freed lock is left on the waiting list and corrupted the list.
3308 Frequency : all core dumps
3310 Description: Broke core dumps to lustre
3311 Details : Negative dentry may be unhashed if parent does not have UPDATE
3312 lock, but some callers, e.g. do_coredump, expect dentry to be
3313 hashed after successful create, hash it in ll_create_it.
3315 ------------------------------------------------------------------------------
3317 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3319 * Support for kernels:
3320 2.6.9-42.0.2.EL (RHEL 4)
3321 2.6.5-7.276 (SLES 9)
3322 2.4.21-40.EL (RHEL 3)
3323 2.6.12.6 vanilla (kernel.org)
3327 Frequency : always on RHEL 3
3329 Description: Number of open files grows over time
3330 Details : The number of open files grows over time, whether or not
3331 Lustre is started. This was due to a filp leak introduced
3332 by one of our kernel patches.
3334 ------------------------------------------------------------------------------
3336 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3338 * Support for kernels:
3339 2.6.9-42.EL (RHEL 4)
3340 2.6.5-7.276 (SLES 9)
3341 2.4.21-40.EL (RHEL 3)
3342 2.6.12.6 vanilla (kernel.org)
3347 Bugzilla : 5719, 9635, 9792, 9684
3348 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3349 Details : If a server is under heavy load and cannot reply to new
3350 connection requests before the client resends the (re)connect,
3351 the connection handling code can behave badly if two service
3352 threads are concurrently handing separate (re)connections from
3353 the same client. Add better locking to the connection handling
3354 code, and ensure that only a single connection will be processed
3355 for a given client UUID, even if the lock is dropped.
3357 Severity : enhancement
3359 Description: add TCP zero-copy support to kernel
3360 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3361 sends if the hardware supports scatter-gather and checksumming.
3362 This allows socklnd to do client-write and server-read more
3363 efficiently and reduce CPU utilization from skbuf copying.
3366 Frequency : only if NFS exporting from client
3368 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3369 Details : If setting attributes on a file created under NFS that had
3370 never been opened it would be possible to oops the client
3371 if the file had no objects.
3374 Frequency : always for liblustre
3376 Description: liblustre client does MDS+OSTs setattr RPC for each write
3377 Details : When doing a write from a liblustre client, the client
3378 incorrectly issued an RPC to the MDS and each OST the file was
3379 striped over in order to update the timestamps. When writing
3380 with small chunks and many clients this could overwhelm the MDS
3381 with RPCs. In all cases it would slow down the write because
3382 these RPCs are unnecessary.
3384 Severity : enhancement
3386 Description: allow number of MDS service threads to be changed at module load
3387 Details : It is now possible to change the number of MDS service threads
3388 running. Adding "options mds mds_num_threads={N}" to the MDS's
3389 /etc/modprobe.conf will set the number of threads for the next
3390 time Lustre is restarted (assuming the "mds" module is also
3391 reloaded at that time). The default number of threads will
3392 stay the same, 32 for most systems.
3397 Description: OST crash if filesystem is unformatted or corrupt
3398 Details : If an OST is started on a device that has never been formatted
3399 or if the filesystem is corrupt and cannot even mount then the
3400 error handling cleanup routines would dereference a NULL pointer.
3405 Description: NULL pointer deref in llap_from_page.
3406 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3407 incorrect) mapping if the page was truncated/reclaimed while it was
3408 searched for. Check for this condition and skip such pages when
3409 doing readahead. Introduce extra check to llap_from_page() to
3410 verify page->mapping->host is non-NULL (so page is not anonymous).
3413 Frequency : Sometimes when using sys_sendfile
3415 Description: "page not covered by a lock" warnings from ll_readpage
3416 Details : sendfile called ll_readpage without right page locks present.
3417 Now we introduced ll_file_sendfile that does necessary locking
3418 around call to generic_file_sendfile() much like we do in
3422 Frequency : with certain MDS communication failures at client mount time
3424 Description: NULL pointer deref after failed client mount
3425 Details : a client connection request may delayed by the network layer
3426 and not be sent until after the PTLRPC layer has timed out the
3427 request. If the client fails the mount immediately it will try
3428 to clean up before the network times out the request. Add a
3429 reference from the request import to the obd device and delay
3430 the cleanup until the network drops the request.
3433 Frequency : occasionally during client (re)connect
3435 Description: assertion failure during client (re)connect
3436 Details : processing a client connection request may be delayed by the
3437 client or server longer than the client connect timeout. This
3438 causes the client to resend the connection request. If the
3439 original connection request is replied in this interval, the
3440 client may trip an assertion failure in ptlrpc_connect_interpret()
3441 which thought it would be the only running connect process.
3444 Frequency : only with obd_echo servers and clients that are rebooted
3446 Description: kernel BUG accessing uninitialized data structure
3447 Details : When running an obd_echo server it did not start the ping_evictor
3448 thread, and when a client was evicted an uninitialized data
3449 structure was accessed. Start the ping_evictor in the RPC
3450 service startup instead of the OBD startup.
3452 Severity : enhancement
3453 Bugzilla : 10193 (patchless)
3454 Description: Remove dependency on various unexported kernel interfaces.
3455 Details : No longer need reparent_to_init, exit_mm, exit_files,
3456 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3459 Frequency : rare (only users of deprecated and unsupported LDAP config)
3461 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3462 Details : LDAP apparently contains 'lustreName' attributes instead of
3463 'name'. A simple remapping of the name is sufficient.
3466 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3468 Description: waiting_locks_callback trips kernel BUG if client is evicted
3469 Details : Running with the dump_on_timeout debug flag turned on makes
3470 it possible that the waiting_locks_callback() can try to dump
3471 the Lustre kernel debug logs from an interrupt handler. Defer
3472 this log dumping to the expired_lock_main() thread.
3474 Severity : enhancement
3476 Description: Support NFS exporting on 2.6 kernels.
3477 Details : Implement non-rawops metadata methods for NFS server to use without
3478 changing NFS server code.
3481 Frequency : very rare (synthetic metadata workload only)
3483 Description: two racing renames might cause an MDS thread to deadlock
3484 Details : Running the "racer" program may cause one MDS thread to rename
3485 a file from being the source of a rename to being the target of
3486 a rename at exactly the same time that another thread is doing
3487 so, and the second thread has already enqueued these locks after
3488 doing a lookup of the target and is trying to relock them in
3489 order. Ensure that we don't try to re-lock the same resource.
3492 Frequency : only very large systems with liblustre clients
3494 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3495 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3496 all OSTs in parallel. This allows the eviction of stale liblustre
3497 clients to proceed much faster than if they were done in series,
3498 and also offers similar improvements for other set_info RPCs.
3503 Description: excessive CPU usage during initial read phase on client
3504 Details : During the initial read phase on a client, it would agressively
3505 retry readahead on the file, consuming too much CPU and impacting
3506 performance (since 1.4.5.8). Improve the readahead algorithm
3507 to avoid this, and also improve some other common cases (read
3508 of small files in particular, where "small" is files smaller than
3509 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3514 Description: MDS crash when receiving packet with unknown intent.
3515 Details : Do not LBUG in unknown intent case, just return -EFAULT
3517 Severity : enhancement
3518 Bugzilla : 9293, 9385
3519 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3520 Details : Do not serialize getattr (non-intent version) and statfs.
3523 Frequency : occasional, when OST network is overloaded/intermittent
3525 Description: client evicted by OST after bulk IO timeout
3526 Details : If a client sends a bulk IO request (read or write) the OST
3527 may evict the client if it is unresposive to its data GET/PUT
3528 request. This is incorrect if the network is overloaded (takes
3529 too long to transfer the RPC data) or dropped the OST GET/PUT
3530 request. There is no need to evict the client at all, since
3531 the pinger and/or lock callbacks will handle this, and the
3532 client can restart the bulk request.
3535 Frequency : Always when mmapping file with no objects
3537 Description: client crashes when mmapping file with no objects
3538 Details : Check that we actually have objects in a file before doing any
3539 operations on objects in ll_vm_open, ll_vm_close and
3545 Description: Request leak when working with deleted CWD
3546 Details : Introduce advanced request refcount tracking for requests
3547 referenced from lustre intent.
3549 Severity : Enhancement
3551 Description: Cache open file handles on client.
3552 Details : MDS now will return special lock along with openhandle, if
3553 requested and client is allowed to hold openhandle, even if unused,
3554 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3555 closing files for every read/write openration.
3557 Severity : Enhancement
3559 Description: Cache open negative dentries on client when possible.
3560 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3561 negative dentries on lock revocation.
3566 Description: Remounting a client read-only wasn't possible with a zconf mount
3567 Details : It wasn't possible to remount a client read-only with llmount.
3569 Severity : enhancement
3570 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3571 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3572 MPICH is included to add Lustre-specific ADIO interfaces.
3573 This is based closely on the UFS ADIO layer and only differs
3574 in file creation, in order to allow the OST striping to be set.
3575 This is user-contributed code and not supported by CFS.
3580 Description: extended inode attributes (immutable, append-only) work improperly
3581 when 2.4 and 2.6 kernels are used on client/server or vice versa
3582 Details : Introduce kernel-independent values for these flags.
3584 Severity : enhancement
3587 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3588 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3589 granularity. Now, Fractional MB granularity is be supported,
3590 this is very useful for low memory system.
3592 Severity : enhancement
3594 Description: Getattr by fid
3595 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3596 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3597 have locks and avoids getattr rpc requests for it on every stat.
3600 Frequency : Always, for filesystems larger than 2TB
3602 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3603 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3604 filesystems larger than 2TB to be mounted with mballoc enabled.
3607 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3609 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3610 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3611 filesystems larger than 2TB is mounted then the kernel will
3612 silently corrupt the start of the filesystem. CONFIG_LBD is
3613 enabled for all CFS-supported kernels, but the possibility of
3614 this happening with a modified kernel config exists.
3616 Severity : enhancement
3618 Description: add client O_DIRECT support for 2.6 kernels
3619 Details : It is now possible to do O_DIRECT reads and writes to files
3620 in the Lustre client mountpoint on 2.6 kernel clients.
3622 Severity : enhancement
3624 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3625 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3626 parallel, not waiting for response from every OST before sending
3627 a rpc to the next OST.
3632 Description: setattr vs write race when updating file timestamps
3633 Details : Client processes that update a file timestamp into the past
3634 right after writing to the file (e.g. tar) it is possible that
3635 the updated file modification time can be reset to the current
3636 time due to a race between processing the setattr and write RPC.
3638 Severity : enhancement
3640 Description: Bring 'lfs find' closer in line with regular Linux find.
3641 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3642 -print0 options and obtains all the needed info through the lustre
3645 Severity : enhancement
3647 Description: support up to 1024 configured devices on one node
3648 Details : change obd_dev array from statically allocated to dynamically
3649 allocated structs as they are first used to reduce memory usage
3654 Description: Flush dirty partially truncated pages during truncate
3655 Details : Immediatelly flush partially truncated pages in filter_setattr,
3656 this way we completely avoid having any pages in page cache on OST
3657 and can retire ugly workarounds during writes to flush such pages.
3662 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3663 Details : protect inode from truncation within vfs_unlink() context
3664 just take a reference before calling vfs_unlink() and release it
3665 when parent's i_sem is free.
3670 Description: last_id value checked outside lock on OST caused LASSERT failure
3671 Details : If there were multiple MDS->OST object precreate requests in
3672 flight, it was possible that the OST's last object id was checked
3673 outside a lock and incorrectly tripped an assertion. Move checks
3674 inside locks, and discard old precreate requests.
3677 Frequency : always, if extents are used on OSTs
3679 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3680 Details : OSTs using the extents format would not zero the high 16 bits of
3681 the index physical block number. This is not a problem for any
3682 OST filesystems smaller than 16TB, and no kernels support ext3
3683 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3684 new/modified files) and can be fixed for existing filesystems
3685 with e2fsprogs-1.39-cfs1.
3690 Description: import connection selection may be incorrect if timer wraps
3691 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3692 import connections to be ignored if the 32-bit jiffies counter
3693 wraps. Use a 64-bit jiffies counter.
3696 Frequency : very large clusters immediately after boot
3698 Description: LNET request buffers exhausted under heavy short-term load
3699 Details : If a large number of client requests are generated on a service
3700 that has previously never seen so many requests it is possible
3701 that the request buffer growth cannot keep up with the spike in
3702 demand. Instead of dropping incoming requests, they are held in
3703 the LND until the RPC service can accept more requests.
3706 Frequency : Sometimes during replay
3708 Description: Assertion failure in ll_local_open after replay.
3709 Details : If replay happened on an open request reply before we were able
3710 to set replay handler, reply will become not swabbed tripping the
3711 assertion in ll_local_open. Now we set the handler right after
3712 recognising of open request
3715 Frequency : very rare
3717 Description: kernel reports "badness in vsnprintf"
3718 Details : Reading from the "recovery_status" /proc file in small chunks
3719 may cause a negative length in lprocfs_obd_rd_recovery_status()
3720 call to vsnprintf() (which is otherwise harmless). Exit early
3721 if there is no more space in the output buffer.
3723 Severity : enhancement
3725 Description: clear OBD RPC statistics by writing to them
3726 Details : It is now possible to clear the OBD RPC statistics by writing
3727 to the "stats" file.
3732 Description: Client mtime is not the same on different clients after utimes
3733 Details : In some cases, the client was using the utimes() syscall on
3734 a file cached on another node. The clients now validate the
3735 ctime from the MDS + OSTs to determine which one is right.
3740 Description: Inability to activate failout mode
3741 Details : lconf script incorrectly assumed that in pythong string's numeric
3742 value is used in comparisons.
3745 Frequency : always with multiple stripes per file
3747 Description: Inefficient object allocation for mutli-stripe files
3748 Details : When selecting which OSTs to stripe files over, for files with
3749 a stripe count that divides evenly into the number of OSTs,
3750 the MDS is always picking the same starting OST for each file.
3751 Return the OST selection heuristic to the original design.
3756 Description: mount failures may take full timeout to return an error
3757 Details : Under some heavy load conditions it is possible that a
3758 failed mount can wait for the full obd_timeout interval,
3759 possibly several minutes, before reporting an error.
3760 Instead return an error as soon as the status is known.
3762 Frequency : quota enabled and large files being deleted
3764 Description: releasing more than 4GB of quota at once hangs OST
3765 Details : If a user deletes more than 4GB of files on a single OST it
3766 will cause the OST to spin in an infinite loop. Release
3767 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3772 Description: statfs data retrieved from /proc may be stale or zero
3773 Details : When reading per-device statfs data from /proc, in the
3774 {kbytes,files}_{total,free,avail} files, it may appear
3775 as zero or be out of date.
3778 Frequency : systems with MD RAID1 external journal devices
3780 Description: lconf's call to blkid is confused by RAID1 journal devices
3781 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3782 of returning all block devices that match the journal UUID.
3785 Frequency : always, for aggregate stripe size over 4GB
3787 Description: assertion fails when trying to use 4GB stripe size
3788 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3789 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3792 Frequency : always on ppc64
3794 Description: the first write on an ext3 filesystem with mballoc got stuck
3795 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3796 perform endianness conversion.
3798 ------------------------------------------------------------------------------
3800 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3802 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3803 INTEROPERATE with older versions automatically. Please read the
3804 user documentation before upgrading any part of a live system.
3805 * WARNING: Lustre networking configuration changes are required with
3806 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3809 * Support for kernels:
3810 2.6.9-22.0.2.EL (RHEL 4)
3811 2.6.5-7.244 (SLES 9)
3812 2.6.12.6 vanilla (kernel.org)
3815 Severity : enhancement
3816 Bugzilla : 7981/8208
3817 Description: Introduced Lustre Networking (LNET)
3818 Details : LNET is new networking infrastructure for Lustre, it includes
3819 a reorganized network configuration mode (see the user
3820 documentation for full details) as well as support for routing
3821 between different network fabrics. Lustre Networking Devices
3822 (LNDs) for the supported network fabrics have also been
3823 created for this new infrastructure.
3825 Severity : enhancement
3826 Description: Introduced Access control lists
3827 Details : clients can set ACLs on files and directories in order to have
3828 more fine-grained permissions than the standard Unix UGO+RWX.
3829 The MDS must be started with the "-o acl" mount option.
3831 Severity : enhancement
3832 Description: Introduced filesystem quotas
3833 Details : Administrators may now establish per-user quotas on the
3836 Severity : enhancement
3838 Description: Configuration change for the XT3
3839 The PTLLND is now used to run Lustre over Portals on the XT3
3840 The configure option(s) --with-cray-portals are no longer used.
3841 Rather --with-portals=<path-to-portals-includes> is used to
3842 enable building on the XT3. In addition to enable XT3 specific
3843 features the option --enable-cray-xt3 must be used.
3848 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3849 Details : A race between allocating a new llcd and re-getting the llcd_lock
3850 allowed another thread to grab newly-allocated llcd.
3852 Severity : enhancement
3854 Description: 2.6 OST async journal commit and locking fix to improve performance
3855 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3856 2.6 kernels are now async as they already were in 2.4 kernels so
3857 that they can commit concurrently with the network bulk transfer.
3858 For block-allocated files the filter allocation semaphore is held
3859 to avoid filesystem fragmentation during allocation. BKL lock
3860 removed for 2.6 xattr operations where it is no longer needed.
3865 Description: lconf incorrectly determined whether two IP networks could talk
3866 Details : In some more complicated routing and multiple-network
3867 configurations, lconf will avoid trying to make a network
3868 connection to a disjoint part of the IP space. It was doing the
3869 math incorrectly for one set of cases.
3874 Description: Fix for potential infinite loop processing records in an llog.
3875 Details : If an llog record is corrupted/zeroed, it is possible to loop
3876 forever in llog_process(). Validate the llog record length
3877 and skip the remainder of the block on error.
3880 Frequency : occasional (liblustre only)
3882 Description: liblustre could not open files whose last component is a symlink
3883 Details : sysio_path_walk() would incorrectly pass the open intent to
3884 intermediate path components.
3887 Frequency : rare (liblustre only with non-standard tuning)
3888 Bugzilla : 7201 (7350)
3889 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3890 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3891 after it has been released, only do this for referenced locks
3893 Severity : enhancement
3895 Description: specifying an (invalid) directory default stripe_size of -1
3896 would reset the directory default striping
3897 Details : stripe_size -1 was used internally to signal directory stripe
3898 removal, now use "all default" to signal dir stripe removal
3899 as a directory striping of "all default" is not useful
3902 Frequency : common for large clusters running liblustre clients
3904 Description: doing an ls when liblustre clients are running is slow
3905 Details : sending a glimpse AST to a liblustre client waits for every AST
3906 to time out, as liblustre clients will not respond. Since they
3907 cannot cache data we refresh the OST lock LVB from disk instead.
3909 Severity : enhancement
3911 Description: doing an ls at the same time as file IO can be slow
3912 Details : enqueue and other "small" requests can be blocked behind many
3913 large IO requests. Create a new OST IO portal for non-IO
3914 requests so they can be processed faster.
3917 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3919 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3920 created file with mode 000
3921 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3922 allow an ftruncate() as a non-root user to a file with mode 000.
3923 Lustre now respects this flag to disable mode checks when
3924 truncating a file owned by the user
3927 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3930 Description: Revoking locks from clients that went dead or catatonic might take
3932 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3933 cancellation of such locks instant on servers without waiting for
3934 any reply from clients. Clients drops these locks when cancel
3935 notification from server is received without replying.
3938 Frequency : liblustre-only, when liblustre client dies or becomes busy
3940 Description: Doing ls on Linux clients can take a long time with active
3942 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3943 granting such locks to it in the first place if possible. Locks
3944 are taken by proxy on the OST during the read or write and
3945 dropped immediately afterward. Add connect flags handling, do
3946 not grant locks to liblustre clients for glimpse ASTs.
3948 Severity : enhancement
3950 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3951 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3952 sequential reads and behaves badly with applications doing random
3953 reads. Improve it by reading ahead at least read region, and
3954 avoiding excessive large RPC for small reads.
3956 Severity : enhancement
3958 Description: Creating more than 1000 files for a single job may cause a load
3959 imbalance on the OSTs if there are also a large number of OSTs.
3960 Details : qos_prep_create() uses an OST index reseed value that is an
3961 even multiple of the number of available OSTs so that if the
3962 reseed happens in the middle of the object allocation it will
3963 still utilize the OSTs as uniformly as possible.
3968 Description: OST or MDS may oops in ping_evictor_main()
3969 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3970 but doesn't restart at beginning of obd_exports_timed list
3973 Severity : enhancement
3975 Description: improve by-nid export eviction on the MDS and OST
3976 Details : allow multiple exports with the same NID to be evicted at one
3977 time without re-searching the exports list.
3980 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3982 Description: MDS may oops in groups_free()
3983 Details : in rare race conditions a newly allocated group_info struct is
3984 freed again, and this can be NULL. The 2.4 compatibility code
3985 for groups_free() checked for a NULL pointer, but 2.6 did not.
3988 Frequency : common for liblustre clients doing little filesystem IO
3989 Bugzilla : 9352, 7313
3990 Description: server may evict liblustre clients accessing contended locks
3991 Details : if a client is granted a lock or receives a completion AST
3992 with a blocking AST already set it would not reply to the AST
3993 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3996 Frequency : lfs setstripe, only systems with more than 160 OSTs
3998 Description: unable to set striping with a starting offset beyond OST 160
3999 Details : llapi_create_file() incorrectly limited the starting stripe
4000 index to the maximum single-file stripe count.
4003 Frequency : LDAP users only
4005 Description: lconf did not handle in-kernel recovery with LDAP properly
4006 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4008 Severity : enhancement
4010 Description: bind OST threads to NUMA nodes to improve performance
4011 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4012 node and do their allocations there to localize memory access
4014 Severity : enhancement
4016 Description: llmount can determine client NID directly from Myrinet (GM)
4017 Details : the client NID code from gmnalnid was moved directly into
4018 llmount, removing the need to use this or specifying the
4019 client NID explicitly when mounting GM clients with zeroconf
4022 Frequency : if client is started with down MDS
4024 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4025 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4026 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4027 ptlrpc_import_delayed_req() because we didn't check if the
4028 request was interrupted, and we also didn't break out of the
4029 event loop if there was no timeout
4034 Description: data loss during non-page-aligned writes to a single file from
4035 both multiple nodes and multiple threads on one node at same time
4036 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4037 inconsistency led to false short-reads, that were cached and later
4038 used by ->prepare_write() to fill in partially written page,
4039 leading to data loss.
4042 Frequency : always, if lconf --abort_recovery used
4044 Description: lconf --abort_recovery fails with 'Operation not supported'
4045 Details : lconf was attempting to abort recovery on the MDT device and not
4048 Severity : enhancement
4050 Description: remove cleanup logs
4051 Details : replace lconf-generated cleanup logs with lustre internal
4052 cleanup routines. Eliminates the need for client-cleanup and
4055 Severity : enhancement
4057 Description: add support for EAs (user and system) on lustre filesystems
4058 Details : it is now possible to store extended attributes in the Lustre
4059 client filesystem, and with the user_xattr mount option it
4060 is possible to allow users to store EAs on their files also
4062 Severity : enhancement
4064 Description: Add possibility (config option) to show minimal available OST free
4066 Details : When compiled with --enable-mindf configure option, statfs(2)
4067 (and so, df) will return least minimal free space available from
4068 all OSTs as amount of free space on FS, instead of summary of
4069 free spaces of all OSTs.
4071 Severity : enhancement
4073 Description: do not expand extent locks acquired on OST-side
4074 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4075 by server: they are not cached anyway.
4078 Frequency : when mmap is used/binaries executed from Lustre
4080 Description: Unmmap pages before throwing them away from read cache.
4081 Details : llap_shrink cache now attempts to unmap pages before discarding
4082 them (if unmapping failed - do not discard). SLES9 kernel has
4083 extra checks that trigger if this unmapping is not done first.
4088 Description: lconf didn't resolve symlinks before checking to see whether a
4089 given mountpoint was already in use
4092 Frequency : when migrating failover services
4093 Bugzilla : 6395, 9514
4094 Description: When migrating a subset of services from a node (e.g. failback
4095 from a failover service node) the remaining services would
4096 time out and evict clients.
4097 Details : lconf --force (implied by --failover) sets the global obd_timeout
4098 to 5 seconds in order to quickly disconnect, but this caused
4099 other RPCs to time out too quickly. Do not change the global
4100 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4102 Severity : enhancement
4103 Frequency : if MDS is started with down OST
4104 Bugzilla : 9439,5706
4105 Description: Allow startup/shutdown of an MDS without depending on the
4106 availability of the OSTs.
4107 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4108 Add appropriate locking and lov-osc refcounts for safe
4109 cleaning. Add osc abort_inflight calls in case the
4110 synchronize never started.
4113 Frequency : occasional (Cray XT3 only)
4115 Description: root not authorized to access files in CRAY_PORTALS environment
4116 Details : The client process capabilities were not honoured on the MDS in
4117 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4118 been accessed by an authorized user then root was able to access
4119 the file on the local client also. The root user capabilities
4120 are now allowed on the MDS, as this environment has secure UID.
4123 Frequency : occasional
4125 Description: ldiskfs "too long searching" message happens too often
4126 Details : A debugging message (otherwise harmless) prints too often on
4127 the OST console. This has been reduced to only happen when
4128 there are fragmentation problems on the filesystem.
4133 Description: Division by zero in statfs when all OSCs are inactive
4134 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4135 lov_statfs divides by value returned by lov_get_stripecnt().
4139 Bugzilla : 9489, 3273
4140 Description: First write from each client to each OST was only 4kB in size,
4141 to initialize client writeback cache, which caused sub-optimal
4142 RPCs and poor layout on disk for the first writen file.
4143 Details : Clients now request an initial cache grant at (re)connect time
4144 and so that they can start streaming writes to the cache right
4145 away and always do full-sized RPCs if there is enough data.
4146 If the OST is rebooted the client also re-establishes its grant
4147 so that client cached writes will be honoured under the grant.
4152 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4153 Details : Now I/O RPCs go to different portal number and (presumably) fast
4154 lock requests (and glimses) and other RPCs get their own service
4155 threads pool that should be able to service those RPCs
4158 Severity : enhancement
4160 Description: Ability to exchange lustre version between client and servers and
4161 issue warnings at client side if client is too old. Also for
4162 liblustre clients there is ability to refuse connection of too old
4164 Details : New 'version' field is added to connect data structure that is
4165 filled with version info. That info is later checked by server and
4169 Frequency : rare, liblustre only.
4170 Bugzilla : 9296, 9581
4171 Description: Two simultaneous writes from liblustre at offset within same page
4172 might proceed at the same time overwriting eachother with stale
4174 Details : I/O lock withing llu_file_prwv was released too early, before data
4175 actually was hitting the wire. Extended lock-holding time until
4176 server acknowledges receiving data.
4179 Frequency : extremely rare. Never observed in practice.
4181 Description: avoid generating lustre_handle cookie of 0.
4182 Details : class_handle_hash() generates handle cookies by incrementing
4183 global counter, and can hit 0 occasionaly (this is unlikely, but
4184 not impossible, because initial value of cookie counter is
4185 selected randonly). Value of 0 is used as a sentinel meaning
4186 "unassigned handle" --- avoid it. Also coalesce two critical
4187 sections in this function into one.
4189 Severity : enhancement
4191 Description: allow liblustre clients to delegate truncate locking to OST
4192 Details : To avoid overhead of locking, liblustre client instructs OST to
4193 take extent lock in ost_punch() on client's behalf. New connection
4194 flag is added to handle backward compatibility.
4196 Severity : enhancement
4197 Bugzilla : 4928, 7341, 9758
4198 Description: allow number of OST service threads to be specified
4199 Details : a module parameter allows the number of OST service threads
4200 to be specified via "options ost ost_num_threads={N}" in the
4201 OSS's /etc/modules.conf or /etc/modprobe.conf.
4205 Bugzilla : 6146, 9635, 9895
4206 Description: servers crash with bad pointer in target_handle_connect()
4207 Details : In rare cases when a client is reconnecting it was possible that
4208 the connection request was the last reference for that export.
4209 We would temporarily drop the export reference and get a new
4210 one, but this may have been the last reference and the export
4211 was just destroyed. Get new reference before dropping old one.
4213 Severity : enhancement
4214 Frequency : if client is started with failover MDS
4216 Description: Allow multiple MDS hostnames in the mount command
4217 Details : Try to read the configuration from all specified MDS
4218 hostnames during a client mount in case the "primary"
4221 Severity : enhancement
4223 Description: Stop sending data to evicted clients as soon as possible.
4224 Details : Check if the client we are about to send or are sending data to
4225 was evicted already. (Check is done every second of waiting,
4226 for which l_wait_event interface was extended to allow checking
4227 of exit condition at specified intervals).
4230 Frequency : rare, normally only when NFS exporting is done from client
4232 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4234 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4237 Frequency : always, if srand() is called before liblustre initialization
4239 Description: Liblustre uses system PRNG disturbing its usage by user application
4240 Details : Introduce internal to lustre fast and high-quality PRNG for
4241 lustre usage and make liblustre and some other places in generic
4242 lustre code to use it.
4244 Severity : enhancement
4245 Bugzilla : 9477, 9557, 9870
4246 Description: Verify that the MDS configuration logs are updated when xml is
4247 Details : Check if the .xml configuration logs are newer than the config
4248 logs stored on the MDS and report an error if this is the case.
4249 Request --write-conf, or allow starting with --old_conf.
4251 Severity : enhancement
4253 Description: Handle symlinks in the path when checking if Lustre is mounted.
4254 Details : Resolve intermediate symlinks when checking if a client has
4255 mounted a filesystem to avoid duplicate client mounts.
4260 Description: lconf can hit an error exception but still return success.
4261 Details : The lconf command catches the Command error exception at the top
4262 level script context and will exit with the associated exit
4263 status, but doesn't ensure that this exit status is non-zero.
4268 Description: failure of ptlrpc thread startup can cause oops
4269 Details : Starting a ptlrpc service thread can fail if there are a large
4270 number of threads or the server memory is very fragmented.
4271 Handle this without oopsing.
4274 Frequency : always, only if liblustre and non-default acceptor port was used
4276 Description: liblustre cannot connect to servers with non-default acceptor port
4277 Details : tcpnal_set_default_params() was not called and was therefore
4278 ignoring the environment varaible TCPNAL_PORT, as well as other
4279 TCPNAL_ environment variables
4284 Description: two objects could be created on the same OST for a single file
4285 Details : If an OST is down, in some cases it was possible to create two
4286 objects on a single OST for a single file. No problems other
4287 than potential performance impact and spurious error messages.
4291 Bugzilla : 5681, 9562
4292 Description: Client may oops in ll_unhash_aliases
4293 Details : Client dcache may become inconsistent in race condition.
4294 In some cases "getcwd" can fail if the current directory is
4300 Description: Inode refcounting problems in NFS export code
4301 Details : link_raw functions used to call d_instantiate without obtaining
4302 extra inode reference first.
4306 Bugzilla : 9942, 9903
4307 Description: Referencing freed requests leading to crash, memleaks with NFS.
4308 Details : We used to require that call to ll_revalidate_it was always
4309 followed by ll_lookup_it. Also with revalidate_special() it is
4310 possible to call ll_revalidate_it() twice for the same dentry
4311 even if first occurence returned success. This fix changes semantic
4312 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4313 reference on a request referred from the intent.
4314 ll_intent_release() then releases such a request.
4317 Frequency : rare, normally benchmark loads only
4319 Description: unlinked inodes were kept in memory on the client
4320 Details : If a client is repeatedly creating and unlinking files it
4321 can accumulate a lot of stale inodes in the inode slab cache.
4322 If there is no other client load running this can cause the
4323 client node to run out of memory. Instead flush old inodes
4324 from client cache that have the same inode number as a new inode.
4327 Frequency : SLES9 2.6.5 kernel and long filenames only
4328 Bugzilla : 9969, 10379
4329 Description: utime reports stale NFS file handle
4330 Details : SLES9 uses out-of-dentry names in some cases, which confused
4331 the lustre dentry revalidation. Change it to always use the
4335 Frequency : rare, unless heavy write-truncate concurrency is continuous
4336 Bugzilla : 4180, 6984, 7171, 9963, 9331
4337 Description: OST becomes very slow and/or deadlocked during object unlink
4338 Details : filter_destroy() was holding onto the parent directory lock
4339 while truncating+unlinking objects. For very large objects this
4340 may block other threads for a long time and slow overall OST
4341 responsiveness. It may also be possible to get a lock ordering
4342 deadlock in this case, or run out of journal credits because of
4343 the combined truncate+unlink. Solution is to do object truncate
4344 first in one transaction without parent lock, and then do the
4345 final unlink in a new transaction with the parent lock. This
4346 reduces the lock hold time dramatically.
4349 Frequency : rare, 2.4 kernels only
4351 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4352 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4353 before unmounting their filesystem in order to avoid deadlock.
4354 The kernel_locked() function in 2.4 kernels only checks whether
4355 the kernel lock is held, not whether it is this process that is
4356 holding it as 2.6 kernels do.
4361 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4362 Details : The client ptlrpc code may be trying to reconnect to a down
4363 server before a previous connection attempt has timed out.
4364 Increase the reconnect interval to be longer than the connection
4365 timeout interval to avoid sending duplicate connections to
4369 Frequency : echo_client brw_test command
4371 Description: fix echo_client to work with OST preallocated code
4372 Details : OST preallocation code (5137) didn't take echo_client IO path
4373 into account: echo_client calls filter methods outside of any
4374 OST thread and, hence, there is no per-thread preallocated
4375 pages and buffers to use. Solution: hijack pga pages for IO. As
4376 a byproduct, this avoids unnecessary data copying.
4380 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4381 Description: Client can oops in mdc_commit_close() after open replay
4382 Details : It was possible for the MDS to return an open request with no
4383 transaction number in mds_finish_transno() if the client was
4384 evicted, but without actually returning an error. Clients
4385 would later try to replay that open and may trip an assertion
4386 Simplify the client close codepath, and always return an error
4387 from the MDS in case the open is not successful.
4390 Frequency : rare, 2.6 OSTs only
4392 Description: OST may deadlock under high load on fragmented files
4393 Details : If there was a heavy load and highly-fragmented OST filesystems
4394 it was possible to have all the OST threads deadlock waiting on
4395 allocation of biovecs, because the biovecs were not released
4396 until the entire RPC IO was completed. Instead, release biovecs
4397 as soon as they are complete to ensure forward IO progress.
4399 Severity : enhancement
4401 Description: Support for specifying external journal device at mount
4402 Details : If an OST or MDS device is formatted with an external journal
4403 device, this device major/minor is stored in the ext3 superblock
4404 and may not be valid for failover. Allow detecting and
4405 specifying the external journal at mount time.
4410 Description: Mounting an MDS with pending unlinked files may cause oops
4411 Details : target_finish_recovery() calls mds_postrecov() which returned
4412 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4413 considers this an error and immediately begins cleaning up the
4414 lov, just after starting the mds_lov process
4416 Severity : enhancement
4418 Description: Implement 'lfs df' to report actual free space on per-OST basis
4419 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4420 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4421 usage of objects; '-h' to report in human readable format.
4423 ------------------------------------------------------------------------------
4425 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4432 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4433 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4434 at startup during orphan unlinking. Instead, skip buddy update
4435 before setup, as it will be regenerated after recovery is complete.
4440 Description: If an OST is inactive, its locks might reference stale inodes.
4441 Details : lov_change_cbdata() must iterate over all namespaces, even if
4442 they are inactive to clear inode references from the lock.
4444 Severity : enhancement
4445 Frequency : occasional, if non-standard max_dirty_mb used
4447 Description: Client will block write RPCs if not enough grant
4448 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4449 then the client will block writes while waiting for another RPC
4450 to complete instead of consuming its dirty limit. With change
4451 we get improved performance when max_dirty_mb is small.
4453 Severity : enhancement
4454 Bugzilla : 3389, 6253
4455 Description: Add support for supplementary groups on the MDS.
4456 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4457 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4458 lookups for user supplementary groups into a cache.
4462 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4463 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4464 for liblustre O_EXCL is enforced by MDS.
4468 Description: readdir never returns NULL in liblustre.
4469 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4470 dentry in struct dirent.
4474 Frequency : liblustre only, depends on application IO pattern
4475 Description: liblustre clients evicted if not contacting servers
4476 Details : Don't put liblustre clients into the ping_evictor list, so
4477 they will not be evicted by the pinger ever.
4479 Severity : enhancement
4481 Description: Add ability to evict clients by NID from MDS.
4482 Details : By echoing "nid:$NID" string into
4483 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4484 $NID would be instantly evicted from this MDS and from all active
4485 OSTs connected to it.
4489 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4490 Details : lookup_it_finish() used to query file size from OSTs that was not
4495 Description: service threads change working directory to that of init
4496 Details : Starting lustre service threads may pin the working directory
4497 of the parent thread, making that filesystem busy. Threads
4498 now change to the working directory of init to avoid this.
4502 Frequency : during shutdown only
4503 Description: shutdown with a failed MDS or OST can cause unmount to hang
4504 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4509 Frequency : 2.6 only
4510 Description: chmod/chown may include an extra supplementary group
4511 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4512 supplementary group and if none is specified this is used.
4515 Bugzilla : 5479 (6816)
4517 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4518 Details : If lookup is in progress on a file that is unlinked we might try
4519 to revalidate the inode and fail in revalidate after lookup is
4520 complete and ll_file_open() enqueues the open again but
4521 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4524 Frequency : always, if lconf --abort_recovery used
4526 Description: lconf --abort_recovery fails with 'Operation not supported'
4527 Details : lconf was attempting to abort recovery on the MDT device and not
4530 ------------------------------------------------------------------------------
4532 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4537 Frequency : rare (only unsupported configurations with a node running as an
4539 Bugzilla : 6514, 5137
4540 Description: Mounting a Lustre file system on a node running as an OST could
4542 Details : OSTs now preallocates memory needed to write out data at
4543 startup, instead of when needed, to avoid having to
4544 allocate memory in possibly low memory situations.
4545 Specifically, if the file system is mounted on on OST,
4546 memory pressure could force it to try to write out data,
4547 which it needed to allocate memory to do. Due to the low
4548 memory, it would be unable to do so and the node would
4549 become unresponsive.
4551 Severity : enhancement
4553 Description: Addition of lconf --service command line option
4554 Details : lconf now accepts a '--service <arg>' option, which is
4555 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4557 Severity : enhancement
4559 Description: Failover mode is now the default for OSTs.
4560 Details : By default, OSTs will now run in failover mode. To return to
4561 the old behaviour, add '--failout' to the lmc line for OSTs.
4563 Severity : enhancement
4565 Description: Health checks are now provided for MDS and OSTs
4566 Details : Additional detailed health check information on MSD and OSTs
4567 is now provided through the procfs health_check value.
4570 Frequency : occasional, depends on IO load
4572 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4573 after numerous create/delete cycles
4574 Details : The ext3 inode allocation policy would not allocate new inodes
4575 very well on the OSTs because there are no new directories
4576 being created. Instead we look for groups with free space if
4577 the parent directories are nearly full.
4582 Description: Network or server problems during mount may cause partially
4583 mounted clients instead of returning an error.
4584 Details : The config llog parsing code may overwrite the error return
4585 code during mount error handling, returning success instead
4591 Description: MDS can fail to allocate large reply buffers
4592 Details : After long uptimes the MDS can fail to allocate large reply
4593 buffers (e.g. zconf client mount config records) due to memory
4594 fragmentation or consumption by the buffer cache. Preallocate
4595 some large reply buffers so that these replies can be sent even
4596 under memory pressure.
4600 Frequency : rare (liblustre)
4601 Description: fsx running with liblustre complained that using truncate() to
4602 extend the file doesn't work. This patch corrects that issue.
4603 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4604 ATTR_SIZE and lsm use in llu_setattr_raw.
4608 Frequency : rare, only 2.6 kernels
4609 Description: Unusual file access patterns on the MDS may result in inode
4610 data being lost in very rare circumstances.
4611 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4612 code in the 2.6 kernel caused the inode and/or EA data not to
4613 be read from disk, causing single-file corruption.
4617 Frequency : rare, only 2.6 filesystems using extents
4618 Description: Heavy concurrent write and delete load may cause data corruption.
4619 Details : It was possible under high-load situations to have an extent
4620 metadata block in the block device cache from a just-unlinked
4621 file overwrite a newly-allocated data block. We now unmap any
4622 metadata buffers that alias just-allocated data blocks.
4626 Frequency : filesystems with default stripe_count larger than 77
4627 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4628 Details : lconf specifies an inode size of 4096 bytes when the default
4629 stripe_count is larger than 77. This conflicts with the default
4630 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4631 this case to avoid pinning too much memory for large EAs.
4633 ------------------------------------------------------------------------------
4635 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4640 Frequency : rare (extremely heavy IO load with hundreds of clients)
4642 Description: Client is evicted, gets IO error writing to file
4643 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4644 caused clients to be evicted for AST timeouts. The fixes in
4645 bug 5192 mean we no longer need to have such short AST timeouts
4646 so ldlm_timeout has been increased.
4649 Frequency : occasional during --force or --failover shutdown under load
4650 Bugzilla : 5949, 4834
4651 Description: Server oops/LBUG if stopped with --force or --failover under load
4652 Details : a collection of import/export refcount and cleanup ordering
4653 issues fixed for safer force cleanup
4656 Frequency : only filesystems larger than 120 OSTs
4657 Bugzilla : 5990, 6223
4658 Description: lfs getstripe would oops on a very large filesystem
4659 Details : lov_getconfig used kfree on vmalloc'd memory
4662 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4663 Bugzilla : 6242, 6243
4664 Description: reading from files that had been truncated to a non-zero size
4665 but never opened returned no data
4666 Details : ll_file_read() reads zeros from no-object files to EOF
4671 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4672 Details : There's little harm in aborting MDS/OSS recovery and letting it
4673 try again, so I removed the LASSERT and return an error instead.
4675 Severity : enhancement
4677 Description: New debugging infrastructure for tracking down data corruption
4678 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4679 (b) cover more of the client-side code path, and (c) try to narrow
4680 down where problems occurred
4684 Bugzilla : 3819, 4364, 4397, 6313
4685 Description: Racing close and eviction MDS could cause assertion in mds_close
4686 Details : It was possible to get multiple mfd references during close and
4687 client eviction, leading to one thread referencing a freed mfd.
4689 Severity: : enhancement
4690 Bugzilla : 3262, 6359
4691 Description: Attempts to reconnect to servers are now more aggressive.
4692 Details : This builds on the enhanced upcall-less recovery that was added
4693 in 1.4.2. When trying to reconnect to servers, clients will
4694 now try each server in the failover group every 10 seconds. By
4695 default, clients would previously try one server every 25 seconds.
4700 Description: After recovery, certain operations trigger a failed
4701 assertion on a client.
4702 Details : Failing over an mds, using lconf -d --failover, while a
4703 client was doing a readdir() call would cause the client to
4704 LBUG after recovery completed and the readdir() was resent.
4706 Severity : enhancement
4708 Description: Default groups are now added by lconf
4709 Details : You can now run lconf --group <servicename> without having to
4710 manually add groups with lmc.
4713 Frequency : occasional
4715 Description: Nodes with an elan id of 0 trigger a failed assertion
4718 Frequency : always when accessing e.g. tty/console device nodes
4720 Description: tty and some other devices nodes cannot be used on lustre
4721 Details : file's private_data field is used by device data and lustre
4722 values in there got lost. New field was added to struct file to
4723 store fs-specific private data.
4726 Frequency : when exporting Lustre via NFS
4728 Description: NFSD failed occasionally when looking up a path component
4729 Details : NFSD is looking up ".." which was broken in ext3 directories
4730 that had grown large enough to become hashed.
4733 Frequency : Clusters with multiple interfaces not on the same subnet
4735 Description: Nodes will repeatedly try to reconnect to an interface which it
4736 cannot reach and report an error to the log.
4737 Details : Extra peer list entries will be created by lconf with some peers
4738 unreachable. lconf now validates the peer before adding it.
4741 Frequency : Only if a default stripe is set on the filesystem root.
4743 Description: Setting a default stripe on the filesystem root prevented the
4744 filesystem from being remounted.
4745 Details : The client was sending extra request flags in the root getattr
4746 request and did not allocate a reply buffer for the dir EA.
4749 Frequency : occasional, higher if lots of files are accessed by one client
4750 Bugzilla : 6159, 6097
4751 Description: Client trips assertion regarding lsm mismatch/magic
4752 Details : While revalidating inodes the VFS looks up inodes with ifind()
4753 and in rare cases can find an inode that is being freed.
4754 The ll_test_inode() code will free the lsm during ifind()
4755 when it finds an existing inode and then the VFS later attaches
4756 this free lsm to a new inode.
4760 Bugzilla : 6422, 7030
4761 Description: MDS deadlock between mkdir and client eviction
4762 Details : Creating a new file via mkdir or mknod (starting a transaction
4763 and getting the ns lock) can deadlock with client eviction
4764 (gets ns lock and trying to finish a synchronous transaction).
4767 Frequency : occasional
4768 Description: While starting a server, the fsfilt_ext3 module could not be
4770 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4771 kernels. Previously, lconf would still use the ext3 name
4772 when trying to load modules. Now, it will correctly use
4773 ext3 on 2.4 and ldiskfs on 2.6.
4775 Severity : enhancement
4776 Description: The default stripe count has been changed to 1
4777 Details : The interpretation of the default stripe count (0, to lfs
4778 or lmc) has been changed to mean striping across a single
4779 OST, rather than all available. For general usage we have
4780 found a stripe count of 1 or 2 works best.
4782 Severity : enhancement
4783 Description: Add support for compiling against Cray portals.
4784 Details : Conditional compiling for some areas that are different
4788 Frequency : occasional
4789 Bugzilla : 6409, 6834
4790 Description: Creating files with an explicit stripe count may lead to
4791 a failed assertion on the MDS
4792 Details : If some OSTs are full or unavailable, creating files may
4793 trigger a failed assertion on the MDS. Now, Lustre will
4794 try to use other servers or return an error to the
4798 Frequency : occasional
4800 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4801 caused lock timeout during readahead (since 1.4.2).
4802 Details : Processes doing readahead might match a lock that hasn't been
4803 granted yet if there are overlapping and conflicting lock
4804 requests. The readahead process waits on ungranted lock
4805 (original lock is CBPENDING), while OST waits for that process
4806 to cancel CBPENDING read lock and eventually evicts client.
4808 Severity : enhancement
4810 Description: Initial enabling of flock support for clients
4811 Details : Implements fcntl advisory locking and file status functions.
4812 This feature is provided as an optional mount flag (default
4813 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4814 locking are implemented yet, and those that are are not guaranteed
4815 to be completely correct in production environments.
4816 mount -t lustre -o [flock|noflock] ...
4819 Frequency : occasional
4821 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4822 assertion in the ext3 JBD (journaling) layer.
4823 Details : The b_committed_data struct is protected by the big kernel lock
4824 in 2.4 kernels, serializing journal_commit_transaction() and
4825 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4826 access to this struct. In 2.6 kernels there is finer grained
4827 locking to improve SMP performance of the JBD layer.
4831 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4836 Description: merge and cleanup kernel patches.
4839 -----------------------------------------------------------------------------
4841 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4843 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4844 versions of Lustre. Please update all servers and clients to
4845 version 1.4.2 or later at the same time. You must also run
4846 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4847 to update the configuration logs.
4849 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4850 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4851 - swab llog records of type '0' so we get proper header size/idx (5861)
4852 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4853 - fix rename of one directory over another leaking an inode (5953)
4854 - avoid SetPageDirty on 2.6 (5981)
4855 - don't re-add just-being-destroyed locks to the waiting list (5653)
4856 - when creating new directories, inherit the parent's custom
4857 striping settings if present parent (3048)
4858 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4859 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4860 - don't hold client locks on temporary worklist from l_lru (5666)
4861 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4862 - automatically evict dead clients (5921)
4863 - Update file size properly in create+truncate+fstat case (6196)
4864 - Do not unhash mountpoint dentries, do not allow removal of
4866 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4867 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4868 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4869 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4870 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4871 - Correctly handle reads of files with no objects (6243)
4872 - lctl recover will also mark a device active if deactivate used (5933)
4874 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4875 - allow --write-conf on an MDS with different nettype than client (5619)
4876 - don't write config llogs to MDS for mounts not from that MDS (5617)
4877 - lconf should create multiple TCP connections from a client (5201)
4878 - init scripts are now turned off by default; run chkconfig --on
4879 lustre and chkconfig --on lustrefs to use them
4880 - upcalls are no longer needed for clients to recover to failover
4882 - add --abort-recovery option to lconf to abort recovery on device
4884 - add support for an arbitrary number of OSTs (3026)
4885 - Quota support protocol changes.
4886 - forward compatibility changes to wire structs (6007)
4887 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4888 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4889 - improved "lustre status" script
4890 - initialize blocksize for non-regular files (6062)
4891 - added --disable-server and --disable-client configure options (5782)
4892 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4893 - Vanilla 2.4.29 support
4894 - increase maximum number of obd devices to 520 (6242)
4895 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4896 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4897 SLES 9 2.6.5 kernels
4898 - limit stripes per file to 160 (the maximum EA size) (6093)
4900 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4903 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4904 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4905 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4906 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4907 - small ext3 extents cleanups and fixes (5733)
4908 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4909 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4910 - avoid needless client->OST connect, fix handle mismatch (5317)
4911 - fix DLM error path that led to out-of-sync client, long delays (5779)
4912 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4913 - fix several locking issues related to i_size (5492,5624,5654,5672)
4914 - don't move pending lock onto export if it is already evicted (5683)
4915 - fix kernel oops when creating .foo in unlinked directory (5548)
4916 - fix deadlock in obdfilter statistics vs. object create (5811)
4917 - use time_{before,after} to avoid timer jiffies wrap (5882)
4918 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4919 - Do not leak request if server was not able to process it (5154)
4920 - If mds_open unable to find parent dir, make that negative lookup(5154)
4921 - don't create new directories with extent-mapping (5909, 5936)
4923 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4924 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4925 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4926 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4927 - support for RHEL4 kernel on i686 (5773)
4928 - provide error messages when incompatible logs are encountered (5898)
4930 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4931 * version 1.4.0.10 (1.4.1 release candidate 1)
4933 - don't keep a lock reference when lock is not granted (4238)
4934 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4935 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4936 - drop import inflight refcount on signal_completed_replay error (5255)
4937 - unlock page after async write error during send (3677)
4938 - handle missing objects in filter_preprw_read properly (5265)
4939 - no transno return for symlink open, don't save no-trasno open (3440)
4940 - don't try to complete elan receive that already failed (4012)
4941 - free RPC server reply state on error (5406)
4942 - clean up thread from ptlrpc_start_thread() on error (5160)
4943 - readahead could read extra page into cache that wasn't ejected (5388)
4944 - prevent races in class_attach/setup/cleanup/detach (5260)
4945 - don't dereference de->d_inode after l_dput of de (5458)
4946 - use "int" for stripe value returned from lock_to_stripe (5544)
4947 - mballoc allocation and error-checking fixes in 2.6 (5504)
4948 - block device patches to fix I/O request sizes in 2.6 (5482)
4949 - look up hostnames for IB nals (5602)
4950 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4951 - don't start multiple acceptors for the same port (5277)
4952 - fix incorrect LASSERT in mds_getattr_name (5635)
4953 - export a proc file for general "ping" checking (5628)
4954 - fix "lfs check" to not block when the MDS is down (5628)
4956 - service request history (4965)
4957 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4958 - create an "evict_client" /proc entry on OSTs, like the MDS has
4959 - fix mount usage message, return errors per mount(8) (5168)
4960 - change grep [] to grep "[]" in tests so they work in more UMLs
4961 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4962 - remove ancient LOV_MAGIC_V0 EA support (5047)
4963 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4964 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4965 - allow readahead allocations to fail when low on memory (5383)
4966 - mmap locking landed again, after considerable improvement (2828)
4967 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4969 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4972 - send OST transaction number in read/write reply to free req (4966)
4973 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4974 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4975 - the watchdog thread now runs as interruptible (5246)
4976 - flock/lockf fixes (but it's still disabled, pending 5135)
4977 - don't use EXT3 constants in llite code (5094)
4978 - memory shortage at startup could cause assertion (5176)
4980 - reorganization of lov code
4981 - single portals codebase
4983 - add extents/mballoc support (5025)
4984 - direct I/O reads in the obdfilter (4048)
4985 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4987 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4990 - send OST transaction number in read/write reply to free req (4966)
4991 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4992 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4993 - the watchdog thread now runs as interruptible (5246)
4994 - handle missing objects in filter_preprw_read properly (5265)
4995 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4996 - drop import inflight refcount on signal_completed_replay error (5255)
4997 - unlock page after async write error during send (3677)
4998 - return original error code on reconstructed replies (3761)
4999 - no transno return for symlink open, don't save no-trasno open (3440)
5001 - add pid to ldlm debugging output (4922)
5002 - bump the watchdog timeouts -- we can't handle 30sec yet
5003 - extra debugging for orphan dentry/inode bug (5259)
5005 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5008 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5009 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5010 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5011 - fix NULL dereference in /proc/sys/portals/routes (4827)
5012 - allow failed mdc_close() operations to be interrupted (4561)
5013 - stop precreate on OST before MDS would time out on it (4778)
5014 - don't send partial-page writes before EOF from client (4410)
5015 - discard client grant for sub-page writes on large-page clients (4520)
5016 - don't free dentries not owned by NFS code, check generation (4806)
5017 - fix lsm leak if mds_create_objects() fails (4801)
5018 - limit debug_daemon file size, always print CERROR messages (4789)
5019 - use transno after validating reply (3892)
5020 - process timed out requests if import state changes (3754)
5021 - update mtime on OST during writes, return in glimpse (4829)
5022 - add mkfsoptions to LDAP (4679)
5023 - use ->max_readahead method instead of zapping global ra (5039)
5024 - don't interrupt __l_wait_event() during strace
5026 - add software watchdogs to catch hung threads quickly (4941)
5027 - make lustrefs init script start after nfs is mounted
5028 - fix CWARN/ERROR duplication (4930)
5029 - return async write errors to application if possible (2248)
5030 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5031 - print NAL number in %x format (4645)
5032 - update barely-supported suse-2.4.21-171 series (4842)
5033 - support for sles 9 %post scripts
5034 - support for building 2.6 kernel-source packages
5035 - support for sles km_* packages
5037 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5040 - ignore -ENOENT errors in osc_destroy (3639)
5041 - notify osc create thread that OSC is being cleaned up (4600)
5042 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5043 - reconstruct ost_handle() like mds_handle() (4657)
5044 - create a new thread to do import eviction to avoid deadlock (3969)
5045 - let lconf resolve symlinked-to devices (4629)
5046 - don't unlink "objects" from directory with default EA (4554)
5047 - hold socknal file ref over connect in case target is down (4394)
5048 - allow more than 32000 subdirectories in a single directory (3244)
5049 - fix blocks count for O_DIRECT writes (3751)
5050 - OST returns ENOSPC from object create when no space left (4539)
5051 - don't send truncate RPC if file size isn't changing (4410)
5052 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5053 - bind to privileged port in socknal and tcpnal (3689)
5055 - rate limit CERROR/CWARN console message to avoid overload (4519)
5056 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5057 - basic mmap support (3918)
5058 - kernel patch series update from b1_4 (4711)
5060 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5063 - avoid crash during MDS cleanup with OST shut down (2775)
5064 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5065 - don't use bad inodes on the MDS (3744)
5066 - dynamic object preallocation to improve recovery speed (4236)
5067 - don't hold spinlock over lock dumping or change debug flags (4401)
5068 - don't zero obd_dev when it is force cleaned (3651)
5069 - print grants to console if they go negative (4431)
5070 - "lctl deactivate" will stop automatic recovery attempts (3406)
5071 - look for existing locks in ldlm_handle_enqueue() (3764)
5072 - don't resolve lock handle twice in recovery avoiding race (4401)
5073 - revalidate should check working dir is a directory (4134)
5075 - don't always mark "slow" obdfilter messages as errors (4418)
5077 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5080 - don't close LustreDB during write_conf until it is done (3860)
5081 - fix typo in lconf for_each_profile (3821)
5082 - allow dumping logs from multiple threads at one time (3820)
5083 - don't allow multiple threads in OSC recovery (3812)
5084 - fix debug_size parameters (3864)
5085 - fix mds_postrecov to initialize import for llog ctxt (3121)
5086 - replace config semaphore with spinlock (3306)
5087 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5088 - don't allow enqueue to complete on a destroyed export (3822)
5089 - down write_lock before checking llog header bitmap (3825)
5090 - recover from lock replay timeout (3764)
5091 - up llog sem before sending rpc (3652)
5092 - reduce ns lock hold times when setting kms (3267)
5093 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5094 - fix NULL deref and obd_dev leak on setup error (3312)
5095 - replace some LBUG about llog ops with error handling (3841)
5096 - don't match INVALID dentries from d_lookup and spin (3784)
5097 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5098 - fix invalid assertion in ptlrpc_set_wait (3880)
5100 - add libwrap support for the TCP acceptor (3996)
5101 - add /proc/sys/portals/routes for non-root route listing (3994)
5102 - allow setting MDS UUID in .xml (2580)
5103 - print the stack of a process that LBUGs (4228)
5105 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5108 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5109 - make sure to unset replay flag from failed open requests (3440)
5110 - if default stripe count is 0, use OST count for inode size (3636)
5111 - update parent mtime/ctime on client for create/unlink (2611)
5112 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5113 - free recovery state on server during a forced cleanup (3571)
5114 - unregister_reply for resent reqs (3063)
5115 - loop back devices mounting and status check on 2.6 (3563)
5116 - fix resource-creation race that can provoke i_size == 0 (3513)
5117 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5118 - more debugging for page-accounting assertion (3746)
5119 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5120 - avoid deadlock after precreation failure (3758)
5121 - fix race and lock order deadlock in orphan handling (3450, 3750)
5122 - add validity checks when grabbing inodes from l_ast_data (3599)
5124 - add /proc/.../recovery_status to obdfilter (3428)
5125 - lightweight CDEBUG infrastructure, debug daemon (3668)
5126 - change default OSC RPC parameters to be better on small clusters
5127 - turn off OST read cache for files smaller than 32MB
5128 - install man pages and include them in rpms (3100)
5129 - add new init script for (un)mounting lustre filesystems (2593)
5130 - run chkconfig in %post for init scripts (3701)
5131 - drop scimac NAL (unmaintained)
5133 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5136 - clean kiobufs before and after use (3485)
5137 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5138 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5139 - change init script to fail more clearly if not run as root (1528)
5140 - allow clients to reconnect during replay (1742)
5141 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5142 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5143 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5144 - fix cond_resched (3554)
5145 - only evict unfinished clients after recovery (3515)
5146 - allow bulk resend, prevent data loss (3570)
5147 - dynamic ptlrpc request buffer allocation (2102)
5148 - don't allow unlinking open directory if it isn't empty (2904)
5149 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5150 - remove extraneous obd dereference causing LASSERT failure (3334)
5151 - don't use get_cycles() when creating temp. files on the mds (3156)
5152 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5153 - handle EEXIST for set-stripe, set proper directory name (3336)
5155 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5156 - fix ksocknal_fmb_callback() error messages (2918)
5158 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5161 - don't copy lvb into (possibly NULL) reply on error (2983)
5162 - don't deref dentry after dput, don't free lvb on error (2922)
5163 - use the kms to determine writeback rpc length (2947)
5164 - increment oti_logcookies when osc is inactive (2948)
5165 - update client's i_blocks count via lvb messages (2543)
5166 - handle intent open/close of special files properly (1557)
5167 - mount MDS with errors=remount-ro, like obdfilter (2009)
5168 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5169 - don't use cancelling-locks' kms values (2947)
5170 - use highest lock extent for kms, not last one (2925)
5171 - don't dereference ERR_PTR() dentry in error handling path (3107)
5172 - fix thread race in portals_debug_dumplog() (3122)
5173 - create lprocfs device entries at setup instead of at attach (1519)
5174 - common AST error handler, don't evict client on completion race (3145)
5175 - zero nameidata in detach_mnt in 2.6 (3118)
5176 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5177 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5178 - initialize RPC timeout timer earlier for 2.6 (3219)
5179 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5180 - print nal/nid for unknown nid (3258)
5181 - additional checks for oscc recovery before doing precreate (3284)
5182 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5183 - don't crash in mdc_close for bad permissions on open (3285)
5184 - zero i_rdev for non-device files (3147)
5185 - clear page->private before handing to FS, better assertion (3119)
5186 - tune the read pipeline (3236)
5187 - fix incorrect decref of invalidated dentry (2350)
5188 - provide read-ahead stats and refine rpc in flight stats (3328)
5189 - don't hold journal transaction open across create RPC (3313)
5190 - update atime on MDS at close time (3265)
5191 - close LDAP connection when recovering to avoid server load (3315)
5192 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5193 - don't leak open file on MDS after open resend (3325)
5194 - serialize filter_precreate and filter_destroy_precreated (3329)
5195 - loop device shouldn't call sync_dev() for nul device (3092)
5196 - clear page cache after eviction (2766)
5197 - resynchronize MDS->OST in background (2824)
5198 - refuse to mount the same filesystem twice on same mountpoint (3394)
5199 - allow llmount to create routes for mounting behind routers (3320)
5200 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5201 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5202 - fix rare problem with rename on htree directories (3417)
5204 - allow default OST striping configuration per directory (1414)
5205 - fix compilation for qswnal for 2.6 kernels (3125)
5206 - increase maximum number of MDS request buffers for large systems
5207 - change liblustreapi to be useful for external progs like lfsck (3098)
5208 - increase local configuration timeout for slow disks (3353)
5209 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5211 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5214 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5215 - don't overwrite extent policy data in reply if lock was blocked (2901)
5216 - drop filter export grants atomically with removal from device (2663)
5217 - del obd_self_export from work_list in class_disconnect_exports (2908)
5218 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5219 - swab reply message in mdc_close, other PPC fixes (2464)
5220 - fix destroying of named logs (2325)
5221 - overwrite old logs when running lconf --write_conf (2264)
5222 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5223 - fix race in target_handle_connect (2898)
5224 - mds_reint_create() should take same inode create lock (2926)
5225 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5226 - don't close files for self_export to avoid uninitialized obd (2936)
5227 - allow MDS with the same name as client node (2939)
5228 - hold dentry reference for closed log files for unlink (2325)
5229 - reserve space for all logs during transactions (2059)
5230 - don't evict page beyond end of stripe extent (2925)
5231 - don't oops on a deleted current working directory (2399)
5232 - handle hard links to targets without a parent properly (2517)
5233 - don't dereference NULL lock when racing during eviction (2867)
5234 - don't grow lock extents when lots of conflicting locks (2919)
5236 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5239 - account for cache space usage on clients to avoid data loss (974)
5240 - lfsck support in lustre kernel code (2349)
5241 - reduce journal credits needed for BRW writes (2370)
5242 - orphan handling to avoid losing space on client/server crashes
5243 - ptlrpcd can be blocked, stopping ALL progress (2477)
5244 - use lock value blocks to assist in proper KMS, faster stat (1021)
5245 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5246 - recovery for initial connections (2355)
5247 - fixes for mds_cleanup_orphans (1934)
5248 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5249 - block all file creations until orphan recovery completes (1901)
5250 - client remove rq_connection from request struct (2423)
5251 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5252 - recovery timer race (2670)
5253 - mdc_close recovey bug (2532)
5254 - ptlrpc cleanup bug (2710)
5255 - mds timeout on local locks (2588)
5256 - namespace lock held during RPCs (2431)
5257 - handle interrupted sync write properly (2503)
5258 - don't try to handle a message that hasn't been replied to (2699)
5259 - client assert failure during cleanup after abort recovery (2701)
5260 - leak mdc device after failed mount (2712)
5261 - ptlrpc_check_set allows timedout requests to complete (2714)
5262 - wait for inflight reqs when ptlrpcd finishes (2710)
5263 - make sure unregistered services are removed from the srv_list
5264 - reset bulk XID's when resending them (caught by 1138 test)
5265 - unregister_bulk after timeout
5266 - fix lconf error (2694)
5267 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5268 - readahead locks pages, leaves pending causing memory pressure (2673)
5269 - increase OST request buffers to 4096 on large machines (2729)
5270 - fix up permission of existing directories in simple_mkdir (2661)
5271 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5272 - don't assign transno to errored transactions (2742)
5273 - don't delete objects on OST if given a bogus objid from MDS (2751)
5274 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5275 - if rq_no_resend, then timeout request after recovery (2432)
5276 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5277 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5278 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5279 - MDS recovery shouldn't delete objects that it has given out (2730)
5280 - if enqueue arrives after completion, don't clobber LVB (2819)
5281 - don't unlock pages twice when trigger_group_io returns error (2814)
5282 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5283 - don't write pages to disk if there was an error (1450)
5284 - don't ping imports that have recovery disabled (2676)
5285 - take buffered bytes into account when balancing socknal conn (2817)
5286 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5287 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5288 - remove little-endian swabbing of llog records (1987)
5289 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5290 - retry reposting request buffers if they fail (1191)
5291 - grow extent at grant time to avoid granting a revoked lock (2809)
5292 - lock revoke doesn't evict page if covered by a second lock (2765)
5293 - disable VM readahead to avoid reading outside lock extents (2805)
5295 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5296 - updated kernel patches for hp-2.4.20 kernel (2681)
5298 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5301 - fix truncated write corruption (2366)
5302 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5304 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5305 - ensure len(uuid) < 37 in lmc (1171)
5306 - fix ia64 OOPS in llog_test (2255)
5307 - zero end of page at obdfilter for partial page writes (2648)
5308 - don't leave stale dentries around after renames (bug 2428)
5309 - fix timeouts when evicting a client with a single lock held (2642)
5310 - set deadline for the initial HELLO message to drain (2634)
5311 - print out dotted-quad IP addresses in the socknal (2302)
5313 - additional debugging for MDS client eviction problem (2443)
5314 - fix mkfsoptions support for osts (2603, 2604)
5316 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5319 - add series for the vanilla 2.6.0 kernel
5320 - add series for the vanilla 2.4.24 kernel
5321 - add series for a cray x86/64 UL kernel drop
5322 - fix xattr patches for the vanilla 2.4.19 series
5324 - generate true UUIDs in lmc (1171)
5325 - have portals stack dumping break in UML (2466)
5326 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5327 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5328 - wake callback waiting threads on client eviction (2460)
5329 - Add --ptldebug and --subsystem to lmc (1719)
5330 - update assertion to allow safe interrupt allocation
5331 - set rq_no_resend for cancel requests (2432)
5332 - recalculate ptlrpcd timeout after resend (2494)
5333 - call vfs_rmdir when removing pending directories (2368)
5334 - fix renaming a file to itself (2429)
5335 - lmc creates a default one-stripe lov (2454)
5336 - expand procfs space to handle large clusters (2326)
5337 - increase UML stack to avoid overflow
5338 - update lconf's list of debug and subsystem masks
5339 - fix lfs find --obd (2510)
5340 - /proc tunable for disabling filter read caching (2591)
5341 - stop rpm packages from altering slapd.conf (2301)
5342 - disable nagle in the socknal under 0conf (2578)
5343 - choose mds inode size based on stripe count (2572)
5344 - fix kernel-source rpm problems (2516)
5346 - add --disable-doc to avoid pdf generation (2421)
5347 - update documentation, tests, type-os, comments
5348 - avoid format warnings on ia64
5349 - remove the TOE NAL
5350 - tiny code cleanups by removing unused fields
5352 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5355 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5356 - remove the most blatant lies from BUILDING file (2371)
5357 - change default debug level to reasonable production setting
5358 - reduce client side cache size to reduce cache flush time
5359 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5360 - make TCP zerocopy and pinger support enabled by default (2476)
5361 - sync writes completed after process exits caused crashes (2319)
5362 - maintain correct mount count on the MDS (2356)
5363 - backout 1557, because 2316 wasn't really fixed
5364 - better file I/O statistics gathering in /proc
5365 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5366 - another kernel patch to fix zero-copy TCP function export
5367 - don't take duplicate lock when processing re-sent getattr (2420)
5368 - lctl uses obd_self_export instead of creating new conn (2353)
5369 - MDS/OST recovery case which requires object creation asserted (2425)
5370 - move lfs from /usr/sbin to /usr/bin in packages
5371 - fix race between mds_client_add and mds_client_free (2417)
5372 - use kmalloc instead of slabs in portals (2430)
5373 - don't create duplicate records when a failover MDS is present (2442)
5374 - remove unnecessary mount age check (2332)
5375 - don't remove directory inodes from locks prematurely (2451)
5376 - don't break if MDS service name is the same as hostname (2103)
5377 - fix races in client write RPC generation when cache full (2482)
5379 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5382 - remove now-unused request->rq_obd (278)
5383 - if an allocation fails, print out how much memory we've used (1933)
5384 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5385 - add the "configurable stack size" patch to most series files (1256)
5386 - ability to write large log records, for 100+ OST configs (2306)
5387 - fix NULL deref when filter_prep fails (2314)
5388 - fix operator precedence error in filter_sync
5389 - dynamic allocation of socknal TX descriptors (2315)
5390 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5391 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5392 - ensure that CWARN messages in Portals always get to the syslog
5393 - __init/__exit are not for prototype decls (ldlm_init/exit)
5394 - x86-64 compile warning fixes
5395 - fix gateway LMC keyword conflict (2318)
5396 - fix MDS lock inversions in getattr/reint paths (1844)
5397 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5398 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5399 - DLM race condition prevented some lock evictions (2328)
5400 - ENOMEM detection and retry on socknal sends (2230)
5401 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5402 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5403 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5404 - better error messages when a client is rejected during recovery (1505)
5405 - avoid cancelling locks which were never granted, after failure (2330)
5406 - fix i_sem/journal inversion in mds_client_add (2333)
5407 - fix truncate/getattr lock cycle deadlock (2334)
5408 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5409 - fix two rare exit paths which could leak an l_lock() ref (2321)
5410 - fencepost error in MDS/OST orphan recovery (2226)
5411 - make log record alignment 8 bytes (1988)
5412 - lstripe now fails when requested offset > ost_count (2237)
5413 - ensure that all kernel series have a complete list.h (1607)
5414 - fix crashes in special-file operations (2316)
5415 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5417 - allow configurable automake binary, for testing new versions
5418 - small update to the lfs documentation
5420 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5422 * fix negative export reference count in fsfilt_sync (2312)
5424 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5425 * release candidate 0.9.1
5427 - orphans are moved into the PENDING directory for possible recovery
5428 - replayed opens now open by fid for orphan/rename safety (1042)
5429 - last close of an orphan inode generates a transno (683)
5430 - chdir() and mount() now pin the directory entry (1020)
5431 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5432 - discard very old requests without processing them (1502)
5433 - remove some common, well-understood CERRORs (1505)
5434 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5435 - clear "grant" flags in OST replies until OST grant code lands (1644)
5436 - fix read performance by not clobbering i_blksize on client (1598)
5437 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5438 - make LRU size a /proc tunable, clears locks when reduced (707)
5439 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5440 - server locks take references on exports now (1558)
5441 - build fixes for 2.4.20-rh trees (1663)
5442 - return an error from lov_create if all OSCs are inactive (1751)
5443 - fix import levels when a reconnect happens without a timeout (1597)
5444 - exit early from mds_open if we get a lookup error (1749)
5445 - partial page read at EOF wouldn't wait for disk before sending (1642)
5446 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5447 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5448 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5449 - write/truncate lock inversion (1639)
5450 - Don't auto-load obdclass, portals modules during cleanup (1495)
5451 - fix timestamps from jumping to "now" (1763)
5452 - extra journal assertions (1648)
5453 - add an extra multiunlink test (1771)
5454 - fix read_record/write_record API (1776)
5455 - fix leak of offset_extent, possible incorrect i_size later (1772)
5456 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5457 - Debugging for the kqswnal_get_idle_tx problems (1820)
5458 - Allow recovery to be attempted multiple times (1536)
5459 - Write out MDS last_rcvd file after it is first created (1600)
5460 - Fix tx_descriptor leak in failed transmit situations (1827)
5461 - ext3 journaling fixes for assertion failure after IO error (1871)
5462 - class_export_put() on freed export after completion AST error (1896)
5463 - Fix revalidate looping in VFS (1322)
5464 - Don't access a freed export during MDS_REINT timeout (1521)
5465 - Add open-unlink recovery support on the MDS (1673,1764)
5466 - Return an error if no MDS data was read from last_rcvd (1946)
5467 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5468 - Don't setup a disk device that doesn't match exported UUID (317)
5469 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5470 - avoid committing NULL handle in force close
5471 - local.sh is now a one-stripe LOV configuration
5472 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5473 - fix ext3 htree duplicate directory entry corruption (1516)
5474 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5475 - update to new LOV EA format (2097)
5476 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5478 2003-06-15 Phil Schwan <phil@clusterfs.com>
5481 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5482 - per-import/export recovery handling (958, 931, 959)
5483 - multiple last-rcvd slots, for serving multiple FSes (949)
5484 - connections are again shared between multiple imp/exports (963, 964)
5485 - "umount -f" would hang if any requests needed to be sent (393, 978)
5486 - avoid pinning large req buffer by copying for queued messages (989)
5487 - add "uuid" to "lctl device" command to help upcalls (991)
5488 - "open" RPCs with transnos would confuse recovery counters (1037)
5489 - do proper endian conversion of all wire messages (288, 340, 891)
5490 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5491 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5492 - fix ptlrpc_connection leak in target_handle_connect (1174)
5493 - fix import refcounting bug in OST and MDS cleanup (1134)
5494 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5495 - fix very unlikely obd_types race condition (501)
5496 - remove osc_open hack for echo_client (1187)
5497 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5498 - a failure in read_inode2 leads to deadlock (1139)
5499 - cancel ack-locks as soon as transaction is committed (1072)
5500 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5501 - make sure to commitrw after any preprw to avoid deadlock (1162)
5502 - failing to execute a file in a lustre FS would lock inode (1203)
5503 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5504 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5505 - obd_uuid2tgt would walk off the end of the list (1255)
5506 - on IA64 the osc would give portals incorrect bulk size (1258)
5507 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5508 - fix lock inversion caused by new llite matching code (1282)
5509 - limit the number of dirty pages on a client to 10MB (1286)
5510 - timed out locks were not being corrected cancelled (1289)
5511 - fix O_DIRECT above 4GB on IA-32 (1292)
5512 * major user-visible changes
5513 - fail out/fail over policy now controlled by the upcall (993)
5515 - add OBD_PING to check server availability and failure (954)
5516 - lustre messages are now sent in sending host order (288, 340, 891)
5517 - add eadatalen to MDS getattr reply (340)
5518 - OST read replies may contain second buffer, with per-page status (593)
5520 2003-03-11 Phil Schwan <phil@clusterfs.com>
5523 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5524 - failed open()s could cause deadlock; fixed (867, 869)
5525 - stop cancelling OST locks when files are closed (481)
5526 - overlapping XID spaces caused network corruption (851, 853)
5527 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5528 - setattr_raw added, to do single-RPC, server-side setattrs
5529 - lmc/lconf syntax change for OST UUIDs
5530 - fix crashy race condition between ptlrpc_free_req and osc_close
5531 - don't use request in mdc_enqueue if we hit a timeout (889)
5532 - don't set the inode i_size for regular files from the MDS (896)
5533 - handle out of order completion AST (842)
5534 - don't LBUG if a lock request times out after receiving AST (913)
5535 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5536 - if a bad lock AST arrives, send an error instead of dropping entirely
5537 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5538 - fix leak in bulk IO when only partially completed (899, 900, 926)
5539 - fix O_DIRECT for ia64 (55)
5540 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5541 - C-z after timeout could hang a process forever; fixed (977)
5543 - client-side I/O cache (678, 924, 929, 941, 970)
5545 - READPAGE and SETATTRs which don't take server-side locks get
5548 2003-02-11 Phil Schwan <phil@clusterfs.com>
5551 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5552 - Fix fsfilt_extN_readpage() to read a full page of directory
5553 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5554 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5555 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5556 - Most intent operations were removed, in favour of a new RPC mode
5557 that does a single RPC to the server and bypasses most of the VFS
5558 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5559 - Aggressively cancel local locks on DLM servers
5560 - mds_reint_unlink sends EA to the client if it's the last nlink.
5561 client uses that EA to unlink OST objects.
5562 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5563 - recursive symlinks were fixed (439)
5564 - fixed NULL deref in DEBUG_REQ
5565 - filter_update_lastobjid no longer calls sync, which annoyed extN
5566 - fixed multi-client small-writes to a single file problem (445)
5567 - fixed mtime updates during file writes (607)
5568 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5569 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5570 - fixed Linux OST size reporting problem (444, 656)
5571 - OST now updates object mtime with writes or setattr (607, 619)
5572 - client verifies file size before zeroing page past EOF (445)
5573 - OST now writes last allocated objid to disk with allocation (108)
5574 - LOV on echo now works (409)
5576 - mds_reint_unlink sends a new buffer, with the EA included. this
5577 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5578 set if a regular file was being unlinked, and it was the last link
5579 - use PtlGet from the target for bulk writes (315)
5580 - OST now updates object mtime with writes or setattr (607, 619)
5581 - LDLM now has a grant-time callback to revalidate locked items, if
5583 - Many MDS operations were reorganized to combat race conditions
5585 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5586 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5588 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5591 - Fully reactivate OST imports after reconnection (512, others)
5592 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5593 - More graceful error handling for truncating on dead OST (515)
5594 - Don't error out unless we're actually accessing dead stripes (474)
5595 - Fix garbage sizes when stripes are missing (410)
5596 - LRU counters were broken, causing constant lock purge (433, 432)
5597 - garbage on read from stripes with failed OSTs (441)
5598 - mark OSCs as active before reconnecting during recovery (438)
5599 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5600 - lfind did not preserve OST order in output (443)
5601 - symlinks cause hung clients, incorrect data (439)
5602 - stop dereferencing request after dropping refcount (457)
5603 - don't LASSERT(spin_is_locked) on non-SMP (455)
5604 - fixes for many rename() bugs
5605 - fstat didn't correctly synchronize attributes (399)
5606 - server must handle lock cancellation during blocking AST prep (487)
5607 - bulk descriptors were free()d too soon (511)
5608 - fix paths in lconf, which would load incorrect modules (451, 507)
5609 - fix confusing lconf 'host not found' error message (386)
5610 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5611 - fix race condition in mdc_blocking_ast() for inode access (526)
5612 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5613 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5614 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5615 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5616 - fix lstripe in several ways (kernel side) (527)
5617 - fix request leak in ldlm_cli_enqueue (262)
5618 - incorrect OSC was marked inactive after OST failure
5619 - call mds_fs_cleanup before unmounting filesystem (524)
5620 - fix races between taking ns_lock and ldlm_lock_change_resource
5621 - fix races updating LOV export open file list
5622 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5623 - fix recovery NULL deref in ldlm_cli_cancel_unused
5624 - fix some DLM races by using new hash table for lock handles (419)
5625 - permit the client to specify desired inodes, at replay
5626 - duplicate requests when we queue them for replay reintegration
5627 - fix last_rcvd offset calculation
5628 - sync after each recovered transaction, so we always make progress
5629 - never, not always, ERESTART requests without transnos
5630 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5632 - skip replay if the MDS says that the client is already connected
5633 - don't check for a recovery-enabled export to match lctl's UUID
5634 - don't INC_USE_COUNT for phantom exports
5635 - don't crash when cleaning up phantom exports (567)
5636 - don't double-finish or set replay data for errored mdc_open requests
5637 - abort requests when they time out, so we don't get old replies
5638 - send/receive replies for AST messages again
5639 - if the client says that it doesn't have the lock, cancel it on the
5641 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5642 mark it as destroyed, it will all work out when decref is called
5643 - fix module use counts (22, 581)
5645 - ASTs now expect a reply (server cancels lock on error reply)
5647 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5650 - fix many simultaneous client startup (392)
5651 - fix dentry->d_it clobbering
5652 - credentials weren't being shipped for readdir/getattr operations
5653 - remove invalid assertions triggered during some concurrent MD
5655 - proper Lustre versions added (336, 389)
5656 - fix memory leak for create error case (398)
5657 - fix LOV locking bug that would get cli/srv out of sync
5658 - fix echo client over LOV (409)
5659 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5660 - fix double-O_EXCL intent crash (424)
5661 - avoid sending multiple lock CANCELs (352)
5663 - MDS can do multi-client recovery (modulo bugs in new code)
5665 - many updates, edits, cleanups
5667 2002-11-18 Phil Schwan <phil@clusterfs.com>
5670 - fix null d_it dereference (346)
5671 - fix full OST/dbench hang (333)
5672 - fix permission problem with file removal (286)
5673 - fix removal of OSCs from LOV when they fail
5674 - fix NULL deref during bulk timeout (214)
5675 - fix problems related to multiple filesystems on one MDS (241)
5676 - fixed serious subtle metadata locking bugs
5677 - free locks on clients when inodes are removed due to memory
5679 - fix inode pointer in lock data (285)
5680 - partial support for multiple MDS on a single host (241)
5681 - data locks weren't cancelled at clear_inode time (290, 311)
5682 - intent locks could lead to unbounded lock growth (205)
5683 - added a maximum lock count, an LRU list, and a flusher
5684 - fix multiple rename (365)
5685 - properly abstracted the echo client
5686 - OSC locked 1 byte too many; fixed
5687 - rewrote brw callback code:
5688 - fixed recovery bugs related to LOVs (306)
5689 - fixed too-many-pages-in-one-write crash (191)
5690 - fixed (again) crash in sync_io_timeout (214)
5691 - probably fixed callback-related race (385)
5693 - Add capability to MDS protocol
5694 - LDLM cancellations and callbacks on different portals
5696 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5699 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5700 - defer open object destruction to close time (601981, 138)
5701 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5702 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5703 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5704 - don't call obd_setattr in ll_file_release for destroyed objects
5705 * protocol change to lustre_msg: move |version| and add |flags|
5706 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5707 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5708 * added replay of create, unlink, link and rename operations during
5709 MDS failover; recovery should be much more robust now
5710 * remove failed OSCs from LOVs (only lov_create uses this so far)
5711 * the lustre-HOWTO was brought (more) up to date (582544)
5713 2002-10-23 Phil Schwan <phil@clusterfs.com>
5716 - in-use dentries weren't being reused properly (617851)
5717 - prevent multiple LDLM setup (599178)
5718 - fix LOV size calculations for truncate (617853)
5719 - fix client handling of MDS intent errors (POSIX)
5720 - fix permission bug in lovstripe.c test (624321)
5721 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5722 - truncate past end of file could corrupt data
5723 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5724 - a race in recovery could return ETIMEDOUT to apps (623947)
5725 - building outside the source directory was fixed
5726 * the lustre-HOWTO was brought (more) up to date (582544)
5727 * major progress was made on recovery functionality
5729 2002-10-10 Phil Schwan <phil@clusterfs.com>
5732 - recovery deadlock fix
5733 - rm -rf causes LBUG fix (617817)
5734 - file open by multiple tasks fix (618962)
5735 - directory permissions bugs (602707 and 620007)
5736 - journal_stop fixed with locking (611313)
5737 - O_APPEND failures resolved (618273, perhaps 614459)
5738 - lconf PATH fix (619770)
5739 - IA64 build fix (621450)
5740 - RPC buffer sizes scale with amount of memory
5742 2002-10-01 Phil Schwan <phil@clusterfs.com>
5745 - locks would be cancelled without throwing away data pages,
5746 resulting in inconsistent data (605627)
5747 - inode attributes were not always being refreshed (605627, 612449)
5748 - lconf now continues to cleanup after lctl reports an error
5749 - MDS now enforces user permissions (602707)
5750 - lprocfs cleanup fixed, but not yet enabled (614157)
5751 - fixed infinite server hang, should a client not respond to an AST
5752 - avoid going into recovery if user calls readlink() with a buffer
5753 that's too small (613941)
5754 - AST RPCs no longer require replies (614867) -- this may be changed
5755 - don't crash server if client sends an IOV that's too big (611336)
5756 - fixed lock conversion deadlock (611892)
5757 - fixed the following of symlinks (614622)
5758 * recovery: the server can remove locks from a client that dies, other
5759 clients can make progress
5760 * more extN patch fixes
5761 * compile-time configurable ptlrpc buffer allocations
5763 - collaborative read cache document
5764 - Lustre Lite Performance CDR document-in-progress
5766 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5769 - fix typo in patch-2.4.18
5771 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5774 - clear ptlrpc request each time in handle_incoming_request()
5775 - unlink of files now destroys the object on the OST
5777 2002-09-19 Peter Braam <braam@clusterfs.com>
5779 * add hard link support
5780 * change obdfile creation method
5781 * kernel patch changed
5783 2002-09-19 Peter Braam <braam@clusterfs.com>
5786 - stack overflow bug in extN fixed
5788 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5790 * documentation updates
5791 - add man pages for config tools
5792 - update tests/README to describe testing with new config tools
5793 - finish metadata API descriptions
5794 * bug fixes and cleanups
5795 - statfs workaround for 16TB limit
5796 - LOV stripe allocation improved, can stripe on subset of OSTs
5797 - LOV file size/IO offset was wrong for files > 4GB in size
5798 - object EA data was being dropped, caused files to be unreadable
5799 - memory overflow with non-LOV OST caused memory corruption
5800 - fixed regression tests to work with new config tools, obdfilter
5801 - fixed bug when directory size became larger than 1 block
5802 - fixed bug (for single client case) when PWD was deleted
5803 - invalidate local directory pages when doing intent-based ops
5804 - avoid LDLM oops when lock callback contained bad data
5806 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5808 * documentation updates
5809 * bug fixes and cleanups
5810 - configuration tools
5813 - 64-bit compile warnings
5814 - 64-bit internal statfs data
5816 * test_brw on persistent OST devices
5818 * lprocfs (disabled)
5820 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5822 * documentation updates
5823 * bug fixes and cleanups
5824 * configuration tools
5826 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5829 * many small fixes to 0_5_4
5830 * io/network handling
5831 * thinkos in MDS operations
5833 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5836 * crucial basic fixes to 0.5.3
5837 * IOR, Iozone work over Elan
5840 2002-08-07 Phil Schwan <phil@clusterfs.com>
5841 * version 0_5_3, our first alpha
5842 * we use the new Portals iovs
5843 * documentation updates
5844 * bug fixes and cleanups
5845 * small changes in the DLM wire protocol
5847 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5848 * version 0_5_1 with some initial stability,
5849 * locking on MD and file I/O.
5850 * documentation updates
5851 * several bug fixes since 0.5.0
5852 * small changes in wire protocol
5854 2002-07-18 Phil Schwan <phil@clusterfs.com>
5856 * delivered as Lustre Light Alpha
5857 * fixed a crash after handling invalid MDS requests
5858 * fixed directory pages for architectures with non-4k pages sizes
5860 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5861 * release version v0_4_4
5862 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5864 * Fixes a number of interruption problems with OST operations.
5865 * Update documentation for portals header changes
5866 * Move all wire protocol structs/defines to lustre_idl.h
5867 * Fixes symlink length bug.
5868 * Add tcpdump to repository.
5870 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5871 * release version v0_4_3
5872 * Fixes statfs for inodes on extN.
5873 * Fixes bug in runtests which would delete /etc/hosts.
5874 * Use 64-bit object IDs wherever possible (not into VFS though)
5875 Remove ost_get_info, which is unused by lustre, and out of date.
5877 2002-07-03 Peter Braam <braam@clusterfs.com>
5878 * release version v0_4_2 Fixes a lookup error (type not passed)
5879 * move forward to head of Portals
5880 * move forward to latest Lustre kernel
5882 2002-06-25 Peter Braam <braam@clusterfs.com>
5883 * release version v0_4_1. Hopefully stable on single node use.