1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.39.3 (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.
16 Severity : enhancement
18 Description: Add LustreNetLink, a kernel-userspace communcation path.
20 Severity : enhancement
22 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
27 Description: access to llog context before init.
28 Details : move handling CATALOGS file at osc layer and forbid access to llog
33 Description: Avoid deadlock for local client writes
34 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
35 memory freeing context. This allows OST threads to set the
36 PF_MEMALLOC flag on task structures in order to allocate memory
37 from reserved pools and complete IO.
38 Use GFP_HIGHUSER for OST allocations for non-local client writes,
39 so that the OST threads generate memory pressure and allow
40 inactive pages to be reclaimed.
45 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
46 Details : move ldlm namespace creation in setup phase to avoid grab
47 _lprocfs_lock with cli_sem held.
51 Description: Temporarily disable grant shrink.
52 Details : Disable the feature for debugging.
56 Description: Unable to run several mkfs.lustre on loop devices at the same
58 Details : mkfs.lustre returns error 256 on the concurrent loop devices
59 formatting. The solution is to proper handle the error.
61 Severity : enhancement
63 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
65 Severity : enhancement
67 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
71 Description: resolve race between obd_disconnect and class_disconnect_exports
72 Details : if obd_disconnect will be called to already disconnected export he
73 forget release one reference and osc module can't unloaded.
75 Severity : enhancement
77 Description: Allow tuning service thread via /proc
78 Details : For each service a new
79 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
80 created that can be used to set min/max thread counts, and get the
81 current number of running threads.
85 Descriptoin: don't return error if have particaly created objects for file.
86 Details : lov_update_create_set uses set->set_success as index for created objects,
87 so if some requests will be failed, they will have hole at end of
88 array and we can use qos_shrink_lsm for allocate correct lsm.
90 Severity : enhancement
92 Description: Update OFED support to 1.4.1
96 Description: Reduce small size read RPC
97 Details : Set read-ahead limite for every file and only do read-ahead when
98 available read-ahead pages are bigger than 1M to avoid small size
101 Severity : enhancement
103 Description: move AT tunable parameters for more consistent usage
104 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
106 Severity : enhancement
108 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
109 Details : allow skip disconnected ost for send statfs request and hide error
115 Description: fix racy locking of mballoc block bitmaps causing BUG
116 Details : The locking of the mballoc buddy bitmap and the in-memory
117 block bitmap was using two different spin locks in some
118 cases. This made it possible to incorrectly access the
119 mballoc bitmap while another process was modifying it,
120 causing a sanity assertion to fail. While no on-disk corruption
121 was reported, there was some risk of this happening.
124 Frequency : rare, on llog test 6
126 Descriptoin: don't allow connect to already connected import
127 Details : allowing connect to already connected import is hide connecting problem.
132 Descriptoin: allow kill process which wait statahead result
133 Details : for some reasons 'ls' can stick in waiting result from statahead,
134 in this case need way for kill this process.
136 Severity : enhancement
138 Description: Add state history info file, enhance import info file
139 Details : Track import connection state changes in a new osc/mdc proc file;
140 add overview-type data to the osc/mdc import proc file.
142 Severity : enhancement
144 Description: MDS create should not wait for statfs RPC while holding DLM lock.
146 Severity : enhancement
148 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
153 Descriptoin: OST grant too much space to client even there are not enough space.
154 Details : Client will shrink its grant cache to OST if there are no write
155 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
156 this grant cache if there are already not enough avaible space
157 (left_space < total_clients * 32M).
160 Frequency : start MDS on uncleanly shutdowned MDS device
162 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
163 Details : stay in waiting mds<>ost recovery finished produce random bugs
164 due race between two ll_sync thread for one lov target. send
165 ACTIVATE event only if connect realy finished and import have
169 Frequency : rare, connect and disconnect target at same time
171 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
172 Details : don't call obd_disconnect under lov_lock. this long time
173 operation and can block ptlrpcd which answer to connect request.
178 Descriptoin: don't lose wakeup for imp_recovery_waitq
179 Details : recover_import_no_retry or invalidate_import and import_close can
180 both sleep on imp_recovery_waitq, but we was send only one wakeup
184 Frequency : always with long access acl
186 Descriptoin: mds can't pack reply with long acl.
187 Details : mds don't control size of acl but they limited by reint/getattr
190 Severity : enhancement
192 Description: Update to SLES10 kernel-2.6.16.60-0.33.
194 Severity : enhancement
196 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
199 Frequency : start MDS on uncleanly shutdowned MDS device
201 Descriptoin: aborting recovery hang on MDS
202 Details : don't throttle destroy RPCs for the MDT.
205 Frequency : on remount
207 Description: external journal device not working after the remount
208 Details : clear dev_rdonly flag for external journal devices in
214 Description: shutdown vs evict race
215 Details : client_disconnect_export vs connect request race.
216 if client will evicted at this time - we start invalidate
217 thread without referece to import and import can be freed
221 Frequency : rare, need acl's on inode.
223 Description: client can't handle ost additional correctly
224 Details : if ost was added after client connected to mds client can have
225 hit lnet_try_match_md ... to big messages to wide striped files.
226 in this case need teach client to handle config events about add
227 lov target and update client max ea size at that event.
229 Severity : enhancement
231 Description: Changelogs
232 Details : Changelogs are a lightweight mechanism to track filesystem
233 metadata and namespace changes. The changelog is recorded
234 permanently on the MDTs, and is periodically "consumed" / purged
235 when records are no longer needed.
237 Severity : enhancement
239 Description: compact fld format with extents
240 Details : Store range of seq rather than every seq in FLD. Seq
241 controller update FLD rather than clients. In Case of CMD, mdt0
242 has FLD, all other metadata server act as non persistent proxy
243 for FLD queries and cache fld entries in fld cache.
248 Description: don't skip ost target if they assigned to file
249 Details : Drop slow OSCs if we can, but not for requested start idx.
250 This means "if OSC is slow and it is not the requested
251 start OST, then it can be skipped, otherwise skip it only
252 if it is inactive/recovering/out-of-space.
256 Description: more cleanup in mds_lov
257 Details : not send LOV EA under replay, we can't know about they size at this
258 time. Don't allow client connect to mds before any ost connected,
259 for avoid problems with LOV EA size and returning EIO to client.
261 Severity : enhancement
263 Description: Interoperability at server side (Disk interoperability)
265 Severity : enhancement
267 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
269 Severity : enhancement
271 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
273 Severity : enhancement
275 Description: New client IO stack (CLIO).
277 Severity : enhancement
279 Description: Commit on sharing. Eliminate inter-client dependencies between
280 uncommitted transactions by doing transaction commits.
281 Thereby clients may recovery independently.
284 Frequency : Create a symlink file with a very long name
286 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
287 Details : If there is no extra space in the request for early cancels,
288 ldlm_req_handles_avail() returns 0 instead of a negative value.
290 Severity : enhancement
292 Description: Add /proc entry for import status
293 Details : The mdc, osc, and mgc import directories now have
294 an import directory that contains useful import data for debugging
297 Severity : enhancement
299 Description: Re-disable certain /proc logging
300 Details : Enable and disable client's offset_stats, extents_stats and
301 extents_stats_per_process stats logging on the fly.
304 Frequency : Only on FC kernels 2.6.22+
306 Description: oops in statahead
307 Details : Do not drop reference count for the dentry from VFS when lookup,
308 VFS will do that by itself.
310 Severity : enhancement
312 Description: Generic /proc file permissions
313 Details : Set /Proc file permissions in a more generic way to enable non-
314 root users operate on some /proc files.
318 Description: Hitting mdc_commit_close() ASSERTION
319 Details : Properly handle request reference release in
320 ll_release_openhandle().
324 Description: quota recovery deadlock during mds failover
325 Details : This patch includes att18982, att18236, att18237 in bz14840.
327 1. fix osts hang when mds does failover with quotaon
328 2. prevent watchdog storm when osts threads wait for the
333 Frequency : only patchless client
334 Description: add workaround for race between add/remove dentry from hash
336 Severity : enhancement
338 Description: Allow OST glimpses to return PW locks
342 Description: LBUG when llog conf file is full
343 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
347 Description: Prevent import from entering FULL state when server in recovery
351 Description: service mount cannot take device name with ":"
352 Details : Only when device name contains ":/" will mount treat it as
358 Description: replace ptlrpcd with the statahead thread to interpret the async
359 statahead RPC callback
363 Frequency : on recovery
364 Description: I/O failures after umount during fail back
365 Details : if client reconnected to restarted server we need join to recovery
366 instead of find server handler is changed and process self
367 eviction with cancel all locks.
369 Severity : enhancement
371 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
373 Severity : enhancement
375 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
377 Severity : enhancement
379 Description: Upcall on Lustre log has been dumped
380 Details : Allow for a user mode script to be called once a Lustre log has
381 been dumped. It passes the filename of the dumped log to the
382 script, the location of the script can be specified via
383 /proc/sys/lnet/debug_log_upcall.
388 Description: avoid idr_remove called for id which is not allocated.
389 Details : Move assigment s_dev for clustered nfs to end of initialization,
390 to avoid problem with error handling.
395 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
396 Details : When connection is reused this not moved from CONN_UNUSED_HASH
397 into CONN_USED_HASH and this prodice warning when put connection
398 again in unused hash.
401 Severity : enhancement
403 Description: File striping can now be set to use an arbitrary pool of OSTs.
405 Severity : enhancement
407 Description: Export bytes_read/bytes_write count on OSC/OST.
411 Description: Early reply size mismatch, MGC loses connection
412 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
413 the connect flags are properly negotiated.
418 Description: quotas are not honored with O_DIRECT
419 Details : all writes with the flag O_DIRECT will use grants which leads to
420 this problem. Now using OBD_BRW_SYNC to guard this.
424 Description: add quota statistics
425 Details : 1. sort out quota proc entries and proc code.
426 2. add quota statistics
428 Severity : enhancement
430 Description: enable quota support for HEAD.
434 Description: Properly propagate oinfo flags from lov to osc for statfs
435 Details : restore missing copy oi_flags to lov requests.
437 Severity : enhancement
439 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
440 : l_getgroups(8), lst(8), routerstat(8)
441 Details : included man pages for llobdstat(8), llstat(8),
442 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
444 Severity : enhancement
446 Description: configure's --enable-quota should check the
447 : kernel .config for CONFIG_QUOTA
448 Details : configure is terminated if --enable-quota is passed but
449 : no quota support is in kernel
453 Description: Remove portals compatibility
454 Details : Remove portals compatibility, not interoperable with releases
459 Description: Resolve device initialization race
460 Details : Prevent proc handler from accessing devices added to the
461 obd_devs array but yet be intialized.
463 Severity : enhancement
465 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
467 Severity : enhancement
469 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
474 Description: Using wrong pointer in osc_brw_prep_request
475 Details : Access to array[-1] can produce panic if kernel compiled with
476 CONFIG_PAGE_ALLOC enabled
480 Description: Client runs out of low memory
481 Details : Consider only lowmem when counting initial number of llap pages
485 Description: *optional* service tags registration
486 Details : if the "service tags" package is installed on a Lustre node
487 When the filesystem is mounted, a local-node service tag will
488 be created. See http://inventory.sun.com/ for more information
489 about the Service Tags asset management system.
493 Description: Kernel BUG tries to release flock
494 Details : Lustre does not destroy flock lock before last reference goes
495 away. So always drop flock locks when client is evicted and
496 perform unlock regardless of successfulness of speaking to MDS.
500 Description: add refcount protection for osc callbacks, avoid panic on shutdown
504 Description: sanity test 65a fails if stripecount of -1 is set
505 Details : handle -1 striping on filesystem in ll_dirstripe_verify
510 Description: ASSERTION(CheckWriteback(page,cmd)) failed
511 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
514 Severity : enhancement
516 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
520 Description: do not process already freed flock
521 Details : flock can possibly be freed by another thread before it reaches
522 to ldlm_flock_completion_ast.
526 Description: LBUG during stress test
527 Details : Need properly lock accesses the flock deadlock detection list.
531 Description: oops in page fault handler
532 Details : kernel page fault handler can return two special 'pages' in error
533 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
537 Description: timeout with invalidate import.
538 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
539 should be handled by ptlrpcd. This produce long age waiting and
540 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
542 Severity : enhancement
544 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
548 Description: enable MGS and MDT services start separately
549 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
550 a co-located MGS without starting the MGS, which is a complement
551 to 'nosvc' mount option.
555 Frequency : after recovery
556 Description: precreate to many object's after del orphan.
557 Details : del orphan st in oscc last_id == next_id and this triger growing
558 count of precreated objects. Set flag LOW to skip increase count
559 of precreated objects.
563 Frequency : rare, on clear nid stats
564 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
565 Details : when clean nid stats sometimes try destroy live entry,
566 and this produce panic in free.
570 Description: Stack overflow during MDS log replay
571 ease stack pressure by using a thread dealing llog_process.
575 Description: wait until IO finished before start new when do lock cancel.
576 Details : VM protocol want old IO finished before start new, in this case
577 need wait until PG_writeback is cleared until check dirty flag and
578 call writepages in lock cancel callback.
580 Severity : enhancement
582 Description: using special macro for print time and cleanup in includes.
586 Description: mds_mfd_close() ASSERTION(rc == 0)
587 Details : In mds_mfd_close(), we need protect inode's writecount change
588 within its orphan write semaphore to prevent possible races.
592 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
596 Frequency : rare, on shutdown ost
597 Description: don't hit live lock with umount ost.
598 Details : shrink_dcache_parent can be in long loop with destroy dentries,
599 use shrink_dcache_sb instead.
603 Description: don't panic with use echo client
604 Details : echo client pass NULL as client nid pointer and this produce null
609 Description: fix build on ppc32
610 Details : compile code with -m64 flag produce wrong object file for ppc32.
614 Description: add message levels for liblustreapi
618 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
619 Details : Move the 'good_osts' check before the 'total_bavail' check. This
620 will result in an -EAGAIN and in the exit call path we call
621 alloc_rr() which will with increasing aggressiveness attempt to
622 aquire precreated objects on the minimum number of required OSCs.
626 Description: Use old size assignment to avoid deadlock
627 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
628 the scheduling while holding a spinlock. We do not need locking
629 for size in ll_update_inode() because size is only updated from
630 the MDS for directories or files without objects, so there is no
631 other place to do the update, and concurrent access to such inodes
632 are protected by the inode lock.
636 Description: resolve "_IOWR redefined" build error on SLES10
640 Description: dump the memory debugging after all modules are unloaded to
641 suppress false negative in conf_sanity test 39
643 Severity : enhancement
645 Description: build kernel-ib packages for OFED 1.3 in our release cycle
650 Description: fix SLES kernel versioning
651 Details : the kernel version for our SLES 10 kernel did not include a "-"
652 before the "smp" at the end. while this was not a problem in
653 general, it did mean that software trying to use the kernel
654 version to try to detect a vendor specific kernel would fail.
655 this was most evident by the OFED build scripts.
659 Description: Don't update lov_desc members until making sure they are valid
660 Details : When updating lov_desc members via proc fs, need fix their
661 validities before doing the real update.
665 Description: don't put request into delay list while invalidate in flight.
666 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
667 import in flight. this produce timeout for invalidate and sometimes
668 can cause stale data.
672 Frequency : on ppc only
673 Description: not convert ost objects for directory because it's not exist.
674 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
678 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
680 Severity : enhancement
682 Description: Update to support 2.6.22.14 vanilla kernel.
686 Frequency : rare, on recovery
687 Description: read procfs can produce deadlock in some situation
688 Details : Holding lprocfs lock which send rpc can produce block for destroy
689 obd objects and this also block reconnect with -EALREADY. This
690 isn't fix all lprocfs bugs - but make it rare.
692 Severity : enhancement
694 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
697 Frequency : frequent on X2 node
699 Description: mdc_set_open_replay_data LBUG
700 Details : Set replay data for requests that are eligible for replay.
704 Description: lustre_mgs: operation 101 on unconnected MGS
705 Details : When MGC is disconnected from MGS long enough, MGS will evict the
706 MGC, and late on MGC cannot successfully connect to MGS and a lot
707 of the error messages complaining that MGS is not connected.
711 Frequency : on network error
712 Description: panic with double free request if network error
713 Details : mdc_finish_enqueue is finish request if any network error ocuring,
714 but it's true only for synchronus enqueue, for async enqueue
715 (via ptlrpcd) this incorrect and ptlrpcd want finish request
718 Severity : enhancement
720 Description: client-side metadata stat-ahead during readdir(directory readahead)
721 Details : perform client-side metadata stat-ahead when the client detects
722 readdir and sequential stat of dir entries therein
725 Frequency : on start mds
727 Description: Implement get_info(last_id) in obdfilter.
730 Frequency : occasional
732 Description: Correctly check stale fid, not start epoch if ost not support SOM
733 Details : open with flag O_CREATE need set old fid in op_fid3 because
734 op_fid2 was overwritten with new generated fid, but mds can answer
735 with one of these two fids and both is not stale. Setattr
736 incorrectly started an epoch and assume will be called
737 done_writing, but without SOM done_writing ever being called.
740 Frequency : rare, depends on device drivers and load
742 Description: MDS or OSS nodes crash due to stack overflow
743 Details : Code changes in 1.8.0 increased the stack usage of some functions.
744 In some cases, in conjunction with device drivers that use a lot
745 of stack the MDS (or possibly OSS) service threads could overflow
746 the stack. One change which was identified to consume additional
747 stack has been reworked to avoid the extra stack usage.
750 Frequency : occasional
752 Description: Do not fail import if osc_interpret_create gets -EAGAIN
753 Details : If osc_interpret_create got -EAGAIN it immediately exits and
754 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
755 oscc_has_objects and see OSC has no objests and call
756 oscc_internal_create to resend create request.
758 Severity : enhancement
760 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
762 Severity : enhancement
764 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
767 Frequency : very rare
769 Description: Fix replay if there is an un-replied request and open
770 Details : In some cases, older replay request will revert the
771 mcd->mcd_last_xid on MDS which is used to record the client's
774 Severity : enhancement
776 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
778 Severity : enhancement
780 Description: Add rhel5 support to HEAD.
782 Serverity : enhancement
784 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
789 Description: Don't allow skipping OSTs if index has been specified.
790 Details : Don't allow skipping OSTs if index has been specified, make
791 locking in internal create lots better.
795 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
796 Details : access to bitfield in structure is always rounded to long
797 and this produce problem with not atomic change any bit.
801 Description: Lustre make rpms failed.
802 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
803 builds Lustre rpms from tarball.
806 Frequency : rare on shutdown ost
808 Description: If llog cancel was not send before clean_exports phase, this can
809 produce deadlock in llog code.
810 Details : If llog thread has last reference to obd and call class_import_put
811 this produce deadlock because llog_cleanup_commit_master wait when
812 last llog_commit_thread exited, but this never success because was
813 called from llog_commit_thread.
817 Description: allow userland application know is lost one of stripes.
818 Details : fill lvb_blocks with error code on ost and return it to
819 application if error flag found.
823 Description: NULL lov_tgts causing MDS oops
824 Details : more safe checks for NULL lov_tgts for avoid oops.
826 Severity : enhancement
828 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
832 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
833 Details : If an OST is down the MDS will hang indefinitely in
834 obd_statfs_rqset() waiting for the statfs data. While for
835 MDS QOS usage of statfs, it should not stuck in waiting.
837 Severity : enhancement
839 Description: remote_acl support
840 Details : Support ACL-based permission check for remote user.
841 Support setfacl/getfacl for remote user with the utils
842 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
843 the system "{s,g}etfacl" utils.
845 Severity : enhancement
847 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
849 Severity : enhancement
851 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
855 Description: make mgs_setparam() handle fsname containing dash
856 Details : fsname containing a dash does not work with lctl conf_param
858 Severity : enhancement
860 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
862 Severity : enhancement
864 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
868 Frequency : rare, at shutdown
869 Description: access already free / zero obd_namespace.
870 Details : if client_disconnect_export was called without force flag set,
871 and exist connect request in flight, this can produce access to
872 NULL pointer (or already free pointer) when connect_interpret
873 store ocd flags in obd_namespace.
877 Frequency : only at startup
878 Description: not alloc memory with spinlock held.
879 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
880 if any spinlock held.
882 Severity : enhancement
884 Description: make lustre randomly fail allocating memory
885 Details : Make lustre randomly failed allocating memory for testing purpose.
887 Severity : enhancement
889 Description: lost problems with lov objid file
890 Details : Fixes some scability and access to not inited memory problems
891 in work with lov objdid file.
896 Description: lfs find does not continue on file error
897 Details : Continue other files processing when a file/dir is absent.
901 Description: Inconsistent usage of lustre_pack_reply()
902 Details : Standardize the usage of lustre_pack_reply() such that it
903 always generate a CERROR on failure.
906 Frequency : occasional
908 Description: MDS hang or stay in waiting lock
909 Details : If client receive lock with CBPENDING flag ldlm need send lock
910 cancel as separate rpc, to avoid situation when cancel request
911 can't processed due all i/o threads stay in wait lock.
915 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
919 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
923 Description: Update to SLES9 kernel-2.6.5-7.287.3.
927 Description: Update to RHEL4 latest kernel.
929 Severity : enhancement
931 Description: Build SLES10 patchless client fails
932 Details : The configure was broken by run ./configure with
933 --with-linux-obj=.... argument for patchless client. When the
934 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
935 can't search header adequately. Use absolute path such as
936 -I($LINUX)/include instead.
940 Description: interrupt oig_wait produce painc on resend.
941 Details : brw_redo_request can be used for resend requests from ptlrpcd and
942 private set, and this produce situation when rq_ptlrpcd_data not
943 copyed to new allocated request and triggered LBUG on assert
944 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
945 ptlrpcd set if request is changed and can be safety changed to use
950 Description: LASSERT_{REQ,REP}SWAB macros are buggy
951 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
952 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
953 these macros and replace them with their difinitions instead.
957 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
958 Details : Update which_patch & target file for SLES10 latest kernel.
961 Description: Remove client patches from SLES 10 kernel.
962 Details : This causes SLES 10 clients to behave as patchless clients
963 even on a Lustre-patched (server) kernel.
965 Severity : enhancement
967 Description: self-adjustable client's lru lists
968 Details : use adaptive algorithm for managing client cached locks lru
969 lists according to current server load, other client's work
970 pattern, memory activities, etc. Both, server and client
971 side namespaces provide number of proc tunables for controlling
974 Severity : enhancement
976 Description: light-weight GSS support
977 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
978 and reduce performance overhead.
980 Severity : enhancement
982 Description: Linux keyring support
983 Details : Support using service of Linux keyring for Lustre GSS internal
984 context refresh/cache mechanism.
988 Description: Fix errors in lfs documentation
989 Details : Fixes man pages
993 Description: don't use GFP_* in generic Lustre code.
994 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
998 Description: obdclass is limited by single OBD_ALLOC(idarray)
999 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
1003 Description: warnings with build patchless client with vanila 2.6.19 and up
1004 Details : change old ctl_table style and replace ctl_table/ctl_table_header
1005 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1009 Frequency : Only for SLES
1010 Description: sanity_quota fail test_1
1011 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1012 file, which makes TSTID[2] inunique.
1016 Frequency : Only for SLES9
1017 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1018 Details : Update target/ChangeLog/which_patch .
1022 Frequency : Only for test_10 in sanity-quota.sh
1023 Description: a bug in quota test code
1024 Details : checking a test flag in a wrong place
1028 Frequency : Only for test_13 in sanity-quota.sh
1029 Description: a bug in quota test script
1030 Details : shouldn't check overfull allocated quota
1034 Description: jbd statistics
1035 Details : Port older jbd statistics patch for sles10
1037 Severity : enhancement
1039 Description: Add printing inode info into message about error in writepage.
1044 Description: reply_lock_interpret crash due to race with it and lock cancel.
1045 Details : Do not replay locks that are being cancelled. Do not reference
1046 locks by their address during replay, just by their handle.
1050 Frequency : When flocks are used.
1051 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1052 Details : Flock locks might destroy just granted lock if it could be merged
1053 with another existing flock, this is done in completion handler,
1054 so teach ldlm_cli_enquque_fini that this is a valid case for
1060 Description: Oops in read and write path when failing to allocate lock.
1061 Details : Check if lock allocation failed and return error back.
1065 Description: lstripe command fails for valid OST index
1066 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1067 instead of lov->desc.ld_active_tgt_count.
1071 Description: sanity.sh failed test 103
1072 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1077 Description: df doesn't work properly if diskfs blocksize != 4K
1078 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1082 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1083 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1085 Severity : enhancement
1087 Description: Adaptive timeouts
1088 Details : RPC timeouts adapt to changing server load and network
1089 conditions to reduce resend attempts and improve recovery time.
1093 Description: llapi_file_create() does not allow some changes
1094 Details : add llapi_file_open() that allows specifying the mode and
1095 open flags, and also returns an open file handle.
1099 Description: merge and cleanup kernel patches.
1100 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1104 Description: Add journal checksum support.(Kernel part)
1105 Details : The journal checksum feature adds two new flags i.e
1106 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1107 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1108 indicates that the commit block contains the checksum for
1109 the blocks described by the descriptor blocks. Now commit
1110 record can be sent to disk without waiting for descriptor
1111 blocks to be written to disk. This behavior is controlled
1112 using JBD2_FEATURE_ASYNC_COMMIT flag.
1116 Description: OSS needs mutliple precreate threads
1117 Details : Add ability to start more than one create thread per OSS.
1121 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1122 Details : execute lfs setstripe on client
1126 Description: mds_obd_create error creating tmp object
1127 Details : When the user sets quota on root, llog will be affected and can't
1128 create files and write files.
1131 Frequency : Always on ia64 patchless client, and possibly others.
1133 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1134 Details : This allows the patchless client to be loaded on architectures
1135 without this export.
1139 Description: RedHat Update kernel for RHEL5
1140 Details : Add the kernel config file for RHEL5.
1144 Description: RedHat Update kernel for RHEL5
1145 Details : Modify the kernel config file more closer RHEL5.
1149 Description: Build failure against Centos5 (RHEL5)
1150 Details : Define PAGE_SIZE when it isn't present.
1154 Description: test_7 on sanity-quota.sh can't be used on separate machines
1155 Details : add facet to handle it
1159 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1160 Details : with NFS, the anon dentry's parent was set to itself in
1161 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1162 corresponding dentry other than use rec->ur_name.
1164 Severity : enhancement
1166 Description: lfs setstripe enhancement
1167 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1171 Description: enable data checksumming by default
1172 Details : enable checksum by default, allow --disable-checksum
1173 configure option and "-o nochecksum" mount option. Checksums
1174 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1178 Description: lustre support for RHEL5
1179 Details : Add support for RHEL5.
1183 Description: Client eviction due to failover config
1184 Details : after a connection loss, the lustre client should attempt to
1185 reconnect to the last active server first before trying the
1186 other potential connections.
1190 Description: when mds and osts use different quota unit(32bit and 64bit),
1191 quota will be released repeatly.
1192 Details : void sending multiple quota reqs to mds, which will keep the
1193 status between the reqs.
1197 Description: osts not allocated evenly to files
1198 Details : change the condition to increase offset_idx
1201 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1203 Description: Data corruption for OSTs that are formatted larger than 2TB
1205 Details : When generating the bio request for lustre file writes the
1206 sector number would overflow a temporary variable before being
1207 used for the IO. The data reads correctly from Lustre (which
1208 will overflow in a similar manner) but other file data or
1209 filesystem metadata may be corrupted in some cases.
1213 Description: Tune the kernel for good SCSI performance.
1214 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1215 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1220 Description: rewrite ext2-derived code in obdclass/uuid.c
1221 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1222 from scratch preserving functionality.
1226 Description: "lfs find -obd UUID" prints directories
1227 Details : "lfs find -obd UUID" will return all directory names instead
1228 of just file names. It is incorrect because the directories
1229 do not reside on the OSTs.
1233 Description: use i_size_read and i_size_write in 2.6 port
1234 Details : replace inode->i_size access with i_size_read/write()
1237 Frequency : when removing large files
1239 Description: scheduling issue during removal of large Lustre files
1240 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1241 It causes scheduling issues when removing large files (17TB in the
1245 Frequency : only with liblustre clients on XT3
1247 Description: evictions taking too long
1248 Details : allow llrd to evict clients directly on OSTs
1251 Frequency : only on ppc
1253 Description: /proc/fs/lustre/devices broken on ppc
1254 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1255 We should make sure the type of 'index' is loff_t and then cast
1256 explicitly as needed below. Do not assign an explicitly cast
1260 Frequency : only for rhel5
1262 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1263 Details : Modify the target file & which_kernel.
1265 Severity : enhancement
1267 Description: omit set fsid for export NFS
1268 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1269 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1273 Frequency : Always, for kernels after 2.6.16
1274 Description: Fix warning idr_remove called for id=.. which is not allocated.
1275 Details : Last kernels save old s_dev before kill super and not allow
1276 to restore from callback - restore it before call kill_anon_super.
1280 Description: buffer overruns could theoretically occur
1281 Details : llapi_semantic_traverse() modifies the "path" argument by
1282 appending values to the end of the origin string, and a
1283 overrun may occur. Adding buffer overrun check in liblustreapi.
1287 Description: Fix error on 'ls .' at the top of the Lustre mount.
1288 Details : Don't revalidate dentry if it is a root dentry.
1292 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1293 Details : Modify vm-tunables-rhel4.patch.
1297 Description: Kernel config for 2.6.18-vanilla.
1298 Details : Modify targets/2.6-vanilla.target.in.
1299 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1300 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1301 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1302 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1306 Description: improve handling recoverable errors
1307 Details : if request processig with error which can be recoverable on server
1308 request should be resend, otherwise page released from cache and
1313 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1314 Details : Modify target file & which_patch.
1315 A flaw was found in the IA32 system call emulation provided
1316 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1317 value could be stored in the %RAX register, which could trigger an
1318 out-of-bounds system call table access. An untrusted local user
1319 could exploit this flaw to run code in the kernel
1320 (ie a root privilege escalation). (CVE-2007-4573).
1324 Description: O_DIRECT bypasses client statistics.
1325 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1326 read_ahead_stats were updated. copied the stats section from
1327 osc_send_oap_rpc() into async_internal().
1331 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1335 Description: change order of libsysio includes
1336 Details : '#include sysio.h' should always come before '#include xtio.h'
1340 Description: handle "serious error: objid * already exists" more gracefully
1341 Details : If LAST_ID value on disk is smaller than the objects existing in
1342 the O/0/d* directories, it indicates disk corruption and causes an
1343 LBUG(). If the object is 0-length, then we should use the existing
1344 object. This will help to avoid a full fsck in most cases.
1346 Severity : enhancement
1348 Description: adapt the lustre_config script to support the upgrade case
1349 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1350 server targets to 1.6 in parallel.
1354 Description: To avoid grant space > avaible space when the disk is almost
1355 full. Without this patch you might see the error "grant XXXX >
1356 available" or some LBUG about grant, when the disk is almost
1358 Details : In filter_check_grant, for non_grant cache write, we should
1359 check the left space by if (*left > ungranted + bytes), instead
1360 of (*left > ungranted), because only we are sure the left space
1361 is enough for another "bytes", then the ungrant space should be
1362 increase. In client, we should update cl_avail_grant only there
1363 is OBD_MD_FLGRANT in the reply.
1367 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1368 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1369 described in CVE-2007-4573. This problem could allow untrusted
1370 local users to gain root access.
1373 Frequency : when using O_DIRECT and quotas
1375 Description: Incorrect file ownership on O_DIRECT output files
1376 Details : block usage reported by 'lfs quota' does not take into account
1377 files that have been written with O_DIRECT.
1382 Description: touch file failed when fs is not full
1383 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1384 otherwise we can get ENOSP while fs is not full.
1388 Description: parallel lock callbacks
1389 Details : Instead of sending blocking and completion callbacks as separated
1390 requests, adding them to a set and sending in parallel.
1393 Frequency : only for Cray XT3
1394 Bugzilla : 12829/13455
1395 Description: Changing primary group doesn't change the group lustre assigns to
1397 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1398 overridden with the primary group provided by the group upcall,
1399 whereas the supplied fsgid can be trusted if it is in the list of
1400 supplementary groups returned by the group upcall.
1402 Severity : enhancement
1404 Description: Allow masking D_WARNING, D_ERROR messages from console
1405 Details : Console messages can now be disabled via lnet.printk.
1409 Description: User code with malformed file open parameter crashes client node
1410 Details : Before packing join_file req, all the related reference should be
1411 checked carefully in case some malformed flags cause fake
1412 join_file req on client.
1416 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1417 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1418 mode is equal to the granted mode. However, it should also check
1419 that LDLM_FL_CP_REQD is not set.
1423 Description: Heavy nfs access might result in deadlocks
1424 Details : After ELC code landed, it is now improper to enqueue any mds
1425 locks under och_sem, because enqueue might want to decide to
1426 cancel open locks for same inode we are holding och_sem for.
1430 Description: Client eviction while running blogbench
1431 Details : A lot of unlink operations with concurrent I/O can lead to a
1432 deadlock causing evictions. To address the problem, the number of
1433 oustanding OST_DESTROY requests is now throttled to
1434 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1435 callbacks are processed in priority.
1439 Description: enable ACLs on MDS by default
1440 Details : ACLs must be enabled on MDS by default.
1443 Frequency : PPC/PPC64 only
1445 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1446 Details : fix duplicated definitions between asm-ppc64/types.h and
1447 lustre_types.h on PPC.
1450 Frequency : PPC/PPC64 only
1452 Description: asm-ppc/segment.h does not exist
1453 Details : fix compile issue on PPC.
1457 Description: better handle error messages in extents code
1460 Frequency : RHEL4 only
1462 Description: mkfs is very slow on IA64/RHEL4
1463 Details : A performance regression has been discovered in the MPT Fusion
1464 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1465 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1466 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1468 Severity : enhancement
1470 Description: SNMP support enhancement
1471 Details : Adding total number of sampled request for an MDS node in snmp
1474 Severity : enhancement
1476 Description: Optimize ldlm waiting list processing for PR extent locks
1477 Details : When processing waiting list for read extent lock and meeting read
1478 lock that is same or wider to it that is not contended, skip
1479 processing rest of the list and immediatelly return current
1480 status of conflictness, since we are guaranteed there are no
1481 conflicting locks in the rest of the list.
1485 Description: Time out and refuse to reconnect
1486 Details : When the failover node is the primary node, it is possible
1487 to have two identical connections in imp_conn_list. We must
1488 compare not conn's pointers but NIDs, otherwise we can defeat
1489 connection throttling.
1493 Description: port llog fixes from b1_6 into HEAD
1494 Details : Port llog reference couting and some llog cleanups from b1_6
1495 (bug 10800) into HEAD, for protect from panic and access to already
1496 free llog structures.
1500 Description: Detect stride IO mode in read-ahead
1501 Details : When a client does stride read, read-ahead should detect that and
1502 read-ahead pages according to the detected stride pattern.
1506 Description: data checksumming impacts single node performance
1507 Details : add support for several checksum algorithm. Currently, only CRC32
1508 and Adler-32 are supported. The checksum type can be changed on
1509 the fly via /proc/fs/lustre/osc/*/checksum_type.
1513 Description: use adler32 for page checksums
1514 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1519 Description: build for x2 fails
1520 Details : fix compile issue on Cray systems.
1524 Description: Properly match for duplicate locks
1525 Details : Due to different lock order from skiplists code, we need to
1526 traverse entire list for now
1529 Frequency : only on PPC/SLES10
1531 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1532 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1533 be defined. Add a hack in mkfs_lustre.c to work around this compile
1538 Description: LASSERT on MDS when client holding flock lock dies
1539 Details : ldlm pool logic depends on number of granted locks equal to
1540 number of released locks which is not true for flock locks, so
1541 just exclude such locks from consideration.
1545 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1546 Details : Use fsfilt_sync() for both the whole filesystem sync and
1547 individual file sync to eliminate dangerous inode locking
1548 with I_LOCK that can lead to a deadlock.
1552 Description: performance in 1.6.3
1553 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1554 sure that 1MB requests are not fragmented by the block layer.
1556 Severity : enhancement
1558 Description: organize the server-side client stats on per-nid basis
1559 Details : Change the structure of stats under obdfilter and mds to
1567 The "uuid"s file would list the uuids of _active_ exports.
1568 And the clear entry is to clear all stats and stale nids.
1570 Severity : enhancement
1572 Description: eliminate client locks in face of contention
1573 Details : file contention detection and lockless i/o implementation
1574 for contended files.
1578 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1581 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1583 Description: MDS may lose file striping (and hence file data) in some cases
1584 Details : If there are additional extended attributes stored on the MDS,
1585 in particular ACLs, SELinux, or user attributes (if user_xattr
1586 is specified for the client mount options) then there is a risk
1587 of attribute loss. Additionally, the Lustre file striping
1588 needs to be larger than default (e.g. striped over all OSTs),
1589 and an additional attribute must be stored initially in the
1590 inode and then increase in size enough to be moved to the
1591 external attribute block (e.g. ACL growing in size) for file
1596 Description: skiplist implementation simplification
1597 Details : skiplists are used to group compatible locks on granted list
1598 that was implemented as tracking first and last lock of each
1599 lock group the patch changes that to using doubly linked lists
1603 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1604 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1605 because LASSERT on both the data supplied by a client, and the
1606 data on disk is dangerous and incorrect.
1608 Severity : enhancement
1610 Description: Slow truncate/writes to huge files at high offsets.
1611 Details : Directly associate cached pages to lock that protect those pages,
1612 this allows us to quickly find what pages to write and remove
1613 once lock callback is received.
1617 Description: more ldlm soft lockups
1618 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1619 starve other threads from the resource lock for a long time in
1620 case of long waiting queue, so change the debug level from
1621 D_OTHER to the less frequently used D_INFO.
1623 Severity : enhancement
1625 Description: add -gid, -group, -uid, -user options to lfs find
1629 Description: Hung threads in invalidate_inode_pages2_range
1630 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1631 once one is completed. As a result, some RPCs are stuck in the
1632 queue and are never sent.
1636 Description: filter threads hungs on waiting journal commit
1637 Details : Cleanup filter group llog code, then only filter group llog will
1638 be only created in the MDS/OST syncing process.
1642 Description: Procfs and llog threads access destoryed import sometimes.
1643 Details : Sync the import destoryed process with procfs and llog threads by
1644 the import refcount and semaphore.
1646 Severity : enhancement
1648 Description: openlock cache of b1_6 port to HEAD
1653 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1654 Details : If insertion of an extent fails, then discard the inode
1655 preallocation and free data blocks else it can lead to duplicate
1660 Description: don't always update ctime in ext3_xattr_set_handle()
1661 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1662 In some cases the ctime should not be updated, for example for
1663 2.0->1.8 compatibility it is necessary to delete an xattr and it
1664 should not update the ctime.
1668 Bugzilla : 15713/16362
1669 Description: Assertion in iopen_connect_dentry in 1.6.3
1670 Details : looking up an inode via iopen with the wrong generation number can
1671 populate the dcache with a disconneced dentry while the inode
1672 number is in the process of being reallocated. This causes an
1673 assertion failure in iopen since the inode's dentry list contains
1674 both a connected and disconnected dentry.
1678 Description: assertion failure in ldlm_handle2lock()
1679 Details : fix a race between class_handle_unhash() and class_handle2object()
1680 introduced in lustre 1.6.5 by bug 13622.
1685 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1686 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1691 Description: oops in sd_iostats_seq_show()
1692 Details : unloading/reloading the scsi low level driver triggers a kernel
1693 bug when trying to access the sd iostat file.
1698 Description: Kernel panics during QLogic driver reload
1699 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1700 patch, causing memory corruption.
1705 Description: journal_dev option does not work in b1_6
1706 Details : pass mount option during pre-mount.
1708 Severity : enhancement
1710 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1711 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1712 extent information of a file. It can be used to map logical blocks
1713 in a file to physical blocks in the block device.
1717 Description: LDLM soft lockups - improvement
1718 Details : It is be possible to send the lock handle along with each read
1719 or write request because the client is already doing a lock match
1720 itself so there isn't any reason the OST should have to re-do that
1724 Frequency : only on Cray X2
1726 Description: X2 build failures
1727 Details : fix build failures on Cray X2.
1731 Description: xid & resent requests
1732 Details : Initialize RPC XID from clock at startup (randomly if clock is
1735 Severity : enhancement
1737 Description: Add lustre_start utility to start or stop multiple Lustre servers
1742 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1743 Details : In case of memory pressure, list_del() can be called twice on
1744 req->rq_history_list, causing a kernel oops.
1748 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1749 Details : fix stack overflow in the distributed lock manager by defering
1750 export eviction after a failed AST to the elt thread instead of
1751 handling it in the dlm interpret routine.
1755 Description: Convert some comments to new format.
1756 Details : Update documenting comments to match doxygen conventions.
1760 Description: Grammar fixes.
1761 Details : A couple of trivial spelling fixes.
1765 Description: OSD_COUNTERS-mandatory
1766 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1767 to simplify the code.
1771 Description: simplify cmm_device freeing logic.
1772 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1776 Description: Add lockdep support to dt_object_operations locking interface.
1777 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
1778 parameter indicating lock ordering. Update mdd code to use new
1783 Description: Introduce failloc constants for lockless IO tests.
1784 Details : Add two new failloc constants to test lockless IO. Only one of
1785 them in implemented---another is checked in yet to be landed
1790 Description: Add lockdep support for inode mutex.
1791 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1792 in the situations where Lustre has to lock more than one inode
1797 Description: Add optional invariants checking support.
1798 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1799 configure switch. This macro is to be used for consistency and
1800 sanity checks that are too expensive to be left in `production'
1805 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1806 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1811 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1812 Details : Use spin_lock_nested() in (the only) situation where more than
1813 one ldlm_lock is locked simultaneously. Also, fix possible
1814 dead-lock in ldlm_lock_change_resource() by enforcing particular
1819 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1820 Details : Instead of specifying each ldlm_lock call-back through separate
1821 parameter, wrap them into struct ldlm_callback_suite.
1825 Description: Kill join_lru obd method and its callers.
1826 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
1827 in memory---a requirement implemented through ->o_join_lru() obd
1828 method in HEAD. Remove this method and its users.
1832 Description: Add asynchronous ldlm ENQUEUE completion handler.
1833 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
1834 case that stock ldlm_completion_ast() cannot handle as it waits
1835 until lock is granted. Introduce new ldlm_completion_ast_async()
1836 for this. Also comment ldlm_completion_ast().
1840 Description: ldlm_error <-> errno conversion.
1841 Details : Add functions to map (rather arbitrary) between LDLM error codes
1842 and standard errno values. CLIO needs this to prevent LDLM specific
1843 constants from escaping ldlm and osc.
1847 Description: Kill unused ldlm_handle2lock_ns() function.
1848 Details : Kill unused ldlm_handle2lock_ns() function.
1852 Description: Add lu_ref support to ldlm_lock
1853 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
1854 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
1855 and ldlm_resource. LDLM interface has to be changed, because code that
1856 releases a reference on a lock, has to "know" what reference this is.
1857 In the most frequent case
1859 lock = ldlm_handle2lock(handle);
1861 LDLM_LOCK_PUT(lock);
1863 no changes are required. When any other reference (received _not_ from
1864 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
1865 instead of LDLM_LOCK_PUT().
1867 Arguably, changes are pervasive, and interface requires some discipline
1868 for proper use. On the other hand, it was very instrumental in finding
1869 a few leaked lock references.
1873 Description: Add ldlm_lock_addref_try().
1874 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
1875 attempts to addref a lock that might be being canceled
1880 Description: Add ldlm_weigh_callback().
1881 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
1882 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
1883 hard-coded `number of pages' logic.
1887 Description: Add lockdep annotations to llog code.
1888 Details : Use appropriately tagged _nested() locking calls in the places
1889 where llog takes more than one ->lgh_lock lock.
1893 Description: Add loi_kms_set().
1894 Details : Wrap kms updates into a helper function.
1898 Description: Constify instances of struct lsm_operations.
1899 Details : Constify instances of struct lsm_operations.
1903 Description: lu_conf support.
1904 Details : On a server, a file system object is uniquely identified
1905 by a fid, which is sufficient to locate and load all object
1906 state (inode). On a client, on the other hand, more data are
1907 necessary instantiate an object. Change lu_object_find() and
1908 friends to take additional `lu_conf' argument describing object.
1909 Typically this includes layout information.
1913 Description: lu_context fixes.
1914 Details : Introduce new lu_context functions that are needed on the client
1915 side, where some system threads (ptlrpcd) are shared by multiple
1916 modules, and so cannot be stopped during module shutdown.
1920 Description: Add start and stop methods to lu_device_type_operations.
1921 Details : Introduce two new methods in lu_device_type_operations, that are
1922 invoked when first instance of a given type is created and last one
1923 is destroyed respectively. This is need by CLIO.
1927 Description: Add lu_ref support to struct lu_device.
1928 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
1929 track leaked references.
1933 Description: Introduce lu_kmem_descr.
1934 Details : lu_kmem_descr and its companion interface allow to create
1935 and destroy a number of kmem caches at once.
1939 Description: Fix lu_object finalization race.
1940 Details : Fix a race between lu_object_find() finding an object and its
1941 concurrent finalization. This race is (most likely) not possible
1942 on the server, but might happen on the client.
1946 Description: Introduce lu_ref interface.
1947 Details : lu_ref is a debugging module allowing to track references to
1948 a given object. It is quite cpu expensive, and has to be
1949 explicitly enabled with --enable-lu_ref. See usage description
1954 Description: Factor lu_site procfs stats into a separate function.
1955 Details : Separate lu_site stats printing code into a separate function
1956 that can be reused on a client.
1960 Description: Constify instances of struct {lu,dt,md}_device_operations.
1961 Details : Constify instances of struct {lu,dt,md}_device_operations.
1965 Description: Introduce struct md_site and move meta-data specific parts of
1966 struct lu_site here.
1967 Details : Move md-specific fields out of struct lu_site into special struct
1968 md_site, so that lu_site can be used on a client.
1972 Description: Kill mdd_lov_destroy().
1973 Details : Remove unused mdd code.
1977 Description: Add st_block checking to multistat.c.
1978 Details : Add st_block checking to multistat.c.
1982 Description: Add lu_ref support to struct obd_device.
1983 Details : Add lu_ref tracking to obd_device.
1987 Description: Kill obd_set_fail_loc().
1988 Details : Remove unused code.
1992 Description: Add special type for ptlrpc_request interpret functions.
1993 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
1994 there. Actual usage will be in CLIO.
1998 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
1999 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
2000 doesn't work with lockdep.
2004 Description: Add rwv.c test program.
2005 Details : New testing program exercising readv(2) and writev(2) (Qian).
2009 Description: Add sendfile.c test program.
2010 Details : New testing program exercising sendfile(2) (Jay).
2014 Description: Ratelimit a message that can be very frequent.
2015 Details : Ratelimit a memory allocation failure message that can
2020 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2021 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2025 Description: Liblustre build fixes.
2026 Details : Liblustre build fixes.
2030 Description: libcfs: add cfs_{need,cond}_resched() interface.
2031 Details : libcfs: add cfs_{need,cond}_resched() definition and
2032 implementations for Linux, NT, and liblustre.
2034 Severity : enhancement
2036 Description: More exported tunables for mballoc
2037 Details : Add support for tunable preallocation window and new tunables for
2038 large/small requests
2042 Description: Detect corruption of block bitmap and checking for preallocations
2043 Details : Checks validity of on-disk block bitmap. Also it does better
2044 checking of number of applied preallocations. When corruption is
2045 found, it turns filesystem readonly to prevent further corruptions.
2049 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2050 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2051 window to 0. If it does hit cache miss, and read-pattern is still
2052 stride-io mode, does not reset the stride window, but also does not
2053 increase the stride window length in this case.
2057 Frequency : only for big-endian servers
2058 Description: Check if system is big-endian while mounting fs with extents feature
2059 Details : Mounting a filesystem with extents feature will fail on big-endian
2060 systems since ext3-based ldiskfs is not supported on big-endian
2061 systems. This can be over-riden with "bigendian_extents" mount option.
2063 Severity : enhancement
2065 Description: The root squash functionality
2066 Details : A security feature, which is to prevent users from being able
2067 to mount lustre on their desktop, run as root, and delete
2068 all of the files in the filesystem. The goal is accomplished by
2069 remapping user id (UID) and group id (GID) of the root user to
2070 a UID and GID specified by the system administartor via Lustre
2071 configuration management server (MGS). The functionality also
2072 allows to specify sets of clients for which the remapping does
2077 Description: Excessive recovery window
2078 Details : With AT enabled, the recovery window can be excessively long (6000+
2079 seconds). To address this problem, we no longer use
2080 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2081 timeout no longer depends on the service time, it is set to
2082 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2083 time via pb_service_time.
2087 Description: Watchdog triggered on MDS failover
2088 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2089 the OSTs know that the MDS "UUID" can be reused for the same export
2090 from a different NID, so we do not need to wait for the export to be
2094 Frequency : rare, only if using MMP with Linux RAID
2096 Description: MMP doesn't work with Linux RAID
2097 Details : While using HA for Lustre servers with Linux RAID, it is possible
2098 that MMP will not detect multiple mounts. To make this work we
2099 need to unplug the device queue in RAID when the MMP block is being
2100 written. Also while reading the MMP block, we should read it from
2101 disk and not the cached one.
2103 Severity : enhancement
2105 Description: open file using fid
2106 Details : A file can be opened using just its fid, like
2107 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2110 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2113 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2114 Details : add proper locking for creation of jbd_4k slab cache
2118 Description: MMP check in ext3_remount() fails without displaying any error
2119 Details : When multiple mount protection fails during remount, proper error
2122 Severity : enhancement
2124 Description: Allow stripe size to be up to 4G-64k
2125 Details : Fix math logic to allow large stripe sizes.
2129 Description: add check for >8TB ldiskfs filesystems
2130 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2131 Don't allow >8TB ldiskfs filesystems to be mounted without
2132 force_over_8tb mount option
2134 --------------------------------------------------------------------------------
2136 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2138 * Support for kernels:
2139 2.6.5-7.283 (SLES 9),
2140 2.6.9-55.EL (RHEL 4),
2141 2.6.16.46-0.14 (SLES 10),
2142 2.6.18.8 vanilla (kernel.org)
2143 * Client support for unpatched kernels:
2144 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2145 2.6.16 - 2.6.22 vanilla (kernel.org)
2146 * Due to recently discovered recovery problems, we do not recommend
2147 using patchless RHEL 4 clients with this or any earlier release.
2148 * Recommended e2fsprogs version: 1.39.cfs8
2149 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2150 * Starting with this release, the ldiskfs backing filesystem required
2151 by Lustre is now in its own package, lustre-ldiskfs. This package
2152 should be installed. It is versioned separately from Lustre and
2153 may be released separately in future.
2157 Description: block reactivating mgc import until all deactivates complete
2158 Details : Fix race when failing back MDT/MGS to itself (testing)
2161 Frequency : at statup only
2163 Description: mds_lov_synchronize race leads to various problems
2164 Details : simultaneous MDT->OST connections at startup can cause the
2165 sync to abort, leaving the OSC in a bad state.
2167 Severity : enhancement
2169 Description: add optional extra BUILD_VERSION info
2170 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2171 to override the lustre version.
2174 Frequency : 2.6.18 servers only
2176 Description: ll_kern_mount() doesn't release the module reference
2177 Details : The ldiskfs module reference count never drops down to 0
2178 because ll_kern_mount() doesn't release the module reference.
2183 Description: server LBUG when using old ost_num_threads parameter
2184 Details : Accept the old ost_num_threads parameter but warn that it
2185 is deprecated, and fix an off-by-one error that caused an LBUG.
2190 Description: Transient SCSI error results in persistent IO issue
2191 Details : iobuf->dr_error is not reinitialized to 0 between two
2195 Frequency : sometimes when underlying device returns I/O errors
2197 Description: OSTs not going read-only during write failures
2198 Details : OSTs are not remounted read-only when the journal commit threads
2199 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2200 instead of the ext3 wrappers.
2204 Description: poor connect scaling with increasing client count
2205 Details : Don't run filter_grant_sanity_check for more than 100 exports
2206 to improve scaling for large numbers of clients.
2209 Frequency : SLES10 only
2211 Description: sanity-quota.sh quotacheck failed: rc = -22
2212 Details : Quotas cannot be enabled on SLES10.
2215 Frequency : liblustre clients only
2217 Description: getdirentries does not give error when run on compute nodes
2218 Details : getdirentries does not fail when the size specified as an argument
2219 is too small to contain at least one entry
2221 Severity : enhancement
2223 Description: Add LNET router traceability for debug purposes
2224 Details : If a checksum failure occurs with a router as part of the
2225 IO path, the NID of the last router that forwarded the bulk data
2226 is printed so it can be identified.
2231 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2232 Details : Due to a race condition, liblustre clients were occasionally
2233 evicted incorrectly.
2235 Severity : enhancement
2237 Description: lfs setstripe use optional parameters instead of postional
2240 Severity : enhancement
2242 Description: Nanosecond timestamp support for ldiskfs
2243 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2244 resolution timestamps. There is not yet support for this at
2245 the Lustre filesystem level.
2248 Frequency : during server recovery
2250 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2251 Details : request with rq_no_resend flag not awake l_wait_event if they get
2255 Frequency : nfs export on patchless client
2257 Description: connectathon hang when test nfs export over patchless client
2258 Details : Disconnected dentry cannot be found with lookup, so we do not need
2259 to unhash it or make it invalid
2262 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2263 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2264 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2269 Description: open req refcounting wrong on reconnect
2270 Details : If reconnect happened between getting open reply from server and
2271 call to mdc_set_replay_data in ll_file_open, we will schedule
2272 replay for unreferenced request that we are about to free.
2273 Subsequent close will crash in variety of ways.
2274 Check that request is still eligible for replay in
2275 mdc_set_replay_data().
2280 Description: disable writes to filesystem when reading health_check file
2281 Details : the default for reading the health_check proc file has changed
2282 to NOT do a journal transaction and write to disk, because this
2283 can cause reads of the /proc file to hang and block HA state
2284 checking on a healthy but otherwise heavily loaded system. It
2285 is possible to return to the previous behaviour during configure
2286 with --enable-health-write.
2288 Severity : enhancement
2290 Description: 64-bit inode version
2291 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2292 to the inode. This will be required for version-based recovery.
2297 Description: MDS fails to start if a duplicate client export is detected
2298 Details : in some rare cases it was possible for a client to connect to
2299 an MDS multiple times. Upon recovery the MDS would detect this
2300 and fail during startup. Handle this more gracefully.
2302 Severity : enhancement
2304 Description: Add -o localflock option to simulate old noflock
2306 Details : This will achieve local-only flock/fcntl locks
2312 Description: log_commit_thread vs filter_destroy race leads to crash
2313 Details : Take import reference before releasing llog record semaphore
2318 Description: Wrong request locking in request set processing
2319 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2320 imp_delayed_list, in this place should be used imp_lock.
2323 Frequency : when reconnection
2325 Description: Grant Leak when osc reconnect to OST
2326 Details : When osc reconnect ost, OST(filter) should check whether it
2327 should grant more space to client by comparing fed_grant and
2328 cl_avail_grant, and return the granted space to client instead
2329 of "new granted" space, because client will call osc_init_grant
2330 to update the client grant space info.
2333 Frequency : when client reconnect to OST
2335 Description: Grant Leak when osc do resend and replay bulk write
2336 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2337 bulk write request, because the grant info while be sync between
2338 OSC and OST when reconnect, and we should ignore the grant info
2339 these of resend/replay write req.
2344 Description: Grant space more than avaiable left space sometimes.
2345 Details : When then OST is about to be full, if two bulk writing from
2346 different clients came to OST. Accord the avaliable space of the
2347 OST, the first req should be permitted, and the second one
2348 should be denied by ENOSPC. But if the seconde arrived before
2349 the first one is commited. The OST might wrongly permit second
2350 writing, which will cause grant space > avaiable space.
2353 Frequency : when client is evicted
2355 Description: Grant might be wrongly erased when osc is evicted by OST
2356 Details : when the import is evicted by server, it will fork another
2357 thread ptlrpc_invalidate_import_thread to invalidate the
2358 import, where the grant will be set to 0. While the original
2359 thread will update the grant it got when connecting. So if
2360 the former happened latter, the grant will be wrongly errased
2361 because of this race.
2366 Description: Checking Stale with correct fid
2367 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2368 to check whether it is stale, because sometimes, we want the
2369 enqueue happened anyway, and op_data.fid2 will not be initialized.
2371 Severity : enhancement
2373 Description: update patchless client
2374 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2377 Frequency : only with 2.4 kernel
2379 Description: random memory corruption
2380 Details : size of struct ll_inode_info is to big for union inode.u and this
2381 can be cause of random memory corruption.
2386 Description: Memory leak in recovery
2387 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2388 It should also check obd_fail before fsfilt_start, otherwise if
2389 fsfilt_start return -EROFS,(failover mds during mds recovery).
2390 then the req will return with repmsg->transno = 0 and rc = EROFS.
2391 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2392 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2393 be freed no matter whether fsfilt_commit success or not.
2396 Frequency : only with huge count clients
2398 Description: Prevents from taking the superblock lock in llap_from_page for
2400 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2401 for llap_from_page call in ll_removepage prevents from taking the
2402 superblock lock for a soon died page.
2407 Description: Not check open intent error before release open handle
2408 Details : in some rare cases, the open intent error is not checked before
2409 release open handle, which may cause
2410 ASSERTION(open_req->rq_transno != 0), because it tries to release
2411 the failed open handle.
2416 Description: Set cat log bitmap only after create log success.
2417 Details : in some rare cases, the cat log bitmap is set too early. and it
2418 should be set only after create log success.
2422 Description: Accessing a block bevice can re-enable I/O when Lustre is
2423 tearing down a device.
2424 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2429 Description: service threads may hog cpus when there are a lot of requests
2430 Details : Insert cond_resched to give other threads a chance to use some CPU
2435 Description: the cat log was not initialized in recovery
2436 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2437 unlink log on mds will not be initialized until mds post
2438 recovery. And also in mds post recovery, the unlink log will
2439 initialization will be done asynchronausly, so there will be race
2440 between add unlink log and unlink log initialization.
2444 Description: brw_stats were being printed incorrectly
2445 Details : brw_stats were being printed as log2 but all of them were not
2446 recorded as log2. Also remove some code duplication arising from
2447 filter_tally_{read,write}.
2451 Frequency : rare, only in recovery.
2452 Description: ASSERTION(req->rq_type != LI_POISON) failed
2453 Details : imp_lock should be held while iterating over imp_sending_list for
2454 prevent destroy request after get timeout in ptlrpc_queue_wait.
2458 Description: replay-single.sh test 52 fails
2459 Details : A lock's skiplist need to be cleanup when it being unlinked
2460 from its resource list.
2464 Description: Short directio read returns full requested size rather than
2466 Details : Direct I/O operations should return actual amount of bytes
2467 transferred rather than requested size.
2469 Severity : enhancement
2471 Description: metadata RPC reduction (e.g. for rm performance)
2472 Details : decrease the amount of synchronous RPC between clients and servers
2473 by canceling conflicing lock before the operation on the client
2474 and packing thier handles into the main operation RPC to server.
2476 Severity : enhancement
2478 Description: Async OSC create to avoid the blocking unnecessarily.
2479 Details : If a OST has no remain object, system will block on the creating
2480 when need to create a new object on this OST. Now, ways use
2481 pre-created objects when available, instead of blocking on an
2482 empty osc while others are not empty. If we must block, we block
2483 for the shortest possible period of time.
2487 Description: improve handling recoverable errors
2488 Details : if request processig with error which can be recoverable on server
2489 request should be resend, otherwise page released from cache and
2492 Severity : enhancement
2494 Description: refine locking for avoid write wrong info into lov_objid file
2495 Details : fix possible races with add new target and write/update data in
2498 --------------------------------------------------------------------------------
2500 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2505 Frequency : on some architectures
2507 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2508 Details : Uninitialized flags sometimes cause configuration commands to
2512 Frequency : patchless clients only
2514 Description: missing __iget() symbol export
2515 Details : The __iget() symbol export is missing. To avoid the need for
2516 this on patchless clients the deathrow inode reaper is turned
2517 off, and we depend on the VM to clean up old inodes. This
2518 dependency was during via the fix for bug 12181.
2520 --------------------------------------------------------------------------------
2522 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2524 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2525 INTEROPERATE with older versions automatically. In many cases a
2526 special upgrade step is needed. Please read the
2527 user documentation before upgrading any part of a live system.
2528 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2529 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2530 * WARNING: Lustre configuration and startup changes are required with
2531 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2535 Severity : enhancement
2537 Description: MountConf
2538 Details : Lustre configuration is now managed via mkfs and mount
2539 commands instead of lmc and lconf. New obd types (MGS, MGC)
2540 are added for dynamic configuration management. See
2541 https://mail.clusterfs.com/wikis/lustre/MountConf for
2544 Severity : enhancement
2546 Description: dynamic OST addition
2547 Details : OSTs can now be added to a live filesystem
2549 Severity : enhancement
2551 Description: startup order invariance
2552 Details : MDTs and OSTs can be started in any order. Clients only
2553 require the MDT to complete startup.
2555 Severity : enhancement
2557 Description: parallel, asynchronous orphan cleanup
2558 Details : orphan cleanup is now performed in separate threads for each
2559 OST, allowing parallel non-blocking operation.
2561 Severity : enhancement
2563 Description: optimized stripe assignment
2564 Details : stripe assignments are now made based on ost space available,
2565 ost previous usage, and OSS previous usage, in order to try
2566 to optimize storage space and networking resources.
2568 Severity : enhancement
2570 Description: Permanently set tunables
2571 Details : All writable /proc/fs/lustre tunables can now be permanently
2572 set on a per-server basis, at mkfs time or on a live system.
2574 Severity : enhancement
2576 Description: Lustre message v2
2577 Details : Add lustre message format v2.
2579 Severity : enhancement
2581 Description: client OST exclusion list
2582 Details : Clients can be started with a list of OSTs that should be
2583 declared "inactive" for known non-responsive OSTs.
2587 Description: SPEC SFS validation failure on NFS v2 over lustre.
2588 Details : Changes the blocksize for regular files to be 2x RPC size,
2589 and not depend on stripe size.
2591 Severity : enhancement
2593 Description: Multiple MD RPCs in flight.
2594 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2595 To avoid overly-overloading MDS, introduce a limit on number of
2596 MDS RPCs in flight for a single client and add /proc controls
2597 to adjust this limit.
2599 Severity : enhancement
2601 Description: client read/write statistics
2602 Details : Add client read/write call usage stats for performance
2603 analysis of user processes.
2604 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2605 file access. extents_stats shows chunk size distribution.
2606 extents_stats_per_process show chunk size distribution per
2609 Severity : enhancement
2611 Description: mds statistics
2612 Details : Add detailed mds operations statistics in
2613 /proc/fs/lustre/mds/*/stats.
2617 Description: Failure of copying files with lustre special EAs.
2618 Details : Client side always return success for setxattr call for lustre
2619 special xattr (currently only "trusted.lov").
2624 Description: Refcount LNET uuids
2625 Details : The global LNET uuid list grew linearly with every startup;
2626 refcount repeated list entries instead of always adding to
2629 Severity : enhancement
2631 Description: Dynamic service threads
2632 Details : Within a small range, start extra service threads
2633 automatically when the request queue builds up.
2636 Frequency : mixed-endian client/server environments
2638 Description: mixed-endian crashes
2639 Details : The new msg_v2 system had some failures in mixed-endian
2642 Severity : enhancement
2644 Description: Easy OST removal
2645 Details : OSTs can be permanently deactivated with e.g. 'lctl
2646 conf_param lustre-OST0001.osc.active=0'
2648 Severity : enhancement
2650 Description: MGS proc entries
2651 Details : Added basic proc entries for the MGS showing what filesystems
2654 Severity : enhancement
2656 Description: provide MGS failover
2657 Details : Added config lock reacquisition after MGS server failover.
2659 Severity : enhancement
2661 Description: add Linux 2.4 support
2662 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2666 Description: a large application tries to do I/O to the same resource and dies
2667 in the middle of it.
2668 Details : Check the req->rq_arrival time after the call to
2669 ost_brw_lock_get(), but before we do anything about
2670 processing it & sending the BULK transfer request. This
2671 should help move old stale pending locks off the queue as
2672 quickly as obd_timeout.
2675 Frequency : when an incorrect nid is specified during startup
2677 Description: ptlrpc connect to non-existant node causes kernel crash
2678 Details : LNET can't be re-entered from an event callback, which
2679 happened when we expire a message after the export has been
2680 cleaned up. Instead, hand the zombie cleanup off to another
2683 Severity : enhancement
2685 Description: plain/inodebits lock performance improvement
2686 Details : Grouping plain/inodebits in granted list by their request modes
2687 and bits policy, thus improving the performance of search through
2691 Frequency : only if OST filesystem is corrupted
2693 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2694 Details : for a short time RPCs with bulk IO are in the replay list,
2695 but replay of bulk IOs is unimplemented. If the OST filesystem
2696 is corrupted due to disk cache incoherency and then replay is
2697 started it is possible to trip an assertion. Avoid putting
2698 committed RPCs into the replay list at all to avoid this issue.
2701 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2703 Description: System hang on startup
2704 Details : This bug allowed the liblustre (e.g. catamount) client to
2705 return to the app before handling all startup RPCs. This
2706 could leave the node unresponsive to lustre network traffic
2707 and manifested as a server ptllnd timeout.
2709 Severity : enhancement
2711 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2712 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2713 boolean to control whether to print peer debug info when a
2714 client's RPC times out.
2717 Frequency : only for kernels with patches from Lustre below 1.4.3
2719 Description: Remove old rdonly API
2720 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2723 Frequency : only for devices with external journals
2725 Description: Set external device read-only also
2726 Details : During a commanded failover stop, we set the disk device
2727 read-only while the server shuts down. We now also set any
2728 external journal device read-only at the same time.
2731 Frequency : when upgrading from 1.4 while trying to change parameters
2733 Description: The wrong (new) MDC name was used when setting parameters for
2734 upgraded MDT's. Also allows changing of OSC (and MDC)
2735 parameters if --writeconf is specified at tunefs upgrade time.
2738 Frequency : when setting specific ost indicies
2740 Description: QOS code breaks on skipped indicies
2741 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2742 created with --index need not be sequential.
2744 Severity : enhancement
2746 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2747 Details : The uninit_groups feature works in conjunction with the kernel
2748 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2749 up the pass1 processing of e2fsck. This is a read-only feature
2750 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2751 mount filesystems that have had this feature enabled.
2753 Severity : enhancement
2755 Description: Improve multi-block allocation algorithm to avoid fragmentation
2756 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2757 allocation locality and avoid filesystem fragmentation.
2759 ------------------------------------------------------------------------------
2761 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2763 * Support for kernels:
2764 2.4.21-47.0.1.EL (RHEL 3)
2765 2.6.5-7.283 (SLES 9)
2766 2.6.9-42.0.10.EL (RHEL 4)
2767 2.6.12.6 vanilla (kernel.org)
2768 2.6.16.27-0.9 (SLES 10)
2769 * Recommended e2fsprogs version: 1.39.cfs5
2771 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2775 Frequency : occasional, depends on client load and configuration
2776 Bugzilla : 12181, 12203
2777 Description: data loss for recently-modified files
2779 Details : In some cases it is possible that recently written or created
2780 files may not be written to disk in a timely manner (this should
2781 normally be within 30s unless client IO load is very high).
2782 The problem appears as zero-length files or files that are a
2783 multiple of 1MB in size after a client crash or client eviction
2784 that are missing data at the end of the file.
2786 This problem is more likely to be hit on clients where files are
2787 repeatedly created and unlinked in the same directory, clients
2788 have a large amount of RAM, have many CPUs, the filesystem has
2789 many OSTs, the clients are rebooted frequently, and/or the files
2790 are not accessed by other nodes after being written.
2792 The presence of the problem can be detected by looking at
2793 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2794 smaller than the second (nr_unused) then dirty files will not
2795 be flushed automatically to disk. "sync; sleep 10" should be
2796 run several times on the node before unmounting it to update
2797 Lustre (this is also safe to run on nodes without this problem).
2799 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2800 that can cause this same problem, so customers using that kernel
2801 also need to update the kernel in addition to Lustre. In order
2802 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2804 It is normal that files written just before a client crash (less
2805 than 30s) may not yet have been flushed to disk, even for local
2809 Frequency : frequent on thin XT3 nodes
2811 Description: UUID collision on thin XT3 Linux nodes
2812 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2813 randomly, since we relied on an insufficiently-seeded PRNG.
2818 Description: OSS hangs after "All ost request buffers busy"
2819 Details : A deadlock between quota and journal operations caused OSS
2820 hangs after printing "All ost request buffers busy."
2823 Frequency : always on liblustre builds
2825 Description: Cleanup compiler warnings on liblustre
2828 Frequency : always on liblustre builds on XT3
2830 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2834 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2835 > 32000 subdirectories
2836 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2837 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2838 subdirectory count crosses 32000. This will aid e2fsck to
2839 correctly handle more than 32000 subdirectories.
2842 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2844 Description: System hang on startup
2845 Details : This bug allowed the liblustre (e.g. catamount) client to
2846 return to the app before handling all startup RPCs. This
2847 could leave the node unresponsive to lustre network traffic
2848 and manifested as a server ptllnd timeout.
2850 Severity : enhancement
2852 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2853 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2854 boolean to control whether to print peer debug info when a
2855 client's RPC times out.
2860 Description: make O_SYNC working on 2.6 kernels
2861 Details : 2.6 kernels use different method for mark pages for write,
2862 so need add a code to lustre for O_SYNC work.
2867 Description: Failure to close file and release space on NFS
2868 Details : Put inode details into lock acquired in ll_intent_file_open.
2869 Use mdc_intent_lock in ll_intent_open to properly
2870 detect all kind of errors unhandled by mdc_enqueue.
2875 Description: proc file read during shutdown sometimes raced obd removal,
2877 Details : Add lock to prevent obd access after proc file removal.
2880 Frequency : Only for files larger than 4GB on 32-bit clients.
2882 Description: improperly doing page alignment of locks
2883 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2884 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2889 Description: RPCs being resent when they shouldn't be
2890 Details : Some RPCs that should not be resent are being resent. This
2891 can cause inconsistencies in the RPC state machine. Do not
2892 resend such requests.
2895 Frequency : rare, only with NFS export
2897 Description: Crash on NFS re-export node
2898 Details : under very unusual load conditions an assertion is hit in
2899 ll_intent_file_open()
2902 Frequency : only if OST filesystem is corrupted
2904 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2905 Details : for a short time RPCs with bulk IO are in the replay list,
2906 but replay of bulk IOs is unimplemented. If the OST filesystem
2907 is corrupted due to disk cache incoherency and then replay is
2908 started it is possible to trip an assertion. Avoid putting
2909 committed RPCs into the replay list at all to avoid this issue.
2914 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2915 Details : Large single O_DIRECT read and write calls can fail to allocate
2916 a sufficiently large buffer to process the request. In case of
2917 allocation failure the allocation is retried with a smaller
2918 buffer and broken into smaller requests.
2920 Severity : enhancement
2922 Description: Add -o localflock option to simulate old noflock behaviour.
2923 Details : This will achieve local-only flock/fcntl locks coherentness.
2928 Description: versioning check is incomplete
2929 Details : Checking the version difference of client vs. server, report
2930 error if the gap is too big.
2935 Description: add support PG_writeback bit
2936 Details : add support for PG_writeback bit for Lustre, for more carefull
2937 work with page cache in 2.6 kernel. This also fix some deadlocks
2938 and remove hack for work O_SYNC with 2.6 kernel.
2940 Severity : enhancement
2942 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2943 Details : The uninit_groups feature works in conjunction with the kernel
2944 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2945 up the pass1 processing of e2fsck. This is a read-only feature
2946 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2947 mount filesystems that have had this feature enabled.
2949 Severity : enhancement
2951 Description: Improve multi-block allocation algorithm to avoid fragmentation
2952 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2953 allocation locality and avoid filesystem fragmentation.
2955 ------------------------------------------------------------------------------
2957 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2959 * Support for kernels:
2960 2.6.9-42.0.3.EL (RHEL 4)
2961 2.6.5-7.276 (SLES 9)
2962 2.4.21-47.0.1.EL (RHEL 3)
2963 2.6.12.6 vanilla (kernel.org)
2964 2.6.16.21-0.8 (SLES10)
2965 * Recommended e2fsprogs version: 1.39.cfs2-0
2967 * The backwards-compatible /proc/sys/portals symlink has been removed
2968 in this release. Before upgrading, please ensure that you change
2969 any configuration scripts or /etc/sysctl.conf files that access
2970 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2971 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2972 in advance of the upgrade on any system running Lustre 1.4.6 or
2973 newer, since /proc/sys/lnet was added in that version.
2974 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2978 Frequency : only when quota is used
2980 Description: avoid scanning export list for quota master
2981 Details : Change the algorithms to avoid scanning export list in order
2982 to improve the efficiency.
2985 Frequency : MDS failover only, very rarely
2987 Description: "went back in time" messages on mds failover
2988 Details : The greatest transno may be lost when the current operation
2989 finishes with an error (transno==0) and the client's last_rcvd
2990 record is over-written. Save the greatest transno in the
2991 mds_last_transno for this case.
2994 Frequency : always for specific kernels and striping counts
2996 Description: client may get "Matching packet too big" without ACL support
2997 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2998 when trying to access files in certain configurations. The
2999 clients should in fact be denied when mounting because they do
3000 not understand ACLs.
3003 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3005 Description: many clients connecting with IO in progress causes connect timeouts
3006 Details : Avoid synchronous journal commits to avoid delays caused by many
3007 clients connecting/disconnecting when bulk IO is in progress.
3008 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3009 OST_IO_PORTAL to avoid delays behind potentially many pending
3013 Frequency : occasionally with multiple writers to a single file
3015 Description: shared writes to file may result in wrong size reported by stat()
3016 Details : Allow growing of kms when extent lock is cancelled
3019 Frequency : always with random mmap IO to multi-striped file
3021 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3022 Details : Only if the hole is at the end of OST object so that kms is too
3023 small. Fix is to increase kms accordingly in ll_nopage.
3026 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3028 Description: writes to a missing object would leak memory on the OST
3029 Details : If there is an inconsistency between the MDS and OST filesystems,
3030 such that the MDS references an object that doesn't exist, writes
3031 to that object will leak memory due to incorrect cleanup in the
3032 error handling path, eventually running out of memory on the OST.
3037 Description: Creating too long symlink causes lustre errors
3038 Details : Check symlink and name lengths before sending requests to MDS.
3041 Frequency : only if flock is enabled (not on by default)
3043 Description: posix locks not released on fd closure on 2.6.9+
3044 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3045 kernels, this caused such locks not to be released on fd close and
3046 then assertions on fs unmount about still used locks.
3049 Frequency : MDS failover only, very rarely
3051 Description: clients may get ASSERTION(granted_lock != NULL)
3052 Details : When request was taking a long time, and a client was resending
3053 a getattr by name lock request. The were multiple lock requests
3054 with the same client lock handle and
3055 mds_getattr_name->fixup_handle_for_resent_request found one of the
3056 lock handles but later failed with ASSERTION(granted_lock != NULL).
3061 Description: handle->h_buffer_credits > 0, assertion failure
3062 Details : h_buffer_credits is zero after truncate, causing assertion
3063 failure. This patch extends the transaction or creates a new
3067 Frequency : NFS re-export or patchless client
3068 Bugzilla : 11179, 10796
3069 Description: Crash on NFS re-export node (__d_move)
3070 Details : We do not want to hash the dentry if we don't have a lock.
3071 But if this dentry is later used in d_move, we'd hit uninitialised
3072 list head d_hash, so we just do this to init d_hash field but
3073 leave dentry unhashed.
3076 Frequency : NFS re-export or patchless client
3078 Description: NFS exports has problem with symbolic link
3079 Details : lustre client didn't properly install dentry when re-exported
3080 to NFS or running patchless client.
3083 Frequency : NFS re-export or patchless client
3085 Description: Various nfs/patchless fixes.
3086 Details : fixes reuse disconected alias for lookup process - this fixes
3087 warning "find_exported_dentry: npd != pd",
3088 fix permission error with open files at nfs.
3089 fix apply umask when do revalidate.
3092 Frequency : occasional
3094 Description: Crash on NFS re-export node
3095 Details : calling clear_page() on the wrong pointer triggered oops in
3096 generic_mapping_read().
3099 Frequency : rarely, using O_DIRECT IO
3101 Description: unaligned directio crashes client with LASSERT
3102 Details : check for unaligned buffers before trying any requests.
3105 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3107 Description: stale data returned from RAID cache
3108 Details : If only a small amount of IO is done to the RAID device before
3109 reading it again it is possible to get stale data from the RAID
3110 cache instead of reading it from disk.
3113 Frequency : always for sles10 kernel
3115 Description: sles10 support
3116 Details : ll_follow_link: compile fixes and using of nd_set_link
3117 under newer kernels.
3120 Frequency : depends on arch, kernel and compiler version, always on sles10
3123 Description: recursive or deep enough symlinks cause stack overflow
3124 Details : getting rid of large stack-allocated variable in
3128 Frequency : depends on hardware
3130 Description: lustre write performance loss in the SLES10 kernel
3131 Details : the performance loss is caused by using of write barriers in the
3132 ext3 code. The SLES10 kernel turns barrier support on by
3133 default. The fix is to undo that change for ldiskfs.
3135 ------------------------------------------------------------------------------
3137 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3139 * Support for kernels:
3140 2.6.9-42.0.3EL (RHEL 4)
3141 2.6.5-7.276 (SLES 9)
3142 2.4.21-47.0.1.EL (RHEL 3)
3143 2.6.12.6 vanilla (kernel.org)
3147 Frequency : quota enabled and large files being deleted
3149 Description: releasing more than 4GB of quota at once hangs OST
3150 Details : If a user deletes more than 4GB of files on a single OST it
3151 will cause the OST to spin in an infinite loop. Release
3152 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3157 Description: statfs data retrieved from /proc may be stale or zero
3158 Details : When reading per-device statfs data from /proc, in the
3159 {kbytes,files}_{total,free,avail} files, it may appear
3160 as zero or be out of date.
3163 Frequency : systems with MD RAID1 external journal devices
3165 Description: lconf's call to blkid is confused by RAID1 journal devices
3166 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3167 of returning all block devices that match the journal UUID.
3170 Frequency : always, for aggregate stripe size over 4GB
3172 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3173 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3174 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3177 Frequency : always if "lfs find" used on a local file/directory
3179 Description: "lfs find" segfaults if used on a local file/directory
3180 Details : The case where a directory component was not specified wasn't
3181 handled correctly. Handle this properly.
3184 Frequency : always on ppc64
3186 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3187 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3188 perform endianness conversion.
3191 Frequency : rarely (truncate to non-zero file size after write under load)
3192 Bugzilla : 10730, 10687
3193 Description: Files padded with zeros to next 4K multiple
3194 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3195 it is possible that files that are truncated to a non-zero size
3196 immediately after being written are filled with zero bytes beyond
3197 the truncated size. No file data is lost.
3199 Severity : enhancement
3201 Description: Allow recovery/failover for liblustre clients.
3202 Details : liblustre clients were unaware of failover configurations until
3205 Severity : enhancement
3207 Description: user file locks should fail when not mounting with flock option
3208 Details : Set up an error-returning stub in ll_file_operations.lock field
3209 to prevent incorrect behaviour when client is mounted without
3210 flock option. Also, set up properly f_op->flock field for
3214 Frequency : always on ia64
3216 Description: "lfs df" loops on printing out MDS statfs information
3217 Details : The obd_ioctl_data was not initialized and in some systems
3218 this caused a failure during the ioctl that did not return
3219 an error. Initialize the struct and return an error on failure.
3222 Frequency : SLES 9 only
3224 Description: Error of copying files with lustre special EAs as root
3225 Details : Client side always return success for setxattr call for lustre
3226 special xattr (currently only "trusted.lov").
3229 Frequency : rarely on clusters with both ia64+i386 clients
3231 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3232 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3233 clients are doing concurrent writes to the same file it is
3234 possible that the ia64 clients may LASSERT because the OST
3235 extent locks are not PAGE_SIZE aligned. Ensure that grown
3236 locks are always aligned on the request boundary.
3239 Frequency : specific use, occasional
3241 Description: Overwriting in use executable truncates on-disk binary image
3242 Details : If one node attempts to overwrite an executable in use by
3243 another node, we now correctly return ETXTBSY instead of
3244 truncating the file.
3246 Severity : enhancement
3248 Description: Async OSC create to avoid the blocking unnecessarily.
3249 Details : If a OST has no remain object, system will block on the creating
3250 when need to create a new object on this OST. Now, ways use
3251 pre-created objects when available, instead of blocking on an
3252 empty osc while others are not empty. If we must block, we block
3253 for the shortest possible period of time.
3258 Description: chmod on Lustre root is propagated to other clients
3259 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3260 invalid by the follow_mount time.
3263 Frequency : liblustre clients only
3265 Description: Race in 'instant cancel' lock handling could lead to such locks
3266 never to be granted in case of SMP MDS
3267 Details : Do not destroy not yet granted but cbpending locks in
3271 Frequency : replay/resend of open
3273 Description: non null lock assetion failure in mds_intent_policy
3274 Details : Trying to replay/resend lockless open requests resulted in
3275 mds_open() returning 0 with no lock. Now it sets a flag if
3276 a lock is going to be returned.
3278 Severity : enhancement
3280 Description: Checksum enhancements
3281 Details : New checksum enhancements allow for resending RPCs that failed
3284 Severity : enhancement
3286 Description: Tunables on number of dirty pages in cacche
3287 Details : Allow to set limit on number of dirty pages cached.
3292 Description: client crash on unmount - lock still has references
3293 Details : In some error handling cases it was possible to leak a lock
3294 reference on a client while accessing a file. This was not
3295 harmful to the client during operation, but would cause the
3296 client to crash when the filesystem is unmounted.
3299 Frequency : specific case, rare
3301 Description: ETXTBSY on mds though file not in use by client
3302 Details : ETXTBSY is no longer incorrectly returned when attempting to
3303 chmod or chown a directory that the user previously tried to
3304 execute or a currently-executing binary.
3307 Frequency : extremely rare except on liblustre-based clients
3309 Description: Lustre space not freed when files are deleted
3310 Details : Clean up open-unlinked files after client eviction. Previously
3311 the unlink was skipped and the files remained as orphans.
3316 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3317 Details : In some cases it was possible to send a blocking callback to
3318 a client doing a glimpse, even though that client didn't get
3319 a lock granted. When the glimpse lock is cancelled on the OST
3320 the freed lock is left on the waiting list and corrupted the list.
3323 Frequency : all core dumps
3325 Description: Broke core dumps to lustre
3326 Details : Negative dentry may be unhashed if parent does not have UPDATE
3327 lock, but some callers, e.g. do_coredump, expect dentry to be
3328 hashed after successful create, hash it in ll_create_it.
3330 ------------------------------------------------------------------------------
3332 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3334 * Support for kernels:
3335 2.6.9-42.0.2.EL (RHEL 4)
3336 2.6.5-7.276 (SLES 9)
3337 2.4.21-40.EL (RHEL 3)
3338 2.6.12.6 vanilla (kernel.org)
3342 Frequency : always on RHEL 3
3344 Description: Number of open files grows over time
3345 Details : The number of open files grows over time, whether or not
3346 Lustre is started. This was due to a filp leak introduced
3347 by one of our kernel patches.
3349 ------------------------------------------------------------------------------
3351 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3353 * Support for kernels:
3354 2.6.9-42.EL (RHEL 4)
3355 2.6.5-7.276 (SLES 9)
3356 2.4.21-40.EL (RHEL 3)
3357 2.6.12.6 vanilla (kernel.org)
3362 Bugzilla : 5719, 9635, 9792, 9684
3363 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3364 Details : If a server is under heavy load and cannot reply to new
3365 connection requests before the client resends the (re)connect,
3366 the connection handling code can behave badly if two service
3367 threads are concurrently handing separate (re)connections from
3368 the same client. Add better locking to the connection handling
3369 code, and ensure that only a single connection will be processed
3370 for a given client UUID, even if the lock is dropped.
3372 Severity : enhancement
3374 Description: add TCP zero-copy support to kernel
3375 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3376 sends if the hardware supports scatter-gather and checksumming.
3377 This allows socklnd to do client-write and server-read more
3378 efficiently and reduce CPU utilization from skbuf copying.
3381 Frequency : only if NFS exporting from client
3383 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3384 Details : If setting attributes on a file created under NFS that had
3385 never been opened it would be possible to oops the client
3386 if the file had no objects.
3389 Frequency : always for liblustre
3391 Description: liblustre client does MDS+OSTs setattr RPC for each write
3392 Details : When doing a write from a liblustre client, the client
3393 incorrectly issued an RPC to the MDS and each OST the file was
3394 striped over in order to update the timestamps. When writing
3395 with small chunks and many clients this could overwhelm the MDS
3396 with RPCs. In all cases it would slow down the write because
3397 these RPCs are unnecessary.
3399 Severity : enhancement
3401 Description: allow number of MDS service threads to be changed at module load
3402 Details : It is now possible to change the number of MDS service threads
3403 running. Adding "options mds mds_num_threads={N}" to the MDS's
3404 /etc/modprobe.conf will set the number of threads for the next
3405 time Lustre is restarted (assuming the "mds" module is also
3406 reloaded at that time). The default number of threads will
3407 stay the same, 32 for most systems.
3412 Description: OST crash if filesystem is unformatted or corrupt
3413 Details : If an OST is started on a device that has never been formatted
3414 or if the filesystem is corrupt and cannot even mount then the
3415 error handling cleanup routines would dereference a NULL pointer.
3420 Description: NULL pointer deref in llap_from_page.
3421 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3422 incorrect) mapping if the page was truncated/reclaimed while it was
3423 searched for. Check for this condition and skip such pages when
3424 doing readahead. Introduce extra check to llap_from_page() to
3425 verify page->mapping->host is non-NULL (so page is not anonymous).
3428 Frequency : Sometimes when using sys_sendfile
3430 Description: "page not covered by a lock" warnings from ll_readpage
3431 Details : sendfile called ll_readpage without right page locks present.
3432 Now we introduced ll_file_sendfile that does necessary locking
3433 around call to generic_file_sendfile() much like we do in
3437 Frequency : with certain MDS communication failures at client mount time
3439 Description: NULL pointer deref after failed client mount
3440 Details : a client connection request may delayed by the network layer
3441 and not be sent until after the PTLRPC layer has timed out the
3442 request. If the client fails the mount immediately it will try
3443 to clean up before the network times out the request. Add a
3444 reference from the request import to the obd device and delay
3445 the cleanup until the network drops the request.
3448 Frequency : occasionally during client (re)connect
3450 Description: assertion failure during client (re)connect
3451 Details : processing a client connection request may be delayed by the
3452 client or server longer than the client connect timeout. This
3453 causes the client to resend the connection request. If the
3454 original connection request is replied in this interval, the
3455 client may trip an assertion failure in ptlrpc_connect_interpret()
3456 which thought it would be the only running connect process.
3459 Frequency : only with obd_echo servers and clients that are rebooted
3461 Description: kernel BUG accessing uninitialized data structure
3462 Details : When running an obd_echo server it did not start the ping_evictor
3463 thread, and when a client was evicted an uninitialized data
3464 structure was accessed. Start the ping_evictor in the RPC
3465 service startup instead of the OBD startup.
3467 Severity : enhancement
3468 Bugzilla : 10193 (patchless)
3469 Description: Remove dependency on various unexported kernel interfaces.
3470 Details : No longer need reparent_to_init, exit_mm, exit_files,
3471 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3474 Frequency : rare (only users of deprecated and unsupported LDAP config)
3476 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3477 Details : LDAP apparently contains 'lustreName' attributes instead of
3478 'name'. A simple remapping of the name is sufficient.
3481 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3483 Description: waiting_locks_callback trips kernel BUG if client is evicted
3484 Details : Running with the dump_on_timeout debug flag turned on makes
3485 it possible that the waiting_locks_callback() can try to dump
3486 the Lustre kernel debug logs from an interrupt handler. Defer
3487 this log dumping to the expired_lock_main() thread.
3489 Severity : enhancement
3491 Description: Support NFS exporting on 2.6 kernels.
3492 Details : Implement non-rawops metadata methods for NFS server to use without
3493 changing NFS server code.
3496 Frequency : very rare (synthetic metadata workload only)
3498 Description: two racing renames might cause an MDS thread to deadlock
3499 Details : Running the "racer" program may cause one MDS thread to rename
3500 a file from being the source of a rename to being the target of
3501 a rename at exactly the same time that another thread is doing
3502 so, and the second thread has already enqueued these locks after
3503 doing a lookup of the target and is trying to relock them in
3504 order. Ensure that we don't try to re-lock the same resource.
3507 Frequency : only very large systems with liblustre clients
3509 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3510 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3511 all OSTs in parallel. This allows the eviction of stale liblustre
3512 clients to proceed much faster than if they were done in series,
3513 and also offers similar improvements for other set_info RPCs.
3518 Description: excessive CPU usage during initial read phase on client
3519 Details : During the initial read phase on a client, it would agressively
3520 retry readahead on the file, consuming too much CPU and impacting
3521 performance (since 1.4.5.8). Improve the readahead algorithm
3522 to avoid this, and also improve some other common cases (read
3523 of small files in particular, where "small" is files smaller than
3524 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3529 Description: MDS crash when receiving packet with unknown intent.
3530 Details : Do not LBUG in unknown intent case, just return -EFAULT
3532 Severity : enhancement
3533 Bugzilla : 9293, 9385
3534 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3535 Details : Do not serialize getattr (non-intent version) and statfs.
3538 Frequency : occasional, when OST network is overloaded/intermittent
3540 Description: client evicted by OST after bulk IO timeout
3541 Details : If a client sends a bulk IO request (read or write) the OST
3542 may evict the client if it is unresposive to its data GET/PUT
3543 request. This is incorrect if the network is overloaded (takes
3544 too long to transfer the RPC data) or dropped the OST GET/PUT
3545 request. There is no need to evict the client at all, since
3546 the pinger and/or lock callbacks will handle this, and the
3547 client can restart the bulk request.
3550 Frequency : Always when mmapping file with no objects
3552 Description: client crashes when mmapping file with no objects
3553 Details : Check that we actually have objects in a file before doing any
3554 operations on objects in ll_vm_open, ll_vm_close and
3560 Description: Request leak when working with deleted CWD
3561 Details : Introduce advanced request refcount tracking for requests
3562 referenced from lustre intent.
3564 Severity : Enhancement
3566 Description: Cache open file handles on client.
3567 Details : MDS now will return special lock along with openhandle, if
3568 requested and client is allowed to hold openhandle, even if unused,
3569 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3570 closing files for every read/write openration.
3572 Severity : Enhancement
3574 Description: Cache open negative dentries on client when possible.
3575 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3576 negative dentries on lock revocation.
3581 Description: Remounting a client read-only wasn't possible with a zconf mount
3582 Details : It wasn't possible to remount a client read-only with llmount.
3584 Severity : enhancement
3585 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3586 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3587 MPICH is included to add Lustre-specific ADIO interfaces.
3588 This is based closely on the UFS ADIO layer and only differs
3589 in file creation, in order to allow the OST striping to be set.
3590 This is user-contributed code and not supported by CFS.
3595 Description: extended inode attributes (immutable, append-only) work improperly
3596 when 2.4 and 2.6 kernels are used on client/server or vice versa
3597 Details : Introduce kernel-independent values for these flags.
3599 Severity : enhancement
3602 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3603 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3604 granularity. Now, Fractional MB granularity is be supported,
3605 this is very useful for low memory system.
3607 Severity : enhancement
3609 Description: Getattr by fid
3610 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3611 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3612 have locks and avoids getattr rpc requests for it on every stat.
3615 Frequency : Always, for filesystems larger than 2TB
3617 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3618 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3619 filesystems larger than 2TB to be mounted with mballoc enabled.
3622 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3624 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3625 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3626 filesystems larger than 2TB is mounted then the kernel will
3627 silently corrupt the start of the filesystem. CONFIG_LBD is
3628 enabled for all CFS-supported kernels, but the possibility of
3629 this happening with a modified kernel config exists.
3631 Severity : enhancement
3633 Description: add client O_DIRECT support for 2.6 kernels
3634 Details : It is now possible to do O_DIRECT reads and writes to files
3635 in the Lustre client mountpoint on 2.6 kernel clients.
3637 Severity : enhancement
3639 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3640 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3641 parallel, not waiting for response from every OST before sending
3642 a rpc to the next OST.
3647 Description: setattr vs write race when updating file timestamps
3648 Details : Client processes that update a file timestamp into the past
3649 right after writing to the file (e.g. tar) it is possible that
3650 the updated file modification time can be reset to the current
3651 time due to a race between processing the setattr and write RPC.
3653 Severity : enhancement
3655 Description: Bring 'lfs find' closer in line with regular Linux find.
3656 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3657 -print0 options and obtains all the needed info through the lustre
3660 Severity : enhancement
3662 Description: support up to 1024 configured devices on one node
3663 Details : change obd_dev array from statically allocated to dynamically
3664 allocated structs as they are first used to reduce memory usage
3669 Description: Flush dirty partially truncated pages during truncate
3670 Details : Immediatelly flush partially truncated pages in filter_setattr,
3671 this way we completely avoid having any pages in page cache on OST
3672 and can retire ugly workarounds during writes to flush such pages.
3677 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3678 Details : protect inode from truncation within vfs_unlink() context
3679 just take a reference before calling vfs_unlink() and release it
3680 when parent's i_sem is free.
3685 Description: last_id value checked outside lock on OST caused LASSERT failure
3686 Details : If there were multiple MDS->OST object precreate requests in
3687 flight, it was possible that the OST's last object id was checked
3688 outside a lock and incorrectly tripped an assertion. Move checks
3689 inside locks, and discard old precreate requests.
3692 Frequency : always, if extents are used on OSTs
3694 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3695 Details : OSTs using the extents format would not zero the high 16 bits of
3696 the index physical block number. This is not a problem for any
3697 OST filesystems smaller than 16TB, and no kernels support ext3
3698 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3699 new/modified files) and can be fixed for existing filesystems
3700 with e2fsprogs-1.39-cfs1.
3705 Description: import connection selection may be incorrect if timer wraps
3706 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3707 import connections to be ignored if the 32-bit jiffies counter
3708 wraps. Use a 64-bit jiffies counter.
3711 Frequency : very large clusters immediately after boot
3713 Description: LNET request buffers exhausted under heavy short-term load
3714 Details : If a large number of client requests are generated on a service
3715 that has previously never seen so many requests it is possible
3716 that the request buffer growth cannot keep up with the spike in
3717 demand. Instead of dropping incoming requests, they are held in
3718 the LND until the RPC service can accept more requests.
3721 Frequency : Sometimes during replay
3723 Description: Assertion failure in ll_local_open after replay.
3724 Details : If replay happened on an open request reply before we were able
3725 to set replay handler, reply will become not swabbed tripping the
3726 assertion in ll_local_open. Now we set the handler right after
3727 recognising of open request
3730 Frequency : very rare
3732 Description: kernel reports "badness in vsnprintf"
3733 Details : Reading from the "recovery_status" /proc file in small chunks
3734 may cause a negative length in lprocfs_obd_rd_recovery_status()
3735 call to vsnprintf() (which is otherwise harmless). Exit early
3736 if there is no more space in the output buffer.
3738 Severity : enhancement
3740 Description: clear OBD RPC statistics by writing to them
3741 Details : It is now possible to clear the OBD RPC statistics by writing
3742 to the "stats" file.
3747 Description: Client mtime is not the same on different clients after utimes
3748 Details : In some cases, the client was using the utimes() syscall on
3749 a file cached on another node. The clients now validate the
3750 ctime from the MDS + OSTs to determine which one is right.
3755 Description: Inability to activate failout mode
3756 Details : lconf script incorrectly assumed that in pythong string's numeric
3757 value is used in comparisons.
3760 Frequency : always with multiple stripes per file
3762 Description: Inefficient object allocation for mutli-stripe files
3763 Details : When selecting which OSTs to stripe files over, for files with
3764 a stripe count that divides evenly into the number of OSTs,
3765 the MDS is always picking the same starting OST for each file.
3766 Return the OST selection heuristic to the original design.
3771 Description: mount failures may take full timeout to return an error
3772 Details : Under some heavy load conditions it is possible that a
3773 failed mount can wait for the full obd_timeout interval,
3774 possibly several minutes, before reporting an error.
3775 Instead return an error as soon as the status is known.
3777 Frequency : quota enabled and large files being deleted
3779 Description: releasing more than 4GB of quota at once hangs OST
3780 Details : If a user deletes more than 4GB of files on a single OST it
3781 will cause the OST to spin in an infinite loop. Release
3782 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3787 Description: statfs data retrieved from /proc may be stale or zero
3788 Details : When reading per-device statfs data from /proc, in the
3789 {kbytes,files}_{total,free,avail} files, it may appear
3790 as zero or be out of date.
3793 Frequency : systems with MD RAID1 external journal devices
3795 Description: lconf's call to blkid is confused by RAID1 journal devices
3796 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3797 of returning all block devices that match the journal UUID.
3800 Frequency : always, for aggregate stripe size over 4GB
3802 Description: assertion fails when trying to use 4GB stripe size
3803 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3804 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3807 Frequency : always on ppc64
3809 Description: the first write on an ext3 filesystem with mballoc got stuck
3810 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3811 perform endianness conversion.
3813 ------------------------------------------------------------------------------
3815 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3817 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3818 INTEROPERATE with older versions automatically. Please read the
3819 user documentation before upgrading any part of a live system.
3820 * WARNING: Lustre networking configuration changes are required with
3821 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3824 * Support for kernels:
3825 2.6.9-22.0.2.EL (RHEL 4)
3826 2.6.5-7.244 (SLES 9)
3827 2.6.12.6 vanilla (kernel.org)
3830 Severity : enhancement
3831 Bugzilla : 7981/8208
3832 Description: Introduced Lustre Networking (LNET)
3833 Details : LNET is new networking infrastructure for Lustre, it includes
3834 a reorganized network configuration mode (see the user
3835 documentation for full details) as well as support for routing
3836 between different network fabrics. Lustre Networking Devices
3837 (LNDs) for the supported network fabrics have also been
3838 created for this new infrastructure.
3840 Severity : enhancement
3841 Description: Introduced Access control lists
3842 Details : clients can set ACLs on files and directories in order to have
3843 more fine-grained permissions than the standard Unix UGO+RWX.
3844 The MDS must be started with the "-o acl" mount option.
3846 Severity : enhancement
3847 Description: Introduced filesystem quotas
3848 Details : Administrators may now establish per-user quotas on the
3851 Severity : enhancement
3853 Description: Configuration change for the XT3
3854 The PTLLND is now used to run Lustre over Portals on the XT3
3855 The configure option(s) --with-cray-portals are no longer used.
3856 Rather --with-portals=<path-to-portals-includes> is used to
3857 enable building on the XT3. In addition to enable XT3 specific
3858 features the option --enable-cray-xt3 must be used.
3863 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3864 Details : A race between allocating a new llcd and re-getting the llcd_lock
3865 allowed another thread to grab newly-allocated llcd.
3867 Severity : enhancement
3869 Description: 2.6 OST async journal commit and locking fix to improve performance
3870 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3871 2.6 kernels are now async as they already were in 2.4 kernels so
3872 that they can commit concurrently with the network bulk transfer.
3873 For block-allocated files the filter allocation semaphore is held
3874 to avoid filesystem fragmentation during allocation. BKL lock
3875 removed for 2.6 xattr operations where it is no longer needed.
3880 Description: lconf incorrectly determined whether two IP networks could talk
3881 Details : In some more complicated routing and multiple-network
3882 configurations, lconf will avoid trying to make a network
3883 connection to a disjoint part of the IP space. It was doing the
3884 math incorrectly for one set of cases.
3889 Description: Fix for potential infinite loop processing records in an llog.
3890 Details : If an llog record is corrupted/zeroed, it is possible to loop
3891 forever in llog_process(). Validate the llog record length
3892 and skip the remainder of the block on error.
3895 Frequency : occasional (liblustre only)
3897 Description: liblustre could not open files whose last component is a symlink
3898 Details : sysio_path_walk() would incorrectly pass the open intent to
3899 intermediate path components.
3902 Frequency : rare (liblustre only with non-standard tuning)
3903 Bugzilla : 7201 (7350)
3904 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3905 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3906 after it has been released, only do this for referenced locks
3908 Severity : enhancement
3910 Description: specifying an (invalid) directory default stripe_size of -1
3911 would reset the directory default striping
3912 Details : stripe_size -1 was used internally to signal directory stripe
3913 removal, now use "all default" to signal dir stripe removal
3914 as a directory striping of "all default" is not useful
3917 Frequency : common for large clusters running liblustre clients
3919 Description: doing an ls when liblustre clients are running is slow
3920 Details : sending a glimpse AST to a liblustre client waits for every AST
3921 to time out, as liblustre clients will not respond. Since they
3922 cannot cache data we refresh the OST lock LVB from disk instead.
3924 Severity : enhancement
3926 Description: doing an ls at the same time as file IO can be slow
3927 Details : enqueue and other "small" requests can be blocked behind many
3928 large IO requests. Create a new OST IO portal for non-IO
3929 requests so they can be processed faster.
3932 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3934 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3935 created file with mode 000
3936 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3937 allow an ftruncate() as a non-root user to a file with mode 000.
3938 Lustre now respects this flag to disable mode checks when
3939 truncating a file owned by the user
3942 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3945 Description: Revoking locks from clients that went dead or catatonic might take
3947 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3948 cancellation of such locks instant on servers without waiting for
3949 any reply from clients. Clients drops these locks when cancel
3950 notification from server is received without replying.
3953 Frequency : liblustre-only, when liblustre client dies or becomes busy
3955 Description: Doing ls on Linux clients can take a long time with active
3957 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3958 granting such locks to it in the first place if possible. Locks
3959 are taken by proxy on the OST during the read or write and
3960 dropped immediately afterward. Add connect flags handling, do
3961 not grant locks to liblustre clients for glimpse ASTs.
3963 Severity : enhancement
3965 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3966 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3967 sequential reads and behaves badly with applications doing random
3968 reads. Improve it by reading ahead at least read region, and
3969 avoiding excessive large RPC for small reads.
3971 Severity : enhancement
3973 Description: Creating more than 1000 files for a single job may cause a load
3974 imbalance on the OSTs if there are also a large number of OSTs.
3975 Details : qos_prep_create() uses an OST index reseed value that is an
3976 even multiple of the number of available OSTs so that if the
3977 reseed happens in the middle of the object allocation it will
3978 still utilize the OSTs as uniformly as possible.
3983 Description: OST or MDS may oops in ping_evictor_main()
3984 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3985 but doesn't restart at beginning of obd_exports_timed list
3988 Severity : enhancement
3990 Description: improve by-nid export eviction on the MDS and OST
3991 Details : allow multiple exports with the same NID to be evicted at one
3992 time without re-searching the exports list.
3995 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3997 Description: MDS may oops in groups_free()
3998 Details : in rare race conditions a newly allocated group_info struct is
3999 freed again, and this can be NULL. The 2.4 compatibility code
4000 for groups_free() checked for a NULL pointer, but 2.6 did not.
4003 Frequency : common for liblustre clients doing little filesystem IO
4004 Bugzilla : 9352, 7313
4005 Description: server may evict liblustre clients accessing contended locks
4006 Details : if a client is granted a lock or receives a completion AST
4007 with a blocking AST already set it would not reply to the AST
4008 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4011 Frequency : lfs setstripe, only systems with more than 160 OSTs
4013 Description: unable to set striping with a starting offset beyond OST 160
4014 Details : llapi_create_file() incorrectly limited the starting stripe
4015 index to the maximum single-file stripe count.
4018 Frequency : LDAP users only
4020 Description: lconf did not handle in-kernel recovery with LDAP properly
4021 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4023 Severity : enhancement
4025 Description: bind OST threads to NUMA nodes to improve performance
4026 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4027 node and do their allocations there to localize memory access
4029 Severity : enhancement
4031 Description: llmount can determine client NID directly from Myrinet (GM)
4032 Details : the client NID code from gmnalnid was moved directly into
4033 llmount, removing the need to use this or specifying the
4034 client NID explicitly when mounting GM clients with zeroconf
4037 Frequency : if client is started with down MDS
4039 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4040 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4041 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4042 ptlrpc_import_delayed_req() because we didn't check if the
4043 request was interrupted, and we also didn't break out of the
4044 event loop if there was no timeout
4049 Description: data loss during non-page-aligned writes to a single file from
4050 both multiple nodes and multiple threads on one node at same time
4051 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4052 inconsistency led to false short-reads, that were cached and later
4053 used by ->prepare_write() to fill in partially written page,
4054 leading to data loss.
4057 Frequency : always, if lconf --abort_recovery used
4059 Description: lconf --abort_recovery fails with 'Operation not supported'
4060 Details : lconf was attempting to abort recovery on the MDT device and not
4063 Severity : enhancement
4065 Description: remove cleanup logs
4066 Details : replace lconf-generated cleanup logs with lustre internal
4067 cleanup routines. Eliminates the need for client-cleanup and
4070 Severity : enhancement
4072 Description: add support for EAs (user and system) on lustre filesystems
4073 Details : it is now possible to store extended attributes in the Lustre
4074 client filesystem, and with the user_xattr mount option it
4075 is possible to allow users to store EAs on their files also
4077 Severity : enhancement
4079 Description: Add possibility (config option) to show minimal available OST free
4081 Details : When compiled with --enable-mindf configure option, statfs(2)
4082 (and so, df) will return least minimal free space available from
4083 all OSTs as amount of free space on FS, instead of summary of
4084 free spaces of all OSTs.
4086 Severity : enhancement
4088 Description: do not expand extent locks acquired on OST-side
4089 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4090 by server: they are not cached anyway.
4093 Frequency : when mmap is used/binaries executed from Lustre
4095 Description: Unmmap pages before throwing them away from read cache.
4096 Details : llap_shrink cache now attempts to unmap pages before discarding
4097 them (if unmapping failed - do not discard). SLES9 kernel has
4098 extra checks that trigger if this unmapping is not done first.
4103 Description: lconf didn't resolve symlinks before checking to see whether a
4104 given mountpoint was already in use
4107 Frequency : when migrating failover services
4108 Bugzilla : 6395, 9514
4109 Description: When migrating a subset of services from a node (e.g. failback
4110 from a failover service node) the remaining services would
4111 time out and evict clients.
4112 Details : lconf --force (implied by --failover) sets the global obd_timeout
4113 to 5 seconds in order to quickly disconnect, but this caused
4114 other RPCs to time out too quickly. Do not change the global
4115 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4117 Severity : enhancement
4118 Frequency : if MDS is started with down OST
4119 Bugzilla : 9439,5706
4120 Description: Allow startup/shutdown of an MDS without depending on the
4121 availability of the OSTs.
4122 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4123 Add appropriate locking and lov-osc refcounts for safe
4124 cleaning. Add osc abort_inflight calls in case the
4125 synchronize never started.
4128 Frequency : occasional (Cray XT3 only)
4130 Description: root not authorized to access files in CRAY_PORTALS environment
4131 Details : The client process capabilities were not honoured on the MDS in
4132 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4133 been accessed by an authorized user then root was able to access
4134 the file on the local client also. The root user capabilities
4135 are now allowed on the MDS, as this environment has secure UID.
4138 Frequency : occasional
4140 Description: ldiskfs "too long searching" message happens too often
4141 Details : A debugging message (otherwise harmless) prints too often on
4142 the OST console. This has been reduced to only happen when
4143 there are fragmentation problems on the filesystem.
4148 Description: Division by zero in statfs when all OSCs are inactive
4149 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4150 lov_statfs divides by value returned by lov_get_stripecnt().
4154 Bugzilla : 9489, 3273
4155 Description: First write from each client to each OST was only 4kB in size,
4156 to initialize client writeback cache, which caused sub-optimal
4157 RPCs and poor layout on disk for the first writen file.
4158 Details : Clients now request an initial cache grant at (re)connect time
4159 and so that they can start streaming writes to the cache right
4160 away and always do full-sized RPCs if there is enough data.
4161 If the OST is rebooted the client also re-establishes its grant
4162 so that client cached writes will be honoured under the grant.
4167 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4168 Details : Now I/O RPCs go to different portal number and (presumably) fast
4169 lock requests (and glimses) and other RPCs get their own service
4170 threads pool that should be able to service those RPCs
4173 Severity : enhancement
4175 Description: Ability to exchange lustre version between client and servers and
4176 issue warnings at client side if client is too old. Also for
4177 liblustre clients there is ability to refuse connection of too old
4179 Details : New 'version' field is added to connect data structure that is
4180 filled with version info. That info is later checked by server and
4184 Frequency : rare, liblustre only.
4185 Bugzilla : 9296, 9581
4186 Description: Two simultaneous writes from liblustre at offset within same page
4187 might proceed at the same time overwriting eachother with stale
4189 Details : I/O lock withing llu_file_prwv was released too early, before data
4190 actually was hitting the wire. Extended lock-holding time until
4191 server acknowledges receiving data.
4194 Frequency : extremely rare. Never observed in practice.
4196 Description: avoid generating lustre_handle cookie of 0.
4197 Details : class_handle_hash() generates handle cookies by incrementing
4198 global counter, and can hit 0 occasionaly (this is unlikely, but
4199 not impossible, because initial value of cookie counter is
4200 selected randonly). Value of 0 is used as a sentinel meaning
4201 "unassigned handle" --- avoid it. Also coalesce two critical
4202 sections in this function into one.
4204 Severity : enhancement
4206 Description: allow liblustre clients to delegate truncate locking to OST
4207 Details : To avoid overhead of locking, liblustre client instructs OST to
4208 take extent lock in ost_punch() on client's behalf. New connection
4209 flag is added to handle backward compatibility.
4211 Severity : enhancement
4212 Bugzilla : 4928, 7341, 9758
4213 Description: allow number of OST service threads to be specified
4214 Details : a module parameter allows the number of OST service threads
4215 to be specified via "options ost ost_num_threads={N}" in the
4216 OSS's /etc/modules.conf or /etc/modprobe.conf.
4220 Bugzilla : 6146, 9635, 9895
4221 Description: servers crash with bad pointer in target_handle_connect()
4222 Details : In rare cases when a client is reconnecting it was possible that
4223 the connection request was the last reference for that export.
4224 We would temporarily drop the export reference and get a new
4225 one, but this may have been the last reference and the export
4226 was just destroyed. Get new reference before dropping old one.
4228 Severity : enhancement
4229 Frequency : if client is started with failover MDS
4231 Description: Allow multiple MDS hostnames in the mount command
4232 Details : Try to read the configuration from all specified MDS
4233 hostnames during a client mount in case the "primary"
4236 Severity : enhancement
4238 Description: Stop sending data to evicted clients as soon as possible.
4239 Details : Check if the client we are about to send or are sending data to
4240 was evicted already. (Check is done every second of waiting,
4241 for which l_wait_event interface was extended to allow checking
4242 of exit condition at specified intervals).
4245 Frequency : rare, normally only when NFS exporting is done from client
4247 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4249 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4252 Frequency : always, if srand() is called before liblustre initialization
4254 Description: Liblustre uses system PRNG disturbing its usage by user application
4255 Details : Introduce internal to lustre fast and high-quality PRNG for
4256 lustre usage and make liblustre and some other places in generic
4257 lustre code to use it.
4259 Severity : enhancement
4260 Bugzilla : 9477, 9557, 9870
4261 Description: Verify that the MDS configuration logs are updated when xml is
4262 Details : Check if the .xml configuration logs are newer than the config
4263 logs stored on the MDS and report an error if this is the case.
4264 Request --write-conf, or allow starting with --old_conf.
4266 Severity : enhancement
4268 Description: Handle symlinks in the path when checking if Lustre is mounted.
4269 Details : Resolve intermediate symlinks when checking if a client has
4270 mounted a filesystem to avoid duplicate client mounts.
4275 Description: lconf can hit an error exception but still return success.
4276 Details : The lconf command catches the Command error exception at the top
4277 level script context and will exit with the associated exit
4278 status, but doesn't ensure that this exit status is non-zero.
4283 Description: failure of ptlrpc thread startup can cause oops
4284 Details : Starting a ptlrpc service thread can fail if there are a large
4285 number of threads or the server memory is very fragmented.
4286 Handle this without oopsing.
4289 Frequency : always, only if liblustre and non-default acceptor port was used
4291 Description: liblustre cannot connect to servers with non-default acceptor port
4292 Details : tcpnal_set_default_params() was not called and was therefore
4293 ignoring the environment varaible TCPNAL_PORT, as well as other
4294 TCPNAL_ environment variables
4299 Description: two objects could be created on the same OST for a single file
4300 Details : If an OST is down, in some cases it was possible to create two
4301 objects on a single OST for a single file. No problems other
4302 than potential performance impact and spurious error messages.
4306 Bugzilla : 5681, 9562
4307 Description: Client may oops in ll_unhash_aliases
4308 Details : Client dcache may become inconsistent in race condition.
4309 In some cases "getcwd" can fail if the current directory is
4315 Description: Inode refcounting problems in NFS export code
4316 Details : link_raw functions used to call d_instantiate without obtaining
4317 extra inode reference first.
4321 Bugzilla : 9942, 9903
4322 Description: Referencing freed requests leading to crash, memleaks with NFS.
4323 Details : We used to require that call to ll_revalidate_it was always
4324 followed by ll_lookup_it. Also with revalidate_special() it is
4325 possible to call ll_revalidate_it() twice for the same dentry
4326 even if first occurence returned success. This fix changes semantic
4327 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4328 reference on a request referred from the intent.
4329 ll_intent_release() then releases such a request.
4332 Frequency : rare, normally benchmark loads only
4334 Description: unlinked inodes were kept in memory on the client
4335 Details : If a client is repeatedly creating and unlinking files it
4336 can accumulate a lot of stale inodes in the inode slab cache.
4337 If there is no other client load running this can cause the
4338 client node to run out of memory. Instead flush old inodes
4339 from client cache that have the same inode number as a new inode.
4342 Frequency : SLES9 2.6.5 kernel and long filenames only
4343 Bugzilla : 9969, 10379
4344 Description: utime reports stale NFS file handle
4345 Details : SLES9 uses out-of-dentry names in some cases, which confused
4346 the lustre dentry revalidation. Change it to always use the
4350 Frequency : rare, unless heavy write-truncate concurrency is continuous
4351 Bugzilla : 4180, 6984, 7171, 9963, 9331
4352 Description: OST becomes very slow and/or deadlocked during object unlink
4353 Details : filter_destroy() was holding onto the parent directory lock
4354 while truncating+unlinking objects. For very large objects this
4355 may block other threads for a long time and slow overall OST
4356 responsiveness. It may also be possible to get a lock ordering
4357 deadlock in this case, or run out of journal credits because of
4358 the combined truncate+unlink. Solution is to do object truncate
4359 first in one transaction without parent lock, and then do the
4360 final unlink in a new transaction with the parent lock. This
4361 reduces the lock hold time dramatically.
4364 Frequency : rare, 2.4 kernels only
4366 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4367 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4368 before unmounting their filesystem in order to avoid deadlock.
4369 The kernel_locked() function in 2.4 kernels only checks whether
4370 the kernel lock is held, not whether it is this process that is
4371 holding it as 2.6 kernels do.
4376 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4377 Details : The client ptlrpc code may be trying to reconnect to a down
4378 server before a previous connection attempt has timed out.
4379 Increase the reconnect interval to be longer than the connection
4380 timeout interval to avoid sending duplicate connections to
4384 Frequency : echo_client brw_test command
4386 Description: fix echo_client to work with OST preallocated code
4387 Details : OST preallocation code (5137) didn't take echo_client IO path
4388 into account: echo_client calls filter methods outside of any
4389 OST thread and, hence, there is no per-thread preallocated
4390 pages and buffers to use. Solution: hijack pga pages for IO. As
4391 a byproduct, this avoids unnecessary data copying.
4395 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4396 Description: Client can oops in mdc_commit_close() after open replay
4397 Details : It was possible for the MDS to return an open request with no
4398 transaction number in mds_finish_transno() if the client was
4399 evicted, but without actually returning an error. Clients
4400 would later try to replay that open and may trip an assertion
4401 Simplify the client close codepath, and always return an error
4402 from the MDS in case the open is not successful.
4405 Frequency : rare, 2.6 OSTs only
4407 Description: OST may deadlock under high load on fragmented files
4408 Details : If there was a heavy load and highly-fragmented OST filesystems
4409 it was possible to have all the OST threads deadlock waiting on
4410 allocation of biovecs, because the biovecs were not released
4411 until the entire RPC IO was completed. Instead, release biovecs
4412 as soon as they are complete to ensure forward IO progress.
4414 Severity : enhancement
4416 Description: Support for specifying external journal device at mount
4417 Details : If an OST or MDS device is formatted with an external journal
4418 device, this device major/minor is stored in the ext3 superblock
4419 and may not be valid for failover. Allow detecting and
4420 specifying the external journal at mount time.
4425 Description: Mounting an MDS with pending unlinked files may cause oops
4426 Details : target_finish_recovery() calls mds_postrecov() which returned
4427 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4428 considers this an error and immediately begins cleaning up the
4429 lov, just after starting the mds_lov process
4431 Severity : enhancement
4433 Description: Implement 'lfs df' to report actual free space on per-OST basis
4434 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4435 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4436 usage of objects; '-h' to report in human readable format.
4438 ------------------------------------------------------------------------------
4440 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4447 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4448 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4449 at startup during orphan unlinking. Instead, skip buddy update
4450 before setup, as it will be regenerated after recovery is complete.
4455 Description: If an OST is inactive, its locks might reference stale inodes.
4456 Details : lov_change_cbdata() must iterate over all namespaces, even if
4457 they are inactive to clear inode references from the lock.
4459 Severity : enhancement
4460 Frequency : occasional, if non-standard max_dirty_mb used
4462 Description: Client will block write RPCs if not enough grant
4463 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4464 then the client will block writes while waiting for another RPC
4465 to complete instead of consuming its dirty limit. With change
4466 we get improved performance when max_dirty_mb is small.
4468 Severity : enhancement
4469 Bugzilla : 3389, 6253
4470 Description: Add support for supplementary groups on the MDS.
4471 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4472 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4473 lookups for user supplementary groups into a cache.
4477 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4478 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4479 for liblustre O_EXCL is enforced by MDS.
4483 Description: readdir never returns NULL in liblustre.
4484 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4485 dentry in struct dirent.
4489 Frequency : liblustre only, depends on application IO pattern
4490 Description: liblustre clients evicted if not contacting servers
4491 Details : Don't put liblustre clients into the ping_evictor list, so
4492 they will not be evicted by the pinger ever.
4494 Severity : enhancement
4496 Description: Add ability to evict clients by NID from MDS.
4497 Details : By echoing "nid:$NID" string into
4498 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4499 $NID would be instantly evicted from this MDS and from all active
4500 OSTs connected to it.
4504 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4505 Details : lookup_it_finish() used to query file size from OSTs that was not
4510 Description: service threads change working directory to that of init
4511 Details : Starting lustre service threads may pin the working directory
4512 of the parent thread, making that filesystem busy. Threads
4513 now change to the working directory of init to avoid this.
4517 Frequency : during shutdown only
4518 Description: shutdown with a failed MDS or OST can cause unmount to hang
4519 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4524 Frequency : 2.6 only
4525 Description: chmod/chown may include an extra supplementary group
4526 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4527 supplementary group and if none is specified this is used.
4530 Bugzilla : 5479 (6816)
4532 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4533 Details : If lookup is in progress on a file that is unlinked we might try
4534 to revalidate the inode and fail in revalidate after lookup is
4535 complete and ll_file_open() enqueues the open again but
4536 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4539 Frequency : always, if lconf --abort_recovery used
4541 Description: lconf --abort_recovery fails with 'Operation not supported'
4542 Details : lconf was attempting to abort recovery on the MDT device and not
4545 ------------------------------------------------------------------------------
4547 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4552 Frequency : rare (only unsupported configurations with a node running as an
4554 Bugzilla : 6514, 5137
4555 Description: Mounting a Lustre file system on a node running as an OST could
4557 Details : OSTs now preallocates memory needed to write out data at
4558 startup, instead of when needed, to avoid having to
4559 allocate memory in possibly low memory situations.
4560 Specifically, if the file system is mounted on on OST,
4561 memory pressure could force it to try to write out data,
4562 which it needed to allocate memory to do. Due to the low
4563 memory, it would be unable to do so and the node would
4564 become unresponsive.
4566 Severity : enhancement
4568 Description: Addition of lconf --service command line option
4569 Details : lconf now accepts a '--service <arg>' option, which is
4570 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4572 Severity : enhancement
4574 Description: Failover mode is now the default for OSTs.
4575 Details : By default, OSTs will now run in failover mode. To return to
4576 the old behaviour, add '--failout' to the lmc line for OSTs.
4578 Severity : enhancement
4580 Description: Health checks are now provided for MDS and OSTs
4581 Details : Additional detailed health check information on MSD and OSTs
4582 is now provided through the procfs health_check value.
4585 Frequency : occasional, depends on IO load
4587 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4588 after numerous create/delete cycles
4589 Details : The ext3 inode allocation policy would not allocate new inodes
4590 very well on the OSTs because there are no new directories
4591 being created. Instead we look for groups with free space if
4592 the parent directories are nearly full.
4597 Description: Network or server problems during mount may cause partially
4598 mounted clients instead of returning an error.
4599 Details : The config llog parsing code may overwrite the error return
4600 code during mount error handling, returning success instead
4606 Description: MDS can fail to allocate large reply buffers
4607 Details : After long uptimes the MDS can fail to allocate large reply
4608 buffers (e.g. zconf client mount config records) due to memory
4609 fragmentation or consumption by the buffer cache. Preallocate
4610 some large reply buffers so that these replies can be sent even
4611 under memory pressure.
4615 Frequency : rare (liblustre)
4616 Description: fsx running with liblustre complained that using truncate() to
4617 extend the file doesn't work. This patch corrects that issue.
4618 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4619 ATTR_SIZE and lsm use in llu_setattr_raw.
4623 Frequency : rare, only 2.6 kernels
4624 Description: Unusual file access patterns on the MDS may result in inode
4625 data being lost in very rare circumstances.
4626 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4627 code in the 2.6 kernel caused the inode and/or EA data not to
4628 be read from disk, causing single-file corruption.
4632 Frequency : rare, only 2.6 filesystems using extents
4633 Description: Heavy concurrent write and delete load may cause data corruption.
4634 Details : It was possible under high-load situations to have an extent
4635 metadata block in the block device cache from a just-unlinked
4636 file overwrite a newly-allocated data block. We now unmap any
4637 metadata buffers that alias just-allocated data blocks.
4641 Frequency : filesystems with default stripe_count larger than 77
4642 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4643 Details : lconf specifies an inode size of 4096 bytes when the default
4644 stripe_count is larger than 77. This conflicts with the default
4645 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4646 this case to avoid pinning too much memory for large EAs.
4648 ------------------------------------------------------------------------------
4650 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4655 Frequency : rare (extremely heavy IO load with hundreds of clients)
4657 Description: Client is evicted, gets IO error writing to file
4658 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4659 caused clients to be evicted for AST timeouts. The fixes in
4660 bug 5192 mean we no longer need to have such short AST timeouts
4661 so ldlm_timeout has been increased.
4664 Frequency : occasional during --force or --failover shutdown under load
4665 Bugzilla : 5949, 4834
4666 Description: Server oops/LBUG if stopped with --force or --failover under load
4667 Details : a collection of import/export refcount and cleanup ordering
4668 issues fixed for safer force cleanup
4671 Frequency : only filesystems larger than 120 OSTs
4672 Bugzilla : 5990, 6223
4673 Description: lfs getstripe would oops on a very large filesystem
4674 Details : lov_getconfig used kfree on vmalloc'd memory
4677 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4678 Bugzilla : 6242, 6243
4679 Description: reading from files that had been truncated to a non-zero size
4680 but never opened returned no data
4681 Details : ll_file_read() reads zeros from no-object files to EOF
4686 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4687 Details : There's little harm in aborting MDS/OSS recovery and letting it
4688 try again, so I removed the LASSERT and return an error instead.
4690 Severity : enhancement
4692 Description: New debugging infrastructure for tracking down data corruption
4693 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4694 (b) cover more of the client-side code path, and (c) try to narrow
4695 down where problems occurred
4699 Bugzilla : 3819, 4364, 4397, 6313
4700 Description: Racing close and eviction MDS could cause assertion in mds_close
4701 Details : It was possible to get multiple mfd references during close and
4702 client eviction, leading to one thread referencing a freed mfd.
4704 Severity: : enhancement
4705 Bugzilla : 3262, 6359
4706 Description: Attempts to reconnect to servers are now more aggressive.
4707 Details : This builds on the enhanced upcall-less recovery that was added
4708 in 1.4.2. When trying to reconnect to servers, clients will
4709 now try each server in the failover group every 10 seconds. By
4710 default, clients would previously try one server every 25 seconds.
4715 Description: After recovery, certain operations trigger a failed
4716 assertion on a client.
4717 Details : Failing over an mds, using lconf -d --failover, while a
4718 client was doing a readdir() call would cause the client to
4719 LBUG after recovery completed and the readdir() was resent.
4721 Severity : enhancement
4723 Description: Default groups are now added by lconf
4724 Details : You can now run lconf --group <servicename> without having to
4725 manually add groups with lmc.
4728 Frequency : occasional
4730 Description: Nodes with an elan id of 0 trigger a failed assertion
4733 Frequency : always when accessing e.g. tty/console device nodes
4735 Description: tty and some other devices nodes cannot be used on lustre
4736 Details : file's private_data field is used by device data and lustre
4737 values in there got lost. New field was added to struct file to
4738 store fs-specific private data.
4741 Frequency : when exporting Lustre via NFS
4743 Description: NFSD failed occasionally when looking up a path component
4744 Details : NFSD is looking up ".." which was broken in ext3 directories
4745 that had grown large enough to become hashed.
4748 Frequency : Clusters with multiple interfaces not on the same subnet
4750 Description: Nodes will repeatedly try to reconnect to an interface which it
4751 cannot reach and report an error to the log.
4752 Details : Extra peer list entries will be created by lconf with some peers
4753 unreachable. lconf now validates the peer before adding it.
4756 Frequency : Only if a default stripe is set on the filesystem root.
4758 Description: Setting a default stripe on the filesystem root prevented the
4759 filesystem from being remounted.
4760 Details : The client was sending extra request flags in the root getattr
4761 request and did not allocate a reply buffer for the dir EA.
4764 Frequency : occasional, higher if lots of files are accessed by one client
4765 Bugzilla : 6159, 6097
4766 Description: Client trips assertion regarding lsm mismatch/magic
4767 Details : While revalidating inodes the VFS looks up inodes with ifind()
4768 and in rare cases can find an inode that is being freed.
4769 The ll_test_inode() code will free the lsm during ifind()
4770 when it finds an existing inode and then the VFS later attaches
4771 this free lsm to a new inode.
4775 Bugzilla : 6422, 7030
4776 Description: MDS deadlock between mkdir and client eviction
4777 Details : Creating a new file via mkdir or mknod (starting a transaction
4778 and getting the ns lock) can deadlock with client eviction
4779 (gets ns lock and trying to finish a synchronous transaction).
4782 Frequency : occasional
4783 Description: While starting a server, the fsfilt_ext3 module could not be
4785 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4786 kernels. Previously, lconf would still use the ext3 name
4787 when trying to load modules. Now, it will correctly use
4788 ext3 on 2.4 and ldiskfs on 2.6.
4790 Severity : enhancement
4791 Description: The default stripe count has been changed to 1
4792 Details : The interpretation of the default stripe count (0, to lfs
4793 or lmc) has been changed to mean striping across a single
4794 OST, rather than all available. For general usage we have
4795 found a stripe count of 1 or 2 works best.
4797 Severity : enhancement
4798 Description: Add support for compiling against Cray portals.
4799 Details : Conditional compiling for some areas that are different
4803 Frequency : occasional
4804 Bugzilla : 6409, 6834
4805 Description: Creating files with an explicit stripe count may lead to
4806 a failed assertion on the MDS
4807 Details : If some OSTs are full or unavailable, creating files may
4808 trigger a failed assertion on the MDS. Now, Lustre will
4809 try to use other servers or return an error to the
4813 Frequency : occasional
4815 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4816 caused lock timeout during readahead (since 1.4.2).
4817 Details : Processes doing readahead might match a lock that hasn't been
4818 granted yet if there are overlapping and conflicting lock
4819 requests. The readahead process waits on ungranted lock
4820 (original lock is CBPENDING), while OST waits for that process
4821 to cancel CBPENDING read lock and eventually evicts client.
4823 Severity : enhancement
4825 Description: Initial enabling of flock support for clients
4826 Details : Implements fcntl advisory locking and file status functions.
4827 This feature is provided as an optional mount flag (default
4828 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4829 locking are implemented yet, and those that are are not guaranteed
4830 to be completely correct in production environments.
4831 mount -t lustre -o [flock|noflock] ...
4834 Frequency : occasional
4836 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4837 assertion in the ext3 JBD (journaling) layer.
4838 Details : The b_committed_data struct is protected by the big kernel lock
4839 in 2.4 kernels, serializing journal_commit_transaction() and
4840 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4841 access to this struct. In 2.6 kernels there is finer grained
4842 locking to improve SMP performance of the JBD layer.
4846 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4851 Description: merge and cleanup kernel patches.
4854 -----------------------------------------------------------------------------
4856 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4858 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4859 versions of Lustre. Please update all servers and clients to
4860 version 1.4.2 or later at the same time. You must also run
4861 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4862 to update the configuration logs.
4864 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4865 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4866 - swab llog records of type '0' so we get proper header size/idx (5861)
4867 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4868 - fix rename of one directory over another leaking an inode (5953)
4869 - avoid SetPageDirty on 2.6 (5981)
4870 - don't re-add just-being-destroyed locks to the waiting list (5653)
4871 - when creating new directories, inherit the parent's custom
4872 striping settings if present parent (3048)
4873 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4874 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4875 - don't hold client locks on temporary worklist from l_lru (5666)
4876 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4877 - automatically evict dead clients (5921)
4878 - Update file size properly in create+truncate+fstat case (6196)
4879 - Do not unhash mountpoint dentries, do not allow removal of
4881 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4882 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4883 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4884 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4885 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4886 - Correctly handle reads of files with no objects (6243)
4887 - lctl recover will also mark a device active if deactivate used (5933)
4889 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4890 - allow --write-conf on an MDS with different nettype than client (5619)
4891 - don't write config llogs to MDS for mounts not from that MDS (5617)
4892 - lconf should create multiple TCP connections from a client (5201)
4893 - init scripts are now turned off by default; run chkconfig --on
4894 lustre and chkconfig --on lustrefs to use them
4895 - upcalls are no longer needed for clients to recover to failover
4897 - add --abort-recovery option to lconf to abort recovery on device
4899 - add support for an arbitrary number of OSTs (3026)
4900 - Quota support protocol changes.
4901 - forward compatibility changes to wire structs (6007)
4902 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4903 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4904 - improved "lustre status" script
4905 - initialize blocksize for non-regular files (6062)
4906 - added --disable-server and --disable-client configure options (5782)
4907 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4908 - Vanilla 2.4.29 support
4909 - increase maximum number of obd devices to 520 (6242)
4910 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4911 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4912 SLES 9 2.6.5 kernels
4913 - limit stripes per file to 160 (the maximum EA size) (6093)
4915 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4918 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4919 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4920 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4921 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4922 - small ext3 extents cleanups and fixes (5733)
4923 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4924 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4925 - avoid needless client->OST connect, fix handle mismatch (5317)
4926 - fix DLM error path that led to out-of-sync client, long delays (5779)
4927 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4928 - fix several locking issues related to i_size (5492,5624,5654,5672)
4929 - don't move pending lock onto export if it is already evicted (5683)
4930 - fix kernel oops when creating .foo in unlinked directory (5548)
4931 - fix deadlock in obdfilter statistics vs. object create (5811)
4932 - use time_{before,after} to avoid timer jiffies wrap (5882)
4933 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4934 - Do not leak request if server was not able to process it (5154)
4935 - If mds_open unable to find parent dir, make that negative lookup(5154)
4936 - don't create new directories with extent-mapping (5909, 5936)
4938 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4939 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4940 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4941 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4942 - support for RHEL4 kernel on i686 (5773)
4943 - provide error messages when incompatible logs are encountered (5898)
4945 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4946 * version 1.4.0.10 (1.4.1 release candidate 1)
4948 - don't keep a lock reference when lock is not granted (4238)
4949 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4950 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4951 - drop import inflight refcount on signal_completed_replay error (5255)
4952 - unlock page after async write error during send (3677)
4953 - handle missing objects in filter_preprw_read properly (5265)
4954 - no transno return for symlink open, don't save no-trasno open (3440)
4955 - don't try to complete elan receive that already failed (4012)
4956 - free RPC server reply state on error (5406)
4957 - clean up thread from ptlrpc_start_thread() on error (5160)
4958 - readahead could read extra page into cache that wasn't ejected (5388)
4959 - prevent races in class_attach/setup/cleanup/detach (5260)
4960 - don't dereference de->d_inode after l_dput of de (5458)
4961 - use "int" for stripe value returned from lock_to_stripe (5544)
4962 - mballoc allocation and error-checking fixes in 2.6 (5504)
4963 - block device patches to fix I/O request sizes in 2.6 (5482)
4964 - look up hostnames for IB nals (5602)
4965 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4966 - don't start multiple acceptors for the same port (5277)
4967 - fix incorrect LASSERT in mds_getattr_name (5635)
4968 - export a proc file for general "ping" checking (5628)
4969 - fix "lfs check" to not block when the MDS is down (5628)
4971 - service request history (4965)
4972 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4973 - create an "evict_client" /proc entry on OSTs, like the MDS has
4974 - fix mount usage message, return errors per mount(8) (5168)
4975 - change grep [] to grep "[]" in tests so they work in more UMLs
4976 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4977 - remove ancient LOV_MAGIC_V0 EA support (5047)
4978 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4979 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4980 - allow readahead allocations to fail when low on memory (5383)
4981 - mmap locking landed again, after considerable improvement (2828)
4982 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4984 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4987 - send OST transaction number in read/write reply to free req (4966)
4988 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4989 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4990 - the watchdog thread now runs as interruptible (5246)
4991 - flock/lockf fixes (but it's still disabled, pending 5135)
4992 - don't use EXT3 constants in llite code (5094)
4993 - memory shortage at startup could cause assertion (5176)
4995 - reorganization of lov code
4996 - single portals codebase
4998 - add extents/mballoc support (5025)
4999 - direct I/O reads in the obdfilter (4048)
5000 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5002 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5005 - send OST transaction number in read/write reply to free req (4966)
5006 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5007 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5008 - the watchdog thread now runs as interruptible (5246)
5009 - handle missing objects in filter_preprw_read properly (5265)
5010 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5011 - drop import inflight refcount on signal_completed_replay error (5255)
5012 - unlock page after async write error during send (3677)
5013 - return original error code on reconstructed replies (3761)
5014 - no transno return for symlink open, don't save no-trasno open (3440)
5016 - add pid to ldlm debugging output (4922)
5017 - bump the watchdog timeouts -- we can't handle 30sec yet
5018 - extra debugging for orphan dentry/inode bug (5259)
5020 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5023 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5024 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5025 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5026 - fix NULL dereference in /proc/sys/portals/routes (4827)
5027 - allow failed mdc_close() operations to be interrupted (4561)
5028 - stop precreate on OST before MDS would time out on it (4778)
5029 - don't send partial-page writes before EOF from client (4410)
5030 - discard client grant for sub-page writes on large-page clients (4520)
5031 - don't free dentries not owned by NFS code, check generation (4806)
5032 - fix lsm leak if mds_create_objects() fails (4801)
5033 - limit debug_daemon file size, always print CERROR messages (4789)
5034 - use transno after validating reply (3892)
5035 - process timed out requests if import state changes (3754)
5036 - update mtime on OST during writes, return in glimpse (4829)
5037 - add mkfsoptions to LDAP (4679)
5038 - use ->max_readahead method instead of zapping global ra (5039)
5039 - don't interrupt __l_wait_event() during strace
5041 - add software watchdogs to catch hung threads quickly (4941)
5042 - make lustrefs init script start after nfs is mounted
5043 - fix CWARN/ERROR duplication (4930)
5044 - return async write errors to application if possible (2248)
5045 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5046 - print NAL number in %x format (4645)
5047 - update barely-supported suse-2.4.21-171 series (4842)
5048 - support for sles 9 %post scripts
5049 - support for building 2.6 kernel-source packages
5050 - support for sles km_* packages
5052 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5055 - ignore -ENOENT errors in osc_destroy (3639)
5056 - notify osc create thread that OSC is being cleaned up (4600)
5057 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5058 - reconstruct ost_handle() like mds_handle() (4657)
5059 - create a new thread to do import eviction to avoid deadlock (3969)
5060 - let lconf resolve symlinked-to devices (4629)
5061 - don't unlink "objects" from directory with default EA (4554)
5062 - hold socknal file ref over connect in case target is down (4394)
5063 - allow more than 32000 subdirectories in a single directory (3244)
5064 - fix blocks count for O_DIRECT writes (3751)
5065 - OST returns ENOSPC from object create when no space left (4539)
5066 - don't send truncate RPC if file size isn't changing (4410)
5067 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5068 - bind to privileged port in socknal and tcpnal (3689)
5070 - rate limit CERROR/CWARN console message to avoid overload (4519)
5071 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5072 - basic mmap support (3918)
5073 - kernel patch series update from b1_4 (4711)
5075 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5078 - avoid crash during MDS cleanup with OST shut down (2775)
5079 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5080 - don't use bad inodes on the MDS (3744)
5081 - dynamic object preallocation to improve recovery speed (4236)
5082 - don't hold spinlock over lock dumping or change debug flags (4401)
5083 - don't zero obd_dev when it is force cleaned (3651)
5084 - print grants to console if they go negative (4431)
5085 - "lctl deactivate" will stop automatic recovery attempts (3406)
5086 - look for existing locks in ldlm_handle_enqueue() (3764)
5087 - don't resolve lock handle twice in recovery avoiding race (4401)
5088 - revalidate should check working dir is a directory (4134)
5090 - don't always mark "slow" obdfilter messages as errors (4418)
5092 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5095 - don't close LustreDB during write_conf until it is done (3860)
5096 - fix typo in lconf for_each_profile (3821)
5097 - allow dumping logs from multiple threads at one time (3820)
5098 - don't allow multiple threads in OSC recovery (3812)
5099 - fix debug_size parameters (3864)
5100 - fix mds_postrecov to initialize import for llog ctxt (3121)
5101 - replace config semaphore with spinlock (3306)
5102 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5103 - don't allow enqueue to complete on a destroyed export (3822)
5104 - down write_lock before checking llog header bitmap (3825)
5105 - recover from lock replay timeout (3764)
5106 - up llog sem before sending rpc (3652)
5107 - reduce ns lock hold times when setting kms (3267)
5108 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5109 - fix NULL deref and obd_dev leak on setup error (3312)
5110 - replace some LBUG about llog ops with error handling (3841)
5111 - don't match INVALID dentries from d_lookup and spin (3784)
5112 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5113 - fix invalid assertion in ptlrpc_set_wait (3880)
5115 - add libwrap support for the TCP acceptor (3996)
5116 - add /proc/sys/portals/routes for non-root route listing (3994)
5117 - allow setting MDS UUID in .xml (2580)
5118 - print the stack of a process that LBUGs (4228)
5120 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5123 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5124 - make sure to unset replay flag from failed open requests (3440)
5125 - if default stripe count is 0, use OST count for inode size (3636)
5126 - update parent mtime/ctime on client for create/unlink (2611)
5127 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5128 - free recovery state on server during a forced cleanup (3571)
5129 - unregister_reply for resent reqs (3063)
5130 - loop back devices mounting and status check on 2.6 (3563)
5131 - fix resource-creation race that can provoke i_size == 0 (3513)
5132 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5133 - more debugging for page-accounting assertion (3746)
5134 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5135 - avoid deadlock after precreation failure (3758)
5136 - fix race and lock order deadlock in orphan handling (3450, 3750)
5137 - add validity checks when grabbing inodes from l_ast_data (3599)
5139 - add /proc/.../recovery_status to obdfilter (3428)
5140 - lightweight CDEBUG infrastructure, debug daemon (3668)
5141 - change default OSC RPC parameters to be better on small clusters
5142 - turn off OST read cache for files smaller than 32MB
5143 - install man pages and include them in rpms (3100)
5144 - add new init script for (un)mounting lustre filesystems (2593)
5145 - run chkconfig in %post for init scripts (3701)
5146 - drop scimac NAL (unmaintained)
5148 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5151 - clean kiobufs before and after use (3485)
5152 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5153 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5154 - change init script to fail more clearly if not run as root (1528)
5155 - allow clients to reconnect during replay (1742)
5156 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5157 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5158 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5159 - fix cond_resched (3554)
5160 - only evict unfinished clients after recovery (3515)
5161 - allow bulk resend, prevent data loss (3570)
5162 - dynamic ptlrpc request buffer allocation (2102)
5163 - don't allow unlinking open directory if it isn't empty (2904)
5164 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5165 - remove extraneous obd dereference causing LASSERT failure (3334)
5166 - don't use get_cycles() when creating temp. files on the mds (3156)
5167 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5168 - handle EEXIST for set-stripe, set proper directory name (3336)
5170 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5171 - fix ksocknal_fmb_callback() error messages (2918)
5173 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5176 - don't copy lvb into (possibly NULL) reply on error (2983)
5177 - don't deref dentry after dput, don't free lvb on error (2922)
5178 - use the kms to determine writeback rpc length (2947)
5179 - increment oti_logcookies when osc is inactive (2948)
5180 - update client's i_blocks count via lvb messages (2543)
5181 - handle intent open/close of special files properly (1557)
5182 - mount MDS with errors=remount-ro, like obdfilter (2009)
5183 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5184 - don't use cancelling-locks' kms values (2947)
5185 - use highest lock extent for kms, not last one (2925)
5186 - don't dereference ERR_PTR() dentry in error handling path (3107)
5187 - fix thread race in portals_debug_dumplog() (3122)
5188 - create lprocfs device entries at setup instead of at attach (1519)
5189 - common AST error handler, don't evict client on completion race (3145)
5190 - zero nameidata in detach_mnt in 2.6 (3118)
5191 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5192 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5193 - initialize RPC timeout timer earlier for 2.6 (3219)
5194 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5195 - print nal/nid for unknown nid (3258)
5196 - additional checks for oscc recovery before doing precreate (3284)
5197 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5198 - don't crash in mdc_close for bad permissions on open (3285)
5199 - zero i_rdev for non-device files (3147)
5200 - clear page->private before handing to FS, better assertion (3119)
5201 - tune the read pipeline (3236)
5202 - fix incorrect decref of invalidated dentry (2350)
5203 - provide read-ahead stats and refine rpc in flight stats (3328)
5204 - don't hold journal transaction open across create RPC (3313)
5205 - update atime on MDS at close time (3265)
5206 - close LDAP connection when recovering to avoid server load (3315)
5207 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5208 - don't leak open file on MDS after open resend (3325)
5209 - serialize filter_precreate and filter_destroy_precreated (3329)
5210 - loop device shouldn't call sync_dev() for nul device (3092)
5211 - clear page cache after eviction (2766)
5212 - resynchronize MDS->OST in background (2824)
5213 - refuse to mount the same filesystem twice on same mountpoint (3394)
5214 - allow llmount to create routes for mounting behind routers (3320)
5215 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5216 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5217 - fix rare problem with rename on htree directories (3417)
5219 - allow default OST striping configuration per directory (1414)
5220 - fix compilation for qswnal for 2.6 kernels (3125)
5221 - increase maximum number of MDS request buffers for large systems
5222 - change liblustreapi to be useful for external progs like lfsck (3098)
5223 - increase local configuration timeout for slow disks (3353)
5224 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5226 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5229 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5230 - don't overwrite extent policy data in reply if lock was blocked (2901)
5231 - drop filter export grants atomically with removal from device (2663)
5232 - del obd_self_export from work_list in class_disconnect_exports (2908)
5233 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5234 - swab reply message in mdc_close, other PPC fixes (2464)
5235 - fix destroying of named logs (2325)
5236 - overwrite old logs when running lconf --write_conf (2264)
5237 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5238 - fix race in target_handle_connect (2898)
5239 - mds_reint_create() should take same inode create lock (2926)
5240 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5241 - don't close files for self_export to avoid uninitialized obd (2936)
5242 - allow MDS with the same name as client node (2939)
5243 - hold dentry reference for closed log files for unlink (2325)
5244 - reserve space for all logs during transactions (2059)
5245 - don't evict page beyond end of stripe extent (2925)
5246 - don't oops on a deleted current working directory (2399)
5247 - handle hard links to targets without a parent properly (2517)
5248 - don't dereference NULL lock when racing during eviction (2867)
5249 - don't grow lock extents when lots of conflicting locks (2919)
5251 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5254 - account for cache space usage on clients to avoid data loss (974)
5255 - lfsck support in lustre kernel code (2349)
5256 - reduce journal credits needed for BRW writes (2370)
5257 - orphan handling to avoid losing space on client/server crashes
5258 - ptlrpcd can be blocked, stopping ALL progress (2477)
5259 - use lock value blocks to assist in proper KMS, faster stat (1021)
5260 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5261 - recovery for initial connections (2355)
5262 - fixes for mds_cleanup_orphans (1934)
5263 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5264 - block all file creations until orphan recovery completes (1901)
5265 - client remove rq_connection from request struct (2423)
5266 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5267 - recovery timer race (2670)
5268 - mdc_close recovey bug (2532)
5269 - ptlrpc cleanup bug (2710)
5270 - mds timeout on local locks (2588)
5271 - namespace lock held during RPCs (2431)
5272 - handle interrupted sync write properly (2503)
5273 - don't try to handle a message that hasn't been replied to (2699)
5274 - client assert failure during cleanup after abort recovery (2701)
5275 - leak mdc device after failed mount (2712)
5276 - ptlrpc_check_set allows timedout requests to complete (2714)
5277 - wait for inflight reqs when ptlrpcd finishes (2710)
5278 - make sure unregistered services are removed from the srv_list
5279 - reset bulk XID's when resending them (caught by 1138 test)
5280 - unregister_bulk after timeout
5281 - fix lconf error (2694)
5282 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5283 - readahead locks pages, leaves pending causing memory pressure (2673)
5284 - increase OST request buffers to 4096 on large machines (2729)
5285 - fix up permission of existing directories in simple_mkdir (2661)
5286 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5287 - don't assign transno to errored transactions (2742)
5288 - don't delete objects on OST if given a bogus objid from MDS (2751)
5289 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5290 - if rq_no_resend, then timeout request after recovery (2432)
5291 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5292 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5293 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5294 - MDS recovery shouldn't delete objects that it has given out (2730)
5295 - if enqueue arrives after completion, don't clobber LVB (2819)
5296 - don't unlock pages twice when trigger_group_io returns error (2814)
5297 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5298 - don't write pages to disk if there was an error (1450)
5299 - don't ping imports that have recovery disabled (2676)
5300 - take buffered bytes into account when balancing socknal conn (2817)
5301 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5302 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5303 - remove little-endian swabbing of llog records (1987)
5304 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5305 - retry reposting request buffers if they fail (1191)
5306 - grow extent at grant time to avoid granting a revoked lock (2809)
5307 - lock revoke doesn't evict page if covered by a second lock (2765)
5308 - disable VM readahead to avoid reading outside lock extents (2805)
5310 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5311 - updated kernel patches for hp-2.4.20 kernel (2681)
5313 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5316 - fix truncated write corruption (2366)
5317 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5319 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5320 - ensure len(uuid) < 37 in lmc (1171)
5321 - fix ia64 OOPS in llog_test (2255)
5322 - zero end of page at obdfilter for partial page writes (2648)
5323 - don't leave stale dentries around after renames (bug 2428)
5324 - fix timeouts when evicting a client with a single lock held (2642)
5325 - set deadline for the initial HELLO message to drain (2634)
5326 - print out dotted-quad IP addresses in the socknal (2302)
5328 - additional debugging for MDS client eviction problem (2443)
5329 - fix mkfsoptions support for osts (2603, 2604)
5331 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5334 - add series for the vanilla 2.6.0 kernel
5335 - add series for the vanilla 2.4.24 kernel
5336 - add series for a cray x86/64 UL kernel drop
5337 - fix xattr patches for the vanilla 2.4.19 series
5339 - generate true UUIDs in lmc (1171)
5340 - have portals stack dumping break in UML (2466)
5341 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5342 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5343 - wake callback waiting threads on client eviction (2460)
5344 - Add --ptldebug and --subsystem to lmc (1719)
5345 - update assertion to allow safe interrupt allocation
5346 - set rq_no_resend for cancel requests (2432)
5347 - recalculate ptlrpcd timeout after resend (2494)
5348 - call vfs_rmdir when removing pending directories (2368)
5349 - fix renaming a file to itself (2429)
5350 - lmc creates a default one-stripe lov (2454)
5351 - expand procfs space to handle large clusters (2326)
5352 - increase UML stack to avoid overflow
5353 - update lconf's list of debug and subsystem masks
5354 - fix lfs find --obd (2510)
5355 - /proc tunable for disabling filter read caching (2591)
5356 - stop rpm packages from altering slapd.conf (2301)
5357 - disable nagle in the socknal under 0conf (2578)
5358 - choose mds inode size based on stripe count (2572)
5359 - fix kernel-source rpm problems (2516)
5361 - add --disable-doc to avoid pdf generation (2421)
5362 - update documentation, tests, type-os, comments
5363 - avoid format warnings on ia64
5364 - remove the TOE NAL
5365 - tiny code cleanups by removing unused fields
5367 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5370 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5371 - remove the most blatant lies from BUILDING file (2371)
5372 - change default debug level to reasonable production setting
5373 - reduce client side cache size to reduce cache flush time
5374 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5375 - make TCP zerocopy and pinger support enabled by default (2476)
5376 - sync writes completed after process exits caused crashes (2319)
5377 - maintain correct mount count on the MDS (2356)
5378 - backout 1557, because 2316 wasn't really fixed
5379 - better file I/O statistics gathering in /proc
5380 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5381 - another kernel patch to fix zero-copy TCP function export
5382 - don't take duplicate lock when processing re-sent getattr (2420)
5383 - lctl uses obd_self_export instead of creating new conn (2353)
5384 - MDS/OST recovery case which requires object creation asserted (2425)
5385 - move lfs from /usr/sbin to /usr/bin in packages
5386 - fix race between mds_client_add and mds_client_free (2417)
5387 - use kmalloc instead of slabs in portals (2430)
5388 - don't create duplicate records when a failover MDS is present (2442)
5389 - remove unnecessary mount age check (2332)
5390 - don't remove directory inodes from locks prematurely (2451)
5391 - don't break if MDS service name is the same as hostname (2103)
5392 - fix races in client write RPC generation when cache full (2482)
5394 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5397 - remove now-unused request->rq_obd (278)
5398 - if an allocation fails, print out how much memory we've used (1933)
5399 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5400 - add the "configurable stack size" patch to most series files (1256)
5401 - ability to write large log records, for 100+ OST configs (2306)
5402 - fix NULL deref when filter_prep fails (2314)
5403 - fix operator precedence error in filter_sync
5404 - dynamic allocation of socknal TX descriptors (2315)
5405 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5406 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5407 - ensure that CWARN messages in Portals always get to the syslog
5408 - __init/__exit are not for prototype decls (ldlm_init/exit)
5409 - x86-64 compile warning fixes
5410 - fix gateway LMC keyword conflict (2318)
5411 - fix MDS lock inversions in getattr/reint paths (1844)
5412 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5413 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5414 - DLM race condition prevented some lock evictions (2328)
5415 - ENOMEM detection and retry on socknal sends (2230)
5416 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5417 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5418 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5419 - better error messages when a client is rejected during recovery (1505)
5420 - avoid cancelling locks which were never granted, after failure (2330)
5421 - fix i_sem/journal inversion in mds_client_add (2333)
5422 - fix truncate/getattr lock cycle deadlock (2334)
5423 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5424 - fix two rare exit paths which could leak an l_lock() ref (2321)
5425 - fencepost error in MDS/OST orphan recovery (2226)
5426 - make log record alignment 8 bytes (1988)
5427 - lstripe now fails when requested offset > ost_count (2237)
5428 - ensure that all kernel series have a complete list.h (1607)
5429 - fix crashes in special-file operations (2316)
5430 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5432 - allow configurable automake binary, for testing new versions
5433 - small update to the lfs documentation
5435 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5437 * fix negative export reference count in fsfilt_sync (2312)
5439 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5440 * release candidate 0.9.1
5442 - orphans are moved into the PENDING directory for possible recovery
5443 - replayed opens now open by fid for orphan/rename safety (1042)
5444 - last close of an orphan inode generates a transno (683)
5445 - chdir() and mount() now pin the directory entry (1020)
5446 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5447 - discard very old requests without processing them (1502)
5448 - remove some common, well-understood CERRORs (1505)
5449 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5450 - clear "grant" flags in OST replies until OST grant code lands (1644)
5451 - fix read performance by not clobbering i_blksize on client (1598)
5452 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5453 - make LRU size a /proc tunable, clears locks when reduced (707)
5454 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5455 - server locks take references on exports now (1558)
5456 - build fixes for 2.4.20-rh trees (1663)
5457 - return an error from lov_create if all OSCs are inactive (1751)
5458 - fix import levels when a reconnect happens without a timeout (1597)
5459 - exit early from mds_open if we get a lookup error (1749)
5460 - partial page read at EOF wouldn't wait for disk before sending (1642)
5461 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5462 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5463 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5464 - write/truncate lock inversion (1639)
5465 - Don't auto-load obdclass, portals modules during cleanup (1495)
5466 - fix timestamps from jumping to "now" (1763)
5467 - extra journal assertions (1648)
5468 - add an extra multiunlink test (1771)
5469 - fix read_record/write_record API (1776)
5470 - fix leak of offset_extent, possible incorrect i_size later (1772)
5471 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5472 - Debugging for the kqswnal_get_idle_tx problems (1820)
5473 - Allow recovery to be attempted multiple times (1536)
5474 - Write out MDS last_rcvd file after it is first created (1600)
5475 - Fix tx_descriptor leak in failed transmit situations (1827)
5476 - ext3 journaling fixes for assertion failure after IO error (1871)
5477 - class_export_put() on freed export after completion AST error (1896)
5478 - Fix revalidate looping in VFS (1322)
5479 - Don't access a freed export during MDS_REINT timeout (1521)
5480 - Add open-unlink recovery support on the MDS (1673,1764)
5481 - Return an error if no MDS data was read from last_rcvd (1946)
5482 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5483 - Don't setup a disk device that doesn't match exported UUID (317)
5484 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5485 - avoid committing NULL handle in force close
5486 - local.sh is now a one-stripe LOV configuration
5487 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5488 - fix ext3 htree duplicate directory entry corruption (1516)
5489 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5490 - update to new LOV EA format (2097)
5491 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5493 2003-06-15 Phil Schwan <phil@clusterfs.com>
5496 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5497 - per-import/export recovery handling (958, 931, 959)
5498 - multiple last-rcvd slots, for serving multiple FSes (949)
5499 - connections are again shared between multiple imp/exports (963, 964)
5500 - "umount -f" would hang if any requests needed to be sent (393, 978)
5501 - avoid pinning large req buffer by copying for queued messages (989)
5502 - add "uuid" to "lctl device" command to help upcalls (991)
5503 - "open" RPCs with transnos would confuse recovery counters (1037)
5504 - do proper endian conversion of all wire messages (288, 340, 891)
5505 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5506 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5507 - fix ptlrpc_connection leak in target_handle_connect (1174)
5508 - fix import refcounting bug in OST and MDS cleanup (1134)
5509 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5510 - fix very unlikely obd_types race condition (501)
5511 - remove osc_open hack for echo_client (1187)
5512 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5513 - a failure in read_inode2 leads to deadlock (1139)
5514 - cancel ack-locks as soon as transaction is committed (1072)
5515 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5516 - make sure to commitrw after any preprw to avoid deadlock (1162)
5517 - failing to execute a file in a lustre FS would lock inode (1203)
5518 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5519 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5520 - obd_uuid2tgt would walk off the end of the list (1255)
5521 - on IA64 the osc would give portals incorrect bulk size (1258)
5522 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5523 - fix lock inversion caused by new llite matching code (1282)
5524 - limit the number of dirty pages on a client to 10MB (1286)
5525 - timed out locks were not being corrected cancelled (1289)
5526 - fix O_DIRECT above 4GB on IA-32 (1292)
5527 * major user-visible changes
5528 - fail out/fail over policy now controlled by the upcall (993)
5530 - add OBD_PING to check server availability and failure (954)
5531 - lustre messages are now sent in sending host order (288, 340, 891)
5532 - add eadatalen to MDS getattr reply (340)
5533 - OST read replies may contain second buffer, with per-page status (593)
5535 2003-03-11 Phil Schwan <phil@clusterfs.com>
5538 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5539 - failed open()s could cause deadlock; fixed (867, 869)
5540 - stop cancelling OST locks when files are closed (481)
5541 - overlapping XID spaces caused network corruption (851, 853)
5542 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5543 - setattr_raw added, to do single-RPC, server-side setattrs
5544 - lmc/lconf syntax change for OST UUIDs
5545 - fix crashy race condition between ptlrpc_free_req and osc_close
5546 - don't use request in mdc_enqueue if we hit a timeout (889)
5547 - don't set the inode i_size for regular files from the MDS (896)
5548 - handle out of order completion AST (842)
5549 - don't LBUG if a lock request times out after receiving AST (913)
5550 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5551 - if a bad lock AST arrives, send an error instead of dropping entirely
5552 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5553 - fix leak in bulk IO when only partially completed (899, 900, 926)
5554 - fix O_DIRECT for ia64 (55)
5555 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5556 - C-z after timeout could hang a process forever; fixed (977)
5558 - client-side I/O cache (678, 924, 929, 941, 970)
5560 - READPAGE and SETATTRs which don't take server-side locks get
5563 2003-02-11 Phil Schwan <phil@clusterfs.com>
5566 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5567 - Fix fsfilt_extN_readpage() to read a full page of directory
5568 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5569 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5570 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5571 - Most intent operations were removed, in favour of a new RPC mode
5572 that does a single RPC to the server and bypasses most of the VFS
5573 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5574 - Aggressively cancel local locks on DLM servers
5575 - mds_reint_unlink sends EA to the client if it's the last nlink.
5576 client uses that EA to unlink OST objects.
5577 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5578 - recursive symlinks were fixed (439)
5579 - fixed NULL deref in DEBUG_REQ
5580 - filter_update_lastobjid no longer calls sync, which annoyed extN
5581 - fixed multi-client small-writes to a single file problem (445)
5582 - fixed mtime updates during file writes (607)
5583 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5584 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5585 - fixed Linux OST size reporting problem (444, 656)
5586 - OST now updates object mtime with writes or setattr (607, 619)
5587 - client verifies file size before zeroing page past EOF (445)
5588 - OST now writes last allocated objid to disk with allocation (108)
5589 - LOV on echo now works (409)
5591 - mds_reint_unlink sends a new buffer, with the EA included. this
5592 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5593 set if a regular file was being unlinked, and it was the last link
5594 - use PtlGet from the target for bulk writes (315)
5595 - OST now updates object mtime with writes or setattr (607, 619)
5596 - LDLM now has a grant-time callback to revalidate locked items, if
5598 - Many MDS operations were reorganized to combat race conditions
5600 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5601 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5603 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5606 - Fully reactivate OST imports after reconnection (512, others)
5607 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5608 - More graceful error handling for truncating on dead OST (515)
5609 - Don't error out unless we're actually accessing dead stripes (474)
5610 - Fix garbage sizes when stripes are missing (410)
5611 - LRU counters were broken, causing constant lock purge (433, 432)
5612 - garbage on read from stripes with failed OSTs (441)
5613 - mark OSCs as active before reconnecting during recovery (438)
5614 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5615 - lfind did not preserve OST order in output (443)
5616 - symlinks cause hung clients, incorrect data (439)
5617 - stop dereferencing request after dropping refcount (457)
5618 - don't LASSERT(spin_is_locked) on non-SMP (455)
5619 - fixes for many rename() bugs
5620 - fstat didn't correctly synchronize attributes (399)
5621 - server must handle lock cancellation during blocking AST prep (487)
5622 - bulk descriptors were free()d too soon (511)
5623 - fix paths in lconf, which would load incorrect modules (451, 507)
5624 - fix confusing lconf 'host not found' error message (386)
5625 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5626 - fix race condition in mdc_blocking_ast() for inode access (526)
5627 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5628 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5629 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5630 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5631 - fix lstripe in several ways (kernel side) (527)
5632 - fix request leak in ldlm_cli_enqueue (262)
5633 - incorrect OSC was marked inactive after OST failure
5634 - call mds_fs_cleanup before unmounting filesystem (524)
5635 - fix races between taking ns_lock and ldlm_lock_change_resource
5636 - fix races updating LOV export open file list
5637 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5638 - fix recovery NULL deref in ldlm_cli_cancel_unused
5639 - fix some DLM races by using new hash table for lock handles (419)
5640 - permit the client to specify desired inodes, at replay
5641 - duplicate requests when we queue them for replay reintegration
5642 - fix last_rcvd offset calculation
5643 - sync after each recovered transaction, so we always make progress
5644 - never, not always, ERESTART requests without transnos
5645 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5647 - skip replay if the MDS says that the client is already connected
5648 - don't check for a recovery-enabled export to match lctl's UUID
5649 - don't INC_USE_COUNT for phantom exports
5650 - don't crash when cleaning up phantom exports (567)
5651 - don't double-finish or set replay data for errored mdc_open requests
5652 - abort requests when they time out, so we don't get old replies
5653 - send/receive replies for AST messages again
5654 - if the client says that it doesn't have the lock, cancel it on the
5656 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5657 mark it as destroyed, it will all work out when decref is called
5658 - fix module use counts (22, 581)
5660 - ASTs now expect a reply (server cancels lock on error reply)
5662 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5665 - fix many simultaneous client startup (392)
5666 - fix dentry->d_it clobbering
5667 - credentials weren't being shipped for readdir/getattr operations
5668 - remove invalid assertions triggered during some concurrent MD
5670 - proper Lustre versions added (336, 389)
5671 - fix memory leak for create error case (398)
5672 - fix LOV locking bug that would get cli/srv out of sync
5673 - fix echo client over LOV (409)
5674 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5675 - fix double-O_EXCL intent crash (424)
5676 - avoid sending multiple lock CANCELs (352)
5678 - MDS can do multi-client recovery (modulo bugs in new code)
5680 - many updates, edits, cleanups
5682 2002-11-18 Phil Schwan <phil@clusterfs.com>
5685 - fix null d_it dereference (346)
5686 - fix full OST/dbench hang (333)
5687 - fix permission problem with file removal (286)
5688 - fix removal of OSCs from LOV when they fail
5689 - fix NULL deref during bulk timeout (214)
5690 - fix problems related to multiple filesystems on one MDS (241)
5691 - fixed serious subtle metadata locking bugs
5692 - free locks on clients when inodes are removed due to memory
5694 - fix inode pointer in lock data (285)
5695 - partial support for multiple MDS on a single host (241)
5696 - data locks weren't cancelled at clear_inode time (290, 311)
5697 - intent locks could lead to unbounded lock growth (205)
5698 - added a maximum lock count, an LRU list, and a flusher
5699 - fix multiple rename (365)
5700 - properly abstracted the echo client
5701 - OSC locked 1 byte too many; fixed
5702 - rewrote brw callback code:
5703 - fixed recovery bugs related to LOVs (306)
5704 - fixed too-many-pages-in-one-write crash (191)
5705 - fixed (again) crash in sync_io_timeout (214)
5706 - probably fixed callback-related race (385)
5708 - Add capability to MDS protocol
5709 - LDLM cancellations and callbacks on different portals
5711 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5714 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5715 - defer open object destruction to close time (601981, 138)
5716 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5717 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5718 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5719 - don't call obd_setattr in ll_file_release for destroyed objects
5720 * protocol change to lustre_msg: move |version| and add |flags|
5721 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5722 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5723 * added replay of create, unlink, link and rename operations during
5724 MDS failover; recovery should be much more robust now
5725 * remove failed OSCs from LOVs (only lov_create uses this so far)
5726 * the lustre-HOWTO was brought (more) up to date (582544)
5728 2002-10-23 Phil Schwan <phil@clusterfs.com>
5731 - in-use dentries weren't being reused properly (617851)
5732 - prevent multiple LDLM setup (599178)
5733 - fix LOV size calculations for truncate (617853)
5734 - fix client handling of MDS intent errors (POSIX)
5735 - fix permission bug in lovstripe.c test (624321)
5736 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5737 - truncate past end of file could corrupt data
5738 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5739 - a race in recovery could return ETIMEDOUT to apps (623947)
5740 - building outside the source directory was fixed
5741 * the lustre-HOWTO was brought (more) up to date (582544)
5742 * major progress was made on recovery functionality
5744 2002-10-10 Phil Schwan <phil@clusterfs.com>
5747 - recovery deadlock fix
5748 - rm -rf causes LBUG fix (617817)
5749 - file open by multiple tasks fix (618962)
5750 - directory permissions bugs (602707 and 620007)
5751 - journal_stop fixed with locking (611313)
5752 - O_APPEND failures resolved (618273, perhaps 614459)
5753 - lconf PATH fix (619770)
5754 - IA64 build fix (621450)
5755 - RPC buffer sizes scale with amount of memory
5757 2002-10-01 Phil Schwan <phil@clusterfs.com>
5760 - locks would be cancelled without throwing away data pages,
5761 resulting in inconsistent data (605627)
5762 - inode attributes were not always being refreshed (605627, 612449)
5763 - lconf now continues to cleanup after lctl reports an error
5764 - MDS now enforces user permissions (602707)
5765 - lprocfs cleanup fixed, but not yet enabled (614157)
5766 - fixed infinite server hang, should a client not respond to an AST
5767 - avoid going into recovery if user calls readlink() with a buffer
5768 that's too small (613941)
5769 - AST RPCs no longer require replies (614867) -- this may be changed
5770 - don't crash server if client sends an IOV that's too big (611336)
5771 - fixed lock conversion deadlock (611892)
5772 - fixed the following of symlinks (614622)
5773 * recovery: the server can remove locks from a client that dies, other
5774 clients can make progress
5775 * more extN patch fixes
5776 * compile-time configurable ptlrpc buffer allocations
5778 - collaborative read cache document
5779 - Lustre Lite Performance CDR document-in-progress
5781 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5784 - fix typo in patch-2.4.18
5786 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5789 - clear ptlrpc request each time in handle_incoming_request()
5790 - unlink of files now destroys the object on the OST
5792 2002-09-19 Peter Braam <braam@clusterfs.com>
5794 * add hard link support
5795 * change obdfile creation method
5796 * kernel patch changed
5798 2002-09-19 Peter Braam <braam@clusterfs.com>
5801 - stack overflow bug in extN fixed
5803 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5805 * documentation updates
5806 - add man pages for config tools
5807 - update tests/README to describe testing with new config tools
5808 - finish metadata API descriptions
5809 * bug fixes and cleanups
5810 - statfs workaround for 16TB limit
5811 - LOV stripe allocation improved, can stripe on subset of OSTs
5812 - LOV file size/IO offset was wrong for files > 4GB in size
5813 - object EA data was being dropped, caused files to be unreadable
5814 - memory overflow with non-LOV OST caused memory corruption
5815 - fixed regression tests to work with new config tools, obdfilter
5816 - fixed bug when directory size became larger than 1 block
5817 - fixed bug (for single client case) when PWD was deleted
5818 - invalidate local directory pages when doing intent-based ops
5819 - avoid LDLM oops when lock callback contained bad data
5821 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5823 * documentation updates
5824 * bug fixes and cleanups
5825 - configuration tools
5828 - 64-bit compile warnings
5829 - 64-bit internal statfs data
5831 * test_brw on persistent OST devices
5833 * lprocfs (disabled)
5835 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5837 * documentation updates
5838 * bug fixes and cleanups
5839 * configuration tools
5841 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5844 * many small fixes to 0_5_4
5845 * io/network handling
5846 * thinkos in MDS operations
5848 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5851 * crucial basic fixes to 0.5.3
5852 * IOR, Iozone work over Elan
5855 2002-08-07 Phil Schwan <phil@clusterfs.com>
5856 * version 0_5_3, our first alpha
5857 * we use the new Portals iovs
5858 * documentation updates
5859 * bug fixes and cleanups
5860 * small changes in the DLM wire protocol
5862 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5863 * version 0_5_1 with some initial stability,
5864 * locking on MD and file I/O.
5865 * documentation updates
5866 * several bug fixes since 0.5.0
5867 * small changes in wire protocol
5869 2002-07-18 Phil Schwan <phil@clusterfs.com>
5871 * delivered as Lustre Light Alpha
5872 * fixed a crash after handling invalid MDS requests
5873 * fixed directory pages for architectures with non-4k pages sizes
5875 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5876 * release version v0_4_4
5877 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5879 * Fixes a number of interruption problems with OST operations.
5880 * Update documentation for portals header changes
5881 * Move all wire protocol structs/defines to lustre_idl.h
5882 * Fixes symlink length bug.
5883 * Add tcpdump to repository.
5885 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5886 * release version v0_4_3
5887 * Fixes statfs for inodes on extN.
5888 * Fixes bug in runtests which would delete /etc/hosts.
5889 * Use 64-bit object IDs wherever possible (not into VFS though)
5890 Remove ost_get_info, which is unused by lustre, and out of date.
5892 2002-07-03 Peter Braam <braam@clusterfs.com>
5893 * release version v0_4_2 Fixes a lookup error (type not passed)
5894 * move forward to head of Portals
5895 * move forward to latest Lustre kernel
5897 2002-06-25 Peter Braam <braam@clusterfs.com>
5898 * release version v0_4_1. Hopefully stable on single node use.