1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.37 (SLES 10),
5 2.6.18-128.1.6.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.41.6.sun1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
14 * File join has been disabled in this release, refer to Bugzilla 16929.
19 Description: access to llog context before init.
20 Details : move handling CATALOGS file at osc layer and forbid access to llog
25 Description: Avoid deadlock for local client writes
26 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
27 memory freeing context. This allows OST threads to set the
28 PF_MEMALLOC flag on task structures in order to allocate memory
29 from reserved pools and complete IO.
30 Use GFP_HIGHUSER for OST allocations for non-local client writes,
31 so that the OST threads generate memory pressure and allow
32 inactive pages to be reclaimed.
37 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
38 Details : move ldlm namespace creation in setup phase to avoid grab
39 _lprocfs_lock with cli_sem held.
43 Description: Temporarily disable grant shrink.
44 Details : Disable the feature for debugging.
48 Description: Unable to run several mkfs.lustre on loop devices at the same
50 Details : mkfs.lustre returns error 256 on the concurrent loop devices
51 formatting. The solution is to proper handle the error.
53 Severity : enhancement
55 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
57 Severity : enhancement
59 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
63 Description: resolve race between obd_disconnect and class_disconnect_exports
64 Details : if obd_disconnect will be called to already disconnected export he
65 forget release one reference and osc module can't unloaded.
67 Severity : enhancement
69 Description: Allow tuning service thread via /proc
70 Details : For each service a new
71 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
72 created that can be used to set min/max thread counts, and get the
73 current number of running threads.
77 Descriptoin: don't return error if have particaly created objects for file.
78 Details : lov_update_create_set uses set->set_success as index for created objects,
79 so if some requests will be failed, they will have hole at end of
80 array and we can use qos_shrink_lsm for allocate correct lsm.
82 Severity : enhancement
84 Description: Update OFED support to 1.4.1
88 Description: Reduce small size read RPC
89 Details : Set read-ahead limite for every file and only do read-ahead when
90 available read-ahead pages are bigger than 1M to avoid small size
93 Severity : enhancement
95 Description: move AT tunable parameters for more consistent usage
96 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
98 Severity : enhancement
100 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
101 Details : allow skip disconnected ost for send statfs request and hide error
107 Description: fix racy locking of mballoc block bitmaps causing BUG
108 Details : The locking of the mballoc buddy bitmap and the in-memory
109 block bitmap was using two different spin locks in some
110 cases. This made it possible to incorrectly access the
111 mballoc bitmap while another process was modifying it,
112 causing a sanity assertion to fail. While no on-disk corruption
113 was reported, there was some risk of this happening.
116 Frequency : rare, on llog test 6
118 Descriptoin: don't allow connect to already connected import
119 Details : allowing connect to already connected import is hide connecting problem.
124 Descriptoin: allow kill process which wait statahead result
125 Details : for some reasons 'ls' can stick in waiting result from statahead,
126 in this case need way for kill this process.
128 Severity : enhancement
130 Description: Add state history info file, enhance import info file
131 Details : Track import connection state changes in a new osc/mdc proc file;
132 add overview-type data to the osc/mdc import proc file.
134 Severity : enhancement
136 Description: MDS create should not wait for statfs RPC while holding DLM lock.
138 Severity : enhancement
140 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
145 Descriptoin: OST grant too much space to client even there are not enough space.
146 Details : Client will shrink its grant cache to OST if there are no write
147 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
148 this grant cache if there are already not enough avaible space
149 (left_space < total_clients * 32M).
152 Frequency : start MDS on uncleanly shutdowned MDS device
154 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
155 Details : stay in waiting mds<>ost recovery finished produce random bugs
156 due race between two ll_sync thread for one lov target. send
157 ACTIVATE event only if connect realy finished and import have
161 Frequency : rare, connect and disconnect target at same time
163 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
164 Details : don't call obd_disconnect under lov_lock. this long time
165 operation and can block ptlrpcd which answer to connect request.
170 Descriptoin: don't lose wakeup for imp_recovery_waitq
171 Details : recover_import_no_retry or invalidate_import and import_close can
172 both sleep on imp_recovery_waitq, but we was send only one wakeup
176 Frequency : always with long access acl
178 Descriptoin: mds can't pack reply with long acl.
179 Details : mds don't control size of acl but they limited by reint/getattr
182 Severity : enhancement
184 Description: Update to SLES10 kernel-2.6.16.60-0.33.
186 Severity : enhancement
188 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
191 Frequency : start MDS on uncleanly shutdowned MDS device
193 Descriptoin: aborting recovery hang on MDS
194 Details : don't throttle destroy RPCs for the MDT.
197 Frequency : on remount
199 Description: external journal device not working after the remount
200 Details : clear dev_rdonly flag for external journal devices in
206 Description: shutdown vs evict race
207 Details : client_disconnect_export vs connect request race.
208 if client will evicted at this time - we start invalidate
209 thread without referece to import and import can be freed
213 Frequency : rare, need acl's on inode.
215 Description: client can't handle ost additional correctly
216 Details : if ost was added after client connected to mds client can have
217 hit lnet_try_match_md ... to big messages to wide striped files.
218 in this case need teach client to handle config events about add
219 lov target and update client max ea size at that event.
221 Severity : enhancement
223 Description: Changelogs
224 Details : Changelogs are a lightweight mechanism to track filesystem
225 metadata and namespace changes. The changelog is recorded
226 permanently on the MDTs, and is periodically "consumed" / purged
227 when records are no longer needed.
229 Severity : enhancement
231 Description: compact fld format with extents
232 Details : Store range of seq rather than every seq in FLD. Seq
233 controller update FLD rather than clients. In Case of CMD, mdt0
234 has FLD, all other metadata server act as non persistent proxy
235 for FLD queries and cache fld entries in fld cache.
240 Description: don't skip ost target if they assigned to file
241 Details : Drop slow OSCs if we can, but not for requested start idx.
242 This means "if OSC is slow and it is not the requested
243 start OST, then it can be skipped, otherwise skip it only
244 if it is inactive/recovering/out-of-space.
248 Description: more cleanup in mds_lov
249 Details : not send LOV EA under replay, we can't know about they size at this
250 time. Don't allow client connect to mds before any ost connected,
251 for avoid problems with LOV EA size and returning EIO to client.
253 Severity : enhancement
255 Description: Interoperability at server side (Disk interoperability)
257 Severity : enhancement
259 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
261 Severity : enhancement
263 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
265 Severity : enhancement
267 Description: New client IO stack (CLIO).
269 Severity : enhancement
271 Description: Commit on sharing. Eliminate inter-client dependencies between
272 uncommitted transactions by doing transaction commits.
273 Thereby clients may recovery independently.
276 Frequency : Create a symlink file with a very long name
278 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
279 Details : If there is no extra space in the request for early cancels,
280 ldlm_req_handles_avail() returns 0 instead of a negative value.
282 Severity : enhancement
284 Description: Add /proc entry for import status
285 Details : The mdc, osc, and mgc import directories now have
286 an import directory that contains useful import data for debugging
289 Severity : enhancement
291 Description: Re-disable certain /proc logging
292 Details : Enable and disable client's offset_stats, extents_stats and
293 extents_stats_per_process stats logging on the fly.
296 Frequency : Only on FC kernels 2.6.22+
298 Description: oops in statahead
299 Details : Do not drop reference count for the dentry from VFS when lookup,
300 VFS will do that by itself.
302 Severity : enhancement
304 Description: Generic /proc file permissions
305 Details : Set /Proc file permissions in a more generic way to enable non-
306 root users operate on some /proc files.
310 Description: Hitting mdc_commit_close() ASSERTION
311 Details : Properly handle request reference release in
312 ll_release_openhandle().
316 Description: quota recovery deadlock during mds failover
317 Details : This patch includes att18982, att18236, att18237 in bz14840.
319 1. fix osts hang when mds does failover with quotaon
320 2. prevent watchdog storm when osts threads wait for the
325 Frequency : only patchless client
326 Description: add workaround for race between add/remove dentry from hash
328 Severity : enhancement
330 Description: Allow OST glimpses to return PW locks
334 Description: LBUG when llog conf file is full
335 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
339 Description: Prevent import from entering FULL state when server in recovery
343 Description: service mount cannot take device name with ":"
344 Details : Only when device name contains ":/" will mount treat it as
350 Description: replace ptlrpcd with the statahead thread to interpret the async
351 statahead RPC callback
355 Frequency : on recovery
356 Description: I/O failures after umount during fail back
357 Details : if client reconnected to restarted server we need join to recovery
358 instead of find server handler is changed and process self
359 eviction with cancel all locks.
361 Severity : enhancement
363 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
365 Severity : enhancement
367 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
369 Severity : enhancement
371 Description: Upcall on Lustre log has been dumped
372 Details : Allow for a user mode script to be called once a Lustre log has
373 been dumped. It passes the filename of the dumped log to the
374 script, the location of the script can be specified via
375 /proc/sys/lnet/debug_log_upcall.
380 Description: avoid idr_remove called for id which is not allocated.
381 Details : Move assigment s_dev for clustered nfs to end of initialization,
382 to avoid problem with error handling.
387 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
388 Details : When connection is reused this not moved from CONN_UNUSED_HASH
389 into CONN_USED_HASH and this prodice warning when put connection
390 again in unused hash.
393 Severity : enhancement
395 Description: File striping can now be set to use an arbitrary pool of OSTs.
397 Severity : enhancement
399 Description: Export bytes_read/bytes_write count on OSC/OST.
403 Description: Early reply size mismatch, MGC loses connection
404 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
405 the connect flags are properly negotiated.
410 Description: quotas are not honored with O_DIRECT
411 Details : all writes with the flag O_DIRECT will use grants which leads to
412 this problem. Now using OBD_BRW_SYNC to guard this.
416 Description: add quota statistics
417 Details : 1. sort out quota proc entries and proc code.
418 2. add quota statistics
420 Severity : enhancement
422 Description: enable quota support for HEAD.
426 Description: Properly propagate oinfo flags from lov to osc for statfs
427 Details : restore missing copy oi_flags to lov requests.
429 Severity : enhancement
431 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
432 : l_getgroups(8), lst(8), routerstat(8)
433 Details : included man pages for llobdstat(8), llstat(8),
434 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
436 Severity : enhancement
438 Description: configure's --enable-quota should check the
439 : kernel .config for CONFIG_QUOTA
440 Details : configure is terminated if --enable-quota is passed but
441 : no quota support is in kernel
445 Description: Remove portals compatibility
446 Details : Remove portals compatibility, not interoperable with releases
451 Description: Resolve device initialization race
452 Details : Prevent proc handler from accessing devices added to the
453 obd_devs array but yet be intialized.
455 Severity : enhancement
457 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
459 Severity : enhancement
461 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
466 Description: Using wrong pointer in osc_brw_prep_request
467 Details : Access to array[-1] can produce panic if kernel compiled with
468 CONFIG_PAGE_ALLOC enabled
472 Description: Client runs out of low memory
473 Details : Consider only lowmem when counting initial number of llap pages
477 Description: *optional* service tags registration
478 Details : if the "service tags" package is installed on a Lustre node
479 When the filesystem is mounted, a local-node service tag will
480 be created. See http://inventory.sun.com/ for more information
481 about the Service Tags asset management system.
485 Description: Kernel BUG tries to release flock
486 Details : Lustre does not destroy flock lock before last reference goes
487 away. So always drop flock locks when client is evicted and
488 perform unlock regardless of successfulness of speaking to MDS.
492 Description: add refcount protection for osc callbacks, avoid panic on shutdown
496 Description: sanity test 65a fails if stripecount of -1 is set
497 Details : handle -1 striping on filesystem in ll_dirstripe_verify
502 Description: ASSERTION(CheckWriteback(page,cmd)) failed
503 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
506 Severity : enhancement
508 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
512 Description: do not process already freed flock
513 Details : flock can possibly be freed by another thread before it reaches
514 to ldlm_flock_completion_ast.
518 Description: LBUG during stress test
519 Details : Need properly lock accesses the flock deadlock detection list.
523 Description: oops in page fault handler
524 Details : kernel page fault handler can return two special 'pages' in error
525 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
529 Description: timeout with invalidate import.
530 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
531 should be handled by ptlrpcd. This produce long age waiting and
532 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
534 Severity : enhancement
536 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
540 Description: enable MGS and MDT services start separately
541 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
542 a co-located MGS without starting the MGS, which is a complement
543 to 'nosvc' mount option.
547 Frequency : after recovery
548 Description: precreate to many object's after del orphan.
549 Details : del orphan st in oscc last_id == next_id and this triger growing
550 count of precreated objects. Set flag LOW to skip increase count
551 of precreated objects.
555 Frequency : rare, on clear nid stats
556 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
557 Details : when clean nid stats sometimes try destroy live entry,
558 and this produce panic in free.
562 Description: Stack overflow during MDS log replay
563 ease stack pressure by using a thread dealing llog_process.
567 Description: wait until IO finished before start new when do lock cancel.
568 Details : VM protocol want old IO finished before start new, in this case
569 need wait until PG_writeback is cleared until check dirty flag and
570 call writepages in lock cancel callback.
572 Severity : enhancement
574 Description: using special macro for print time and cleanup in includes.
578 Description: mds_mfd_close() ASSERTION(rc == 0)
579 Details : In mds_mfd_close(), we need protect inode's writecount change
580 within its orphan write semaphore to prevent possible races.
584 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
588 Frequency : rare, on shutdown ost
589 Description: don't hit live lock with umount ost.
590 Details : shrink_dcache_parent can be in long loop with destroy dentries,
591 use shrink_dcache_sb instead.
595 Description: don't panic with use echo client
596 Details : echo client pass NULL as client nid pointer and this produce null
601 Description: fix build on ppc32
602 Details : compile code with -m64 flag produce wrong object file for ppc32.
606 Description: add message levels for liblustreapi
610 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
611 Details : Move the 'good_osts' check before the 'total_bavail' check. This
612 will result in an -EAGAIN and in the exit call path we call
613 alloc_rr() which will with increasing aggressiveness attempt to
614 aquire precreated objects on the minimum number of required OSCs.
618 Description: Use old size assignment to avoid deadlock
619 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
620 the scheduling while holding a spinlock. We do not need locking
621 for size in ll_update_inode() because size is only updated from
622 the MDS for directories or files without objects, so there is no
623 other place to do the update, and concurrent access to such inodes
624 are protected by the inode lock.
628 Description: resolve "_IOWR redefined" build error on SLES10
632 Description: dump the memory debugging after all modules are unloaded to
633 suppress false negative in conf_sanity test 39
635 Severity : enhancement
637 Description: build kernel-ib packages for OFED 1.3 in our release cycle
642 Description: fix SLES kernel versioning
643 Details : the kernel version for our SLES 10 kernel did not include a "-"
644 before the "smp" at the end. while this was not a problem in
645 general, it did mean that software trying to use the kernel
646 version to try to detect a vendor specific kernel would fail.
647 this was most evident by the OFED build scripts.
651 Description: Don't update lov_desc members until making sure they are valid
652 Details : When updating lov_desc members via proc fs, need fix their
653 validities before doing the real update.
657 Description: don't put request into delay list while invalidate in flight.
658 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
659 import in flight. this produce timeout for invalidate and sometimes
660 can cause stale data.
664 Frequency : on ppc only
665 Description: not convert ost objects for directory because it's not exist.
666 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
670 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
672 Severity : enhancement
674 Description: Update to support 2.6.22.14 vanilla kernel.
678 Frequency : rare, on recovery
679 Description: read procfs can produce deadlock in some situation
680 Details : Holding lprocfs lock which send rpc can produce block for destroy
681 obd objects and this also block reconnect with -EALREADY. This
682 isn't fix all lprocfs bugs - but make it rare.
684 Severity : enhancement
686 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
689 Frequency : frequent on X2 node
691 Description: mdc_set_open_replay_data LBUG
692 Details : Set replay data for requests that are eligible for replay.
696 Description: lustre_mgs: operation 101 on unconnected MGS
697 Details : When MGC is disconnected from MGS long enough, MGS will evict the
698 MGC, and late on MGC cannot successfully connect to MGS and a lot
699 of the error messages complaining that MGS is not connected.
703 Frequency : on network error
704 Description: panic with double free request if network error
705 Details : mdc_finish_enqueue is finish request if any network error ocuring,
706 but it's true only for synchronus enqueue, for async enqueue
707 (via ptlrpcd) this incorrect and ptlrpcd want finish request
710 Severity : enhancement
712 Description: client-side metadata stat-ahead during readdir(directory readahead)
713 Details : perform client-side metadata stat-ahead when the client detects
714 readdir and sequential stat of dir entries therein
717 Frequency : on start mds
719 Description: Implement get_info(last_id) in obdfilter.
722 Frequency : occasional
724 Description: Correctly check stale fid, not start epoch if ost not support SOM
725 Details : open with flag O_CREATE need set old fid in op_fid3 because
726 op_fid2 was overwritten with new generated fid, but mds can answer
727 with one of these two fids and both is not stale. Setattr
728 incorrectly started an epoch and assume will be called
729 done_writing, but without SOM done_writing ever being called.
732 Frequency : rare, depends on device drivers and load
734 Description: MDS or OSS nodes crash due to stack overflow
735 Details : Code changes in 1.8.0 increased the stack usage of some functions.
736 In some cases, in conjunction with device drivers that use a lot
737 of stack the MDS (or possibly OSS) service threads could overflow
738 the stack. One change which was identified to consume additional
739 stack has been reworked to avoid the extra stack usage.
742 Frequency : occasional
744 Description: Do not fail import if osc_interpret_create gets -EAGAIN
745 Details : If osc_interpret_create got -EAGAIN it immediately exits and
746 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
747 oscc_has_objects and see OSC has no objests and call
748 oscc_internal_create to resend create request.
750 Severity : enhancement
752 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
754 Severity : enhancement
756 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
759 Frequency : very rare
761 Description: Fix replay if there is an un-replied request and open
762 Details : In some cases, older replay request will revert the
763 mcd->mcd_last_xid on MDS which is used to record the client's
766 Severity : enhancement
768 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
770 Severity : enhancement
772 Description: Add rhel5 support to HEAD.
774 Serverity : enhancement
776 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
781 Description: Don't allow skipping OSTs if index has been specified.
782 Details : Don't allow skipping OSTs if index has been specified, make
783 locking in internal create lots better.
787 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
788 Details : access to bitfield in structure is always rounded to long
789 and this produce problem with not atomic change any bit.
793 Description: Lustre make rpms failed.
794 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
795 builds Lustre rpms from tarball.
798 Frequency : rare on shutdown ost
800 Description: If llog cancel was not send before clean_exports phase, this can
801 produce deadlock in llog code.
802 Details : If llog thread has last reference to obd and call class_import_put
803 this produce deadlock because llog_cleanup_commit_master wait when
804 last llog_commit_thread exited, but this never success because was
805 called from llog_commit_thread.
809 Description: allow userland application know is lost one of stripes.
810 Details : fill lvb_blocks with error code on ost and return it to
811 application if error flag found.
815 Description: NULL lov_tgts causing MDS oops
816 Details : more safe checks for NULL lov_tgts for avoid oops.
818 Severity : enhancement
820 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
824 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
825 Details : If an OST is down the MDS will hang indefinitely in
826 obd_statfs_rqset() waiting for the statfs data. While for
827 MDS QOS usage of statfs, it should not stuck in waiting.
829 Severity : enhancement
831 Description: remote_acl support
832 Details : Support ACL-based permission check for remote user.
833 Support setfacl/getfacl for remote user with the utils
834 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
835 the system "{s,g}etfacl" utils.
837 Severity : enhancement
839 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
841 Severity : enhancement
843 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
847 Description: make mgs_setparam() handle fsname containing dash
848 Details : fsname containing a dash does not work with lctl conf_param
850 Severity : enhancement
852 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
854 Severity : enhancement
856 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
860 Frequency : rare, at shutdown
861 Description: access already free / zero obd_namespace.
862 Details : if client_disconnect_export was called without force flag set,
863 and exist connect request in flight, this can produce access to
864 NULL pointer (or already free pointer) when connect_interpret
865 store ocd flags in obd_namespace.
869 Frequency : only at startup
870 Description: not alloc memory with spinlock held.
871 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
872 if any spinlock held.
874 Severity : enhancement
876 Description: make lustre randomly fail allocating memory
877 Details : Make lustre randomly failed allocating memory for testing purpose.
879 Severity : enhancement
881 Description: lost problems with lov objid file
882 Details : Fixes some scability and access to not inited memory problems
883 in work with lov objdid file.
888 Description: lfs find does not continue on file error
889 Details : Continue other files processing when a file/dir is absent.
893 Description: Inconsistent usage of lustre_pack_reply()
894 Details : Standardize the usage of lustre_pack_reply() such that it
895 always generate a CERROR on failure.
898 Frequency : occasional
900 Description: MDS hang or stay in waiting lock
901 Details : If client receive lock with CBPENDING flag ldlm need send lock
902 cancel as separate rpc, to avoid situation when cancel request
903 can't processed due all i/o threads stay in wait lock.
907 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
911 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
915 Description: Update to SLES9 kernel-2.6.5-7.287.3.
919 Description: Update to RHEL4 latest kernel.
921 Severity : enhancement
923 Description: Build SLES10 patchless client fails
924 Details : The configure was broken by run ./configure with
925 --with-linux-obj=.... argument for patchless client. When the
926 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
927 can't search header adequately. Use absolute path such as
928 -I($LINUX)/include instead.
932 Description: interrupt oig_wait produce painc on resend.
933 Details : brw_redo_request can be used for resend requests from ptlrpcd and
934 private set, and this produce situation when rq_ptlrpcd_data not
935 copyed to new allocated request and triggered LBUG on assert
936 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
937 ptlrpcd set if request is changed and can be safety changed to use
942 Description: LASSERT_{REQ,REP}SWAB macros are buggy
943 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
944 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
945 these macros and replace them with their difinitions instead.
949 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
950 Details : Update which_patch & target file for SLES10 latest kernel.
953 Description: Remove client patches from SLES 10 kernel.
954 Details : This causes SLES 10 clients to behave as patchless clients
955 even on a Lustre-patched (server) kernel.
957 Severity : enhancement
959 Description: self-adjustable client's lru lists
960 Details : use adaptive algorithm for managing client cached locks lru
961 lists according to current server load, other client's work
962 pattern, memory activities, etc. Both, server and client
963 side namespaces provide number of proc tunables for controlling
966 Severity : enhancement
968 Description: light-weight GSS support
969 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
970 and reduce performance overhead.
972 Severity : enhancement
974 Description: Linux keyring support
975 Details : Support using service of Linux keyring for Lustre GSS internal
976 context refresh/cache mechanism.
980 Description: Fix errors in lfs documentation
981 Details : Fixes man pages
985 Description: don't use GFP_* in generic Lustre code.
986 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
990 Description: obdclass is limited by single OBD_ALLOC(idarray)
991 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
995 Description: warnings with build patchless client with vanila 2.6.19 and up
996 Details : change old ctl_table style and replace ctl_table/ctl_table_header
997 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1001 Frequency : Only for SLES
1002 Description: sanity_quota fail test_1
1003 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1004 file, which makes TSTID[2] inunique.
1008 Frequency : Only for SLES9
1009 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1010 Details : Update target/ChangeLog/which_patch .
1014 Frequency : Only for test_10 in sanity-quota.sh
1015 Description: a bug in quota test code
1016 Details : checking a test flag in a wrong place
1020 Frequency : Only for test_13 in sanity-quota.sh
1021 Description: a bug in quota test script
1022 Details : shouldn't check overfull allocated quota
1026 Description: jbd statistics
1027 Details : Port older jbd statistics patch for sles10
1029 Severity : enhancement
1031 Description: Add printing inode info into message about error in writepage.
1036 Description: reply_lock_interpret crash due to race with it and lock cancel.
1037 Details : Do not replay locks that are being cancelled. Do not reference
1038 locks by their address during replay, just by their handle.
1042 Frequency : When flocks are used.
1043 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1044 Details : Flock locks might destroy just granted lock if it could be merged
1045 with another existing flock, this is done in completion handler,
1046 so teach ldlm_cli_enquque_fini that this is a valid case for
1052 Description: Oops in read and write path when failing to allocate lock.
1053 Details : Check if lock allocation failed and return error back.
1057 Description: lstripe command fails for valid OST index
1058 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1059 instead of lov->desc.ld_active_tgt_count.
1063 Description: sanity.sh failed test 103
1064 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1069 Description: df doesn't work properly if diskfs blocksize != 4K
1070 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1074 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1075 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1077 Severity : enhancement
1079 Description: Adaptive timeouts
1080 Details : RPC timeouts adapt to changing server load and network
1081 conditions to reduce resend attempts and improve recovery time.
1085 Description: llapi_file_create() does not allow some changes
1086 Details : add llapi_file_open() that allows specifying the mode and
1087 open flags, and also returns an open file handle.
1091 Description: merge and cleanup kernel patches.
1092 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1096 Description: Add journal checksum support.(Kernel part)
1097 Details : The journal checksum feature adds two new flags i.e
1098 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1099 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1100 indicates that the commit block contains the checksum for
1101 the blocks described by the descriptor blocks. Now commit
1102 record can be sent to disk without waiting for descriptor
1103 blocks to be written to disk. This behavior is controlled
1104 using JBD2_FEATURE_ASYNC_COMMIT flag.
1108 Description: OSS needs mutliple precreate threads
1109 Details : Add ability to start more than one create thread per OSS.
1113 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1114 Details : execute lfs setstripe on client
1118 Description: mds_obd_create error creating tmp object
1119 Details : When the user sets quota on root, llog will be affected and can't
1120 create files and write files.
1123 Frequency : Always on ia64 patchless client, and possibly others.
1125 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1126 Details : This allows the patchless client to be loaded on architectures
1127 without this export.
1131 Description: RedHat Update kernel for RHEL5
1132 Details : Add the kernel config file for RHEL5.
1136 Description: RedHat Update kernel for RHEL5
1137 Details : Modify the kernel config file more closer RHEL5.
1141 Description: Build failure against Centos5 (RHEL5)
1142 Details : Define PAGE_SIZE when it isn't present.
1146 Description: test_7 on sanity-quota.sh can't be used on separate machines
1147 Details : add facet to handle it
1151 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1152 Details : with NFS, the anon dentry's parent was set to itself in
1153 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1154 corresponding dentry other than use rec->ur_name.
1156 Severity : enhancement
1158 Description: lfs setstripe enhancement
1159 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1163 Description: enable data checksumming by default
1164 Details : enable checksum by default, allow --disable-checksum
1165 configure option and "-o nochecksum" mount option. Checksums
1166 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1170 Description: lustre support for RHEL5
1171 Details : Add support for RHEL5.
1175 Description: Client eviction due to failover config
1176 Details : after a connection loss, the lustre client should attempt to
1177 reconnect to the last active server first before trying the
1178 other potential connections.
1182 Description: when mds and osts use different quota unit(32bit and 64bit),
1183 quota will be released repeatly.
1184 Details : void sending multiple quota reqs to mds, which will keep the
1185 status between the reqs.
1189 Description: osts not allocated evenly to files
1190 Details : change the condition to increase offset_idx
1193 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1195 Description: Data corruption for OSTs that are formatted larger than 2TB
1197 Details : When generating the bio request for lustre file writes the
1198 sector number would overflow a temporary variable before being
1199 used for the IO. The data reads correctly from Lustre (which
1200 will overflow in a similar manner) but other file data or
1201 filesystem metadata may be corrupted in some cases.
1205 Description: Tune the kernel for good SCSI performance.
1206 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1207 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1212 Description: rewrite ext2-derived code in obdclass/uuid.c
1213 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1214 from scratch preserving functionality.
1218 Description: "lfs find -obd UUID" prints directories
1219 Details : "lfs find -obd UUID" will return all directory names instead
1220 of just file names. It is incorrect because the directories
1221 do not reside on the OSTs.
1225 Description: use i_size_read and i_size_write in 2.6 port
1226 Details : replace inode->i_size access with i_size_read/write()
1229 Frequency : when removing large files
1231 Description: scheduling issue during removal of large Lustre files
1232 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1233 It causes scheduling issues when removing large files (17TB in the
1237 Frequency : only with liblustre clients on XT3
1239 Description: evictions taking too long
1240 Details : allow llrd to evict clients directly on OSTs
1243 Frequency : only on ppc
1245 Description: /proc/fs/lustre/devices broken on ppc
1246 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1247 We should make sure the type of 'index' is loff_t and then cast
1248 explicitly as needed below. Do not assign an explicitly cast
1252 Frequency : only for rhel5
1254 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1255 Details : Modify the target file & which_kernel.
1257 Severity : enhancement
1259 Description: omit set fsid for export NFS
1260 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1261 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1265 Frequency : Always, for kernels after 2.6.16
1266 Description: Fix warning idr_remove called for id=.. which is not allocated.
1267 Details : Last kernels save old s_dev before kill super and not allow
1268 to restore from callback - restore it before call kill_anon_super.
1272 Description: buffer overruns could theoretically occur
1273 Details : llapi_semantic_traverse() modifies the "path" argument by
1274 appending values to the end of the origin string, and a
1275 overrun may occur. Adding buffer overrun check in liblustreapi.
1279 Description: Fix error on 'ls .' at the top of the Lustre mount.
1280 Details : Don't revalidate dentry if it is a root dentry.
1284 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1285 Details : Modify vm-tunables-rhel4.patch.
1289 Description: Kernel config for 2.6.18-vanilla.
1290 Details : Modify targets/2.6-vanilla.target.in.
1291 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1292 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1293 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1294 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1298 Description: improve handling recoverable errors
1299 Details : if request processig with error which can be recoverable on server
1300 request should be resend, otherwise page released from cache and
1305 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1306 Details : Modify target file & which_patch.
1307 A flaw was found in the IA32 system call emulation provided
1308 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1309 value could be stored in the %RAX register, which could trigger an
1310 out-of-bounds system call table access. An untrusted local user
1311 could exploit this flaw to run code in the kernel
1312 (ie a root privilege escalation). (CVE-2007-4573).
1316 Description: O_DIRECT bypasses client statistics.
1317 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1318 read_ahead_stats were updated. copied the stats section from
1319 osc_send_oap_rpc() into async_internal().
1323 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1327 Description: change order of libsysio includes
1328 Details : '#include sysio.h' should always come before '#include xtio.h'
1332 Description: handle "serious error: objid * already exists" more gracefully
1333 Details : If LAST_ID value on disk is smaller than the objects existing in
1334 the O/0/d* directories, it indicates disk corruption and causes an
1335 LBUG(). If the object is 0-length, then we should use the existing
1336 object. This will help to avoid a full fsck in most cases.
1338 Severity : enhancement
1340 Description: adapt the lustre_config script to support the upgrade case
1341 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1342 server targets to 1.6 in parallel.
1346 Description: To avoid grant space > avaible space when the disk is almost
1347 full. Without this patch you might see the error "grant XXXX >
1348 available" or some LBUG about grant, when the disk is almost
1350 Details : In filter_check_grant, for non_grant cache write, we should
1351 check the left space by if (*left > ungranted + bytes), instead
1352 of (*left > ungranted), because only we are sure the left space
1353 is enough for another "bytes", then the ungrant space should be
1354 increase. In client, we should update cl_avail_grant only there
1355 is OBD_MD_FLGRANT in the reply.
1359 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1360 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1361 described in CVE-2007-4573. This problem could allow untrusted
1362 local users to gain root access.
1365 Frequency : when using O_DIRECT and quotas
1367 Description: Incorrect file ownership on O_DIRECT output files
1368 Details : block usage reported by 'lfs quota' does not take into account
1369 files that have been written with O_DIRECT.
1374 Description: touch file failed when fs is not full
1375 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1376 otherwise we can get ENOSP while fs is not full.
1380 Description: parallel lock callbacks
1381 Details : Instead of sending blocking and completion callbacks as separated
1382 requests, adding them to a set and sending in parallel.
1385 Frequency : only for Cray XT3
1386 Bugzilla : 12829/13455
1387 Description: Changing primary group doesn't change the group lustre assigns to
1389 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1390 overridden with the primary group provided by the group upcall,
1391 whereas the supplied fsgid can be trusted if it is in the list of
1392 supplementary groups returned by the group upcall.
1394 Severity : enhancement
1396 Description: Allow masking D_WARNING, D_ERROR messages from console
1397 Details : Console messages can now be disabled via lnet.printk.
1401 Description: User code with malformed file open parameter crashes client node
1402 Details : Before packing join_file req, all the related reference should be
1403 checked carefully in case some malformed flags cause fake
1404 join_file req on client.
1408 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1409 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1410 mode is equal to the granted mode. However, it should also check
1411 that LDLM_FL_CP_REQD is not set.
1415 Description: Heavy nfs access might result in deadlocks
1416 Details : After ELC code landed, it is now improper to enqueue any mds
1417 locks under och_sem, because enqueue might want to decide to
1418 cancel open locks for same inode we are holding och_sem for.
1422 Description: Client eviction while running blogbench
1423 Details : A lot of unlink operations with concurrent I/O can lead to a
1424 deadlock causing evictions. To address the problem, the number of
1425 oustanding OST_DESTROY requests is now throttled to
1426 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1427 callbacks are processed in priority.
1431 Description: enable ACLs on MDS by default
1432 Details : ACLs must be enabled on MDS by default.
1435 Frequency : PPC/PPC64 only
1437 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1438 Details : fix duplicated definitions between asm-ppc64/types.h and
1439 lustre_types.h on PPC.
1442 Frequency : PPC/PPC64 only
1444 Description: asm-ppc/segment.h does not exist
1445 Details : fix compile issue on PPC.
1449 Description: better handle error messages in extents code
1452 Frequency : RHEL4 only
1454 Description: mkfs is very slow on IA64/RHEL4
1455 Details : A performance regression has been discovered in the MPT Fusion
1456 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1457 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1458 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1460 Severity : enhancement
1462 Description: SNMP support enhancement
1463 Details : Adding total number of sampled request for an MDS node in snmp
1466 Severity : enhancement
1468 Description: Optimize ldlm waiting list processing for PR extent locks
1469 Details : When processing waiting list for read extent lock and meeting read
1470 lock that is same or wider to it that is not contended, skip
1471 processing rest of the list and immediatelly return current
1472 status of conflictness, since we are guaranteed there are no
1473 conflicting locks in the rest of the list.
1477 Description: Time out and refuse to reconnect
1478 Details : When the failover node is the primary node, it is possible
1479 to have two identical connections in imp_conn_list. We must
1480 compare not conn's pointers but NIDs, otherwise we can defeat
1481 connection throttling.
1485 Description: port llog fixes from b1_6 into HEAD
1486 Details : Port llog reference couting and some llog cleanups from b1_6
1487 (bug 10800) into HEAD, for protect from panic and access to already
1488 free llog structures.
1492 Description: Detect stride IO mode in read-ahead
1493 Details : When a client does stride read, read-ahead should detect that and
1494 read-ahead pages according to the detected stride pattern.
1498 Description: data checksumming impacts single node performance
1499 Details : add support for several checksum algorithm. Currently, only CRC32
1500 and Adler-32 are supported. The checksum type can be changed on
1501 the fly via /proc/fs/lustre/osc/*/checksum_type.
1505 Description: use adler32 for page checksums
1506 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1511 Description: build for x2 fails
1512 Details : fix compile issue on Cray systems.
1516 Description: Properly match for duplicate locks
1517 Details : Due to different lock order from skiplists code, we need to
1518 traverse entire list for now
1521 Frequency : only on PPC/SLES10
1523 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1524 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1525 be defined. Add a hack in mkfs_lustre.c to work around this compile
1530 Description: LASSERT on MDS when client holding flock lock dies
1531 Details : ldlm pool logic depends on number of granted locks equal to
1532 number of released locks which is not true for flock locks, so
1533 just exclude such locks from consideration.
1537 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1538 Details : Use fsfilt_sync() for both the whole filesystem sync and
1539 individual file sync to eliminate dangerous inode locking
1540 with I_LOCK that can lead to a deadlock.
1544 Description: performance in 1.6.3
1545 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1546 sure that 1MB requests are not fragmented by the block layer.
1548 Severity : enhancement
1550 Description: organize the server-side client stats on per-nid basis
1551 Details : Change the structure of stats under obdfilter and mds to
1559 The "uuid"s file would list the uuids of _active_ exports.
1560 And the clear entry is to clear all stats and stale nids.
1562 Severity : enhancement
1564 Description: eliminate client locks in face of contention
1565 Details : file contention detection and lockless i/o implementation
1566 for contended files.
1570 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1573 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1575 Description: MDS may lose file striping (and hence file data) in some cases
1576 Details : If there are additional extended attributes stored on the MDS,
1577 in particular ACLs, SELinux, or user attributes (if user_xattr
1578 is specified for the client mount options) then there is a risk
1579 of attribute loss. Additionally, the Lustre file striping
1580 needs to be larger than default (e.g. striped over all OSTs),
1581 and an additional attribute must be stored initially in the
1582 inode and then increase in size enough to be moved to the
1583 external attribute block (e.g. ACL growing in size) for file
1588 Description: skiplist implementation simplification
1589 Details : skiplists are used to group compatible locks on granted list
1590 that was implemented as tracking first and last lock of each
1591 lock group the patch changes that to using doubly linked lists
1595 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1596 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1597 because LASSERT on both the data supplied by a client, and the
1598 data on disk is dangerous and incorrect.
1600 Severity : enhancement
1602 Description: Slow truncate/writes to huge files at high offsets.
1603 Details : Directly associate cached pages to lock that protect those pages,
1604 this allows us to quickly find what pages to write and remove
1605 once lock callback is received.
1609 Description: more ldlm soft lockups
1610 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1611 starve other threads from the resource lock for a long time in
1612 case of long waiting queue, so change the debug level from
1613 D_OTHER to the less frequently used D_INFO.
1615 Severity : enhancement
1617 Description: add -gid, -group, -uid, -user options to lfs find
1621 Description: Hung threads in invalidate_inode_pages2_range
1622 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1623 once one is completed. As a result, some RPCs are stuck in the
1624 queue and are never sent.
1628 Description: filter threads hungs on waiting journal commit
1629 Details : Cleanup filter group llog code, then only filter group llog will
1630 be only created in the MDS/OST syncing process.
1634 Description: Procfs and llog threads access destoryed import sometimes.
1635 Details : Sync the import destoryed process with procfs and llog threads by
1636 the import refcount and semaphore.
1638 Severity : enhancement
1640 Description: openlock cache of b1_6 port to HEAD
1645 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1646 Details : If insertion of an extent fails, then discard the inode
1647 preallocation and free data blocks else it can lead to duplicate
1652 Description: don't always update ctime in ext3_xattr_set_handle()
1653 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1654 In some cases the ctime should not be updated, for example for
1655 2.0->1.8 compatibility it is necessary to delete an xattr and it
1656 should not update the ctime.
1660 Bugzilla : 15713/16362
1661 Description: Assertion in iopen_connect_dentry in 1.6.3
1662 Details : looking up an inode via iopen with the wrong generation number can
1663 populate the dcache with a disconneced dentry while the inode
1664 number is in the process of being reallocated. This causes an
1665 assertion failure in iopen since the inode's dentry list contains
1666 both a connected and disconnected dentry.
1670 Description: assertion failure in ldlm_handle2lock()
1671 Details : fix a race between class_handle_unhash() and class_handle2object()
1672 introduced in lustre 1.6.5 by bug 13622.
1677 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1678 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1683 Description: oops in sd_iostats_seq_show()
1684 Details : unloading/reloading the scsi low level driver triggers a kernel
1685 bug when trying to access the sd iostat file.
1690 Description: Kernel panics during QLogic driver reload
1691 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1692 patch, causing memory corruption.
1697 Description: journal_dev option does not work in b1_6
1698 Details : pass mount option during pre-mount.
1700 Severity : enhancement
1702 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1703 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1704 extent information of a file. It can be used to map logical blocks
1705 in a file to physical blocks in the block device.
1709 Description: LDLM soft lockups - improvement
1710 Details : It is be possible to send the lock handle along with each read
1711 or write request because the client is already doing a lock match
1712 itself so there isn't any reason the OST should have to re-do that
1716 Frequency : only on Cray X2
1718 Description: X2 build failures
1719 Details : fix build failures on Cray X2.
1723 Description: xid & resent requests
1724 Details : Initialize RPC XID from clock at startup (randomly if clock is
1727 Severity : enhancement
1729 Description: Add lustre_start utility to start or stop multiple Lustre servers
1734 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1735 Details : In case of memory pressure, list_del() can be called twice on
1736 req->rq_history_list, causing a kernel oops.
1740 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1741 Details : fix stack overflow in the distributed lock manager by defering
1742 export eviction after a failed AST to the elt thread instead of
1743 handling it in the dlm interpret routine.
1747 Description: Convert some comments to new format.
1748 Details : Update documenting comments to match doxygen conventions.
1752 Description: Grammar fixes.
1753 Details : A couple of trivial spelling fixes.
1757 Description: OSD_COUNTERS-mandatory
1758 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1759 to simplify the code.
1763 Description: simplify cmm_device freeing logic.
1764 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1768 Description: Add lockdep support to dt_object_operations locking interface.
1769 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
1770 parameter indicating lock ordering. Update mdd code to use new
1775 Description: Introduce failloc constants for lockless IO tests.
1776 Details : Add two new failloc constants to test lockless IO. Only one of
1777 them in implemented---another is checked in yet to be landed
1782 Description: Add lockdep support for inode mutex.
1783 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1784 in the situations where Lustre has to lock more than one inode
1789 Description: Add optional invariants checking support.
1790 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1791 configure switch. This macro is to be used for consistency and
1792 sanity checks that are too expensive to be left in `production'
1797 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1798 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1803 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1804 Details : Use spin_lock_nested() in (the only) situation where more than
1805 one ldlm_lock is locked simultaneously. Also, fix possible
1806 dead-lock in ldlm_lock_change_resource() by enforcing particular
1811 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1812 Details : Instead of specifying each ldlm_lock call-back through separate
1813 parameter, wrap them into struct ldlm_callback_suite.
1817 Description: Kill join_lru obd method and its callers.
1818 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
1819 in memory---a requirement implemented through ->o_join_lru() obd
1820 method in HEAD. Remove this method and its users.
1824 Description: Add asynchronous ldlm ENQUEUE completion handler.
1825 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
1826 case that stock ldlm_completion_ast() cannot handle as it waits
1827 until lock is granted. Introduce new ldlm_completion_ast_async()
1828 for this. Also comment ldlm_completion_ast().
1832 Description: ldlm_error <-> errno conversion.
1833 Details : Add functions to map (rather arbitrary) between LDLM error codes
1834 and standard errno values. CLIO needs this to prevent LDLM specific
1835 constants from escaping ldlm and osc.
1839 Description: Kill unused ldlm_handle2lock_ns() function.
1840 Details : Kill unused ldlm_handle2lock_ns() function.
1844 Description: Add lu_ref support to ldlm_lock
1845 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
1846 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
1847 and ldlm_resource. LDLM interface has to be changed, because code that
1848 releases a reference on a lock, has to "know" what reference this is.
1849 In the most frequent case
1851 lock = ldlm_handle2lock(handle);
1853 LDLM_LOCK_PUT(lock);
1855 no changes are required. When any other reference (received _not_ from
1856 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
1857 instead of LDLM_LOCK_PUT().
1859 Arguably, changes are pervasive, and interface requires some discipline
1860 for proper use. On the other hand, it was very instrumental in finding
1861 a few leaked lock references.
1865 Description: Add ldlm_lock_addref_try().
1866 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
1867 attempts to addref a lock that might be being canceled
1872 Description: Add ldlm_weigh_callback().
1873 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
1874 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
1875 hard-coded `number of pages' logic.
1879 Description: Add lockdep annotations to llog code.
1880 Details : Use appropriately tagged _nested() locking calls in the places
1881 where llog takes more than one ->lgh_lock lock.
1885 Description: Add loi_kms_set().
1886 Details : Wrap kms updates into a helper function.
1890 Description: Constify instances of struct lsm_operations.
1891 Details : Constify instances of struct lsm_operations.
1895 Description: lu_conf support.
1896 Details : On a server, a file system object is uniquely identified
1897 by a fid, which is sufficient to locate and load all object
1898 state (inode). On a client, on the other hand, more data are
1899 necessary instantiate an object. Change lu_object_find() and
1900 friends to take additional `lu_conf' argument describing object.
1901 Typically this includes layout information.
1905 Description: lu_context fixes.
1906 Details : Introduce new lu_context functions that are needed on the client
1907 side, where some system threads (ptlrpcd) are shared by multiple
1908 modules, and so cannot be stopped during module shutdown.
1912 Description: Add start and stop methods to lu_device_type_operations.
1913 Details : Introduce two new methods in lu_device_type_operations, that are
1914 invoked when first instance of a given type is created and last one
1915 is destroyed respectively. This is need by CLIO.
1919 Description: Add lu_ref support to struct lu_device.
1920 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
1921 track leaked references.
1925 Description: Introduce lu_kmem_descr.
1926 Details : lu_kmem_descr and its companion interface allow to create
1927 and destroy a number of kmem caches at once.
1931 Description: Fix lu_object finalization race.
1932 Details : Fix a race between lu_object_find() finding an object and its
1933 concurrent finalization. This race is (most likely) not possible
1934 on the server, but might happen on the client.
1938 Description: Introduce lu_ref interface.
1939 Details : lu_ref is a debugging module allowing to track references to
1940 a given object. It is quite cpu expensive, and has to be
1941 explicitly enabled with --enable-lu_ref. See usage description
1946 Description: Factor lu_site procfs stats into a separate function.
1947 Details : Separate lu_site stats printing code into a separate function
1948 that can be reused on a client.
1952 Description: Constify instances of struct {lu,dt,md}_device_operations.
1953 Details : Constify instances of struct {lu,dt,md}_device_operations.
1957 Description: Introduce struct md_site and move meta-data specific parts of
1958 struct lu_site here.
1959 Details : Move md-specific fields out of struct lu_site into special struct
1960 md_site, so that lu_site can be used on a client.
1964 Description: Kill mdd_lov_destroy().
1965 Details : Remove unused mdd code.
1969 Description: Add st_block checking to multistat.c.
1970 Details : Add st_block checking to multistat.c.
1974 Description: Add lu_ref support to struct obd_device.
1975 Details : Add lu_ref tracking to obd_device.
1979 Description: Kill obd_set_fail_loc().
1980 Details : Remove unused code.
1984 Description: Add special type for ptlrpc_request interpret functions.
1985 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
1986 there. Actual usage will be in CLIO.
1990 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
1991 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
1992 doesn't work with lockdep.
1996 Description: Add rwv.c test program.
1997 Details : New testing program exercising readv(2) and writev(2) (Qian).
2001 Description: Add sendfile.c test program.
2002 Details : New testing program exercising sendfile(2) (Jay).
2006 Description: Ratelimit a message that can be very frequent.
2007 Details : Ratelimit a memory allocation failure message that can
2012 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2013 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2017 Description: Liblustre build fixes.
2018 Details : Liblustre build fixes.
2022 Description: libcfs: add cfs_{need,cond}_resched() interface.
2023 Details : libcfs: add cfs_{need,cond}_resched() definition and
2024 implementations for Linux, NT, and liblustre.
2026 Severity : enhancement
2028 Description: More exported tunables for mballoc
2029 Details : Add support for tunable preallocation window and new tunables for
2030 large/small requests
2034 Description: Detect corruption of block bitmap and checking for preallocations
2035 Details : Checks validity of on-disk block bitmap. Also it does better
2036 checking of number of applied preallocations. When corruption is
2037 found, it turns filesystem readonly to prevent further corruptions.
2041 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2042 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2043 window to 0. If it does hit cache miss, and read-pattern is still
2044 stride-io mode, does not reset the stride window, but also does not
2045 increase the stride window length in this case.
2049 Frequency : only for big-endian servers
2050 Description: Check if system is big-endian while mounting fs with extents feature
2051 Details : Mounting a filesystem with extents feature will fail on big-endian
2052 systems since ext3-based ldiskfs is not supported on big-endian
2053 systems. This can be over-riden with "bigendian_extents" mount option.
2055 Severity : enhancement
2057 Description: The root squash functionality
2058 Details : A security feature, which is to prevent users from being able
2059 to mount lustre on their desktop, run as root, and delete
2060 all of the files in the filesystem. The goal is accomplished by
2061 remapping user id (UID) and group id (GID) of the root user to
2062 a UID and GID specified by the system administartor via Lustre
2063 configuration management server (MGS). The functionality also
2064 allows to specify sets of clients for which the remapping does
2069 Description: Excessive recovery window
2070 Details : With AT enabled, the recovery window can be excessively long (6000+
2071 seconds). To address this problem, we no longer use
2072 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2073 timeout no longer depends on the service time, it is set to
2074 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2075 time via pb_service_time.
2079 Description: Watchdog triggered on MDS failover
2080 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2081 the OSTs know that the MDS "UUID" can be reused for the same export
2082 from a different NID, so we do not need to wait for the export to be
2086 Frequency : rare, only if using MMP with Linux RAID
2088 Description: MMP doesn't work with Linux RAID
2089 Details : While using HA for Lustre servers with Linux RAID, it is possible
2090 that MMP will not detect multiple mounts. To make this work we
2091 need to unplug the device queue in RAID when the MMP block is being
2092 written. Also while reading the MMP block, we should read it from
2093 disk and not the cached one.
2095 Severity : enhancement
2097 Description: open file using fid
2098 Details : A file can be opened using just its fid, like
2099 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2102 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2105 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2106 Details : add proper locking for creation of jbd_4k slab cache
2110 Description: MMP check in ext3_remount() fails without displaying any error
2111 Details : When multiple mount protection fails during remount, proper error
2114 Severity : enhancement
2116 Description: Allow stripe size to be up to 4G-64k
2117 Details : Fix math logic to allow large stripe sizes.
2121 Description: add check for >8TB ldiskfs filesystems
2122 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2123 Don't allow >8TB ldiskfs filesystems to be mounted without
2124 force_over_8tb mount option
2126 --------------------------------------------------------------------------------
2128 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2130 * Support for kernels:
2131 2.6.5-7.283 (SLES 9),
2132 2.6.9-55.EL (RHEL 4),
2133 2.6.16.46-0.14 (SLES 10),
2134 2.6.18.8 vanilla (kernel.org)
2135 * Client support for unpatched kernels:
2136 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2137 2.6.16 - 2.6.22 vanilla (kernel.org)
2138 * Due to recently discovered recovery problems, we do not recommend
2139 using patchless RHEL 4 clients with this or any earlier release.
2140 * Recommended e2fsprogs version: 1.39.cfs8
2141 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2142 * Starting with this release, the ldiskfs backing filesystem required
2143 by Lustre is now in its own package, lustre-ldiskfs. This package
2144 should be installed. It is versioned separately from Lustre and
2145 may be released separately in future.
2149 Description: block reactivating mgc import until all deactivates complete
2150 Details : Fix race when failing back MDT/MGS to itself (testing)
2153 Frequency : at statup only
2155 Description: mds_lov_synchronize race leads to various problems
2156 Details : simultaneous MDT->OST connections at startup can cause the
2157 sync to abort, leaving the OSC in a bad state.
2159 Severity : enhancement
2161 Description: add optional extra BUILD_VERSION info
2162 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2163 to override the lustre version.
2166 Frequency : 2.6.18 servers only
2168 Description: ll_kern_mount() doesn't release the module reference
2169 Details : The ldiskfs module reference count never drops down to 0
2170 because ll_kern_mount() doesn't release the module reference.
2175 Description: server LBUG when using old ost_num_threads parameter
2176 Details : Accept the old ost_num_threads parameter but warn that it
2177 is deprecated, and fix an off-by-one error that caused an LBUG.
2182 Description: Transient SCSI error results in persistent IO issue
2183 Details : iobuf->dr_error is not reinitialized to 0 between two
2187 Frequency : sometimes when underlying device returns I/O errors
2189 Description: OSTs not going read-only during write failures
2190 Details : OSTs are not remounted read-only when the journal commit threads
2191 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2192 instead of the ext3 wrappers.
2196 Description: poor connect scaling with increasing client count
2197 Details : Don't run filter_grant_sanity_check for more than 100 exports
2198 to improve scaling for large numbers of clients.
2201 Frequency : SLES10 only
2203 Description: sanity-quota.sh quotacheck failed: rc = -22
2204 Details : Quotas cannot be enabled on SLES10.
2207 Frequency : liblustre clients only
2209 Description: getdirentries does not give error when run on compute nodes
2210 Details : getdirentries does not fail when the size specified as an argument
2211 is too small to contain at least one entry
2213 Severity : enhancement
2215 Description: Add LNET router traceability for debug purposes
2216 Details : If a checksum failure occurs with a router as part of the
2217 IO path, the NID of the last router that forwarded the bulk data
2218 is printed so it can be identified.
2223 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2224 Details : Due to a race condition, liblustre clients were occasionally
2225 evicted incorrectly.
2227 Severity : enhancement
2229 Description: lfs setstripe use optional parameters instead of postional
2232 Severity : enhancement
2234 Description: Nanosecond timestamp support for ldiskfs
2235 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2236 resolution timestamps. There is not yet support for this at
2237 the Lustre filesystem level.
2240 Frequency : during server recovery
2242 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2243 Details : request with rq_no_resend flag not awake l_wait_event if they get
2247 Frequency : nfs export on patchless client
2249 Description: connectathon hang when test nfs export over patchless client
2250 Details : Disconnected dentry cannot be found with lookup, so we do not need
2251 to unhash it or make it invalid
2254 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2255 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2256 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2261 Description: open req refcounting wrong on reconnect
2262 Details : If reconnect happened between getting open reply from server and
2263 call to mdc_set_replay_data in ll_file_open, we will schedule
2264 replay for unreferenced request that we are about to free.
2265 Subsequent close will crash in variety of ways.
2266 Check that request is still eligible for replay in
2267 mdc_set_replay_data().
2272 Description: disable writes to filesystem when reading health_check file
2273 Details : the default for reading the health_check proc file has changed
2274 to NOT do a journal transaction and write to disk, because this
2275 can cause reads of the /proc file to hang and block HA state
2276 checking on a healthy but otherwise heavily loaded system. It
2277 is possible to return to the previous behaviour during configure
2278 with --enable-health-write.
2280 Severity : enhancement
2282 Description: 64-bit inode version
2283 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2284 to the inode. This will be required for version-based recovery.
2289 Description: MDS fails to start if a duplicate client export is detected
2290 Details : in some rare cases it was possible for a client to connect to
2291 an MDS multiple times. Upon recovery the MDS would detect this
2292 and fail during startup. Handle this more gracefully.
2294 Severity : enhancement
2296 Description: Add -o localflock option to simulate old noflock
2298 Details : This will achieve local-only flock/fcntl locks
2304 Description: log_commit_thread vs filter_destroy race leads to crash
2305 Details : Take import reference before releasing llog record semaphore
2310 Description: Wrong request locking in request set processing
2311 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2312 imp_delayed_list, in this place should be used imp_lock.
2315 Frequency : when reconnection
2317 Description: Grant Leak when osc reconnect to OST
2318 Details : When osc reconnect ost, OST(filter) should check whether it
2319 should grant more space to client by comparing fed_grant and
2320 cl_avail_grant, and return the granted space to client instead
2321 of "new granted" space, because client will call osc_init_grant
2322 to update the client grant space info.
2325 Frequency : when client reconnect to OST
2327 Description: Grant Leak when osc do resend and replay bulk write
2328 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2329 bulk write request, because the grant info while be sync between
2330 OSC and OST when reconnect, and we should ignore the grant info
2331 these of resend/replay write req.
2336 Description: Grant space more than avaiable left space sometimes.
2337 Details : When then OST is about to be full, if two bulk writing from
2338 different clients came to OST. Accord the avaliable space of the
2339 OST, the first req should be permitted, and the second one
2340 should be denied by ENOSPC. But if the seconde arrived before
2341 the first one is commited. The OST might wrongly permit second
2342 writing, which will cause grant space > avaiable space.
2345 Frequency : when client is evicted
2347 Description: Grant might be wrongly erased when osc is evicted by OST
2348 Details : when the import is evicted by server, it will fork another
2349 thread ptlrpc_invalidate_import_thread to invalidate the
2350 import, where the grant will be set to 0. While the original
2351 thread will update the grant it got when connecting. So if
2352 the former happened latter, the grant will be wrongly errased
2353 because of this race.
2358 Description: Checking Stale with correct fid
2359 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2360 to check whether it is stale, because sometimes, we want the
2361 enqueue happened anyway, and op_data.fid2 will not be initialized.
2363 Severity : enhancement
2365 Description: update patchless client
2366 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2369 Frequency : only with 2.4 kernel
2371 Description: random memory corruption
2372 Details : size of struct ll_inode_info is to big for union inode.u and this
2373 can be cause of random memory corruption.
2378 Description: Memory leak in recovery
2379 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2380 It should also check obd_fail before fsfilt_start, otherwise if
2381 fsfilt_start return -EROFS,(failover mds during mds recovery).
2382 then the req will return with repmsg->transno = 0 and rc = EROFS.
2383 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2384 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2385 be freed no matter whether fsfilt_commit success or not.
2388 Frequency : only with huge count clients
2390 Description: Prevents from taking the superblock lock in llap_from_page for
2392 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2393 for llap_from_page call in ll_removepage prevents from taking the
2394 superblock lock for a soon died page.
2399 Description: Not check open intent error before release open handle
2400 Details : in some rare cases, the open intent error is not checked before
2401 release open handle, which may cause
2402 ASSERTION(open_req->rq_transno != 0), because it tries to release
2403 the failed open handle.
2408 Description: Set cat log bitmap only after create log success.
2409 Details : in some rare cases, the cat log bitmap is set too early. and it
2410 should be set only after create log success.
2414 Description: Accessing a block bevice can re-enable I/O when Lustre is
2415 tearing down a device.
2416 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2421 Description: service threads may hog cpus when there are a lot of requests
2422 Details : Insert cond_resched to give other threads a chance to use some CPU
2427 Description: the cat log was not initialized in recovery
2428 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2429 unlink log on mds will not be initialized until mds post
2430 recovery. And also in mds post recovery, the unlink log will
2431 initialization will be done asynchronausly, so there will be race
2432 between add unlink log and unlink log initialization.
2436 Description: brw_stats were being printed incorrectly
2437 Details : brw_stats were being printed as log2 but all of them were not
2438 recorded as log2. Also remove some code duplication arising from
2439 filter_tally_{read,write}.
2443 Frequency : rare, only in recovery.
2444 Description: ASSERTION(req->rq_type != LI_POISON) failed
2445 Details : imp_lock should be held while iterating over imp_sending_list for
2446 prevent destroy request after get timeout in ptlrpc_queue_wait.
2450 Description: replay-single.sh test 52 fails
2451 Details : A lock's skiplist need to be cleanup when it being unlinked
2452 from its resource list.
2456 Description: Short directio read returns full requested size rather than
2458 Details : Direct I/O operations should return actual amount of bytes
2459 transferred rather than requested size.
2461 Severity : enhancement
2463 Description: metadata RPC reduction (e.g. for rm performance)
2464 Details : decrease the amount of synchronous RPC between clients and servers
2465 by canceling conflicing lock before the operation on the client
2466 and packing thier handles into the main operation RPC to server.
2468 Severity : enhancement
2470 Description: Async OSC create to avoid the blocking unnecessarily.
2471 Details : If a OST has no remain object, system will block on the creating
2472 when need to create a new object on this OST. Now, ways use
2473 pre-created objects when available, instead of blocking on an
2474 empty osc while others are not empty. If we must block, we block
2475 for the shortest possible period of time.
2479 Description: improve handling recoverable errors
2480 Details : if request processig with error which can be recoverable on server
2481 request should be resend, otherwise page released from cache and
2484 Severity : enhancement
2486 Description: refine locking for avoid write wrong info into lov_objid file
2487 Details : fix possible races with add new target and write/update data in
2490 --------------------------------------------------------------------------------
2492 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2497 Frequency : on some architectures
2499 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2500 Details : Uninitialized flags sometimes cause configuration commands to
2504 Frequency : patchless clients only
2506 Description: missing __iget() symbol export
2507 Details : The __iget() symbol export is missing. To avoid the need for
2508 this on patchless clients the deathrow inode reaper is turned
2509 off, and we depend on the VM to clean up old inodes. This
2510 dependency was during via the fix for bug 12181.
2512 --------------------------------------------------------------------------------
2514 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2516 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2517 INTEROPERATE with older versions automatically. In many cases a
2518 special upgrade step is needed. Please read the
2519 user documentation before upgrading any part of a live system.
2520 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2521 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2522 * WARNING: Lustre configuration and startup changes are required with
2523 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2527 Severity : enhancement
2529 Description: MountConf
2530 Details : Lustre configuration is now managed via mkfs and mount
2531 commands instead of lmc and lconf. New obd types (MGS, MGC)
2532 are added for dynamic configuration management. See
2533 https://mail.clusterfs.com/wikis/lustre/MountConf for
2536 Severity : enhancement
2538 Description: dynamic OST addition
2539 Details : OSTs can now be added to a live filesystem
2541 Severity : enhancement
2543 Description: startup order invariance
2544 Details : MDTs and OSTs can be started in any order. Clients only
2545 require the MDT to complete startup.
2547 Severity : enhancement
2549 Description: parallel, asynchronous orphan cleanup
2550 Details : orphan cleanup is now performed in separate threads for each
2551 OST, allowing parallel non-blocking operation.
2553 Severity : enhancement
2555 Description: optimized stripe assignment
2556 Details : stripe assignments are now made based on ost space available,
2557 ost previous usage, and OSS previous usage, in order to try
2558 to optimize storage space and networking resources.
2560 Severity : enhancement
2562 Description: Permanently set tunables
2563 Details : All writable /proc/fs/lustre tunables can now be permanently
2564 set on a per-server basis, at mkfs time or on a live system.
2566 Severity : enhancement
2568 Description: Lustre message v2
2569 Details : Add lustre message format v2.
2571 Severity : enhancement
2573 Description: client OST exclusion list
2574 Details : Clients can be started with a list of OSTs that should be
2575 declared "inactive" for known non-responsive OSTs.
2579 Description: SPEC SFS validation failure on NFS v2 over lustre.
2580 Details : Changes the blocksize for regular files to be 2x RPC size,
2581 and not depend on stripe size.
2583 Severity : enhancement
2585 Description: Multiple MD RPCs in flight.
2586 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2587 To avoid overly-overloading MDS, introduce a limit on number of
2588 MDS RPCs in flight for a single client and add /proc controls
2589 to adjust this limit.
2591 Severity : enhancement
2593 Description: client read/write statistics
2594 Details : Add client read/write call usage stats for performance
2595 analysis of user processes.
2596 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2597 file access. extents_stats shows chunk size distribution.
2598 extents_stats_per_process show chunk size distribution per
2601 Severity : enhancement
2603 Description: mds statistics
2604 Details : Add detailed mds operations statistics in
2605 /proc/fs/lustre/mds/*/stats.
2609 Description: Failure of copying files with lustre special EAs.
2610 Details : Client side always return success for setxattr call for lustre
2611 special xattr (currently only "trusted.lov").
2616 Description: Refcount LNET uuids
2617 Details : The global LNET uuid list grew linearly with every startup;
2618 refcount repeated list entries instead of always adding to
2621 Severity : enhancement
2623 Description: Dynamic service threads
2624 Details : Within a small range, start extra service threads
2625 automatically when the request queue builds up.
2628 Frequency : mixed-endian client/server environments
2630 Description: mixed-endian crashes
2631 Details : The new msg_v2 system had some failures in mixed-endian
2634 Severity : enhancement
2636 Description: Easy OST removal
2637 Details : OSTs can be permanently deactivated with e.g. 'lctl
2638 conf_param lustre-OST0001.osc.active=0'
2640 Severity : enhancement
2642 Description: MGS proc entries
2643 Details : Added basic proc entries for the MGS showing what filesystems
2646 Severity : enhancement
2648 Description: provide MGS failover
2649 Details : Added config lock reacquisition after MGS server failover.
2651 Severity : enhancement
2653 Description: add Linux 2.4 support
2654 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2658 Description: a large application tries to do I/O to the same resource and dies
2659 in the middle of it.
2660 Details : Check the req->rq_arrival time after the call to
2661 ost_brw_lock_get(), but before we do anything about
2662 processing it & sending the BULK transfer request. This
2663 should help move old stale pending locks off the queue as
2664 quickly as obd_timeout.
2667 Frequency : when an incorrect nid is specified during startup
2669 Description: ptlrpc connect to non-existant node causes kernel crash
2670 Details : LNET can't be re-entered from an event callback, which
2671 happened when we expire a message after the export has been
2672 cleaned up. Instead, hand the zombie cleanup off to another
2675 Severity : enhancement
2677 Description: plain/inodebits lock performance improvement
2678 Details : Grouping plain/inodebits in granted list by their request modes
2679 and bits policy, thus improving the performance of search through
2683 Frequency : only if OST filesystem is corrupted
2685 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2686 Details : for a short time RPCs with bulk IO are in the replay list,
2687 but replay of bulk IOs is unimplemented. If the OST filesystem
2688 is corrupted due to disk cache incoherency and then replay is
2689 started it is possible to trip an assertion. Avoid putting
2690 committed RPCs into the replay list at all to avoid this issue.
2693 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2695 Description: System hang on startup
2696 Details : This bug allowed the liblustre (e.g. catamount) client to
2697 return to the app before handling all startup RPCs. This
2698 could leave the node unresponsive to lustre network traffic
2699 and manifested as a server ptllnd timeout.
2701 Severity : enhancement
2703 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2704 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2705 boolean to control whether to print peer debug info when a
2706 client's RPC times out.
2709 Frequency : only for kernels with patches from Lustre below 1.4.3
2711 Description: Remove old rdonly API
2712 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2715 Frequency : only for devices with external journals
2717 Description: Set external device read-only also
2718 Details : During a commanded failover stop, we set the disk device
2719 read-only while the server shuts down. We now also set any
2720 external journal device read-only at the same time.
2723 Frequency : when upgrading from 1.4 while trying to change parameters
2725 Description: The wrong (new) MDC name was used when setting parameters for
2726 upgraded MDT's. Also allows changing of OSC (and MDC)
2727 parameters if --writeconf is specified at tunefs upgrade time.
2730 Frequency : when setting specific ost indicies
2732 Description: QOS code breaks on skipped indicies
2733 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2734 created with --index need not be sequential.
2736 Severity : enhancement
2738 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2739 Details : The uninit_groups feature works in conjunction with the kernel
2740 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2741 up the pass1 processing of e2fsck. This is a read-only feature
2742 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2743 mount filesystems that have had this feature enabled.
2745 Severity : enhancement
2747 Description: Improve multi-block allocation algorithm to avoid fragmentation
2748 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2749 allocation locality and avoid filesystem fragmentation.
2751 ------------------------------------------------------------------------------
2753 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2755 * Support for kernels:
2756 2.4.21-47.0.1.EL (RHEL 3)
2757 2.6.5-7.283 (SLES 9)
2758 2.6.9-42.0.10.EL (RHEL 4)
2759 2.6.12.6 vanilla (kernel.org)
2760 2.6.16.27-0.9 (SLES 10)
2761 * Recommended e2fsprogs version: 1.39.cfs5
2763 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2767 Frequency : occasional, depends on client load and configuration
2768 Bugzilla : 12181, 12203
2769 Description: data loss for recently-modified files
2771 Details : In some cases it is possible that recently written or created
2772 files may not be written to disk in a timely manner (this should
2773 normally be within 30s unless client IO load is very high).
2774 The problem appears as zero-length files or files that are a
2775 multiple of 1MB in size after a client crash or client eviction
2776 that are missing data at the end of the file.
2778 This problem is more likely to be hit on clients where files are
2779 repeatedly created and unlinked in the same directory, clients
2780 have a large amount of RAM, have many CPUs, the filesystem has
2781 many OSTs, the clients are rebooted frequently, and/or the files
2782 are not accessed by other nodes after being written.
2784 The presence of the problem can be detected by looking at
2785 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2786 smaller than the second (nr_unused) then dirty files will not
2787 be flushed automatically to disk. "sync; sleep 10" should be
2788 run several times on the node before unmounting it to update
2789 Lustre (this is also safe to run on nodes without this problem).
2791 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2792 that can cause this same problem, so customers using that kernel
2793 also need to update the kernel in addition to Lustre. In order
2794 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2796 It is normal that files written just before a client crash (less
2797 than 30s) may not yet have been flushed to disk, even for local
2801 Frequency : frequent on thin XT3 nodes
2803 Description: UUID collision on thin XT3 Linux nodes
2804 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2805 randomly, since we relied on an insufficiently-seeded PRNG.
2810 Description: OSS hangs after "All ost request buffers busy"
2811 Details : A deadlock between quota and journal operations caused OSS
2812 hangs after printing "All ost request buffers busy."
2815 Frequency : always on liblustre builds
2817 Description: Cleanup compiler warnings on liblustre
2820 Frequency : always on liblustre builds on XT3
2822 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2826 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2827 > 32000 subdirectories
2828 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2829 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2830 subdirectory count crosses 32000. This will aid e2fsck to
2831 correctly handle more than 32000 subdirectories.
2834 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2836 Description: System hang on startup
2837 Details : This bug allowed the liblustre (e.g. catamount) client to
2838 return to the app before handling all startup RPCs. This
2839 could leave the node unresponsive to lustre network traffic
2840 and manifested as a server ptllnd timeout.
2842 Severity : enhancement
2844 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2845 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2846 boolean to control whether to print peer debug info when a
2847 client's RPC times out.
2852 Description: make O_SYNC working on 2.6 kernels
2853 Details : 2.6 kernels use different method for mark pages for write,
2854 so need add a code to lustre for O_SYNC work.
2859 Description: Failure to close file and release space on NFS
2860 Details : Put inode details into lock acquired in ll_intent_file_open.
2861 Use mdc_intent_lock in ll_intent_open to properly
2862 detect all kind of errors unhandled by mdc_enqueue.
2867 Description: proc file read during shutdown sometimes raced obd removal,
2869 Details : Add lock to prevent obd access after proc file removal.
2872 Frequency : Only for files larger than 4GB on 32-bit clients.
2874 Description: improperly doing page alignment of locks
2875 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2876 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2881 Description: RPCs being resent when they shouldn't be
2882 Details : Some RPCs that should not be resent are being resent. This
2883 can cause inconsistencies in the RPC state machine. Do not
2884 resend such requests.
2887 Frequency : rare, only with NFS export
2889 Description: Crash on NFS re-export node
2890 Details : under very unusual load conditions an assertion is hit in
2891 ll_intent_file_open()
2894 Frequency : only if OST filesystem is corrupted
2896 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2897 Details : for a short time RPCs with bulk IO are in the replay list,
2898 but replay of bulk IOs is unimplemented. If the OST filesystem
2899 is corrupted due to disk cache incoherency and then replay is
2900 started it is possible to trip an assertion. Avoid putting
2901 committed RPCs into the replay list at all to avoid this issue.
2906 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2907 Details : Large single O_DIRECT read and write calls can fail to allocate
2908 a sufficiently large buffer to process the request. In case of
2909 allocation failure the allocation is retried with a smaller
2910 buffer and broken into smaller requests.
2912 Severity : enhancement
2914 Description: Add -o localflock option to simulate old noflock behaviour.
2915 Details : This will achieve local-only flock/fcntl locks coherentness.
2920 Description: versioning check is incomplete
2921 Details : Checking the version difference of client vs. server, report
2922 error if the gap is too big.
2927 Description: add support PG_writeback bit
2928 Details : add support for PG_writeback bit for Lustre, for more carefull
2929 work with page cache in 2.6 kernel. This also fix some deadlocks
2930 and remove hack for work O_SYNC with 2.6 kernel.
2932 Severity : enhancement
2934 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2935 Details : The uninit_groups feature works in conjunction with the kernel
2936 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2937 up the pass1 processing of e2fsck. This is a read-only feature
2938 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2939 mount filesystems that have had this feature enabled.
2941 Severity : enhancement
2943 Description: Improve multi-block allocation algorithm to avoid fragmentation
2944 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2945 allocation locality and avoid filesystem fragmentation.
2947 ------------------------------------------------------------------------------
2949 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2951 * Support for kernels:
2952 2.6.9-42.0.3.EL (RHEL 4)
2953 2.6.5-7.276 (SLES 9)
2954 2.4.21-47.0.1.EL (RHEL 3)
2955 2.6.12.6 vanilla (kernel.org)
2956 2.6.16.21-0.8 (SLES10)
2957 * Recommended e2fsprogs version: 1.39.cfs2-0
2959 * The backwards-compatible /proc/sys/portals symlink has been removed
2960 in this release. Before upgrading, please ensure that you change
2961 any configuration scripts or /etc/sysctl.conf files that access
2962 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2963 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2964 in advance of the upgrade on any system running Lustre 1.4.6 or
2965 newer, since /proc/sys/lnet was added in that version.
2966 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2970 Frequency : only when quota is used
2972 Description: avoid scanning export list for quota master
2973 Details : Change the algorithms to avoid scanning export list in order
2974 to improve the efficiency.
2977 Frequency : MDS failover only, very rarely
2979 Description: "went back in time" messages on mds failover
2980 Details : The greatest transno may be lost when the current operation
2981 finishes with an error (transno==0) and the client's last_rcvd
2982 record is over-written. Save the greatest transno in the
2983 mds_last_transno for this case.
2986 Frequency : always for specific kernels and striping counts
2988 Description: client may get "Matching packet too big" without ACL support
2989 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2990 when trying to access files in certain configurations. The
2991 clients should in fact be denied when mounting because they do
2992 not understand ACLs.
2995 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2997 Description: many clients connecting with IO in progress causes connect timeouts
2998 Details : Avoid synchronous journal commits to avoid delays caused by many
2999 clients connecting/disconnecting when bulk IO is in progress.
3000 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3001 OST_IO_PORTAL to avoid delays behind potentially many pending
3005 Frequency : occasionally with multiple writers to a single file
3007 Description: shared writes to file may result in wrong size reported by stat()
3008 Details : Allow growing of kms when extent lock is cancelled
3011 Frequency : always with random mmap IO to multi-striped file
3013 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3014 Details : Only if the hole is at the end of OST object so that kms is too
3015 small. Fix is to increase kms accordingly in ll_nopage.
3018 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3020 Description: writes to a missing object would leak memory on the OST
3021 Details : If there is an inconsistency between the MDS and OST filesystems,
3022 such that the MDS references an object that doesn't exist, writes
3023 to that object will leak memory due to incorrect cleanup in the
3024 error handling path, eventually running out of memory on the OST.
3029 Description: Creating too long symlink causes lustre errors
3030 Details : Check symlink and name lengths before sending requests to MDS.
3033 Frequency : only if flock is enabled (not on by default)
3035 Description: posix locks not released on fd closure on 2.6.9+
3036 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3037 kernels, this caused such locks not to be released on fd close and
3038 then assertions on fs unmount about still used locks.
3041 Frequency : MDS failover only, very rarely
3043 Description: clients may get ASSERTION(granted_lock != NULL)
3044 Details : When request was taking a long time, and a client was resending
3045 a getattr by name lock request. The were multiple lock requests
3046 with the same client lock handle and
3047 mds_getattr_name->fixup_handle_for_resent_request found one of the
3048 lock handles but later failed with ASSERTION(granted_lock != NULL).
3053 Description: handle->h_buffer_credits > 0, assertion failure
3054 Details : h_buffer_credits is zero after truncate, causing assertion
3055 failure. This patch extends the transaction or creates a new
3059 Frequency : NFS re-export or patchless client
3060 Bugzilla : 11179, 10796
3061 Description: Crash on NFS re-export node (__d_move)
3062 Details : We do not want to hash the dentry if we don't have a lock.
3063 But if this dentry is later used in d_move, we'd hit uninitialised
3064 list head d_hash, so we just do this to init d_hash field but
3065 leave dentry unhashed.
3068 Frequency : NFS re-export or patchless client
3070 Description: NFS exports has problem with symbolic link
3071 Details : lustre client didn't properly install dentry when re-exported
3072 to NFS or running patchless client.
3075 Frequency : NFS re-export or patchless client
3077 Description: Various nfs/patchless fixes.
3078 Details : fixes reuse disconected alias for lookup process - this fixes
3079 warning "find_exported_dentry: npd != pd",
3080 fix permission error with open files at nfs.
3081 fix apply umask when do revalidate.
3084 Frequency : occasional
3086 Description: Crash on NFS re-export node
3087 Details : calling clear_page() on the wrong pointer triggered oops in
3088 generic_mapping_read().
3091 Frequency : rarely, using O_DIRECT IO
3093 Description: unaligned directio crashes client with LASSERT
3094 Details : check for unaligned buffers before trying any requests.
3097 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3099 Description: stale data returned from RAID cache
3100 Details : If only a small amount of IO is done to the RAID device before
3101 reading it again it is possible to get stale data from the RAID
3102 cache instead of reading it from disk.
3105 Frequency : always for sles10 kernel
3107 Description: sles10 support
3108 Details : ll_follow_link: compile fixes and using of nd_set_link
3109 under newer kernels.
3112 Frequency : depends on arch, kernel and compiler version, always on sles10
3115 Description: recursive or deep enough symlinks cause stack overflow
3116 Details : getting rid of large stack-allocated variable in
3120 Frequency : depends on hardware
3122 Description: lustre write performance loss in the SLES10 kernel
3123 Details : the performance loss is caused by using of write barriers in the
3124 ext3 code. The SLES10 kernel turns barrier support on by
3125 default. The fix is to undo that change for ldiskfs.
3127 ------------------------------------------------------------------------------
3129 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3131 * Support for kernels:
3132 2.6.9-42.0.3EL (RHEL 4)
3133 2.6.5-7.276 (SLES 9)
3134 2.4.21-47.0.1.EL (RHEL 3)
3135 2.6.12.6 vanilla (kernel.org)
3139 Frequency : quota enabled and large files being deleted
3141 Description: releasing more than 4GB of quota at once hangs OST
3142 Details : If a user deletes more than 4GB of files on a single OST it
3143 will cause the OST to spin in an infinite loop. Release
3144 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3149 Description: statfs data retrieved from /proc may be stale or zero
3150 Details : When reading per-device statfs data from /proc, in the
3151 {kbytes,files}_{total,free,avail} files, it may appear
3152 as zero or be out of date.
3155 Frequency : systems with MD RAID1 external journal devices
3157 Description: lconf's call to blkid is confused by RAID1 journal devices
3158 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3159 of returning all block devices that match the journal UUID.
3162 Frequency : always, for aggregate stripe size over 4GB
3164 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3165 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3166 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3169 Frequency : always if "lfs find" used on a local file/directory
3171 Description: "lfs find" segfaults if used on a local file/directory
3172 Details : The case where a directory component was not specified wasn't
3173 handled correctly. Handle this properly.
3176 Frequency : always on ppc64
3178 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3179 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3180 perform endianness conversion.
3183 Frequency : rarely (truncate to non-zero file size after write under load)
3184 Bugzilla : 10730, 10687
3185 Description: Files padded with zeros to next 4K multiple
3186 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3187 it is possible that files that are truncated to a non-zero size
3188 immediately after being written are filled with zero bytes beyond
3189 the truncated size. No file data is lost.
3191 Severity : enhancement
3193 Description: Allow recovery/failover for liblustre clients.
3194 Details : liblustre clients were unaware of failover configurations until
3197 Severity : enhancement
3199 Description: user file locks should fail when not mounting with flock option
3200 Details : Set up an error-returning stub in ll_file_operations.lock field
3201 to prevent incorrect behaviour when client is mounted without
3202 flock option. Also, set up properly f_op->flock field for
3206 Frequency : always on ia64
3208 Description: "lfs df" loops on printing out MDS statfs information
3209 Details : The obd_ioctl_data was not initialized and in some systems
3210 this caused a failure during the ioctl that did not return
3211 an error. Initialize the struct and return an error on failure.
3214 Frequency : SLES 9 only
3216 Description: Error of copying files with lustre special EAs as root
3217 Details : Client side always return success for setxattr call for lustre
3218 special xattr (currently only "trusted.lov").
3221 Frequency : rarely on clusters with both ia64+i386 clients
3223 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3224 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3225 clients are doing concurrent writes to the same file it is
3226 possible that the ia64 clients may LASSERT because the OST
3227 extent locks are not PAGE_SIZE aligned. Ensure that grown
3228 locks are always aligned on the request boundary.
3231 Frequency : specific use, occasional
3233 Description: Overwriting in use executable truncates on-disk binary image
3234 Details : If one node attempts to overwrite an executable in use by
3235 another node, we now correctly return ETXTBSY instead of
3236 truncating the file.
3238 Severity : enhancement
3240 Description: Async OSC create to avoid the blocking unnecessarily.
3241 Details : If a OST has no remain object, system will block on the creating
3242 when need to create a new object on this OST. Now, ways use
3243 pre-created objects when available, instead of blocking on an
3244 empty osc while others are not empty. If we must block, we block
3245 for the shortest possible period of time.
3250 Description: chmod on Lustre root is propagated to other clients
3251 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3252 invalid by the follow_mount time.
3255 Frequency : liblustre clients only
3257 Description: Race in 'instant cancel' lock handling could lead to such locks
3258 never to be granted in case of SMP MDS
3259 Details : Do not destroy not yet granted but cbpending locks in
3263 Frequency : replay/resend of open
3265 Description: non null lock assetion failure in mds_intent_policy
3266 Details : Trying to replay/resend lockless open requests resulted in
3267 mds_open() returning 0 with no lock. Now it sets a flag if
3268 a lock is going to be returned.
3270 Severity : enhancement
3272 Description: Checksum enhancements
3273 Details : New checksum enhancements allow for resending RPCs that failed
3276 Severity : enhancement
3278 Description: Tunables on number of dirty pages in cacche
3279 Details : Allow to set limit on number of dirty pages cached.
3284 Description: client crash on unmount - lock still has references
3285 Details : In some error handling cases it was possible to leak a lock
3286 reference on a client while accessing a file. This was not
3287 harmful to the client during operation, but would cause the
3288 client to crash when the filesystem is unmounted.
3291 Frequency : specific case, rare
3293 Description: ETXTBSY on mds though file not in use by client
3294 Details : ETXTBSY is no longer incorrectly returned when attempting to
3295 chmod or chown a directory that the user previously tried to
3296 execute or a currently-executing binary.
3299 Frequency : extremely rare except on liblustre-based clients
3301 Description: Lustre space not freed when files are deleted
3302 Details : Clean up open-unlinked files after client eviction. Previously
3303 the unlink was skipped and the files remained as orphans.
3308 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3309 Details : In some cases it was possible to send a blocking callback to
3310 a client doing a glimpse, even though that client didn't get
3311 a lock granted. When the glimpse lock is cancelled on the OST
3312 the freed lock is left on the waiting list and corrupted the list.
3315 Frequency : all core dumps
3317 Description: Broke core dumps to lustre
3318 Details : Negative dentry may be unhashed if parent does not have UPDATE
3319 lock, but some callers, e.g. do_coredump, expect dentry to be
3320 hashed after successful create, hash it in ll_create_it.
3322 ------------------------------------------------------------------------------
3324 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3326 * Support for kernels:
3327 2.6.9-42.0.2.EL (RHEL 4)
3328 2.6.5-7.276 (SLES 9)
3329 2.4.21-40.EL (RHEL 3)
3330 2.6.12.6 vanilla (kernel.org)
3334 Frequency : always on RHEL 3
3336 Description: Number of open files grows over time
3337 Details : The number of open files grows over time, whether or not
3338 Lustre is started. This was due to a filp leak introduced
3339 by one of our kernel patches.
3341 ------------------------------------------------------------------------------
3343 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3345 * Support for kernels:
3346 2.6.9-42.EL (RHEL 4)
3347 2.6.5-7.276 (SLES 9)
3348 2.4.21-40.EL (RHEL 3)
3349 2.6.12.6 vanilla (kernel.org)
3354 Bugzilla : 5719, 9635, 9792, 9684
3355 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3356 Details : If a server is under heavy load and cannot reply to new
3357 connection requests before the client resends the (re)connect,
3358 the connection handling code can behave badly if two service
3359 threads are concurrently handing separate (re)connections from
3360 the same client. Add better locking to the connection handling
3361 code, and ensure that only a single connection will be processed
3362 for a given client UUID, even if the lock is dropped.
3364 Severity : enhancement
3366 Description: add TCP zero-copy support to kernel
3367 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3368 sends if the hardware supports scatter-gather and checksumming.
3369 This allows socklnd to do client-write and server-read more
3370 efficiently and reduce CPU utilization from skbuf copying.
3373 Frequency : only if NFS exporting from client
3375 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3376 Details : If setting attributes on a file created under NFS that had
3377 never been opened it would be possible to oops the client
3378 if the file had no objects.
3381 Frequency : always for liblustre
3383 Description: liblustre client does MDS+OSTs setattr RPC for each write
3384 Details : When doing a write from a liblustre client, the client
3385 incorrectly issued an RPC to the MDS and each OST the file was
3386 striped over in order to update the timestamps. When writing
3387 with small chunks and many clients this could overwhelm the MDS
3388 with RPCs. In all cases it would slow down the write because
3389 these RPCs are unnecessary.
3391 Severity : enhancement
3393 Description: allow number of MDS service threads to be changed at module load
3394 Details : It is now possible to change the number of MDS service threads
3395 running. Adding "options mds mds_num_threads={N}" to the MDS's
3396 /etc/modprobe.conf will set the number of threads for the next
3397 time Lustre is restarted (assuming the "mds" module is also
3398 reloaded at that time). The default number of threads will
3399 stay the same, 32 for most systems.
3404 Description: OST crash if filesystem is unformatted or corrupt
3405 Details : If an OST is started on a device that has never been formatted
3406 or if the filesystem is corrupt and cannot even mount then the
3407 error handling cleanup routines would dereference a NULL pointer.
3412 Description: NULL pointer deref in llap_from_page.
3413 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3414 incorrect) mapping if the page was truncated/reclaimed while it was
3415 searched for. Check for this condition and skip such pages when
3416 doing readahead. Introduce extra check to llap_from_page() to
3417 verify page->mapping->host is non-NULL (so page is not anonymous).
3420 Frequency : Sometimes when using sys_sendfile
3422 Description: "page not covered by a lock" warnings from ll_readpage
3423 Details : sendfile called ll_readpage without right page locks present.
3424 Now we introduced ll_file_sendfile that does necessary locking
3425 around call to generic_file_sendfile() much like we do in
3429 Frequency : with certain MDS communication failures at client mount time
3431 Description: NULL pointer deref after failed client mount
3432 Details : a client connection request may delayed by the network layer
3433 and not be sent until after the PTLRPC layer has timed out the
3434 request. If the client fails the mount immediately it will try
3435 to clean up before the network times out the request. Add a
3436 reference from the request import to the obd device and delay
3437 the cleanup until the network drops the request.
3440 Frequency : occasionally during client (re)connect
3442 Description: assertion failure during client (re)connect
3443 Details : processing a client connection request may be delayed by the
3444 client or server longer than the client connect timeout. This
3445 causes the client to resend the connection request. If the
3446 original connection request is replied in this interval, the
3447 client may trip an assertion failure in ptlrpc_connect_interpret()
3448 which thought it would be the only running connect process.
3451 Frequency : only with obd_echo servers and clients that are rebooted
3453 Description: kernel BUG accessing uninitialized data structure
3454 Details : When running an obd_echo server it did not start the ping_evictor
3455 thread, and when a client was evicted an uninitialized data
3456 structure was accessed. Start the ping_evictor in the RPC
3457 service startup instead of the OBD startup.
3459 Severity : enhancement
3460 Bugzilla : 10193 (patchless)
3461 Description: Remove dependency on various unexported kernel interfaces.
3462 Details : No longer need reparent_to_init, exit_mm, exit_files,
3463 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3466 Frequency : rare (only users of deprecated and unsupported LDAP config)
3468 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3469 Details : LDAP apparently contains 'lustreName' attributes instead of
3470 'name'. A simple remapping of the name is sufficient.
3473 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3475 Description: waiting_locks_callback trips kernel BUG if client is evicted
3476 Details : Running with the dump_on_timeout debug flag turned on makes
3477 it possible that the waiting_locks_callback() can try to dump
3478 the Lustre kernel debug logs from an interrupt handler. Defer
3479 this log dumping to the expired_lock_main() thread.
3481 Severity : enhancement
3483 Description: Support NFS exporting on 2.6 kernels.
3484 Details : Implement non-rawops metadata methods for NFS server to use without
3485 changing NFS server code.
3488 Frequency : very rare (synthetic metadata workload only)
3490 Description: two racing renames might cause an MDS thread to deadlock
3491 Details : Running the "racer" program may cause one MDS thread to rename
3492 a file from being the source of a rename to being the target of
3493 a rename at exactly the same time that another thread is doing
3494 so, and the second thread has already enqueued these locks after
3495 doing a lookup of the target and is trying to relock them in
3496 order. Ensure that we don't try to re-lock the same resource.
3499 Frequency : only very large systems with liblustre clients
3501 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3502 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3503 all OSTs in parallel. This allows the eviction of stale liblustre
3504 clients to proceed much faster than if they were done in series,
3505 and also offers similar improvements for other set_info RPCs.
3510 Description: excessive CPU usage during initial read phase on client
3511 Details : During the initial read phase on a client, it would agressively
3512 retry readahead on the file, consuming too much CPU and impacting
3513 performance (since 1.4.5.8). Improve the readahead algorithm
3514 to avoid this, and also improve some other common cases (read
3515 of small files in particular, where "small" is files smaller than
3516 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3521 Description: MDS crash when receiving packet with unknown intent.
3522 Details : Do not LBUG in unknown intent case, just return -EFAULT
3524 Severity : enhancement
3525 Bugzilla : 9293, 9385
3526 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3527 Details : Do not serialize getattr (non-intent version) and statfs.
3530 Frequency : occasional, when OST network is overloaded/intermittent
3532 Description: client evicted by OST after bulk IO timeout
3533 Details : If a client sends a bulk IO request (read or write) the OST
3534 may evict the client if it is unresposive to its data GET/PUT
3535 request. This is incorrect if the network is overloaded (takes
3536 too long to transfer the RPC data) or dropped the OST GET/PUT
3537 request. There is no need to evict the client at all, since
3538 the pinger and/or lock callbacks will handle this, and the
3539 client can restart the bulk request.
3542 Frequency : Always when mmapping file with no objects
3544 Description: client crashes when mmapping file with no objects
3545 Details : Check that we actually have objects in a file before doing any
3546 operations on objects in ll_vm_open, ll_vm_close and
3552 Description: Request leak when working with deleted CWD
3553 Details : Introduce advanced request refcount tracking for requests
3554 referenced from lustre intent.
3556 Severity : Enhancement
3558 Description: Cache open file handles on client.
3559 Details : MDS now will return special lock along with openhandle, if
3560 requested and client is allowed to hold openhandle, even if unused,
3561 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3562 closing files for every read/write openration.
3564 Severity : Enhancement
3566 Description: Cache open negative dentries on client when possible.
3567 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3568 negative dentries on lock revocation.
3573 Description: Remounting a client read-only wasn't possible with a zconf mount
3574 Details : It wasn't possible to remount a client read-only with llmount.
3576 Severity : enhancement
3577 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3578 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3579 MPICH is included to add Lustre-specific ADIO interfaces.
3580 This is based closely on the UFS ADIO layer and only differs
3581 in file creation, in order to allow the OST striping to be set.
3582 This is user-contributed code and not supported by CFS.
3587 Description: extended inode attributes (immutable, append-only) work improperly
3588 when 2.4 and 2.6 kernels are used on client/server or vice versa
3589 Details : Introduce kernel-independent values for these flags.
3591 Severity : enhancement
3594 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3595 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3596 granularity. Now, Fractional MB granularity is be supported,
3597 this is very useful for low memory system.
3599 Severity : enhancement
3601 Description: Getattr by fid
3602 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3603 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3604 have locks and avoids getattr rpc requests for it on every stat.
3607 Frequency : Always, for filesystems larger than 2TB
3609 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3610 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3611 filesystems larger than 2TB to be mounted with mballoc enabled.
3614 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3616 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3617 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3618 filesystems larger than 2TB is mounted then the kernel will
3619 silently corrupt the start of the filesystem. CONFIG_LBD is
3620 enabled for all CFS-supported kernels, but the possibility of
3621 this happening with a modified kernel config exists.
3623 Severity : enhancement
3625 Description: add client O_DIRECT support for 2.6 kernels
3626 Details : It is now possible to do O_DIRECT reads and writes to files
3627 in the Lustre client mountpoint on 2.6 kernel clients.
3629 Severity : enhancement
3631 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3632 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3633 parallel, not waiting for response from every OST before sending
3634 a rpc to the next OST.
3639 Description: setattr vs write race when updating file timestamps
3640 Details : Client processes that update a file timestamp into the past
3641 right after writing to the file (e.g. tar) it is possible that
3642 the updated file modification time can be reset to the current
3643 time due to a race between processing the setattr and write RPC.
3645 Severity : enhancement
3647 Description: Bring 'lfs find' closer in line with regular Linux find.
3648 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3649 -print0 options and obtains all the needed info through the lustre
3652 Severity : enhancement
3654 Description: support up to 1024 configured devices on one node
3655 Details : change obd_dev array from statically allocated to dynamically
3656 allocated structs as they are first used to reduce memory usage
3661 Description: Flush dirty partially truncated pages during truncate
3662 Details : Immediatelly flush partially truncated pages in filter_setattr,
3663 this way we completely avoid having any pages in page cache on OST
3664 and can retire ugly workarounds during writes to flush such pages.
3669 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3670 Details : protect inode from truncation within vfs_unlink() context
3671 just take a reference before calling vfs_unlink() and release it
3672 when parent's i_sem is free.
3677 Description: last_id value checked outside lock on OST caused LASSERT failure
3678 Details : If there were multiple MDS->OST object precreate requests in
3679 flight, it was possible that the OST's last object id was checked
3680 outside a lock and incorrectly tripped an assertion. Move checks
3681 inside locks, and discard old precreate requests.
3684 Frequency : always, if extents are used on OSTs
3686 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3687 Details : OSTs using the extents format would not zero the high 16 bits of
3688 the index physical block number. This is not a problem for any
3689 OST filesystems smaller than 16TB, and no kernels support ext3
3690 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3691 new/modified files) and can be fixed for existing filesystems
3692 with e2fsprogs-1.39-cfs1.
3697 Description: import connection selection may be incorrect if timer wraps
3698 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3699 import connections to be ignored if the 32-bit jiffies counter
3700 wraps. Use a 64-bit jiffies counter.
3703 Frequency : very large clusters immediately after boot
3705 Description: LNET request buffers exhausted under heavy short-term load
3706 Details : If a large number of client requests are generated on a service
3707 that has previously never seen so many requests it is possible
3708 that the request buffer growth cannot keep up with the spike in
3709 demand. Instead of dropping incoming requests, they are held in
3710 the LND until the RPC service can accept more requests.
3713 Frequency : Sometimes during replay
3715 Description: Assertion failure in ll_local_open after replay.
3716 Details : If replay happened on an open request reply before we were able
3717 to set replay handler, reply will become not swabbed tripping the
3718 assertion in ll_local_open. Now we set the handler right after
3719 recognising of open request
3722 Frequency : very rare
3724 Description: kernel reports "badness in vsnprintf"
3725 Details : Reading from the "recovery_status" /proc file in small chunks
3726 may cause a negative length in lprocfs_obd_rd_recovery_status()
3727 call to vsnprintf() (which is otherwise harmless). Exit early
3728 if there is no more space in the output buffer.
3730 Severity : enhancement
3732 Description: clear OBD RPC statistics by writing to them
3733 Details : It is now possible to clear the OBD RPC statistics by writing
3734 to the "stats" file.
3739 Description: Client mtime is not the same on different clients after utimes
3740 Details : In some cases, the client was using the utimes() syscall on
3741 a file cached on another node. The clients now validate the
3742 ctime from the MDS + OSTs to determine which one is right.
3747 Description: Inability to activate failout mode
3748 Details : lconf script incorrectly assumed that in pythong string's numeric
3749 value is used in comparisons.
3752 Frequency : always with multiple stripes per file
3754 Description: Inefficient object allocation for mutli-stripe files
3755 Details : When selecting which OSTs to stripe files over, for files with
3756 a stripe count that divides evenly into the number of OSTs,
3757 the MDS is always picking the same starting OST for each file.
3758 Return the OST selection heuristic to the original design.
3763 Description: mount failures may take full timeout to return an error
3764 Details : Under some heavy load conditions it is possible that a
3765 failed mount can wait for the full obd_timeout interval,
3766 possibly several minutes, before reporting an error.
3767 Instead return an error as soon as the status is known.
3769 Frequency : quota enabled and large files being deleted
3771 Description: releasing more than 4GB of quota at once hangs OST
3772 Details : If a user deletes more than 4GB of files on a single OST it
3773 will cause the OST to spin in an infinite loop. Release
3774 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3779 Description: statfs data retrieved from /proc may be stale or zero
3780 Details : When reading per-device statfs data from /proc, in the
3781 {kbytes,files}_{total,free,avail} files, it may appear
3782 as zero or be out of date.
3785 Frequency : systems with MD RAID1 external journal devices
3787 Description: lconf's call to blkid is confused by RAID1 journal devices
3788 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3789 of returning all block devices that match the journal UUID.
3792 Frequency : always, for aggregate stripe size over 4GB
3794 Description: assertion fails when trying to use 4GB stripe size
3795 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3796 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3799 Frequency : always on ppc64
3801 Description: the first write on an ext3 filesystem with mballoc got stuck
3802 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3803 perform endianness conversion.
3805 ------------------------------------------------------------------------------
3807 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3809 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3810 INTEROPERATE with older versions automatically. Please read the
3811 user documentation before upgrading any part of a live system.
3812 * WARNING: Lustre networking configuration changes are required with
3813 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3816 * Support for kernels:
3817 2.6.9-22.0.2.EL (RHEL 4)
3818 2.6.5-7.244 (SLES 9)
3819 2.6.12.6 vanilla (kernel.org)
3822 Severity : enhancement
3823 Bugzilla : 7981/8208
3824 Description: Introduced Lustre Networking (LNET)
3825 Details : LNET is new networking infrastructure for Lustre, it includes
3826 a reorganized network configuration mode (see the user
3827 documentation for full details) as well as support for routing
3828 between different network fabrics. Lustre Networking Devices
3829 (LNDs) for the supported network fabrics have also been
3830 created for this new infrastructure.
3832 Severity : enhancement
3833 Description: Introduced Access control lists
3834 Details : clients can set ACLs on files and directories in order to have
3835 more fine-grained permissions than the standard Unix UGO+RWX.
3836 The MDS must be started with the "-o acl" mount option.
3838 Severity : enhancement
3839 Description: Introduced filesystem quotas
3840 Details : Administrators may now establish per-user quotas on the
3843 Severity : enhancement
3845 Description: Configuration change for the XT3
3846 The PTLLND is now used to run Lustre over Portals on the XT3
3847 The configure option(s) --with-cray-portals are no longer used.
3848 Rather --with-portals=<path-to-portals-includes> is used to
3849 enable building on the XT3. In addition to enable XT3 specific
3850 features the option --enable-cray-xt3 must be used.
3855 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3856 Details : A race between allocating a new llcd and re-getting the llcd_lock
3857 allowed another thread to grab newly-allocated llcd.
3859 Severity : enhancement
3861 Description: 2.6 OST async journal commit and locking fix to improve performance
3862 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3863 2.6 kernels are now async as they already were in 2.4 kernels so
3864 that they can commit concurrently with the network bulk transfer.
3865 For block-allocated files the filter allocation semaphore is held
3866 to avoid filesystem fragmentation during allocation. BKL lock
3867 removed for 2.6 xattr operations where it is no longer needed.
3872 Description: lconf incorrectly determined whether two IP networks could talk
3873 Details : In some more complicated routing and multiple-network
3874 configurations, lconf will avoid trying to make a network
3875 connection to a disjoint part of the IP space. It was doing the
3876 math incorrectly for one set of cases.
3881 Description: Fix for potential infinite loop processing records in an llog.
3882 Details : If an llog record is corrupted/zeroed, it is possible to loop
3883 forever in llog_process(). Validate the llog record length
3884 and skip the remainder of the block on error.
3887 Frequency : occasional (liblustre only)
3889 Description: liblustre could not open files whose last component is a symlink
3890 Details : sysio_path_walk() would incorrectly pass the open intent to
3891 intermediate path components.
3894 Frequency : rare (liblustre only with non-standard tuning)
3895 Bugzilla : 7201 (7350)
3896 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3897 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3898 after it has been released, only do this for referenced locks
3900 Severity : enhancement
3902 Description: specifying an (invalid) directory default stripe_size of -1
3903 would reset the directory default striping
3904 Details : stripe_size -1 was used internally to signal directory stripe
3905 removal, now use "all default" to signal dir stripe removal
3906 as a directory striping of "all default" is not useful
3909 Frequency : common for large clusters running liblustre clients
3911 Description: doing an ls when liblustre clients are running is slow
3912 Details : sending a glimpse AST to a liblustre client waits for every AST
3913 to time out, as liblustre clients will not respond. Since they
3914 cannot cache data we refresh the OST lock LVB from disk instead.
3916 Severity : enhancement
3918 Description: doing an ls at the same time as file IO can be slow
3919 Details : enqueue and other "small" requests can be blocked behind many
3920 large IO requests. Create a new OST IO portal for non-IO
3921 requests so they can be processed faster.
3924 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3926 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3927 created file with mode 000
3928 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3929 allow an ftruncate() as a non-root user to a file with mode 000.
3930 Lustre now respects this flag to disable mode checks when
3931 truncating a file owned by the user
3934 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3937 Description: Revoking locks from clients that went dead or catatonic might take
3939 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3940 cancellation of such locks instant on servers without waiting for
3941 any reply from clients. Clients drops these locks when cancel
3942 notification from server is received without replying.
3945 Frequency : liblustre-only, when liblustre client dies or becomes busy
3947 Description: Doing ls on Linux clients can take a long time with active
3949 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3950 granting such locks to it in the first place if possible. Locks
3951 are taken by proxy on the OST during the read or write and
3952 dropped immediately afterward. Add connect flags handling, do
3953 not grant locks to liblustre clients for glimpse ASTs.
3955 Severity : enhancement
3957 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3958 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3959 sequential reads and behaves badly with applications doing random
3960 reads. Improve it by reading ahead at least read region, and
3961 avoiding excessive large RPC for small reads.
3963 Severity : enhancement
3965 Description: Creating more than 1000 files for a single job may cause a load
3966 imbalance on the OSTs if there are also a large number of OSTs.
3967 Details : qos_prep_create() uses an OST index reseed value that is an
3968 even multiple of the number of available OSTs so that if the
3969 reseed happens in the middle of the object allocation it will
3970 still utilize the OSTs as uniformly as possible.
3975 Description: OST or MDS may oops in ping_evictor_main()
3976 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3977 but doesn't restart at beginning of obd_exports_timed list
3980 Severity : enhancement
3982 Description: improve by-nid export eviction on the MDS and OST
3983 Details : allow multiple exports with the same NID to be evicted at one
3984 time without re-searching the exports list.
3987 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3989 Description: MDS may oops in groups_free()
3990 Details : in rare race conditions a newly allocated group_info struct is
3991 freed again, and this can be NULL. The 2.4 compatibility code
3992 for groups_free() checked for a NULL pointer, but 2.6 did not.
3995 Frequency : common for liblustre clients doing little filesystem IO
3996 Bugzilla : 9352, 7313
3997 Description: server may evict liblustre clients accessing contended locks
3998 Details : if a client is granted a lock or receives a completion AST
3999 with a blocking AST already set it would not reply to the AST
4000 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4003 Frequency : lfs setstripe, only systems with more than 160 OSTs
4005 Description: unable to set striping with a starting offset beyond OST 160
4006 Details : llapi_create_file() incorrectly limited the starting stripe
4007 index to the maximum single-file stripe count.
4010 Frequency : LDAP users only
4012 Description: lconf did not handle in-kernel recovery with LDAP properly
4013 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4015 Severity : enhancement
4017 Description: bind OST threads to NUMA nodes to improve performance
4018 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4019 node and do their allocations there to localize memory access
4021 Severity : enhancement
4023 Description: llmount can determine client NID directly from Myrinet (GM)
4024 Details : the client NID code from gmnalnid was moved directly into
4025 llmount, removing the need to use this or specifying the
4026 client NID explicitly when mounting GM clients with zeroconf
4029 Frequency : if client is started with down MDS
4031 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4032 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4033 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4034 ptlrpc_import_delayed_req() because we didn't check if the
4035 request was interrupted, and we also didn't break out of the
4036 event loop if there was no timeout
4041 Description: data loss during non-page-aligned writes to a single file from
4042 both multiple nodes and multiple threads on one node at same time
4043 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4044 inconsistency led to false short-reads, that were cached and later
4045 used by ->prepare_write() to fill in partially written page,
4046 leading to data loss.
4049 Frequency : always, if lconf --abort_recovery used
4051 Description: lconf --abort_recovery fails with 'Operation not supported'
4052 Details : lconf was attempting to abort recovery on the MDT device and not
4055 Severity : enhancement
4057 Description: remove cleanup logs
4058 Details : replace lconf-generated cleanup logs with lustre internal
4059 cleanup routines. Eliminates the need for client-cleanup and
4062 Severity : enhancement
4064 Description: add support for EAs (user and system) on lustre filesystems
4065 Details : it is now possible to store extended attributes in the Lustre
4066 client filesystem, and with the user_xattr mount option it
4067 is possible to allow users to store EAs on their files also
4069 Severity : enhancement
4071 Description: Add possibility (config option) to show minimal available OST free
4073 Details : When compiled with --enable-mindf configure option, statfs(2)
4074 (and so, df) will return least minimal free space available from
4075 all OSTs as amount of free space on FS, instead of summary of
4076 free spaces of all OSTs.
4078 Severity : enhancement
4080 Description: do not expand extent locks acquired on OST-side
4081 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4082 by server: they are not cached anyway.
4085 Frequency : when mmap is used/binaries executed from Lustre
4087 Description: Unmmap pages before throwing them away from read cache.
4088 Details : llap_shrink cache now attempts to unmap pages before discarding
4089 them (if unmapping failed - do not discard). SLES9 kernel has
4090 extra checks that trigger if this unmapping is not done first.
4095 Description: lconf didn't resolve symlinks before checking to see whether a
4096 given mountpoint was already in use
4099 Frequency : when migrating failover services
4100 Bugzilla : 6395, 9514
4101 Description: When migrating a subset of services from a node (e.g. failback
4102 from a failover service node) the remaining services would
4103 time out and evict clients.
git://git.whamcloud.com / fs / lustre-release.git / blob